e-edition

First published in Great Britain by

Peter Peregrinus Ltd, London, UK 1983

e-edition published by M.L. Meade 2013

Copyright 2013 M.L. Meade

M.L. Meade has asserted his right under the Copyright, Designs
and Patents Act to be identified as the author of this work.
Contents

The Contents list is for reference only.

Individual chapters can be navigated using the
Table of Contents displayed in the PDF sidebar.

Preface
Acknowledgments
Chapter 1 Introduction
1.1 References
Chapter 2 Basic concepts in lock-in recovery
2.1 Introduction
2.2 Evaluating the signal recovery problem
2.3 Demodulators for signal recovery
2.4 Operation of synchronous detectors
2.4.1 Introduction
2.4.2 Demodulation with a synchronous reference
2.4.3 Amplitude demodulation
2.4.4 Phase demodulation
2.4.5 Mixed modulations
2.4.6 Noise rejection
2.5 Basic lock-in amplifiers
2.5.1 Introduction
2.5.2 The signal channel
2.5.3 Signal conditioning
2.5.4 The multiplier
2.5.5 The reference channel
2.5.6 The low-pass filter
2.6 Signal recovery 'capability'
Chapter 3 Phase-sensitive detectors
3.1 Introduction
3.2 Principles of operation
3.3 Harmonic transmission windows
3.4 Noise bandwidth of phase-sensitive detectors
3.5 Non-sinusoidal signals
3.5.1 Introduction
3.5.2 General considerations
3.5.3 Symmetrical periodic signals

iii
 3.5.4 Asymmetrical periodic signals
3.5.5 Squarewave signals: a special case
3.6 Phase-sensitive detector specifications
3.6.1 Introduction
3.6.2 Full-scale sensitivity
3.6.3 Linearity and out-of-phase rejection
3.6.4 Dynamic reserve
3 6.5 Output stability and minimum detectable signal
3.6.6 Dynamic-reserve/output-stability trade-off
3.6.7 Dynamic range
3.6.8 Summary of specifications
3.7 References
Chapter 4 Lock-in amplifier specifications
4.1 Introduction
4.2 Calibration: full-scale sensitivity
4.3 Phase-sensitive detector related specifications
4.3.1 Introduction
4.3.2 System dynamic reserve
4.3.3 System overload capability
4.3.4 Dynamic reserve and output stability trade-off
4.3.5 Overload capability and output stability trade-off
4.3.6 Dynamic range and linearity trade-off
4.4 Using a tuned filter in the signal channel of a conventional
lock-in amplifier
4.4.1 Influence on overload capability
4.4.2 Suppression of harmonic responses
4.5 Reference-channel specifications
4.5.1 Introduction
4.5.2 Phase accuracy: points of specification
4.5.3 Phase noise (phase jitter) and phase drift
4.5.4 Reference-channel slew rate
4.6 Measurement of phase accuracy
4.6.1 Introduction
4.6.2 Trigger phase errors
4.6.3 Defining in-phase signal and reference
4.6.4 Errors due to oscillator distortion
4.6.5 Phase noise, phase drift and phase-sensitive detector
instability
4.7 References
Chapter 5 Two-phase lock-in amplifiers
5.1 Introduction
5.2 Examples of 'classic' two-phase applications
5.2.1 A.C. bridge balancing
5.2.2 A.C. impedance measurements

iv
 8.3.5 Phase-shifting
8.4 Practical limitations
8.4.1 The frequency synthesizer
8.4.2 The image filter
8.4.3 The signal mixer
8.4.4 The phase-sensitive detector
8.5 Overload capability of heterodyne systems
8.6 Double heterodyne lock-in amplifiers
8.7 Brief comparison of single and double heterodyne systems
8.8 8ynchronous heterodyning
8.8.1 Introduction
8.8.2 Dynamic range improvement
8.8.3 Application to heterodyne lock-in amplifiers
8.9 Conclusions
8.10 References
Chapter 9 P.W.M. systems
9.1 Introduction
9.2 Principles of operation
9.3 Frequency composition of the p.w.m. waveform
9.4 Basic design considerations
9.4.1 Dynamic range
9.4.2 Spurious responses
9.4.3 Choice of switching frequency
9.5 Reference phase-shifting
9.6 Two-phase systems
9.7 Analogue correlation
9.7.1 Matched detection
9.7.2 Double frequency lock-in analysis
9.7.3 High slew rate applications
9.8 Interference rejection filters
9.9 Comparison of p.w.m. systems with heterodyne lock-in amplifiers
9.10 References
Chapter 10 Computer-controlled lock-in amplifiers
10.1 Introduction
10.2 Programmable lock-in amplifiers
10.3 Microprocessor-based systems
10.4 Automatic sensitivity selection
10.5 Automatic phase selection
Appendix 1 Principal Applications

vi
Appendix 2 Selected topics on signals and noise
A2.1 Introduction
A2.2 Voltage noise and current noise spectra
A2.3 Signal spectra
A2.4 Thermal noise and shot noise
A2.5 Noise bandwidth
A2.6 Signal-to-noise-ratio improvement by filtering
A2.7 Low-frequency noise
A2.8 More about narrowband noise
Appendix 3 Synchronous detection and noise
A3.1 Signal-to-noise-ratio improvement
A3.2 Noise measurements
Appendix 4 Signal-conditioning filters
A4.1 Low-pass filters
A4.1.1 First order
A4.1.2 Second order
A4.2 High-pass filters
A4.2.1 First order
A4.2.2 Second order
A4.3 Active tuned filters
A4.3.1 Band-pass
A4.3.2 Low pass
A4.4 Active notch filter
Appendix 5 Amplifier selection and noise matching
A5.1 Introduction
A5.2 What type of amplifier?
A5.3 Noise in voltage amplifiers
A5.3.1 Introduction
A5.3.2 Noise-figure calculations
A5.3.3 Minimum noise figure and optimum source resistance
A5.3.4 Noise-figure contours
A5.3.5 Cryogenic sources
A5.3.6 Transformer noise matching
A5.4 Noise in current amplifiers
A5.5 References
Appendix 6 Interference and ground loop suppression
A6.1 Introduction
A6.2 Ground loops: single-ended amplifiers
A6.3 Ground loops: differential amplifiers
A6.4 Ground loops and lock-in recovery
A6.5 References

vii
Preface

To the first edition

This book has been written for users of lock-in equipment and for those with an
interest in the practical aspects of signal recovery and measurement using
synchronous detection. The subject matter has been tackled for the most part at a
systems level on the understanding that this is the approach most appreciated by
research workers whose main specialization is in an area other than electronic
instrumentation. Circuit designers will therefore look in vain for detailed circuit
implementations; extensive mathematical developments are similarly avoided in
favour of a more qualitative approach to identifying the essential features of lock-
in detection systems. A basic familiarity with Fourier series and transforms
provides adequate preparation for the main part of the text and it is hoped that the
review of system configurations and specifications given there will prove both
interesting and useful to specialists and generalists alike.

To the e-edition
It is now thirty years since Lock-in amplifiers: principles and applications was
first published. In the intervening period it has become established as a minor
classic, being amongst the most widely cited text books of its kind. Of even
greater importance to me personally is that, despite a lapse of almost 25 years
since the final printing, I continue to receive requests from researchers and
students seeking a copy hence my wish to create an 'authorised' electronic
version and make it freely available.
Unlike the PDF held by Google Books (and numerous plagiarised versions
circulating elsewhere) this e-edition has been newly compiled from the original
typescript and diagrams. I make no apologies for leaving the content substantially
the same as before, with no extensive revisions or additions. This reflects my
view that, while there have been significant developments in the technology and
implementation of lock-in systems, the greater part of the book, dealing with
principles and guides to good practice, remains valid and useful. I might also
confess that I am enjoying my retirement too much to spend more than the time
necessary to correct obvious mistakes and to improve on the type-setting of
equations which was less than satisfactory in the printed edition. There is, of
course, the danger that this reworking is prone to fresh errors and, here, I am
more than happy to apologise for any difficulty caused and fully prepared to
amend and reissue any pages where errors are reported.

Mike Meade
Carlton
Bedford, UK
November 2013

viii
Acknowledgements

From the first edition

The following chapters represent a greatly expanded version of a review article I
prepared for the Institute of Physics in 19811. A search of the literature at that
time revealed that very little had been written about lock-in amplifiers beyond the
technical notes published by the leading manufacturers. The task of writing this
book was therefore greatly eased by the co-operation of friends and sometime
colleagues at the E.G. & G. companies, Brookdeal and Princeton Applied
Research, who made available a wealth of applications material and provided
additional information.
At a more detailed level, the treatment in later chapters relating to heterodyne and
p.w.m. systems owes much to discussions with Dr. Simon Carter of Hewlett
Packard Ltd., South Queensferry. The formulation of spurious responses in these
systems follows the lines developed by Dr. Carter in his Ph.D. thesis which is a
prime source of reference on all aspects of lock-in systems.
I am greatly indebted to Chris Meredith of Aquarius Electronics, Beaconsfield,
for providing a valuable perspective on lock-in systems and providing much
painstaking, constructive, criticism on the manuscript. Also, Dr. David Crecraft
and other colleagues in the Electronics Discipline of the Open University, UK,
provided both assistance and encouragement throughout the writing period.
Finally, I should like to make special mention of Christine Martindale and Jane
Barden for managing the production of the manuscript.

For the e-edition

The e-edition was made possible by the combined efforts of the secretarial staff
in the former Electronics Discipline of the Open University, UK, who undertook
the scanning and hand-crafted OCR rendition of my original manuscripts during
the summer of 2002. I am entirely to blame for the fact that the resulting files
have only now seen the light of day.

1
References
1 MEADE, M.L. (1982): 'Advances in lock-in amplifiers', J.Phys E: Sci
Instrum., 15, pp. 395-403
2 CARTER, S.F. (1982): 'A systems approach to the design of lock-in
amplifiers'. Ph.D.Thesis (University of Reading, England)

ix
CHAPTER 1

Introduction

Signal recovery instruments will always be in demand as long as experiments are

attempted with progressively smaller samples, weaker concentrations and fainter
excitations. Nowadays it is more or less standard practice to design experiments
to take advantage of signal-recovery techniques. Only then does it become
possible to carry out measurements in the crucial stage of an investigation where
the signals of interest become obscured by high levels of noise and interference.
Of all the techniques that have been developed for signal recovery, methods
based on the phase-sensitive detector and its modern counterpart, the lock-in
amplifier, are by far the most widely applied in all fields of scientific research.
Indeed, for many research workers, the terms 'signal recovery' and lock-in
recovery' are virtually interchangeable.

Fig, 1.1 A general experimental system

To see why this should be the case, let us begin with the general experimental
system shown in Fig. 1.1. The system could be electrical, mechanical, optical,
biological or any combination of such systems. The excitation source evokes a
response in the output and the response is converted to an electrical signal in the
transducer. We can suppose that the person who devised the experiment has a
clear idea of how to interpret the signal.
In some experiments it may be essential to recover the entire output signal so that
its waveform can be made available for analysis. When the signal is obscured by
high levels of noise, some form of a signal averaging will be necessary and the
experimenter might specify a multipoint averager or Fourier transform analyser
for this purpose. Signals of a transient nature, triggered by repetitive pulses from
the excitation source, are often dealt with in this fashion.
In experiments using 'static' or 'd.c.' excitations the output signal usually appears
in the form of a slowly varying direct voltage proportional to an experimental
parameter of interest. Severe measurement problems then result when the voltage
falls to a level comparable with the error voltages due to offsets and drift in the
transducer and its associated amplifier.
The temperature-dependent drift of d.c. coupled equipment is usually treated as a
component in the 1/f noise or flicker noise that plagues low-frequency
measurements. The effect of flicker noise on the determination of a fixed voltage
level Vo is shown schematically in Fig. 1.2. The record shows a characteristic
deviation from the initial condition established at time t = 0, resulting in an
increasing measurement error as longer observation times are taken.

Chapter 11
Fig. 1.2 D.C. output from an experimental system perturbed by low-
frequency noise

In the light of these problems it is now established practice to interrupt normally

'static' excitations by providing some form of modulator at the input to an
experiment. A typical research institute is usually a fruitful source of vibrating
reeds, rotating discs and other electromechanical 'choppers' that have been
engineered for this purpose. In most cases the net result is an ON/OFF
modulation of the excitation source. The output voltage that originally had the
form shown in Fig. 1.2 is then transformed to a 'chopped' voltage with amplitude
Vo superimposed on the flicker noise voltage fluctuation. This new situation is
depicted in Fig. 1.3(a).

Fig 1.3 (a) Switched output voltage obtained by periodically interrupting the
excitation source; (b) Output voltage with residual noise after
high-pass filtering

In practice, the modulation or chopping frequency is usually made as high as

possible to facilitate separation of the chopped output voltage from low-
frequency noise components. This separation is achieved by using a high-pass
filter and results in the a.c. signal shown in Fig. 1.3(b). The amplitude of this
signal is proportional to the experimental parameter of interest and will usually
vary in the course of an experiment. Note that in Fig. 1.3(b) the short-term noise

Chapter 12
fluctuations that appeared on the original signal have been transmitted by the
high-pass filter and so appear in the final output. One way to overcome this
residual fluctuation, when measuring the signal amplitude Vo, is to apply a
differential measurement procedure to the output voltage. The idea is to measure
the mean voltage difference between successive ON/OFF intervals and then
average results over a number of modulation cycles. An improved estimate of the
amplitude Vo will be obtained as the overall observation time is increased and a
greater number of modulation cycles is taken into account.
In 1946, Dicke [1] showed how this type of measurement could be carried out
automatically by using a phase-sensitive detector. A phase-sensitive detector
measures the difference voltage of interest by using a synchronous reference
voltage derived from the input modulator. We shall find that detection with
respect to a synchronous reference enables the use of very long averaging times
for the purpose of signal-to-noise ratio improvement and that practical systems
are capable of operating with signals well below the background noise level.
The importance of this capability cannot be overstated since in many experiments
the noise level due to thermal noise alone may be of the order of several
millivolts peak-to-peak while the signal of interest has an amplitude measured in
microvolts. Add to this the effect of incidental pick-up and interference and the
result is a real signal recovery 'problem' awaiting solution.
In this context, phase-sensitive detectors offer a significant advance over
alternative amplitude-demodulation schemes employing non-linear devices such
as envelope detectors. The latter make no fundamental distinction between signal
and noise components whereas a phase-sensitive detector is engineered to
respond specifically to the information-bearing signal. If the term signal recovery
implies that we have some prior knowledge of a signal, then the phase-sensitive
detector is a true signal recovery device in that it takes account of the distinctive
structure of the signal imposed by the use of a modulated excitation.
A phase-sensitive detector is responsive to the amplitude of a signal but is also
sensitive to the phase difference between a signal and the derived reference.
Phase-sensitive detector-based systems can therefore be devised to measure
variations in both the amplitude and phase of periodic signals in the presence of
noise and interference. Systems operating on the phase-sensitive detector
principle are termed lock-in systems and the usual way of introducing a phase-
sensitive detector into an experiment is to use a lock-in amplifier. This term has
come to mean a free-standing instrument that incorporates a phase-sensitive
detector, supported by preamplifiers, post-detection amplifiers and a
comprehensive reference processing section.
A number of lock-in amplifier applications are listed in Appendix 1 which also
serves to emphasize the widespread use of periodic excitations in experimental
research. In the majority of cases the signal of interest appears in the output of the
experiment at the same frequency as the fundamental excitation frequency. This
certainly includes all measurements where the experimental processes are
essentially linear insofar as no new frequencies are generated between input and
output; for example, in a wide range of optical and electronic systems. To this
broad class of experiments we can add examples where additional frequencies are
produced, but where, once again, the frequency of interest is identified with that
of the excitation source or bears a harmonic relationship to it. In any of these
cases the phase-shift introduced by the experiment might be of interest, but, very
often, it is sufficient to monitor changes in the magnitude of the output signal. On
further examination we find that the applications can be broadly divided into two
main categories. First of all, we have those many areas of activity where lock-in
systems are used in their long-established role as signal-recovery tools for the

Chapter 13
measurement of modulated signals in noise. Secondly, there are many examples
where lock-in systems are used for the precision measurement of signals, in
situations where signal-recovery capability does not appear to be a prime
consideration.
This essentially dual function of modern lock-in systems will be emphasized
throughout the following chapters. When dealing with the principles of lock-in
detection in Chapter 2 we shall begin along the lines followed by most authors in
the field of telecommunications under the heading 'synchronous' or 'coherent'
detection [2 ,3] Beyond this, however, we must improve upon the standard text-
book treatment which is usually very disappointing to those whose interest is in
experimental applications where signals are very slowly varying or may even be
'fixed' for the time available for measurement. Also, while some authors
acknowledge that synchronous detectors are mathematically capable of
withstanding adverse signal- to-noise ratios we find that very little attention is
given to the practical aspects of demodulation under such conditions.
This is scarcely surprising since in many communication systems, signal-to-noise
ratios of less than 10:1 or 20 dB would be considered quite unacceptable, while in
signal-recovery work a typical starting point is with signal-to-noise ratios of less
than 1:10, that is 20 dB or worse. We shall therefore find it necessary to discuss
aspects of linearity and dynamic range and to make the transition from an ideal
detector model to practical devices furnished with a range of specialist
specifications. Most of the principles and techniques to be described could apply
to lock-in amplifiers operating in almost any frequency range. Furthermore, the
definitions of key specifications are independent of the particular technology used
in the implementation of the lock-in amplifier. When dealing with 'typical'
specifications, however, we shall take examples from commercial lock-in
amplifiers optimized for the low-frequency range extending from less than 1 Hz
up to a maximum of about 1 MHz. This corresponds to the frequency range in
which the most significant developments in phase-sensitive detector technology
have occurred and which satisfies the greatest number of applications.
The object throughout is to present information against a background of
experimental work and to develop an awareness of the nature of signals and noise
in experimental systems. The archetypal measurement system introduced at the
beginning of this chapter will prove to be useful in this respect and is used as the
basis for a general discussion about signals and noise in Chapter 2 which aims to
put lock-in detection into a proper perspective. In Chapter 2, and throughout the
following chapters, the treatment is mainly qualitative, with the principal
mathematical developments left to the numerous Appendices.
Discussion on lock-in systems in the early chapters is confined to the so-called
'traditional' or 'conventional' variety of lock-in amplifier where the overall
handling characteristics are essentially those of the switching phase-sensitive
detector. These include some undesirable features, notably the susceptibility of
the phase- sensitive detector to interference signals at the odd harmonics of the
reference frequency.
The drawbacks of conventional systems are briefly reviewed in Chapter 6 in
preparation for the following chapters which deal with various 'advanced' systems
operating on the heterodyne and the pulse-width-modulation principles. Here we
shall find that the odd harmonic responses referred to above are suppressed
through more extensive processing of the signal and reference voltages, but that
the switching phase-sensitive detector is retained on account of its ability to
maintain linear operation under the most adverse noise conditions. At the same
time, the more complex system configurations are found to be characterized by a
number of additional spurious responses which must be minimized at the design

Chapter 14
stage. There are also certain inevitable trade-offs, notably with respect to
dynamic range and frequency coverage, which must be taken into consideration,
and we shall take account of these where appropriate.
Heterodyne and pulse-width-modulation lock-in amplifiers are almost invariably
supplied and used as self-contained units. In Chapter 7, however, we find that the
topic of phase-locking bridges the gap between 'conventional' and 'advanced'
systems in that reference processing can be arranged by using standard modules
such as phase-sensitive detectors and voltage-controlled oscillators. Readers
familiar with the literature on this subject will find the approach here heavily
biased towards locking with noisy signals, the object being to derive a local
reference voltage when this is otherwise unavailable. The treatment is
unavoidably mathematical in this case, but it is nevertheless intended that non-
specialist readers will benefit from a review of the problems of locking in noisy
conditions. The optimization procedures described represent but one way of
approaching the phase- lock problem. They do, however, take account of long-
term variations in signal amplitude, a feature which is noticeably lacking in the
general literature.
Chapter 10 deals with some of the problems inherent in bringing signal-recovery
equipment under computer control. Here, as elsewhere, it is hoped that the
development of ideas will be accessible to readers with an interest in lock-in
systems as measurement tools but who are otherwise non-specialists in the
general areas of electronics and telecommunications.
The same remarks apply to Appendices 5 and 6 which give an appraisal of noise
in amplifiers and the problems associated with signal connections, in particular
the avoidance of ground loops. It is noted there that lock-in recovery can be a
more-or-less straightforward business provided that proper attention is paid to
signal handling. While acknowledging that familiarity with specifications and the
basic rules of instrument management are best learned 'at the bench', it is hoped
that the guidelines established in this book are of the sort which make lock-in
recovery a reasonably exact science and that they will ensure that many common
pitfalls leading to erroneous or misleading results will be avoided.

1.1 References
1 DICKE, R.H. (1946): "The measurement of thermal radiation at microwave
frequencies', Rev. Sci. Instrum., 17, (7), pp. 268-275
2 TAUB, H., and. SCHILLING, D.L. (1971): 'Principles of Communication
Systems' (New York, McGraw Hill)
3 BETTS, J.A. (1970): 'Signal processing, modulation and noise' (London,
English Univ. Press)

Chapter 15
CHAPTER 2

Basic concepts in lock-in recovery

2.1 Introduction
Lock-in amplifiers are characterized by a wide dynamic range which gives the
ability to measure signals accompanied by relatively high levels of noise and
interference. It is appropriate therefore to begin with an examination of the signal
and noise voltages which might appear at the output of an experimental system of
the general type introduced in Chapter I and, in particular, to identify the
combinations of signals and noise which give rise to a signal recovery 'problem'.
As we shall see, a fairly close examination is necessary in order to give substance
to the vague, if popular, notion of signals 'buried in noise'. We shall then turn our
attention to methods of signal measurement and outline the principles of
synchronous detection which underlie the operation of lock-in amplifiers.
It was shown in the introductory chapter that signal recovery applications of lock-
in amplifiers usually involve the measurement of amplitude variations and - to a
lesser extent - phase variations of periodic signals. In support of this, Appendix 2
gives some attention to the structure of modulated signals and gives methods of
estimating signal bandwidth. In many signal recovery applications, however, the
signal modulations are relatively slowly varying, not just with respect to the
excitation frequency but also with respect to observation intervals ranging from
several seconds up to many hours. Indeed, in many circumstances the signal may
have fixed characteristics throughout the time available for measurement. We
shall find in this chapter that many of the important differences between
synchronous detection methods and other, non-linear, detection methods can be
readily demonstrated on this assumption.
This chapter includes an introduction to basic lock-in amplifiers. A brief appraisal
of the various system components highlights some of the main topics for
discussion in the chapters to follow.

2.2 Evaluating the signal recovery 'problem'

In the following we shall suppose that signals are obtained in the form of a
voltage variation either directly from an electrical transducer or from a suitable
combination of transducer and preamplifier. Although of extreme importance, the
technicalities of amplifier selection and the practicalities of signal connections
will not be a major concern at this stage; our objective will be to examine the
general characteristics of signals and noise which might be encountered in the
course of a typical measurement.
We define noise as all unwanted signals, so all sources of random disturbance
must be included in its description, together with the effect of more-or-less well-
defined interference originating from neighbouring experiments and installations.
The broad spectrum of interference phenomena is charted in Appendix 2 and
includes several well-known - if not notorious - sources such as mains-frequency
hum, r.f. breakthrough from pulsed experiments and pick-up from t.v. and radio
transmissions. Interference phenomena in general have a different status to the

Chapter 21
noise sources that are inherent in the experimental processes under investigation,
in the amplifying devices employed and in the transducers used to provide the
output signals. For example, interference effects can often be suppressed - if not
entirely eliminated - by careful experiment design, so it is useful to make a
distinction between noise which is fundamental to the system and noise of
external origin.

Fig. 2.1 Typical oscilloscope view of a signal contaminated by noise and

discrete interference

At the same time we should not make the mistake of supposing that the system
noise is so fundamental that it cannot be reduced further. In practice there is
often scope for improvement through selection of amplifiers, a topic which is
treated at some length in Appendix 5.
The first direct contact with signals in a laboratory is usually made via an
oscilloscope. In the case of noisy signals the resulting display is often a confused
jumble for example. Fig. 2.1 which shows the effect of noise, including discrete
interference, on a low-level signal.
The term buried in noise seems to be most appropriate here, but in many cases
this conclusion tells us more about the method of display than it does about the
signal. Let us look therefore at an alternative way of displaying the characteristic
of the signal, by using a spectrum analyser.
The spectrum-analyser approach gives a more explicit and graphical
interpretation of the relationships between signals and noise in a system. We find
that by making the transformation from time domain to frequency domain we can
often sort the signal and noise into their respective categories, a process which
overcomes the confusing superposition of the time domain picture.
Fig. 2.2 is such an example and uses the same input voltage as displayed on the
oscilloscope. The signal is now very much in evidence and well separated in
terms of frequency from sources of interference such as high-frequency
breakthrough from a neighbouring experiment. The noise which appeared to
dominate the oscilloscope display is now seen to be spread quite thinly over a
wide range of frequencies.

Chapter 22
Fig. 2.2 Spectrum-analyser display of the signal shown in Fig. 2.1. On
this scale the mains frequency pick-up region is immediately
adjacent to the zero frequency marker

The spectrum analyser picture also serves to remind us that many sources of
interference can be described quite adequately in terms of a line spectrum,
representing the concentration of power at discrete frequencies while the
fundamental noise and other interference sources give rise to a continuous
spectrum of noise up to the upper cut-off frequency of the output transducer or
output amplifier. Appendix 2 gives a breakdown of the principal interference
mechanisms according to their characteristic frequency ranges and spectral
signatures, and then gives attention to providing simple mathematical models for
the various kinds of spectrum which might be encountered in practice. These
include broadband spectra resulting from thermal noise and shot noise,
narrowband spectra and the ubiquitous flicker-noise or 1/f noise. The latter is
associated with a rise in noise power density at low frequencies. These models
provide the means of estimating the contribution to an observed fluctuation which
emanates from different parts of the noise spectrum in preparation for the first
step in signal recovery; this is the elimination of unwanted noise by filtering.
Thus, given a combination of signal and noise spectra such as that shown in
Fig. 2.2, we find that there is ample scope for improvement by introducing filters
to clean-up the signal prior to detection. The elimination of unwanted
components in this way is an important aspect of signal conditioning and is often
effective in bringing about a substantial increase in signal-to-noise ratio. In this
example, the use of a bandpass filter centred on the signal component indicated in
Fig. 2.2 results in the displays shown in Fig. 2.3. The signal of interest is now
relatively noise-free so that variations in, say, signal amplitude could be
measured using a conventional a.c. voltmeter without incurring a serious noise
penalty.
The idea of noise reduction by bandwidth reduction is of course central to an;
discussion of signal recovery. If, however, we restrict ourselves to the question of
signal-to-noise improvement before demodulation, then we find that the benefit
obtained by signal conditioning can be very limited.

Chapter 23
Fig. 2.3 (a) Signal of Fig. 2.2 following the introduction of a bandpass
filter centred on the signal frequency; (b) corresponding
frequency-domain picture

To illustrate this let us look at the idealized case shown in Fig. 2.4 where a
bandpass filter with bandwidth Bo is used to reduce the noise power appearing
with the signal, leaving the signal unaffected. It is shown in Appendix 2 that the
noise power is reduced by a fraction Bo/BI where BI is the input noise bandwidth
to the filter. If, at some later stage in the measurement, the signal amplitude
should be reduced, then the output signal-to-noise ratio can only be maintained
by a further reduction in filter bandwidth. For example, if the signal amplitude
falls by a factor of 10 (signal power reduced by 100) the filter bandwidth must be
reduced to 1/100 of its former value in order to restore the signal-to-noise ratio.
Clearly, this process cannot be repeated indefinitely. First of all, if the signal is
carrying modulation it will occupy a finite bandwidth, and this, in turn, will
determine the smallest filter bandwidth that can be used. Secondly, there are
practical limits to designing filters with very high selectivity, which, in any case,
results in a tightly tuned measurement system. This, almost invariably, gives
rise to additional problems such as susceptibility to drift and the inability to
follow even small variations in signal frequency.

Chapter 24
Fig 2.4 (a) Spectrum of signal and noise; (b) amplitude response of
bandpass filter used for noise reduction. In this ideal case, the
filter signal bandwidth, Bo is equal to the noise equivalent
bandwidth

Bearing these points in mind we can make a clear distinction between our
original example and the case shown in Fig. 2.5. The noise power is now
concentrated with high density in the region of the signal frequency and the
signal is over-shadowed by a massive interference component nearby. We are
now faced with a signal recovery problem of quite a different order, where
adequate separation of a signal and noise cannot be achieved by filtering and a
substantial fraction of the noise and interference must filter through to the
equipment used for demodulation.

Fig. 2.5 Frequency-domain view of a signal recovery problem

Chapter 25
In evaluating the effect of noise on a signal, therefore, we are interested more in
the distribution of the noise components with frequency rather than in the total
noise power which accompanies the signal. Thus, the total peak-to-peak
fluctuation which can so seriously obscure an oscilloscope display is not in itself
sufficient to prevent measurement of a signal, provided that adequate noise
reduction can be achieved through signal conditioning. If this approach fails, then
the final outcome of an experiment will depend entirely on the ability of the
demodulator to function in the presence of noise.

2.3 Demodulators for signal recovery

It was suggested in Chapter 1 that the term signal recovery implies some prior
knowledge of a signal and that this knowledge is used to advantage at the point of
detection.
The use of signal conditioning filters is an example of how prior knowledge of
the signal can aid the process of detection. Let us now turn our attention to the
provision of a demodulator and ask: are there any techniques available which take
specific account of the character of the signal imposed by the choice of
excitation source? In answering this question we would separate the true signal
recovery techniques from those which make no fundamental distinction between
signals and noise. We would find an essential difference in that demodulators for
signal recovery are almost invariably supplied with a reference signal which is
precisely synchronized with the signal of interest.

Fig. 2.6 (a) Incorporation of a synchronous detector in an experimental

system; (b) multiplier model for a synchronous detector

It is the availability of the reference which enables us to exploit the principle of

synchronous detection referred to in Chapter 1. The use of the term lock-in
systems in this context reminds us that we are dealing with demodulators which
are locked to a signal of interest by virtue of a synchronous or coherent
reference voltage. Fig. 2.6(a) is fairly typical in that the reference has been
obtained directly from the excitation voltage at a fixed level.
The response of a synchronous detector to variations in the amplitude and phase
of a synchronous signal will be considered in the next section. In general terms,
operation depends on the high degree of correlation which is known to exist
between a periodic signal of interest, s(t), and the reference r(t). The presence of
correlation is tested using the scheme outlined in Fig. 2.6(b), by first multiplying
the two inputs to form the product
vp(t) = r(t) [s(t) + n(t)]
where n(t) represents the entire disturbance due to additive noise and interference.

Chapter 26
When r(t) and s(t) are closely correlated, their product gives rise to a distinctive
response which depends upon the amplitude of the signal and its phase relative to
the reference. For example, in the important case of practical interest where s(t) is
a sinewave having fixed characteristics during some stage of a measurement, the
product term r(t)s(t) will give rise to a voltage which includes a constant term
proportional to signal level. The purpose of the low-pass filter shown in Fig.
2.6(b) is to separate this voltage from the higher-order products of multiplication
and allow it to filter to the final output. Regarding the noise, there should be no
correlation with the reference in this case and the average value of the noise
product r(t)n(t) is always zero in the final output. The response to a fixed signal is
quite unambiguous because there is no error due to rectified noise components:
any residual fluctuations due to noise appear as an a.c. variation which does not
affect the average value of the true output voltage due to the signal. In principle,
these residual fluctuations can always be attenuated to an acceptable level by
reducing the bandwidth of the output low-pass filter. This represents the major
mechanism for signal-to-noise improvement in synchronous detection systems.

Fig. 2.7 Meter indications at the output of: (a) a synchronous detection
system; (b) an envelope detector. The true deflection due to
signal corresponds to half-scale in each case

The response to a signal with changing amplitude will be a varying output

voltage with additive noise. When the amplitude is changing very slowly we can
interpret the response as a slowly varying d.c. level. In this case an output filter
with very small bandwidth can be used to smooth the fluctuation due to noise.
The problem is then to estimate the value of the d.c. component. This might be
achieved by observing the output on a voltmeter as shown in Fig. 2.7(a). With a
synchronous detector there is no residual deflection due to rectified noise and the
desired d.c. response is given by observing the average deflection of the meter.
Other types of detector, for example envelope detectors, give a meter indication
characterized by Fig. 2.7(b). Here the response is subject to an error due to
rectified noise components which contribute to the net deflection of the meter.
This source of error cannot be removed by using a low-pass filter. The only
recourse in this case is to eliminate noise components before detection by the use
of filters tuned to the signal frequency.
A further point to be noted is that rectifier-detectors are subject to inter-
modulation effects whereby signals and noise become multiplied together.
Intermodulation is inevitable in any process which depends on a non-linear
operation such as squaring or envelope detection. The effect can be safely
ignored at high input signal-to-noise ratios, but when the signal is reduced the

Chapter 27
 intermodulation products dominate and give rise to the phenomenon of threshold
or signal suppression. This generally occurs at input signal-to-noise ratios below
about 1:1 and corresponds to a loss of information in that the detector output no
longer contains a term which is simply proportional to the desired modulation.
In the light of this general discussion it is evident that ideal multiplier
synchronous detectors do not generate products of the type s(t)n(t); a distinct
separation between signal and noise is maintained throughout. As a result, the
residual noise output adds to the desired response. We can say that synchronous
detectors are linear insofar as the principle of superposition can be applied in
order to combine the responses due to individual components in the signal and
noise voltages.
Let us now turn from these general considerations and look at some specific
relationships between the reference, signal and noise in a synchronous detection
system.

2.4 Operation of synchronous detectors

2.4.1 Introduction
We shall begin by summarizing basic mathematical relationships which apply to
the ideal multiplier followed by a low-pass filter shown in Fig. 2.8.

Fig, 2.8 Synchronous detector with sinewave signal and reference

A sinewave reference will be used and we shall calculate the response to a single
sinewave component in the signal path. Signal and reference are conveniently
expressed in terms of their r.m.s. values Vs and VR:
s(t) = 2 Vs cos [st + s]
r(t) = 2 VR cos [Rt + R]
If we now form the product of signal and reference we can separate the result into
sum and difference components:
v p (t) = VsVR cos[(s + R )t + s + R ]
+VsVR cos[(s R )t + s R ]

This is an operation which will recur in succeeding chapters and which is almost
invariably linked to the assumption that the low-pass filter cuts off at a frequency
much less than R. In this case the sum-frequency component is effectively
eliminated from the final output. The fate of the other component will depend on
the magnitude of the difference frequency = s R . If this is less than or
comparable with the bandwidth of the low-pass filter we find that the output
appears in the form of an alternating or beat response at frequency . To
calculate the magnitude of this response we require the frequency-response

Chapter 28
function HL(j) of the low-pass filter. In general, the beat component will have
amplitude
vo = VsVRAL()
where
AL() = HL (j)
Since AL() has a cut-off well below R we find that the system is able to accept
only those signal components which lie very close to the reference frequency.
This can be described in terms of a transmission window centered on the
reference frequency with a characteristic dependent on AL() as shown in
Fig. 2.9. We can thus argue that the combination of reference, multiplier and low-
pass filter functions as a band-pass system giving a response only to signals in the
vicinity of the reference frequency.

Fig. 2.9 (a) Frequency-response magnitude of low-pass filter;

(b) transmission window derived from the characteristics of the
low-pass filter and centred on the reference frequency, R

2.4.2 Demodulation with a synchronous reference

Let us now turn to the case of greatest practical significance, where the signal and
reference are derived from the same source, for example in the experimental
scheme shown in Fig. 2.10 below.
The excitation of the experiment is sinusoidal at frequency R and provides a
reference which is to be used for detection of the output signal. The output signal
appears at the same frequency as the excitation and suffers a phase-shift s in the
experiment. The reference is applied to the multiplier via a variable phase-shift
network.

Fig. 2.10 Using a synchronous detector with a variable phase-shift

network in the reference path

Chapter 29
At this stage let us assume that the signal is noise-free. The output from the low-
pass filter can then be calculated by putting s = R in the results derived earlier.
For signals with fixed amplitude and phase we obtain the classic phase-sensitive
response in the form of a d.c. indication:
vo = kRVs cos
where
= s R
kR = VRAL(0)
Usually the reference amplitude is fixed so that kR is a constant scaling factor.
The response is then simply proportional to the signal amplitude and exhibits a
phase dependence through the cosine term. Let us now see how the reference
phase- shifter can be used to measure some specific modulations on the signal.
2.4.3 Amplitude demodulation
The reference phase is adjusted to bring the signal and reference in phase at the
multiplier to give an output:
V o = k RV s
for
R = s
The output will follow variations in signal amplitude provided that the low-pass
filter has a bandwidth wide enough to transmit the modulation signal without
distortion. For example, when the signal has the form:
vs(t) = m(t) cos ot
the output voltage will be
vo(t) = kRmF (t)
where mF(t) is a low-pass filtered version of the modulation signal m(t).1
In this case the output voltage of the synchronous detector will have a spectrum;
Vo(j) = kRM(j)HL(j)
= M(j)HD(j)
Here, M(j) is the Fourier transform of the modulation signal and we identify
HD(j) = kRHL(j)
as the frequency-response function of the synchronous detector.
2.4.4 Phase demodulation
To use the synchronous detector as a phase demodulator we must assume that the
signal amplitude is constant. The first step is to null the output of the detector by
bringing the signal and reference into quadrature at the multiplier, giving;

1
If we denote the impulse-response of the low-pass filter by hL(t), the filtered version of the
modulation function will be given by:

where denotes convolution.

Chapter 210
 R = s /2
If the signal phase subsequently changes by an amount m the response of the
detector will be
vo = kRvs sinm
The synchronous detector will operate as a linear phase detector only for small
phase variations. In this case we restrict the magnitude of m, so that sinm m
and obtain the approximately linear response:
v o k R v s m
Let us now take the general case where m is a time-varying phase-shift m(t).
When m(t) << l this corresponds to low-index phase modulation. We must now
consider the effect of the low-pass filter on the frequency components of m(t). If
the phase modulation has a Fourier transform m(j) the required relationship is
Vo(j) = kRVsm(j)HL(j)
= Vs m(j)HD(j)
or, in terms of a time variation;
vo(t) = kRVsm(t) hL(t)
The sensitivity of the system to phase variations is thus proportional to the
amplitude of the signal. Otherwise, the detector frequency-response function
HD(j) plays the same role in amplitude and phase detection.
2.4.5 Mixed modulations
In general, we must expect that the signal of interest appears with both amplitude
and phase modulations, in the form
s(t) = m(t) cos [t + s + m(t)]
If the reference phase is adjusted to bring signal and reference in phase at the
multiplier such that s = R the system response becomes:
Vo(t) = kRm(t) cosm(t) hL(t)
= kRm(t) hL(t)
for m(t) << l radian
We see that amplitude detection is first-order independent of phase variations on
the signal. The implication is that small errors in setting-up the in-phase
condition do not seriously affect accuracy when signal amplitude is to be
measured. Indeed an error of 10 in the in-phase condition leads to an error of
only 1.5 % when measuring the amplitude of a fixed signal.
It is worth noting that the operation of a synchronous amplitude detector does not
depend on m(t) being constrained to take only positive values. Negative values of
m(t) correspond to a phase reversal of the carrier signal which will be faithfully
reproduced as negative voltages in the output of the detection system.
Turning now to operation as a phase detector, we have noted that the overall
response is proportional to signal amplitude. This simple system is therefore
unsuited to operation as a phase detector when the signal carries amplitude
modulation.
More complex systems which involve two synchronous detectors operated in
quadrature will be considered in later chapters. These systems allow amplitude

Chapter 211
 and phase variations to be measured simultaneously without restrictions on the
maximum allowable phase shift. The key to this mode of operation lies in the
phasor representation of the signal and reference voltages given in Fig. 2.11,
drawn with respect to the reference phase. The output of a synchronous detector
is now seen to be proportional to the in-phase component of the signal, while
changing the reference phase by /2 gives us a measure of the quadrature
component of the signal. With the possibility of generating these two pieces of
information we begin to see the potential of synchronous detection systems for
phasor analysis and signal characterization beyond their traditional role in signal
recovery.

Fig. 2.11 Phasor representation of a sinusoidal signal drawn with respect

to reference phase

2.4.6 Noise rejection

The response of a synchronous detector to a synchronous signal of fixed
amplitude and phase is always obtained in the form of a d.c. indication. In
contrast, asynchronous signals such as noise and discrete interference
components always give rise to an alternating response in the form of beat
frequency products as discussed in Section 2.4.1. Furthermore, we have seen that
the only unwanted components which can give rise to spurious outputs are those
which originate in the immediate vicinity of the reference frequency, confined to
the transmission window defined by the characteristics of the low-pass filter.
We can conclude, therefore, that the rejection of a large part of the background
noise spectrum is inherent in the operation of a synchronous detector and that the
frequency selectivity of the detector is governed by the properties of the low-pass
filter. In practice, the experimental system itself will set limits to the maximum
rate of change in the amplitude and phase of the signal of interest. This in turn
will determine the minimum bandwidth which can be tolerated in the low-pass
filter if the recovered information is to be transmitted without distortion. In many
applications this maximum rate of change is deliberately restricted in order to
achieve the smallest possible value of output bandwidth and so enhance the noise
rejection properties of the detector (for example, by reducing the scan rate of
spectrometers and swept-response analysers). The case of a signal accompanied
by a wide band of white noise is given consideration in Appendix 3. Because of
the inherent linearity of synchronous detectors we find that the classic noise
reduction formula can be applied to calculate the signal-to-noise improvement
obtained by demodulation. This is defined in terms of the output signal-to-noise
ratio (SNRo) of the recovered information signal and the input signal-to-noise
ratio (SNRI) of the modulated signal. The latter is assumed to appear in white
noise with bandwidth BI. For the recovery of amplitude modulation we have
SNRo BI
Improvement factor = =
SNRI Bo

Chapter 212

where Bo is the (noise) bandwidth of the low-pass filter, set to a value just wide
enough to pass the modulation signal.
The use of the filter noise bandwidth ensures that noise transmitted in the tails
of the filter beyond the signal cut-off frequency is accounted for in the
calculation. The noise bandwidths of several important filter types are given in
Appendix 4.
To conclude this section let us review some of the practical advantages of
synchronous detection schemes compared with methods of non-linear detection.
We can begin with Fig. 2.12 which compares the signal-to-noise improvement
obtained for amplitude demodulation using a synchronous demodulator and an
envelope detector.

Fig. 2.12 Comparison of input and output signal-to-noise ratios for (i) an
envelope detector and (ii) a synchronous detector

For a valid comparison to be made we must ensure that the systems are identical
with respect to the noise-rejection filters used before and after demodulation. We
find that for strong signal conditions the two methods are comparable. The
improvement factor is constant here, with SNRo strictly proportional to SNRI (a
discrepancy of a few dB is barely significant in signal recovery terms). The
situation is quite different, however, when the input signal-to-noise ratio is low.
The synchronous demodulator is capable of maintaining a constant improvement
factor for all levels of SNRI whereas the envelope detector deteriorates rapidly
when the input ratio falls to about 1:1. In the region below threshold the output
signal-to-noise ratio of the envelope detector falls faster than SNRI, and the
demodulated output becomes grossly distorted.
In addition, the performance of the envelope detector is degraded for signal-to-
noise ratios which would normally be considered quite favourable in a signal
recovery context. We therefore find it necessary to provide noise suppression
filters centred on the signal frequency to ensure a high signal-to-noise ratio before
detection. In difficult conditions this probably means that we require highly
selective tuned circuits which are subject to drift and other temperature effects
and which render the system incapable of operation if the signal frequency is
changing by accident or design.
Synchronous demodulators, and hence lock-in systems, are normally operated
without front-end filters and the final noise rejection takes place by averaging in a

Chapter 213
low-pass filter which follows the multiplier. This final bandwidth need be no
wider than to transmit the information signal without distortion. In many
applications the signal variations are of such a long-term nature that a final
bandwidth of a few hertz may be quite adequate. The practical significance of
defining such a narrow bandwidth by means of a low-pass filter, rather than by
using a highly selective bandpass filter at the signal frequency, cannot be
overstated. As a direct result we find that lock-in systems are ideally suited to
swept-response measurements. By using a swept reference, it is possible to track
signals over many decades of frequency.
A plot similar to Fig. 2.12 can be drawn to show the occurrence of a noise
threshold in systems used for phase demodulation. Lock-in systems are free of
this effect for small index phase variations but this aspect is rather overshadowed
by the growth in importance of lock-in systems used for precision measurements
of amplitude and phase on relatively clean signals.
This is a relatively new development brought about by improvements in
electronic circuit and system design during the last decade and which will be
given due attention in the following chapters.

2.5 Basic lock-in amplifiers

2.5.1 Introduction

Fig. 2.13 A basic lock-in amplifier

The requirements of a basic lock-in system are shown in Fig. 2.13. Although such
a system could be built up by interconnecting individual units, the purchase of a
lock-in amplifier usually represents a more cost-effective approach. Lock-in
amplifiers incorporate all the features of Fig. 2.13 in a single unit which is
optimized for operation over a range of frequencies. The major advantages of
using an integrated system of this type are that the controls are calibrated directly
in terms of full-scale sensitivity for a synchronous signal and that the relative
phase of the signal and reference channels can be maintained to within close
limits over the recommended frequency range.
Let us now identify the essential requirements of each main block in the lock-in
amplifier and take the opportunity to review some practical aspects of lock-in
operation.

Chapter 214
2.5.2 The signal channel
An amplifier is necessary to bring the signal to a level sufficient to overcome the
self-noise of the multiplier, and the provision of switched gain permits the
sensitivity of the system to be varied. As shown in Fig. 2.13 lock-in amplifiers
are usually provided with an optional range of preamplifiers. The objective here
is not merely to boost the gain but to provide an optimum noise match to the
signal source. In this way we can ensure that the spectrum of noise in the vicinity
of the signal frequency is not enhanced at the expense of the signal due to an
excessive noise contribution from the amplifier. The main considerations in
choosing a suitably low-noise amplifier are reviewed in Appendix 5. An
associated problem is the occurrence of ground loops and pick-up in a complex
experimental system which serve to degrade the signal-to-noise ratio even
further. This topic is discussed further in Appendix 6 which outlines a basic
stratagem for making interconnections between instruments.
2.5.3 Signal conditioning
As we have seen, there is no fundamental requirement to clean up the signal
prior to detection in a synchronous system. It might seem surprising, therefore,
that many lock-in amplifiers are provided with an array of filters for this very
purpose.
The reason is that signal conditioning is often an essential step when a high
amplification factor is required to obtain the required system sensitivity. If a
substantial fraction of the noise and interference were not eliminated in this way
the amplifier could be driven into saturation at an undesirably low gain factor.
The resulting distortion and intermodulation would then seriously degrade system
performance.
Practical multiplier circuits are similarly designed to cope with a specified range
of signal and noise voltages, so that a degree of signal filtering might be
necessary to protect the vital detection process against overload. In either case,
the main target in providing filters is often the spectrum of discrete interference
components. If an electronic circuit is to saturate on noise it will most likely be
the discrete frequency components occurring at large peak-to-peak values which
make the greatest contribution.
A review of the main types of signal-conditioning filter is given in Appendix 4,
which describes their frequency-response characteristics and catalogues their
noise bandwidths. Of these, one of the most useful is a sharply tuned notch filter
which can be used to suppress a dominant interference component, say at mains
frequency. The remaining components can then be reduced along with the out-of-
band noise by the use of high- and low-pass filters as indicated in Fig. 2.14.
The attenuation of the remaining discrete frequency components will be
obtainable from the amplitude responses of the filters which combine to form a
band-pass response extending from fL to fH. In an extreme case a resonant filter
tuned to the signal frequency might be used, but, as we have seen, this would
introduce undesirable features which we have been so anxious to avoid in
electing to use a lock-in system. The effects of introducing a tuned filter in the
signal channel are discussed further in Chapter 4. For the moment it is sufficient
to note that the inclusion of filters in the signal channel introduces undesirable
restrictions on a signal recovery system and even the most simple high-pass or
low-pass filter introduces a phase shift which must be compensated in the
reference path. Fortunately, the capability of modern systems is such that filtering
can be kept to a minimum in all but the most demanding of applications.

Chapter 215
Fig 2.14 (a) Spectrum of fundamental noise; (b) amplitude spectrum of
signal and discrete interference; (c) combined amplitude-
response characteristic of a mains-frequency notch filter, a high-
pass filter cutting off at fL and a low-pass filter with cut-off
frequency fH

2.5.4 The multiplier

The basic principles of synchronous detection have been established in terms of
an ideal multiplier which can maintain its performance under all applied levels
of signals and noise. If the advantages cited in Section 2.4 are to be realized,
therefore, practical multipliers must possess exceptional dynamic range.
Otherwise, departures from linearity under conditions of even moderate signal-to-
noise ratio will give rise to intermodulation effects and the eventual suppression
of the desired response.
A major objective in the past has been to identify those circuit configurations
which offer the widest dynamic range in order to maximize the signal recovery
capability of practical systems. These efforts have resulted in the almost universal
adoption of the switching multiplier or phase-sensitive detector which is found at
the heart of all lock-in recovery systems. The essential difference in this case is
that the multiplier is now driven by a squarewave switching waveform that is
precisely synchronized to the applied reference waveform as described in the next
section.
The operating characteristics of phase-sensitive detectors are dealt with in some
detail in the next chapter and it is shown there how the properties of a
synchronous detection system are modified by the inclusion of a switching
multiplier. On the basis of our discussions so far we can emphasize the need for
good dynamic performance and note that linearity at the point of detection must
be supported by linearity elsewhere in the system. As we have seen, this applies
mainly to the signal-channel amplifier, which is expected to handle small signals
accompanied by relatively large noise and interference voltages. Not surprisingly,
we find that commercial systems are rigorously monitored to detect the onset of
overload at all critical points and that the allowable range of voltages for linear
operation is clearly specified.

Chapter 216
2.5.5 The reference channel
The phase-sensitive detector is supported by a reference channel which supplies
the precise switching waveform required for signal detection. The switching
waveform is triggered in the first instance from the positive zero crossings of an
applied reference waveform as shown in Fig. 2.15, and is always arranged to be
precisely symmetrical irrespective of the symmetry of the original reference
input. The displacement R is usually introduced by means of a calibrated phase
control. In a broadband reference channel this phase-shift can be maintained to
a high degree of precision over a wide range of reference frequencies and the
symmetry of the switching waveform is rigidly controlled.

Fig. 2.15 (a) Reference input; (b) Symmetrical switching waveform

triggered from (a) and phase-shifted R

Provided the phase conventions are observed there is no ambiguity in defining the
phase shift of the (internal) reference switching waveform with respect to any
applied (external) waveform. Fig. 2.16 gives an example where a non-sinusoidal
reference waveform is available. The zero phase switching waveform is
generated in synchronism with the positive zero crossings of the applied
reference and a phase shift of 90 corresponds to displacing the reference

Fig- 2.16 (a) Non-sinusoidal reference waveform; (b) zero phase

reference switching wave form; (c) switching waveform
displaced by 90

Chapter 217
switching waveform by one quarter of a reference cycle. Phase-shift controls are
usually provided in the form of a continuously variable adjustment covering the
range 0100, together with 90 and 180 pushbuttons for quadrant selection.
In addition, even the most basic lock-in amplifier is usually provided with a '2f'
facility whereby the system becomes synchronized at the second harmonic of the
applied reference waveform. The 2f mode is normally made available by
pushbutton selection up to reference frequencies of one half the maximum value
allowed in normal operation
2.5.6 The low-pass filter
The low-pass filter provided with the majority of lock-in amplifiers is based on
either a single-section or two-section RC filter giving a roll-off of 6 dB or 12 dB
per octave beyond the cut-off frequency. A range of bandwidths is supplied and
the range switch is almost invariably presented as a time-constant control, that is
in terms of the characteristic time T0 = RC of the filter. The characteristics of
these filters are presented in Appendix 4. In calculations involving noise, the
noise bandwidth of the filter must be used, the appropriate values being
1/(4T0) Hz for a single section filter and 1/(8T0) Hz for a two-section filter. The
smaller value of noise bandwidth for the filter with sharper roll-off reminds us
that it is more effective in suppressing noise at frequencies beyond cut-off. Also,
it should not be forgotten that the signal bandwidth of a two-section filter is
smaller than that of the corresponding single-section filter; the bandwidth
B = 1/(2T0) gives the 6 dB frequency in the first case and the 3 dB frequency
in the second.
In the majority of experiments it is advisable to use the smallest possible value of
time constant to ensure that the response of the lock-in amplifier is not too
sluggish. The settling time of the filter is important in this context since it takes
account of the recovery of the filter following a step change in signal level. This
could also apply to an increment in sensitivity caused by range switching or to
switching a phase increment in the reference channel. The step responses of the
two filter types are shown in Fig. 2.17.

Fig. 2.17 Step responses of R-C low-pass filters.

Chapter 218
We find that a settling time of four time constants brings the output of the single-
section filter to within 2% of its final value while the two-section filter output is
in error by 10%. For a given value of time constant, therefore, we must be
prepared to trade noise rejection for the ability to follow a changing signal.
Clearly, for time constants of 10 s or greater, the settling time can become a
major factor.
In this respect it should be noted that both the settling time and the mean-square
fluctuation due to residual noise in the filter output are reduced in proportion to
the time constant setting. Using the results of Appendix 3 with the appropriate
value of noise bandwidth for the low-pass filter, we obtain improvement factors:
I1 = 4BIT0, 6dB/octave filter
I2 = 8B1T0, 12dB/octave filter
where B1 is the input noise bandwidth.
The improvement factors refer to the mean-square fluctuations attending the
signal before and after detection. Since, in most practical cases, the r.m.s value of
a signal is of interest, it is usual to consider the r.m.s. value of the output
fluctuation. Thus, increasing the time constant by a factor x increases the settling
time by the same amount but brings a reduction of only x in the r.m.s. value of
the noise measured on an output meter or chart recorder.

2.6 Signal recovery capability

The word capability is used throughout the literature supplied by lock-in
amplifier manufacturers. Capable systems are those which can withstand the
effect of massive levels of noise while maintaining a linear response to a
synchronous signal. When dealing with theoretical models we can assume infinite
capability. Also we find that the signal-to-noise improvement obtained through
synchronous detection is limited only by the minimum allowable output
bandwidth. As we have seen, this depends upon the nature of the recovered
information signal: if the signal parameters are effectively fixed for the duration
of a measurement the output bandwidth can be made very small and the potential
for signal-to-noise improvement is correspondingly high.
In a practical experiment the improvement factor must bridge the gap between the
input signal-to-noise ratio, SNRI and the desired output signal-to-noise ratio,
SNRo. SNRI must, of course, be restricted in order to preserve linear operation
while the improvement factor should be achievable within the time set aside for
the experiment. If an arbitrarily long time is not available for measurement the
desired improvement factor might not be attainable even though the input signal-
to-noise ratio lies within the capability of the lock-in system. The following
discussion will provide a valuable perspective on system performance when we
consider aspects of specification at a later stage.
Let us begin by supposing that we have a sinusoidal signal appearing against a
background of white noise with bandwidth BI. In order to maximize the prospects
of recovery we shall assume that the amplitude and phase of the signal are fixed.
We can then choose whatever output bandwidth we wish without being
constrained to respond to variations in signal level.
The improvement factor brought about by synchronous detection is B1/Bo, where
Bo is the noise bandwidth of the output filter. The output signal-to-noise ratio is
therefore

Chapter 219
 B1
SNRo = SNR1
Bo
The normal experimental procedure is to select Bo to give an acceptable value of
SNRo for a given value of input signal-to-noise ratio. The problem arises when
the required value of Bo is so small that it results in an excessively sluggish
response and inconveniently long settling time in the output circuit. Under these
circumstances we must either accept a lower value for SNRo or conclude
reluctantly that the input signal-to-noise ratio is too low to permit measurement to
the required precision in the available time.
At a time constant of 100 s the response of a synchronous detector is such that a
settling time approaching 10 minutes is required to recover from the slightest
disturbance in the signal. In many circumstances therefore, an output signal-to-
noise ratio of about 1:1 would represent a reasonable limit to detection in view of
the length of time required to average the response from the residual noise
background. Let us therefore limit the maximum time constant to 10 s and
demand that the output indication appears with a signal-to-noise ratio of about
10:1 (that is about 3:1 in terms of r.m.s. fluctuation). Putting Bo = 1/(8T0) with
T0 = 10 s, we obtain the following bound on the input signal-to-noise ratio:
SNR1 > 1/(8B1) (B1 given in hertz)
For detection of audio-frequency signals in an input bandwidth of about 10 kHz
we find that for reasonable precision at a moderate observation time the input
signal- to-noise ratio must be better than 1/80 000 (50 dB). If the noise appears
in a wider bandwidth (with a correspondingly lower noise density in the vicinity
of the signal) or if a larger time constant can be tolerated, the limit could be
relaxed to about 60 dB. This corresponds to measuring the amplitude of a
100 V signal in a noise background of 100 mV r.m.s. Even at this level, the
achievable performance falls well short of the popular notion of recovering
signals from 100 dB of noise. What is important here is that we have reached this
conclusion without referring to linearity or the ability of electronic circuits to
function correctly with noisy inputs. It would appear that a lock-in amplifier
capable of handling signals in the presence of 60 dB of wideband noise would be
able to fulfil all but the most demanding of measurement tasks. Unfortunately, the
simple calculations given here refer only to disturbance by white noise, whereas
this idealized situation is rarely observed in practice. It can safely be assumed that
the most spectacular claims in respect of lock-in amplifier capability refer to
disturbance by narrowband noise or to large-scale interference components
appearing at frequencies well removed from any transmission windows
associated with the phase-sensitive detector.

Chapter 220
CHAPTER 3

Phase-sensitive detectors

3.1 Introduction
The wide dynamic range of modern lock-in amplifiers results from the use of a switching
multiplier as a synchronous detector. The adoption of a switching circuit leads to a degree of
precision which cannot be matched by true multipliers and, moreover, has the added
advantage of operational simplicity. This is evident from the block diagram given in Fig. 3.1.

Fig 3.1 Phase-sensitive detector: principles of operation

We shall find it convenient at this introductory stage to maintain a distinction between the
switching network and the low-pass filter in the output circuit. For the most part, however, we
will conform with the usual practice of referring to the entire combination as a phase-sensitive
detector1. Thus, when we come to consider the specification of phase-sensitive detectors it
will be the behaviour of the switch/filter combination which is of interest.
In line with comments made in the Preface we shall be concentrating on the systems aspects
of phase-sensitive detectors rather than on detailed circuit implementations. Those with an
interest in circuit techniques are recommended to read the paper by Carter and Faulkner [1]
which contains several circuit configurations of phase-sensitive detectors and examines
sources of error in practical designs. This paper is one of the very few published accounts
where the level of treatment is appropriate to the performance of commercial systems.

1
The alternative forms, mixer and demodulator, are widely used.

Chapter 31
3.2 Principles of operation
The key to the operation of the phase-sensitive detector is the two-state switch which is
controlled electronically from the reference voltage. The switch changes position between
points A and B as the reference changes polarity. This action gives a systematic change of
gain between +1 and 1 in the signal path.
We shall be considering the classic operation in which the phase-sensitive detector spends
equal times in its two states, an arrangement which gives rise to the well known waveforms of
Fig. 3.2

Fig 3.2 Waveforms in a phase-sensitive detector operating with an in-phase

sinewave signal

In Fig 3.2, the sinewave signal and applied reference are precisely in phase. The reference
changes polarity in a symmetrical fashion, in step with the signal, and so causes full-wave
rectification of the signal at the switch output.
The output of the switch is then applied to the low-pass filter which smooths out the ripple
component and delivers a d.c. voltage which is proportional to the amplitude of the signal.

Fig. 3.3 Waveforms in a phase-sensitive detector for different phase conditions:

(a) = 180; (b) = 90; (c) arbitrary phase

Chapter 32
 In most applications the signal and reference will not arrive at the phase-sensitive detector
exactly in phase; hence, Fig. 3.3(a) which shows the effect of a phase reversal, giving a
negative d.c. level, and Fig. 3.3(b) which shows the output when the phase displacement is
90. In this case the output from the switching stage is a bipolar waveform which averages to
zero and gives no net response from the low-pass filter. Finally, Fig. 3.3(c) shows the output
at some intermediate value of phase shift, giving a d.c. level somewhere between the positive
and negative maxima obtained with = 0 and = 180.
To determine the exact relationship between signal and reference, we recognize that the
switching operation is equivalent to multiplication of a signal by a squarewave taking values
of +1 and 1. We can therefore use the ideal multiplier model shown in fig. 3.4 where the
reference waveform has the Fourier series representation
4 1 1
r(t) = [cos(R t + R ) cos 3(R t + R ) + cos 5(R t + R ) ...]
3 5
Fig 3.4(b) shows the switch output for the case where the signal and reference are
asynchronous. The switched signal has zero average value and its general form can be
obtained by forming the product:
v p (t) = r(t)s(t)

where
s(t) = 2Vs cos( s t + s)

Multiplying term by term and separating into sum and difference components we obtain
2 2Vs 1
v p (t) = [cos(R t st + R s ) cos(3R t st + 3 R s )
3
1
+ cos(5R t st + 5 R s ) ...]
5
The development is thus similar to the case of the ideal synchronous detector discussed in
Section 2.4. For synchronous operation we put s = R and, as before, we assume that the
low-pass filter cuts off well below the reference frequency. This eliminates multiplier
products at frequencies 2R, 4R, 6R, etc. from the final output which contains only the
phase-sensitive d.c. component:
2 2
Vo = Vs AL (0) cos( R s )

Here, AL(0) gives the magnitude of the filter response at zero frequency.

Apart from a scaling factor, the response of a phase-sensitive detector to a synchronous

sinewave is identical to that of an ideal multiplier-detector operating with a sinewave
reference. The essential difference in this case is that the phase-sensitive detector will also
give a phase-sensitive d.c. output in response to signals at frequencies 3R, 5R etc. This is
shown in Fig. 3.5 for a sinewave signal at the third harmonic of the reference frequency. The
relative sensitivity of the detection system at these additional frequencies is 1/3, 1/5 and so
on, corresponding to the relative magnitudes of the reference Fourier components.
A detection system with this property is said to be harmonically responding. We shall find
that there is minimal practical advantage in having a harmonically responding system; indeed,
the additional responses are frequently dismissed as anomalous or spurious. In later
chapters, we shall be investigating ways in which the harmonic responses can be suppressed
by improved system design. It is significant, however, that all these improved systems

Chapter 33
continue to rely on a switching phase-sensitive detector to provide the dynamic range
essential for signal recovery operation.

Fig 3.4 (a) Ideal multiplier model for a phase-sensitive detector; (b) output from
switch, before low-pass filter, for a sinusoidal signal with s < R

Fig. 3.5 Waveforms in a phase-sensitive detector for a signal synchronized to the

third reference harmonic

3.3 Harmonic transmission windows

The idea of a transmission window was introduced in Chapter 2 as a convenient way of
representing the response of a synchronous detector to signals with frequencies close to the

Chapter 34
reference frequency. If we now consider the effect of using a squarewave reference waveform
we find that the additional harmonic components lead to the set of transmission windows
illustrated in Fig. 3.6.
The transmission windows are centered on the odd harmonics of the reference frequency and
the maximum magnitude of each window is weighted by the magnitude of its associated
reference Fourier component. Before a signal can produce a response at the output of the
phase-sensitive detector it must lie within one of the transmission windows. In order to
produce a true d.c. response, a signal must be synchronous with one or more of the
reference Fourier components as was shown in Fig. 3.5. Otherwise, the output will appear as
an alternating beat note at the difference frequency between the signal and the centre
frequency of the transmission window.

Fig. 3.6 The first five harmonic transmission windows of a switching phase-sensitive
detector

Fig. 3.7 The Kth transmission window of a phase-sensitive detector corresponding to

a 6 dB/octave low-pass filter with time constant To
The amplitude response of the noise equivalent filter is shown as a dashed
characteristic

Fig. 3.7 shows the form of one of these windows obtained when a low-pass filter having a
roll-off of 6dB/octave is used to follow the switching stage. It is centered on the harmonic

Chapter 35
 frequency KfR where K takes values 1, 3, 5 etc. The 3 dB and noise bandwidth are always
independent of the centre frequency.
The practical importance of the transmission windows stems from the fact that they represent
frequency regions where the phase-sensitive detector is susceptible to large-scale discrete
interference components. Ideal synchronous detectors operating with a sinewave reference
are, of course, relatively immune to such components unless they originate very close to the
reference frequency. The additional susceptibility of switching phase-sensitive detectors to
interference above the reference frequency, together with possible measurement ambiguities
resulting from a harmonically responding system, can be a severe limitation in some
circumstances. Some of the problems associated with the harmonic responses are discussed in
Chapter 6.

3.4 Noise bandwidth of phase-sensitive detector

When a signal in broadband noise is measured using a phase-sensitive detector, the post-
detection noise output is increased over the level which would be calculated for a multiplier-
detector using a sinewave reference. This is because of the effect of the additional
transmission windows which are able to leak noises through to the final output.
The total effect for a band of white noise can be calculated with the help of Figs. 3.7 and 3.8.

Fig. 3.8 Calculation of noise output from a phase-sensitive detector for a white-noise
output. Each transmission window is replaced by a rectangular noise
equivalent window with Bandwidth 2Bo

The noise bandwidth of each transmission window is given by 2B0 where B0 is the noise
bandwidth of the low-pass filter. The noise outputs due to components which fall within each
window will have mean-square values proportional to WN B0, where WN is the spectral
density of the white noise.
The mean-square value of the noise transmitted by the fundamental window can be written as

n12 = aNWN B0
where the constant aN takes account of the scaling factor of the phase-sensitive detector and
any associated amplifiers (see Appendix 3).

Chapter 36
 To calculate the noise output from any other window we must include a weighting factor
which depends on the magnitude of the associated Fourier component. The noise output from
a window from the Kth harmonic is therefore

nK2 = a NWN B0 / K 2 , K = 3, 5, 7
There is no coherence in the noise contributions from the individual transmission windows.
This means that the total noise output can be obtained directly by summing mean-square
values. We obtain

nT2 = a NWN B0 (1+1/9 +1/25 + ...)

Noting that

1/(2n +1) 2
= 2 /8
n =0

We obtain a total mean-square noise voltage:

2
nT2 = a NWN B0 1.23 a NWN B0
8
This result shows an increase of only 23% over the mean-square noise transmitted by the
fundamental window and represents an increase of around 11% in the r.m.s. output
fluctuation. We can conclude that for white noise the effect of noise in the higher-order
windows is negligible in practical terms. However, for exact calculations, the standard noise
reduction formula derived in Appendix 3 for an ideal synchronous detector can be modified to
give a less optimistic improvement factor.
SNRI BI
= 8 / 2
SNRo B0
Here, BI is the input noise bandwidth and 2B0/8 is the exact noise bandwidth of the phase-
sensitive detector.

3.5 Non-sinusoidal signals

3.5.1 Introduction
So far, the properties of synchronous detection systems have been discussed solely in terms of
sinusoidal signals. We shall now extend the discussion to include all types of periodic signal
and pay particular attention to some waveforms which have special practical importance.
We shall assume that a synchronous reference squarewave is available for detection of the
signal. In practice, this would usually be generated by a reference unit of the type described in
Section 2.5.5, triggered in the first instance by the positive zero crossings of an applied
reference voltage.
A problem which is common to many measurements involving phase-sensitive detectors is to
introduce the reference switching waveform in the correct phase to maximize the d.c.
response to a given signal. This must often be achieved under noisy conditions for a signal
where the d.c. response varies with the reference phase setting in a complicated way.
In the special case of sinewave signals we have seen that the switching phase-sensitive
detector response follows a cos law, where represents the relative phase of the signal and
switching waveform measured at the phase-sensitive detector. The following procedure can
therefore be adopted to adjust the phase of the detection system, starting from an arbitrary

Chapter 37
 initial phase condition. First, the phase-sensitive detector output is nulled by adjusting the
reference phase. This enables a quadrature condition at the phase-sensitive detector. The
desired in-phase condition is then obtained by shifting the set phase by 90. For a sinewave
signal the in-phase reference setting arrived at through this procedure maximizes the
response of the detection system. In practice, the null point can be determined with high
accuracy even when the signal is extremely noisy and the procedure defines a precise and
repeatable reference condition for subsequent measurements.
This approach to setting the phase of a phase-sensitive detector is widely applied to periodic
signals of all shapes and forms. This procedure is often justified on the assumption that a
phase displacement of 90 from the null-point automatically maximizes the response of the
phase sensitive detector to all types of periodic signal. Unfortunately, this is not necessarily
the case as we shall see in the following section.
3.5.2 General considerations
Let us take a periodic signal with fundamental frequency 0 having a Fourier description

s(t) = n cos n0 t + n sin n0 t
n =1 n =1

In describing s(t) we have omitted a d.c. component since, in practice, signals are invariably
a.c. coupled prior to phase-sensitive detection.

The signal is switched by a synchronous squarewave with the Fourier series:

4

n cos[(2n +1)(0 t + R )]
r (t ) =

(1) (2n +1)
n =0

where R is the phase of the reference unit defined with respect to an externally applied
reference waveform.

The d.c. response of the phase-sensitive detector is obtained by forming the product s(t)r(t)
and extracting the difference-frequency terms at zero frequency. Since the reference
comprises only odd harmonic components it is only the odd harmonics of the signal which
enter the calculation. We obtain an output voltage:

2 n cos(2n +1) 2 n sin(2n +1)
Vo = (1) 2n +1 2n +1 R (1) 2n +1 2n +1 R
n =0 ( ) n =0 ( )
In general, for any set of n and n, there will be a value of R which leads to Vo = 0 and
which therefore corresponds to the null point referred to earlier. We shall denote this value
by q.
At some other value of R, denoted by i, the magnitude of Vo will be a maximum. This value
is obtained by solving

where

We assume that AL (0) =1 in this case

Chapter 38
We shall not attempt a general solution in either case. It is sufficient to note that if the value
of q is determined, either by calculation or experiment, the value of q 90 will not satisfy
the condition for maximum response unless the signal waveform is subject to certain
constraints. These will now be investigated.
3.5.3 Symmetrical periodic signals
Many non-sinusoidal waveforms of practical importance such as squarewaves, triangle and
rectangular pulse waveforms, possess a high degree of symmetry. In all these cases the time
origin can be chosen to give a waveform which is either an even or an odd function of time.
The corresponding Fourier series then consists only of cosine terms in the first case (n = 0)
and sine terms in the second (n = 0). Inspection of the expressions for Vo and dVo/dR shows
that for waveforms with this essential symmetry we always arrive at values of q and i which
satisfy
i = q 90
Relationships between the reference switching waveform and a number of non-sinusoidal but
symmetrical signals are shown in Fig. 3.9. We can make the following observations about the
types of signal chosen.

Fig. 3.9 (a) - (c) Symmetrical waveforms; (d), (e) reference switching waveforms
introduced at R = i and R = q respectively

i) Setting the reference channel phase by first nulling the phase-sensitive detector output
and then shifting the phase by 90 automatically maximizes the d.c. output of the phase-
sensitive detector. We shall refer to this as the null-shift procedure.
ii) The conditions for zero output and maximum response correspond to bringing the Fourier
components of the reference switching waveform first in quadrature and then in phase
with the corresponding components of the signal.
iii) The null-shift procedure summarized in (i) does not depend on absolute phase calibration.
For recovery work, a variable phase-shifter and a calibrated phase increment of 90 are
sufficient to achieve optimum detection of symmetrical signals.

Chapter 39
iv) When the reference channel is calibrated according to a convention such as that described
in Section 2.5.5 the null-shift procedure provides a basis for measuring the phase-shift of
a sinewave signal relative to that of an applied reference voltage. It is now apparent that
the validity of such a measurement will be in doubt if the sinewave signal is subject to
harmonic distortion. This aspect is discussed further in Chapter 4 in relation to precision
phase measurement.
3.5.4 Asymmetrical periodic signals
In the present context, asymmetrical signals are those for which it is impossible to choose a
time origin such that s(t) = s(t). In practice signals in this category are most likely to occur
when a normally symmetrical signal is subject to linear filtering giving rise to waveforms
such as those shown in Fig 3.10. The waveforms in Figs. 3.10(b) and (c) in particular will
often be encountered in experiments using chopped excitation where a squarewave signal
has been transmitted by a low-pass or a high-pass signal conditioning filter. To this general
class of signals we can add the important case of sinewaves subjected to arbitrary harmonic
distortion.
In all these cases it is possible to determine a value of reference phase-shift which nulls the
phase-sensitive detector output. In general, however, changing the phase by 90 from the null
point fails to maximize the phase-sensitive detector output. Fortunately, in all but the most
extreme cases, the resulting output is normally within 10% to 20% of its maximum possible
value. In addition, the null output obtained for R = q provides an ideal datum point for
setting the phase which can be reproduced on future occasions even under noisy signal
conditions.

Fig. 3.10 Examples of asymmetrical waveforms

In view of these remarks, the procedure of nulling followed by introducing a phase offset of
90 has much in its favour even when the resulting response is less than optimum. As a result,
this approach is almost universally adopted in signal recovery work where it is sufficient to
obtain a consistent measure of signal magnitude in the presence of noise. It is, nevertheless,
worth bearing the following points in mind.
i) With symmetrical signals the derivative dVo/dR is zero for R = q 90. The phase-
sensitive detector output is then maximized and becomes first-order independent of small
phase adjustments in the reference channel. This advantage is lost when the null-shift
procedure is applied to asymmetrical signals.
ii) In the two-phase lock-in systems described in Chapter 5, an automatic phase control loop
is sometimes used to determine the null point R = q. Detection is subsequently carried
out at relative phase R = q 90. In the light of our discussions we must expect that the
overall response of such a system to an asymmetrical signal will have less than its
maximum value.

Chapter 310
 iii) We can conclude that the only way in which the null-shift procedure can be applied to
give a maximum response to all forms of periodic signal is to use a detection system with
fundamental-only response. Some methods of suppressing the harmonic responses of
phase-sensitive detectors are discussed in Chapters 8 and 9.
3.5.5 Squarewave signals: a special case
Fig. 3.11 illustrates the waveforms which result when a squarewave of Vs volts peak is
applied to a switching phase-sensitive detector with an in-phase reference. The switch output
is a constant level of Vs volts, the ripple component which is characteristic of operation with
sinewaves being absent in this case (compare Fig. 3.2).
The net response is larger than that obtained with a sinewave of the same r.m.s. value because
there is now a contribution from each of the fully synchronous Fourier components of the
signal and reference voltages.
Thus, if we begin with a sinewave signal and then convert it to a squarewave by adding
Fourier components of the correct magnitude and phase, the d.c. output voltage will increase
by a factor

[1+1/9 +1/25 + ...] = 2 /8

which, from Section 3.4, is the factor by which the mean-square noise is increased when
white noise is leaked through the harmonic transmission windows. The effect of adding the
additional Fourier components to the sinewave signal is, therefore, to boost the signal
indication at the expense of the noise. The response is increased by 23%, which is more than
compensates for the 11% increase in the r.m.s. noise fluctuation. The reason is, of course, that
the signal contributions add coherently while the individual noise outputs must be added in a
mean-square sense.
In terms of an ideal multiplier model we could reach similar conclusions for any signal
waveform which is perfectly matched to the reference. In a wider sense, this topic belongs
to the realm of matched filter theory which aims to devise schemes for the optimum detection
of signals in white noise.There are more complex lock-in systems to be discussed in
Chapter 9 which are capable of approaching this type of operation; however, as in the case of
squarewave signal and reference, the noise benefits are often marginal and usually
outweighed by benefits in other areas.

Fig. 3.11 Waveforms in a phase-sensitive detector with in-phase squarewave signal

Chapter 311
 For example, when operating a switching phase-sensitive detector with an in-phase
squarewave signal, the principal practical advantage arises because of the absence of
components at 2fR, 4fR and so on in the output of the multiplier. When using sinewave
signals, the time constant of the output signal should always be sufficiently long to suppress
these ripple components, irrespective of whether the signal is noisy or clean. This can often
lead to inconveniently long response times when measuring with low-frequency signals (say,
below 10 Hz). When squarewave signals are used in this regime it may be possible to relax
requirements on the low-pass filter and so obtain results more quickly. At the same time, the
effects of opening wide the transmission windows should be taken into consideration. As we
saw in Section 3.3 the susceptibility to interference is then greatly increased even when
wideband noise is not identified as a major problem.
Finally, let us examine the phase relationships in squarewave operation using Fig. 3.12, which
gives us the waveforms at an arbitrary phase-shift . The phase dependence could be
determined by multiplying together the Fourier series for signal and reference and extracting
the low-frequency output, but it is more convenient to deduce the form of the dependence
directly from Fig. 3.12(a). At arbitrary phase, a low-pass filter is obviously necessary to
smooth the products of multiplication. The d.c. component remaining in the final output is
then
Vo = Vs [1 /90], 0 90

with expressed in degrees.

In lock-in amplifier systems the phase shift introduced in the reference channel can be varied

in the full range 0 360, which gives rise to a characteristic piecewise-linear variation in
the d.c. output as drawn in Fig. 3.12(b). The classic response to a synchronous sinewave is
also shown for comparison, to remind us how the handling characteristics of a phase-sensitive
detector can change with different types of signal.

Fig. 3.12 (a) Waveforms in a phase-sensitive detector with a squarewave signal and
reference at arbitrary phase shift; (b) phase response for a squarewave
signal. The cos dependence is shown for a sinewave signal with the same
r.m.s. value

3.6 Phase-sensitive detector specifications

3.6.1 Introduction
At this stage we shall be concerned only with identifying some key features of the
specifications of phase-sensitive detectors. The extension to more comprehensive systems
will be left until the next chapter. It has been remarked elsewhere that the major

Chapter 312
 developments in lock-in amplifier design have been achieved for instruments operating in the
frequency range up to 1 MHz and the typical specifications given in the following sections
will be for phase-sensitive detectors of this type. In recent years it has become common
practice for manufacturers to quote some specification figures at midband or at a specified
operating frequency, say 1 kHz. The deterioration of such typical specifications towards
lower or higher operating frequencies is not always given. At best it may be possible to obtain
the required information by carefully reading the data sheet. At worst it must be assumed that
some deterioration will occur.
3.6.2 Full-scale sensitivity
Phase-sensitive detectors are general purpose in the sense that they are capable of operating
with a wide range of signal types. However, we have seen that the magnitude of the response
depends on the Fourier composition of the signal and so must differ in each case.
To enable a sensible comparison to be made between alternative designs the scaling factor of
phase-sensitive detectors is almost invariably specified for a synchronous sinewave signal. A
simple scaling factor relates the d.c. output voltage to the r.m.s. value of an in-phase
synchronous sinewave at the phase-sensitive detector input. In practical systems the d.c.
output must be limited to some maximum value. This is the full-scale output which is usually
10 V in modern systems. The full-scale sensitivity of a phase-sensitive detector is defined as
the r.m.s. value of an in-phase synchronous sinewave which gives a full-scale d.c. output, and
is expressed in volts, millivolts or even microvolts.
If the full-scale sensitivity SD is known, the d.c. response to a synchronous sinewave with
r.m.s. value Vs and relative phase-shift is
Vo = VF (Vs / SD ) cos
where VF is the full-scale output voltage.
As is usual with this sort of specification, a phase-sensitive detector with high sensitivity
has a low value of SD and is able to give a full-scale response to signals at low level.
3.6.3 Linearity and out-of-phase rejection
We have seen in Chapter 2 that an ideal synchronous detector is inherently free from non-
linear effects. Thus, in principle, a synchronous signal can be measured in the presence of
noise without incurring errors due to intermodulation and offsets due to rectified noise
components.
In making the transition to practical devices we must first recognize that there is a maximum
level of noise voltage - asynchronous voltage - that a phase-sensitive detector can withstand.
This is determined by the level of asynchronous input that gives rise to gross detection errors
due to distortion and clipping in the electronic circuits. Unfortunately, in practical devices,
the effects of non-linearity, and the resulting detection errors are likely to be discernible when
the signal of interest is accompanied by asynchronous voltages well below the clipping level.
In an attempt to reflect these limitations it used to be common practice for manufacturers to
quote a specification known as out-of-phase rejection. This measure was widely used a
number of years ago for comparing competing instruments.
Out-of-phase rejection is specified and measured with a synchronous sinewave signal
adjusted to give an output at or near full scale. A second, asynchronous, sinewave at a
frequency well removed from any transmission windows is added to the first and increased
until a pre-determined change in the output occurs. This output change is generally attributed
to non-linearity in the phase-sensitive detector transfer function although the precise nature of

Chapter 313
the non-linearity is not usually specified. The standard approach is to calculate the out-of-
phase rejection using:

For example, if we choose a fractional change of 1% and this change corresponds to an

asynchronous/synchronous voltage ratio of 1000:1 the out-of-phase rejection will be
100 000 or 100 dB. The high value of voltage ratio obtained emphasizes that to be of use in
signal recovery applications a phase-sensitive detector should be able to sustain the effect of
asynchronous voltages far in excess of a full-scale synchronous signal.
This fractional change in a near full-scale output increases with increasing asynchronous
voltage in a highly non-linear fashion. Thus, if out-of-phase rejection were to be measured for
a 0.1% or 10% change in output, entirely different values would be obtained.
Users of phase-sensitive detector-based equipment should therefore be aware that there is a
trade-off between linearity and the level of asynchronous voltage permitted at the phase-
sensitive detector. The overall behaviour is such that, for good linearity, the asynchronous
voltage should be maintained at a relatively low level in order to restrict the total voltage
swing at the phase-sensitive detector. In cases where a signal is very noisy, the user might
permit a larger voltage swing due to asynchronous components but would then suffer a
penalty in the form of reduced linearity at the point of detection. Clearly, the maximum
allowed asynchronous input to the phase-sensitive detector would be quite different in these
two extreme examples, determined in each case by the errors due to non-linearity that could
be tolerated by the user. We shall return to this topic when discussing the linearity of different
system configurations in Chapter 4.
3.6.4 Dynamic reserve
When considering the specification of out-of-phase rejection in the last section it became
evident that a small fraction of the total input range of a phase-sensitive detector is in fact
used by synchronous signals while a much larger fraction is held in reserve for noise and
interference. This idea leads to the definition of a prime phase-sensitive detector
specification; this is input dynamic reserve , given by the ratio:

Dynamic reserve thus gives a direct measure of the worst-case signal-to-noise ratio that can
be tolerated at the input to a phase-sensitive detector consistent with maintaining a full-scale
output. The term overload capability is also used in this context, but we shall be using this
description in a more specialized sense in Chapter 4.
Dynamic reserve is a useful concept but only serves as a basis for comparing different
systems when there is general agreement on defining maximum allowable asynchronous
input.
One way to do this is to identify the level of asynchronous input that causes a change of 5%
in the response to a full-scale signal. This approach has its roots in the specification of out-of-
phase rejection and is a way of reaching a compromise between two conflicting requirements:
first of all, to achieve the largest possible input voltage swing, thus maximising the signal
recovery capability of the phase-sensitive detector; secondly, to restrict the total input to a
level where errors due to non-linearity are - for most practical purposes - just discernible.

Usually known simply as "dynamic reserve"

Chapter 314
An alternative approach is simply to equate the maximum allowed level with a value
marginally less than the input clipping level. This is the input overload level quoted by
manufacturers beyond which gross measurement errors will be incurred. Circuit designers
find this definition attractive since the input clipping level is usually a well-defined circuit
parameter, unlike the 5% error limit which can only be determined by painstaking
observations at the phase-sensitive detector output.
In many cases, these two approaches give rise to very similar results for dynamic reserve. It
might be thought, therefore, that the first approach is by far the most satisfactory because it
incorporates the twin ideals of capability and linearity. However, in practice, the second
approach is almost universally adopted since it establishes a predictable relationship between
the internal gain of a phase-sensitive detector and dynamic reserve.
To see this, let us consider some specification figures for a modern phase-sensitive detector.
The input dynamic reserve would typically lie in the range 60 dB 80 dB (1000 to 10 000)
while the input overload level is of the order of several volts peak-to-peak. It follows that the
full-scale sensitivity must be correspondingly high. For example, a phase-sensitive detector
with a dynamic reserve of 60 dB might have a full-scale sensitivity of 1 mV r.m.s. (~ 3 mV
peak-to-peak). This would allow an interference voltage to rise around 3 V peak-to-peak
before the overload indicators gave warning of improper operation.
Let us suppose now that the low-frequency gain of the output low-pass filter is increased by a
factor of 10. The sensitivity, and hence the signal required for a full-scale output, is now
100 V r.m.s., while the input overload level remains at 3 V peak-to-peak.
When dynamic reserve is defined on the basis of input overload level we have no hesitation in
stating that the dynamic reserve of this more sensitive phase-sensitive detector is 10 000 or
80 dB. We thus reach the important conclusion that dynamic reserve increases in proportion
to the d.c. gain of the post-detection filter. This direct relationship would clearly be lacking
when the definition of dynamic reserve involved a detailed evaluation of measurement errors
at different levels of applied signal and noise.
It is usually assumed that the definition of dynamic reserve is valid for a sinusoidal
interference voltage and for interference from a broadband noise source. In the latter case the
phase-sensitive detector output will contain a residual noise fluctuation due to noise
transmitted by the fundamental and higher-order transmission windows. Linear operation of
the phase-sensitive detector then implies that the mean output voltage due to a synchronous
signal is unaffected by the presence of noise at the input. From the discussion given at the end
of Chapter 2, however, it is evident that, if a phase-sensitive detector was operated with a
signal and broadband noise at the dynamic reserve limit, then the main limitation to precision
measurement would result from the relatively large fluctuation remaining in the final output
(unless an extremely long time were available for measurement). The justification for
providing very high values of dynamic reserve is therefore to enable measurement to be
carried out in circumstances where the dominant interference is due to narrowband noise or
discrete frequency pick-up.
In view of this, manufacturers often make use of the alternative definition:

When a sinewave interference voltage is assumed, this definition is equivalent to that given at
the beginning of this section.

Chapter 315
Fig. 3.13 (a) Evaluating the effect of a sinewave interference; (b) dynamic reserve
characteristics of a broadband phase-sensitive detector

It should not be assumed that restriction to a sinewave interference automatically precludes

difficulty in using the specification provided by a manufacturer. Consider, for example, what
happens when the frequency of the interfering sinewave lies close to one of the phase-
sensitive detector transmission windows. When a harmonic window is excited, the phase-
sensitive detector output is perturbed by a difference-frequency beat component which may
be of sufficient peak-to-peak value to cause a severe output overload. When the interference
frequency is fixed, the output overload can be overcome by reducing the bandwidth or the
output low-pass filter. When experimental constraints determine the minimum usable
bandwidth, the output overload can only be prevented by reducing the level of the interfering
component. The result is a loss of dynamic reserve in certain critical frequency ranges. These
effects can be evaluated using the experimental set-up shown in Fig 3.13(a). Fig 3.13(b)
shows the resulting dynamic reserve characteristics of a broadband phase-sensitive detector
referenced at 10 kHz.
In the light of these observations, the standard data-sheet presentation of dynamic reserve as a
single unqualified figure appears to be rather inadequate. At the very least, the figure should
be read as maximum achievable dynamic reserve. Also, wherever possible, operation at the
specification limit should be avoided in the interests of maintaining good linearity. It can be
argued that linearity is often of secondary importance when tackling the more fundamental
problem of obtaining a measure of signal obscured by high levels of noise and interference.
When extreme levels of noise cannot be avoided it is evident that a very detailed
understanding of the actual dynamic characteristics of the phase-sensitive detector under the
proposed measurement conditions is required.
3.6.5 Output stability and minimum detectable signal
If the total signal input to a phase-sensitive detector is removed and the reference waveform is
left connected, the output should, ideally, fall to zero. The application of a millivoltmeter will,
however, reveal a residual offset voltage and long-term observation will show the effect of
drift in the output. The offset is, moreover, dependent upon the reference frequency and will
usually increase quite markedly if the reference frequency is increased beyond about 10 kHz.
This effect is known as h.f.offset and is due to stray reactive coupling of switching spikes
from the reference input through to the final output.
Provision is usually made to trim the offset voltage to zero under a particular set of operating
conditions and, if this is done, the drift component in the output represents the limiting factor

Chapter 316
 to precision measurement. The drift is usually labelled as output stability and quoted as a
fraction of the full-scale output voltage per Kelvin.
Let us now consider the problem of measuring either very small synchronous signals or small
changes in a synchronous signal. In these cases measurement difficulties will result whenever
the corresponding change in output voltage is comparable with the drift component of the
phase-sensitive detector. When operating at full-scale sensitivity SD we accordingly define the
minimum detectable signal smin as
smin = SD
where is the fractional drift/K.
Note that this expression serves as a definition of minimum detectable signal which will be
more or less useful in a given practical situation. Also, when calculating smin, only the
numerical part of the drift specification is used. Thus, by convention, the minimum detectable
signal is expressed in volts or, more usually, in microvolts.
It is worth noting that the d.c. response to a signal at the minimum detectable level can be
separated from the spurious outputs due to offset and drift by introducing a phase reversal of
180 in the reference channel. The d.c. output due to the signal will then reverse its polarity
while the polarity and magnitude of the offset (being phase-insensitive) will be unchanged.
This method of overcoming limitations due to offset has been exploited in systems operating
on the synchronous heterodyne principle, described in Section 8.8.
3.6.6 Dynamic-reserve/output-stability trade-off
It was shown in Section 3.6.4 that both the sensitivity and the dynamic reserve of a phase-
sensitive detector can be increased by increasing the d.c. gain of the post-detection filter. In
practice, this additional gain can always be obtained by incorporating an output d.c. amplifier.
Unfortunately, improved dynamic reserve can only be achieved at the expense of precision
since all errors due to offsets and drift in the phase-sensitive detector are enhanced by the gain
of the output amplifier. If the sensitivity of a phase-sensitive detector was controlled solely by
switching output gain we would find a very uncomfortable situation in which high sensitivity
was linked with high reserve and low sensitivity was required for good output stability.
For maximum flexibility, a phase-sensitive detector should have switched output gain and be
supported by a variable-gain amplifier in the signal path. The balance between dynamic
reserve and output stability can then be adjusted at a given level of sensitivity. This aspect of
system performance will be discussed in Chapter 4.
3.6.7 Dynamic range
By dynamic range of a phase-sensitive detector we usually mean the input dynamic range.
This is defined at one extreme by the maximum allowed input voltage swing and at the other
by the minimum detectable signal. If we denote the maximum allowed voltage swing by v,
the input dynamic range is simply:
DI = v / smin
It was shown in Sections 3.6.3 and 3.6.4 that the maximum allowed input voltage to the
phase-sensitive detector can have different values depending on the errors due to non-linearity
that can be tolerated. In general, the linearity of the phase-sensitive detector is degraded as
larger and larger asynchronous voltage swings are permitted, so we see that there is an
important trade-off to be made between input dynamic range and linearity such that one can
be improved only at the expense of the other. In order to specify input dynamic range, it is
usual to equate the maximum allowed input voltage with the input overload level, that is to

Chapter 317
use the same voltage swing that appears in the specification of dynamic reserve. When this is
done, both v and smin can be expressed in terms of the full-scale sensitivity SD of the phase-
sensitive detector. Using the results of Sections 3.6.3 and 3.6.5 we obtain:

and

where ri is the dynamic reserve and is the fractional output stability. We can therefore arrive
at the following expression for input dynamic range :

The result is quite properly independent of the phase-sensitive detector sensitivity and any
post-detection gain stages: any attempt to increase dynamic reserve by increasing the output
gain is accompanied by a deterioration in output stability. We thus see that, to achieve wide
dynamic range, a phase-sensitive detector must be capable of giving good output stability at a
high level of dynamic reserve and this has proved to be one of the main objectives in phase-
sensitive detector design. As a rough guide, in modern systems, a combination of 2 000
dynamic reserve and 100 p.p.m./K output stability represents good practice corresponding to
a dynamic range of:
DI = 1000 / 104 = 107 (140 dB)
Since the concept of dynamic range incorporates the two key specifications of phase-sensitive
detectors it provides the best figure of merit to be used when comparing competing systems.
It should now be clear that dynamic reserve alone is no guarantee of quality and should
always be viewed in the light of the stability specification.
3.6.8 Summary of specifications
The specifications covered so far can be conveniently summarized in diagram form as shown
in Fig 3.14. The diagram also serves to define two additional quantities, namely output
dynamic reserve and the output dynamic range.

Fig. 3.14 Summary of phase-sensitive detector specifications

Again, we use only the numerical part of , giving DI as a voltage ratio

Chapter 318
To be of any practical use a phase-sensitive detector must be able to give a full-scale response
to a synchronous signal and yet be able to accommodate a residual output fluctuation due to
noise transmitted by the low-pass filter. The residual noise will carry the output beyond full
scale and there should be sufficient margin in the output circuit to allow this to happen
without an overload indicator permanently flashing; hence the provision of output dynamic
reserve which enables the output to exceed full scale by around 20% - 30% without suffering
distortion due to clipping in the output amplifier.
The output dynamic range is given as the ratio of the full-scale output voltage to the output
drift component. If the drift is specified as a fraction of full scale- as we have assumed
throughout - then the output dynamic range is simply 1/. In high-stability operation, phase-
sensitive detectors are capable of operating with a relative drift of less than 10 p.p.m./K. This
corresponds to an output dynamic range of 100 000 (100 dB) and would enable a change of
100 V to be observed in a full-scale output voltage of 10 V. Obviously, such a mode of
operation presupposes a signal which is correspondingly free from noise.
Using the definition of output dynamic range given above, together with the results of Section
3.6.7, we see how dynamic reserve provides a link between input and output dynamic range:
Input dynamic range = Dynamic reserve Output dynamic range
In the following chapter it is shown how the trade-offs between dynamic range and linearity,
and between dynamic reserve and output stability, can be improved by incorporating filters in
the signal channel of an otherwise broadband lock-in system.
3.7 References
1 CARTER, S.F., and FAULKNER, E.A. (1977): "New phase-sensitive rectifier circuit",
Electron. Lett., 3, pp. 339-340.
2 BLAIR, D.P., and SYDENHAM, P.H. (1975): "Phase-sensitive detection as a means to
recover signals buried in noise", J. Phys. E: Sci. Instrum., 8 pp. 621-627.
3 LETZTER, S.G. (1974): "Explore the lock-in amplifier", Electron. Des., 21, pp. 104-108
(also published as Technical Note 115: Princeton Applied Research Corp. Princeton, NJ).
4 "Introduction to lock-in amplifiers". Technical Note 102, Brookdeal Electronics,
Bracknell, England.

Chapter 319
arrangement in a commercial system gives a near-optimum noise performance in
all switched positions and provision is made for signal-channel filtering, both
internal and from external plug-in modules.
We shall assume that the phase-sensitive detector has fixed sensitivity but that the
output gain can be controlled by a switched d.c. amplifier. The amplifier is
labelled as an expand amplifier since it serves to expand the output from the
phase-sensitive detector.
The essentials of the reference channel are also sketched in. These comprise a
precision trigger circuit to respond to the positive zero crossings of an externally
applied reference waveform, a broadband precision phase-shifter, and a
squarewave generator to supply the final drive to the phase-sensitive detector.
The phase-shifter takes the form of a control loop which sets up a phase-shift in
response to a control voltage generated at the front-panel phase dial. Control over
a full 360 range is usually achieved using pushbutton selection of phase
quadrants in conjunction with a 0100 continuous phase control. We shall be
able to identify several critical specification features for the reference channel
from a general review of system behaviour.
Note that the system shown in Fig. 4.1 incorporates two other features that are
commonly found in commercial units. These are a reference-channel frequency
doubler, which enables the detection system to be synchronized at the second
harmonic of the applied reference frequency, and an internal oscillator. The
latter can be used as an excitation source for an experiment and is sometimes
controllable from the front panel of the lock-in amplifier. When switched into
operation, the internal oscillator makes a direct connection to the reference-
channel phase-shifting circuits and so overcomes the need for an external cable
connection.

4.2 Calibration: full-scale sensitivity

As in the case of phase-sensitive detectors, the full-scale sensitivity of a lock-in
amplifier gives the r.m.s. value of an in-phase synchronous signal required to
give a full-scale output. An essential difference between modular lock-in
amplifiers and those supplied as integrated units is that the latter are supplied
with a single sensitivity switch which enables the full-scale sensitivity to be set
up immediately. In this case, the individual gains of the signal channel amplifiers
and the phase-sensitive detector need not be known, whereas in a modular system
these can normally be read from the front panel settings of the individual units.
In either event, the full-scale sensitivity measured from the signal channel input
socket is
Full-scale sensitivity, SF = SD/(GaGe)
where SD denotes the full-scale sensitivity of the phase-sensitive detector,
calibrated on a sinewave, and Ga and Ge are the gains of the signal channel and
expand amplifiers.
In lock-in amplifiers the usual convention is to control only the signal channel
gain from the sensitivity switch which is calibrated at a given expand setting
GCAL. If we denote the indicated sensitivity on the front panel by SI, then we have
Full-scale sensitivity, SF = SI Ge/GCAL
When a synchronous sinewave, with r.m.s. value Vs, is applied to a lock-in
amplifier having a full scale d.c. output of VF, the d.c. response of the system
will be

Chapter 42
The dynamic reserve is measured with a broadband signal channel so that there is
no selective attenuation of the input noise components. The most convenient
starting point is to assume that the lock-in amplifier is operating at its dynamic
reserve limit. Referring to Fig. 4.2, the peak-to-peak voltage *, Ga(vs + vN), at the
phase-sensitive detector input is therefore equal to the maximum allowed level,
v. The ratio vN/vs measured at the signal channel input is thus equal to the
system dynamic reserve. If the system dynamic reserve was now measured using
an interfering sinewave as described in Section 3.6.4, we would find a
characteristic identical to that of the phase-sensitive detector, including the
influence of the harmonic transmission windows.
It is this feature which distinguishes broadband lock-in systems, where the
overall response to asynchronous inputs reflects the properties of the phase-
sensitive detector.
Let us now return to a discussion which was first opened in Section 3.6.6. Any
increase in input noise beyond vN peak-to-peak will overload the phase-
sensitive detector but can be offset by reducing the signal-channel gain. The
reduced response to the synchronous signal can then be expanded using the d.c.
output amplifier in order to restore the overall sensitivity to its original value. For
a fixed level of synchronous input to the instrument we find that the system can
tolerate progressively higher input noise levels as the a.c. gain Ga is reduced and
the expand factor Ge is increased to maintain constant overall gain. In this way,
the dynamic reserve measured at the signal channel input can be controlled by
changing the internal gain distribution.
In the interests of linearity it is essential to limit the voltage swing at the input to
the amplifier to a well defined maximum value. This is the input overload level
which is usually of the order of a few volts peak-to-peak. If a full-scale signal of
100 mV r.m.s. is presented to the lock-in amplifier, then the input noise-to-signal
ratio must be less than about 10:1 to avoid overloading the amplifier input. We
thus see that the full dynamic reserve of the system can only be exploited for
relatively small signal inputs where the lock-in amplifier is operated at
correspondingly high sensitivity. In systems offering dynamic reserves of 1000
(60 dB) and greater this corresponds to operating at a minimum sensitivity of
around 1 mV r.m.s.
4.3.3 System overload capability
The dynamic reserve of a signal recovery system is always measured with a
broadband signal channel and the specification is normally valid over many
decades of frequency subject to the effect of the harmonic transmission
windows discussed in Section 3.6.4.
In a typical measurement the signal will be perturbed by noise which is non-white
in character and a substantial fraction of the input noise voltage might be
attributable to discrete interference components appearing in a well-defined
frequency range. Under these circumstances, the total peak-to-peak value of the
disturbance reaching the phase-sensitive detector can often be reduced
significantly by using signal conditioning filters. By eliminating noise
components before detection the dynamic reserve of the system can appear to be
much greater than when operating with a broadband signal channel. We now

*
When operated in the dynamic reserve limit the noise input to the phase-sensitive detector will
usually exceed a full-scale signal by at least a factor of 10, giving vN>>vs.

Chapter 44
4.3.4 Dynamic reserve and output stability trade-off
We have seen that the effect of incorporating variable gain amplifiers before and
after detection is to produce a lock-in amplifier with controllable dynamic reserve
at a given sensitivity. In order to realize the full practical potential of such a
system we must take into account the effect of gain selection on output stability
and see how dynamic reserve and stability can be jointly controlled to the best
effect.
We can formulate a basic approach to setting up a detection system as follows:
first of all for very noisy signals. Here, the signal channel gain Ga should be
held at a sufficiently low value to avoid premature overload at the input to the
phase-sensitive detector. The desired sensitivity is then obtained by increasing the
gain of the expand amplifier. The output stability is degraded, but we argue that
our main problem in this case is to detect the signal at all. A loss of output
stability is not likely to be significant or even discernible when the final output
contains a relatively large component of residual noise transmitted by the low-
pass filter.
The situation is quite different for relatively clean signals. These can be
subjected to a large amplification factor without exceeding the maximum
allowable voltage swing at the phase-sensitive detector. The minimum value of
expand gain can then be used to ensure that the signal is measured with the best
possible precision.
These two modes of operation can be summarized for a lock-in amplifier offering
two levels of output expansion by means of the example given in Table 4.1.

Table 4.1

Signal Sensitivity Expand gain Output Dynamic

selection stability, reserve
p.p.m./K
clean 100 V 1 < 10 100
noisy 1 mV 10 < 100 1000
Overall sensitivity = 100 V f.s.

In an attempt to combine flexibility with ease of use the principal lock-in

amplifier manufacturers have incorporated pushbutton selection of either high
stability or high reserve operation. One way that this can be achieved is to
retain the expand amplifier, leaving this to be controlled by the user, and to
arrange for the distribution of gain between the signal channel and the phase-
sensitive detector to be switched automatically as shown in Fig. 4.4.

Fig. 4.4 Switch selection of high stability mode and high reserve mode in
a lock-in amplifier

Chapter 46
In high stability mode the phase-sensitive detector sensitivity is reduced at the
expense of signal channel gain. In high reserve mode the signal channel gain is
reduced while that of the phase-sensitive detector is automatically increased so as
to retain the same overall sensitivity. With two levels of output expansion
available the system can be adapted to meet a wide range of measurement
requirements as demonstrated by the example in table 4.2.

Table 4.2

Mode Sensitivity Expand Output Dynamic

selection selection gain stability reserve
p.p.m./K
High 100 V 1 < 10 100
Stability 1 mV 10 < 100 1000
High 100 V 1 < 100 1000
Reserve 1 mV 10 < 1000 10 000
Overall sensitivity = 100 V f.s.

4.3.5 Overload capability and output stability trade-off

Section 4.3.4 dealt only with dynamic reserve. As a result the trade-off between
noise rejection and output stability has been demonstrated only for a broad-band
signal channel which introduces no selective attenuation of noise components.
There is, however, an aspect of noise rejection which applies to the case where a
normally clean signal is perturbed by large-scale interference. Hum pick up on
a low-frequency a.c. bridge provides a good example; the difficulty is to measure
the signal to a high degree of precision.
In the absence of signal conditioning filters, the only approach to this particular
problem would be to select a suitably high value of dynamic reserve. This would
avoid the danger of overload on the interference component, but would
automatically result in a relatively poor output stability and a consequent loss of
precision.
The importance of signal conditioning becomes evident when we realize that for
every 20 dB of suppression brought about by using, say, a notch filter, the system
reserve can be switched down by 20 dB and the output stability improved in
direct proportion. Ultimately it might be possible to achieve the desired precision
with only a modest amount of suppression obtained through signal conditioning,
provided that the relative magnitudes of signal and interference are brought
within the capability of the phase-sensitive detector at the desired level of output
stability.
In the light of earlier discussions it is evident that obtaining improved overload
capability at a given level of output stability is equivalent to increasing the input
dynamic range of a detection system. Furthermore, in circumstances where it is
desirable to obtain high overload capability in conjunction with the best possible
output stability, it is clear that the signal-channel gain should be switched to the
maximum possible value consistent with avoiding overload at the phase-sensitive
detector. Unfortunately, the linearity of the detection system will be degraded
under these conditions, so it is important that we examine the relationship
between dynamic range and linearity in practical lock-in systems.
4.3.6 Dynamic range and linearity trade-off
Suppose we wish to measure the amplitude of a 100 V r.m.s. signal in the
presence of an interference component at a level just below 10 mV r.m.s. A lock-

Chapter 47
in amplifier with the characteristics summarized in Table 4.2 would enable this
signal to be measured at full-scale in a high-stability mode, using a dynamic
reserve of 100. The output stability would then have its best value, < 10 p.p.m.,
however, the linearity of the system would be degraded on account of the large
asynchronous voltage swing presented to the phase-sensitive detector.
If good linearity is required, an obvious step is to reduce the pre-detection, a.c.
gain by, say, a factor of 10. The total input voltage to the phase-sensitive detector
is then reduced, linearity is improved and the sensitivity can be restored by using
a larger expand gain.
The improvement in linearity is observed as an increase in out-of-phase rejection
(Section 3.6.3) since, at an input asynchronous/synchronous voltage ratio of 100,
the error in the full-scale response will be reduced. However, in this new
situation, with the expand gain increased to 10, the output stability will be
degraded while the improved dynamic reserve cannot be exploited without
incurring a larger voltage swing at the phase-sensitive detector. A possible
increase in dynamic reserve must therefore be sacrificed in favour of an
improvement in linearity. Since the output stability is worse, the net result is a
loss of input dynamic range.
The situation can be improved when interference components are rejected in
advance of detection by using signal-channel filters, but a trade-off exists
nevertheless. We have seen that, when wide dynamic range is the main
consideration, the signal-channel gain following a stage of filtering should be
increased to its maximum allowed value. When good linearity is required, the
signal-channel gain must be maintained at a relatively low value so as to restrict
the input swing to the phase-sensitive detector. The overall sensitivity would then
be restored by an increase in expand gain. The linearity, measured in terms of
out-of-phase rejection at the input to the signal channel, is improved while the
system suffers a degradation in output stability. The dynamic range under these
conditions could be greater than that obtainable without filters but will inevitably
be less than the maximum achievable value.

4.4 Using a tuned filter in the signal channel of a

conventional lock-in amplifier
4.4.1 Influence on overload capability
Early lock-in detection systems were traditionally operated with signal channels
tightly tuned to the signal and reference frequency in order to overcome the
limited dynamic range of the then available phase-sensitive detectors. Although
modern systems have much better performance, there are instances where they
can benefit from the rejection properties of a tuned filter, and most commercial
systems are supported by a range of options which includes a tuned filter.
The type of filter most often used has a symmetrical band-pass characteristic of
the type described in Appendix 4. Commercial units are, moreover, almost
invariably active filters which use operational amplifiers with capacitive and
resistive feedback elements to simulate the circuit behaviour of inductors and so
avoid the use of large inductor values at the lower frequencies required. An active
filter usually has a gain greater than unity at frequencies close to the centre
frequency, although this is sometimes compensated by including an output
attenuator to give a net gain of unity.
When dealing with active filters we must be prepared to identify overload
conditions in all the internal amplifiers, and it is usual to specify the maximum
allowable input voltage swing at in-band and out-of-band frequencies. Fig. 4.5

Chapter 48
shows how this can be done for a tuned filter considered in isolation using the
filter frequency-response characteristic. At mid-band, interference components
are subject to the full gain, Go, of the filter, and their magnitude must be
restricted to avoid driving the filter output into saturation. The increased input
capability at frequencies removed from the centre frequency reflects the greater
attenuation provided by the filter. The transition to a flat characteristic at larger
frequency offsets marks the point where the input cannot be increased further
without exceeding the maximum allowable input level to the filter. If a passive
filter had been used, the curve would have continued along the dashed lines. In
most practical active filter designs it turns out that the absolute maximum voltage
swings allowed at input and output are the same, being almost equal to the power
supply voltages applied to the filter.

Fig. 4.5 Maximum allowable voltage swing at the input to an active tuned
filter with midband gain Go.
vi and vo denote the peak-to-peak overload levels at the input
and output of the filter.
The dashed characteristic would be obtained with a passive filter
followed by a voltage amplifier, giving a net gain of Go at
midband.

In normal operation the filter is tuned precisely to the reference frequency and
inserted in the signal path. It now becomes essential to examine the relative
merits of providing amplification before and after the filter in order to determine
its optimum location within the signal channel. To do this, we return to the
concept of dynamic reserve which can be generalized to any combination of
filters and amplifiers. Following the arguments presented in Section 4.3.2 for the
special case of phase-sensitive detectors, we find that the ability to handle large-
scale interference components is enhanced by increasing post-filter gain at the
expense of pre-filter gain. In other words: the allowable input voltage for out-of-
band components reduces in proportion to the amount of gain which is introduced
in front of the filter.
This conclusion is particularly relevant when the user is able to modify the
configuration of the signal channel by interchanging amplifier and filter modules.
The first step in obtaining high overload capability is to obtain the maximum
possible voltage swing at the input to the signal channel. If a degree of flexibility
is available this can be achieved by introducing the filter at the earliest possible
stage. Unfortunately, the resulting improvement in overload capability is usually
obtained at a cost. In this case, the penalty is an increase in system noise. Active
tuned filters are generally more noisy than the high-quality amplifiers used in a

Chapter 49
lock-in signal channel. The result is that the filter makes an increasing
contribution to system noise as the gain distribution is altered in favour of post-
filter amplification.
The final location of the filter always reflects a compromise between system
noise and overload capability. In commercial lock-in amplifiers the filter is
usually introduced immediately after the low-noise input amplifier, the objective
being to maintain acceptable noise performance in the high-sensitivity positions.
In a modular system dedicated to measuring signals in a very noisy environment,
the increase in system noise resulting from a filter directly at the signal input may
well be of no significance. The incorporation of a filter in this position must then
be considered as a valid means of increasing noise-handling capacity.
Whatever combination of amplifiers and filters is used it is possible to arrive at a
maximum allowable input voltage swing for out-of-band frequency components.
If the total available input swing is to be utilized, then the filter must supply
enough attenuation to bridge the gap between the desired overload capability and
the best achievable dynamic reserve at the phase-sensitive detector.
The Q-factor of a tuned filter with a centre frequency fR is defined by
Q = fR/fB
Where fB is the 3 dB bandwidth of the filter. For interference components offset
by several bandwidths, the frequency response of the filter relative to its mid-
band value is (Appendix 4):
H ( ji ) fi / fR
= 2
; fi fR >> fR / Q
H ( jR ) Q1 fi 2 / fR

where fi is the frequency of the interference component.

Suppose we have a lock-in system which can withstand a total voltage of 3 V
peak-to-peak at its signal channel input. The phase-sensitive detector has a
dynamic reserve of 1000 (60 dB) and a signal of 10 V r.m.s. is to be measured
in the presence of 3 V peak-to-peak interference. The signal frequency is 5 kHz
and the interference frequency is 1 kHz.
The interference voltage is at the maximum allowable value of 3 V peak-to-peak
( 1 V r.m.s.). If the signal is to be measured at full-scale we require an overload
capability of 1/(10.106) or 105 while the dynamic reserve is only 1000; the
filter must therefore introduce an attenuation factor of 100. The required Q-factor
can now be calculated from:
fi / fR
1 /100
Q1 fi 2 / fR2

Putting fR = 5 kHz, fi = 1 kHz, we obtain a value of about 20 for the filter

Q-factor. Let us now look at the stability of a system incorporating a filter with
this order of selectivity.
A Q-factor of 20 at a centre frequency of 5 kHz implies an effective operational
bandwidth of 250 Hz. This means that the signal/reference frequency must be
maintained to within 21/2% to stay within the filter bandwidth. If this is not
achieved, the magnitude of the signal reaching the phase-sensitive detector will
be in error by more than 3 dB or 30%.
The next point to be considered is the phase shift introduced by the filter. The
system would normally be set up initially with the filter tuned precisely to the

Chapter 410
reference frequency. The phase-shift introduced by the filter would then, ideally,
be zero, but we can suppose that any small phase error could be compensated by
trimming the phase of the reference channel. The phase-shift of the filter close to
the tuned condition is given in Appendix 4. We find that the change in phase
which occurs when the signal drifts by a relative amount f/fR is given by:
tan 2Q f/fR
Using a Q-factor of 20, we calculate an incremental phase-shift of about 22 for a
one per cent change in signal frequency. The stability of the signal source and of
the filter itself must therefore be of a very high order to avoid excessive
amplitude and phase modulation due to drift. It is unlikely, for example, that
tuned filters with this degree of selectivity could ever be used satisfactorily with
many mechanically derived excitations such as optical choppers.
In view of these constraints, most practical lock-in systems incorporating tuned
filters are designed for a maximum Q-factor of around 5. Provided the
signal/reference source is of reasonable stability, the overload capability of a
broadband detection system can then be substantially improved without incurring
excessive measurement errors. This is especially true when a tuned filter is used
in the signal channel of a two-phase lock-in system. It is shown in the next
chapter that two-phase systems can be used in a phase-independent mode and so
overcome the excess phase-shift introduced by a tuned filter. In addition, the use
of a tuned filter ensures that harmonics of the signal are greatly suppressed in
advance of phase-sensitive detection. The result is a measurement system where
the only significant response to either signals or noise is in the vicinity of the
reference frequency. This aspect will be discussed further in the next section.
In conclusion, it should be stated that the general considerations regarding the
location of a signal-channel tuned filter can be extended to any of the basic high-
pass, low-pass and notch filters which are commonly used with lock-in systems.
In any of these cases an overload characteristic can be drawn using the frequency
response function of the filter or any desired combination of filters. In all cases
the amplitude and phase responses of the filter must be considered as possible
sources of error when the frequency of the signal/reference source is subject to
drift. Ideally, the signal frequency should be as far as possible from the cut-off
frequencies of any filters which are introduced for signal conditioning and, as we
have seen, the tuned filter falls short of this ideal. In many cases, a far better
approach to suppressing discrete interference components is to use a sharply
tuned notch filter. When a dominant interference component is well removed
from the signal frequency the inclusion of a notch filter leaves the signal
substantially undisturbed. It should be remembered that many sources of
interference such as line pick-up and breakthrough from radio transmitters are
extremely stable in terms of their characteristic frequencies. In these
circumstances, a tightly tuned notch filter can be accurately set and maintained
for long periods without further adjustment.
4.4.2 Suppression of-harmonic responses
Subject to the limitations on system flexibility, a tuned filter can also be effective
in suppressing the effect of harmonically related components of the signal which
would normally be applied directly to the phase-sensitive detector.
Again, we shall suppose that the filter has a symmetrical band-pass response of
the type used in the last section.
If we imagine that a filter of this type is tuned to a signal and reference frequency
at angular frequency R then the filter response at the Kth harmonic of R is

Chapter 411
 jK /Q
H(jKR) =
(1 K 2 ) + jK /Q
For Q 5 we can approximate the gain magnitude by
K
| H(jKR)|
(K 1)Q
2

The relative sensitivity of a switching phase-sensitive detector at the K th

harmonic of the reference frequency is I/K; hence the relative sensitivity when a
band-pass filter is included is
1
| H(jKR)| /K=
(K 1)Q
2

Thus, the relative sensitivity at the third harmonic is reduced to 1/40 (32dB) for
a Q-factor of 5.
Note that, in order to approach the 3rd harmonic sensitivity of the order of 60 dB
offered by modern fundamental-only responding instruments, a Q-factor of 100
would be required, a value which would render most systems quite unusable. This
assumes, of course, that a standard second-order band-pass filter is being used.
If, as an alternative, we consider the low-pass tuned filter defined in Appendix 4
we find that the roll-off beyond the centre frequency is now 12 dB/octave rather
than 6 dB/octave as in the band-pass case. Repeating the calculations given
above, using the appropriate frequency response function, we now obtain a
relative sensitivity of approximately 1/(K3Q) at the K th harmonic. In this case, a
Q-factor of 40 or so would achieve the target figure of 60 dB at the 3rd
harmonic, while a Q-factor of 5 would give a relative sensitivity of less than 40
dB.
This aspect of tuned-filter operation is discussed further in Chapter 8 in relation
to the performance of heterodyne lock-in amplifiers. It is shown there that, for a
given level of attenuation, alignment problems can be greatly reduced by using
two filters in cascade, each of which has relatively low Q-factor. There is no
reason in principle why this approach should not be used to improve the
performance of the 'conventional' lock-in systems described so far, however,
implementation is likely to be easiest when operation is confined to a small range
of signal and reference frequencies.

4.5 Reference channel specifications

4.5.1 Introduction
Up to this point we have concentrated almost exclusively on the dynamic
performance of lock-in amplifiers. In practice, this performance will never be
realized unless the reference channel is capable of giving adequate support to the
phase-sensitive detector. It turns out that the lock-in systems with the best all-
round performance are those with genuine broadband reference channels giving
precisely calibrated phase-shifts allied to good stability. In the following sections
we shall be looking at some specifications which are commonly used in relation
to reference channels with these general characteristics.
A typical broadband reference channel operates as a closed-loop control system
and generates a phase shift in response to a control voltage supplied from a front-
panel phase control. By broadband we mean a system able to operate with
reference frequencies covering a range of several decades without adjustment. A
range of 105:1 to 106:1 is quite common in modern lock-in amplifiers, say from
less than 1 Hz to 100 kHz.

Chapter 412
At midband frequencies, good phase accuracy can be obtained with relatively
short response time in the control loop, while the same control loop might not be
so effective at lower reference frequencies. In view of this, most lock-in
amplifiers incorporate an automatic changeover system which selects control
circuitry with longer response times for reference frequencies below about 50 Hz.
If operation is required at very low frequencies, for example down to 0.1 Hz, the
user can often increase the response time of the loop still further by switch-
selecting a slow or low-frequency reference mode. The result is a considerable
improvement in low-frequency phase accuracy obtained at the expense of a still
more sluggish response. The provision of a switch together with an automatic
changeover point ensures that the system can be made to approach its optimum
performance in any frequency range of interest, but it must be expected that
offset errors and noise in the control loop will always conspire to give a phase
shift differing from that indicated by the phase dial. In practice, such errors might
be insignificant or gross as demonstrated by the following extreme modes of
operation.
The first concerns signal recovery work. Here the null-shift procedure defined in
Section 3.5 can be applied to a variety of signal types and overcomes the need for
a continuous phase adjustment with more than a nominal calibration. In this case
a continuous phase adjustment in conjunction with a calibrated increment of 90
is sufficient to reach an optimum detection condition. The phase-sensitive
detector is, moreover, relatively insensitive to small phase changes when adjusted
for maximum output. The overall system is thus tolerant of noise in the reference
control circuits and an error of a few degrees in the 90 phase increment would
not seriously affect the response to a noisy signal.
The second category of measurements includes all those where the phase-shift of
a signal is to be determined relative to the applied reference or where the lock-in
amplifier is required to respond to small phase increments. The reference phase-
shift should now be defined and calibrated in accordance with some agreed
convention and the user expects to have specification limits on phase accuracy
and on stability if small phase changes are to be resolved.
It was shown in Section 2.5.5 that the reference-channel phase-shift can be
defined unambiguously for an applied reference voltage of any waveform. For
the purpose of this section, however, we shall assume that both the signal and the
external reference are of sinewave form. Besides removing any doubt which may
remain about the validity of phase specification for non-sinusoidal signals, this
approach also avoids problems arising from waveform distortion when non-
sinusoidal signals are transmitted by a signal channel of finite bandwidth.
We shall begin by identifying the numerous components of a typical lock-in
amplifier phase specification and then consider briefly the performance of a
broad-band reference channel under swept-frequency conditions. The next step is
to investigate how the phase accuracy of a lock-in system can be checked in
practice. It turns out that the procedures involved are rigorous in the extreme and
serve as useful guidelines to the more general problem of precision phase
measurement with signals and systems. For this reason the principal sources of
error in phase measurement are listed in a self-contained section, together with
ways to minimize their effect. This last section enlarges on some of the more
general properties of phase-sensitive detectors and lock-in amplifiers introduced
so far, and shows how some key specifications can be brought to bear on a
specific type of measurement problem.

Chapter 413
4.5.2 Phase accuracy: points of specification
We suppose that a sinewave signal and an in-phase sinewave reference are
applied to a lock-in amplifier. In any practical system the relative phase of the
signal and reference switching waveform measured at the phase-sensitive
detector will have several components which can be written in the form
= R + R + s + N(t)
Here R represents the phase-shift dialled on the front panel of the lock-in
amplifier, R is a static phase error associated with the reference control circuit
and s represents a residual phase-shift in the signal channel which we assume
is set to its maximum bandwidth. The quantity N(t) represents phase noise
which is attributable to noise in the reference channel control circuit and has
components distributed over a wide frequency range.
In principle, these individual sources of error could be evaluated separately, but
this more likely to be done by the manufacturer who has access to the internal
workings of the lock-in amplifier. Otherwise, it is difficult to devise a
measurement which would isolate the two contributions R and s.
Fortunately, in any application where phase is an important factor, it is usually
sufficient to have a measure of the total phase error given as the sum of a static
error R + s and a fluctuation component N(t). We shall see how this
information can be inferred from measurements carried out at the input and
output terminals of the lock-in amplifier.
When comparing the phase specifications of competing equipments it is clearly
necessary to check manufacturers data sheets very carefully. In some cases the
static phase errors might be specified separately; in others, the figures quoted
could be for composite phase errors. In either case the specifications will be
frequency-dependent and show a deterioration towards the extremes of the
recommended range of signal and reference frequencies. Also, it is usual to
specify only the maximum anticipated value of static phase error at any spot
frequency. If more detailed information is required, for example the phase error
corresponding to a particular phase setting, R, it is up to the user to devise his
own measurement procedures following the guidelines given in later sections.

Fig.4.6 Static phase error of a typical broadband lock-in amplifier

Fig. 4.6 shows the composite, static, phase error of a typical broadband lock-in
amplifier measured over the entire range of operating frequencies. This form of
presentation, allied to a measure of the phase noise, is usually sufficient for all
but the most demanding applications, but the figures given are strictly valid only

Chapter 414
at a quoted value of laboratory temperature. A full specification will also give a
measure of the anticipated phase drift as a function of temperature. It turns out
that the phase drift is usually far more important than the phase noise in
measurements involving the detection of small phase increments in a signal.

4.5.3 Phase noise (phase jitter) and phase drift

Phase noise (or phase jitter) is specified in such a way that its effect on precision
phase measurements can be easily predicted.
We suppose that the lock-in amplifier is supplied with a stable sinewave signal
and reference at the reference frequency of interest. The reference phase-shifter is
then adjusted to null the output from the phase-sensitive detector. The effect of
reference channel phase noise is to give a residual fluctuation measured from the
null point. Using the notation of the last section the net phase shift at the null
point corresponds to a value
= N(t) /2
measured at the phase-sensitive detector.
When the total error due to noise is much less than one radian the phase-sensitive
output of the lock-in amplifier approximates to (Section 4.1.2)
vN(t) = VF(Vs/SF) NF(t)
The variation NF(t) represents the low-frequency components of N(t) which are
transmitted by the output low-pass filter. Following the approach used in Section
2.4, these are given by the convolution
NF(t) = N(t) hL (t)
where hL(t) is the impulse response of the low-pass filter.
The phase noise of the lock-in amplifier is specified in terms of this filtered
fluctuation and measured by measuring the r.m.s. value of the output voltage at a
stated value of filter time constant. The measurement is carried out on a time
scale which precludes the effect of temperature variations on the null position.
The influence of noise in the signal channel is minimized by using a clean
sinewave signal at a low level of system gain. The phase noise is then calculated
from
[NF(t)]r.m.s.= (SF/VFVS) [vN(t)]r.m.s. 180/
Here, the factor 180/ gives a conversion from radians to degrees. This
conversion brings the specification to a standard form, a typical specification
being:
phase noise: 0.01 r.m.s. at a reference frequency of 1 kHz,
0.1 s time constant at 12 dB/octave.
In general, the r.m.s. phase noise will increase for reference frequencies at the
extremes of the lock-in amplifier frequency range but its influence on the final
output can always be reduced by switching to longer time constants. The effect of
phase noise is to introduce noise even when, for practical purposes, the output
from the signal channel is noise free. In any event, its effect, like that of phase
drift, is both phase-sensitive and proportional to signal amplitude and maximized
when signal and reference are brought into quadrature at the phase-sensitive
detector.

Chapter 415
Phase drift is recognized as a temperature-dependent deviation from the
quadrature null point, observed under conditions where the instability of the
phase-sensitive detector has negligible effect.
If any doubt persists, then the output contribution due to phase drift will be
recognized by its variation with sensitivity setting while phase-sensitive detector
drift can be observed under zero signal conditions, being dependent only on the
output or expand gain.
The phase drift of a good-quality lock-in amplifier is of the order 0.03/K. It is
shown in Section 4.7.5 that the phase drift will usually make a far larger
contribution to measurement errors than phase-sensitive detector instability when
a lock-in amplifier is set up for precision phase measurement.
4.5.4 Reference channel slew rate
If a signal and reference are applied to a lock-in amplifier and the frequency is
changed, the inertia of the reference channel control system will result in an
instantaneous phase error known as phase slip. When the frequency stops
changing the phase slip gives way to the pre-existing phase condition. In most
commercial systems the phase slip and the rate of change of reference
frequency are related by:
dfR/dt = AfR
where fR is the instantaneous reference frequency and A is a constant of the
reference channel.
The rate of change of reference frequency corresponding to a phase slip of 5 is a
frequency-dependent quantity known as reference channel slew rate. Slower rates
of change than the slew rate will give less phase slip. Faster rates of change give
increased phase slip up to the point where phase control, and hence reference
channel lock, is lost completely.

Fig. 4.7 Slew rate specification for a typical broadband lock-in amplifier

The phase slip of 5 represents the maximum phase error which could be tolerated
if swept-frequency meaurements are to be carried out to a reasonable level of
precision. In most applications the slew rate is considered to be the maximum
usable rate of change of reference frequency.
When slew rate is specified by means of a graph, the effect of an automatic
changeover point in the reference channel is to divide the graph into two distinct
regions as shown in Fig. 4.7. The system is then characterized by two values of
the constant A, corresponding to operation with reference frequencies above and
below the changeover point. In the light of the discussion given earlier we expect

Chapter 416
the slew rate to be much lower below the changeover point where the response
time of the control system is larger. In Fig. 4.7, the constants differ by a factor of
100, which is typical of' many commercial systems.
The logarithmic dependence of slew rate on reference frequency means that slew
rate can be translated into minimum sweep time per decade. Broadband
reference channels are usually characterized by slew rates in the range
15 seconds/decade in the frequency region above the changeover point.

4.6 Measurement of phase accuracy

4.6.1 Introduction
In measuring phase accuracy it may be sufficient to determine whether the
relative phase of the signal and reference channels, as indicated on a front panel
dial, is correct to within the limits specified by the manufacturer. In other cases,
where phase is of critical importance, it may be necessary to make a detailed
investigation of the relative phase-shift in order to catalogue system phase errors
as a function of frequency and indicated phase.
Problems arise in practice because the phase difference of the two waveforms
appearing at the phase-sensitive detector is not directly accessible in a typical
lock-in amplifier. This essential information must therefore be inferred from a
series of auxiliary measurements made at the input and output sockets of the
instrument. The usual procedure is to apply strictly in-phase sinewaves to the
signal and reference inputs and then to check that the output from the phase-
sensitive detector follows the cosR law as the phase-shift R is varied by means
of the front-panel phase control. Ideally we should be able to interpret deviations
from the cosR law in terms of system phase errors, but this can be done only
when other sources of external error have been eliminated. As we shall see, the
procedures involved are rigorous to the extent that a lock-in system of the highest
precision can be made to appear second-rate if they are not properly applied.
The principal sources of error affecting phase measurements are given as follows:
(i) 'Trigger errors in the reference channel
(ii) Signal and applied reference not strictly in-phase
(iii) Errors due to oscillator distortion
(iv) Errors due to phase noise, phase drift and phase-sensitive detector instability
Ways of minimising these various sources of error are dealt with under separate
headings below. It is hoped that these final sections will provide a useful cheek
list for those with a special interest in phase measurements.
4.6.2 Trigger phase errors
Trigger errors are incurred at the front end of the reference channel, in the input
trigger circuit. The relevant waveforms are shown in Fig. 4.8.

Chapter 417
Fig. 4.8 Waveforms in a broadband trigger circuit

It is usual for the trigger circuit to be designed with a relatively large hysteresis
band, vh of the order of 100 mV. The input is invariably a.c. coupled and we shall
assume that the phase-shift of the reference channel is to be generated with
respect to the instant where the input waveform makes a positive zero crossing.
Once fired, the output from the trigger circuit stays securely at a HIGH level
and does not switch LOW until the input voltage falls below the hysteresis
threshold. This is sufficiently removed from zero voltage to ensure that the
trigger circuit is free from multiple triggering when the reference input is
perturbed by low-level noise (amplitude up to 2vh. peak-to-peak). When the
reference input has relatively low amplitude, the output waveform will be highly
asymmetrical. However, this is corrected in the phase-control system which
follows the trigger circuit.
If phase accuracy is an important consideration we must assume that the
reference waveform is free from noise. In a precisely adjusted trigger circuit the
output will then accurately reflect the positive zero crossings of the input. In
practice, however, we must allow for a small error, , in defining the zero point as
illustrated in Fig. 4.9.
The phase error measured between the reference input and trigger output is
simply
= sin1/VR
/VR radians, << VR

Thus, for a trigger phase error of 0.1 or less we have the condition
VR / 600
In a well designed circuit, can be held at a level of a few millivolts over the
normal range of laboratory temperatures. For sinusoidal reference voltages,
therefore, a peak-to-peak input of a few volts will be sufficient to ensure good
trigger accuracy. Smaller inputs of the order of 100 mV or so will incur trigger
phase errors of up to 1, which is comparable with the expected phase accuracy of
many systems. In fact, most manufacturers specify a reference level of 1 V r.m.s.
for best accuracy.

Chapter 418
Fig.4.9 Trigger phase error

4.6.3 Defining in-phase signal and reference

The next problem is to ensure a strictly in-phase condition at the signal and
reference inputs to the lock-in amplifier. Although oscillators are available giving
signal and reference or SYNC outputs, we cannot, in general, rely on these
for the required phase precision. The only satisfactory way to achieve this is to
derive both inputs from the same oscillator terminal using a resistive potential
divider as shown in Fig. 4.10.

Fig. 4.10 Defining in-phase signal and reference voltages

In the arrangement shown it is assumed that the resistor R is much greater in

value than the source resistance Rs and that R, in turn, is much less in value than
the input resistance of both the signal and reference channels. The idea is that the
screened cable connections to the lock-in amplifier are driven from sources of
roughly equal resistance Rs. Provided that the screened cables are of equal length,
differential phase-shifts due to cable capacitance will then be kept at a minimum .
Cable capacitance is typically of the order of 100 pF per metre run. When
operating at a reference frequency fR, the spurious phase-shift introduced by a
cable of capacitance C connected to a signal source of resistance Rs is
e = tan1 2fRCRs

Chapter 419
Suppose we have C = 200 pF with Rs = 1 k at a frequency of 10 kHz. The
cable-related phase-shift is then about 0.7, increasing to over 7 at fR = 100 kHz.
Obviously an uncompensated phase-shift of this magnitude would be a serious
source of error in any system set up for precision phase measurement.
4.6.4 Errors due to oscillator distortions
Let us now consider the requirements for the signal oscillator. If the accuracy of a
lock-in system is to be measured in terms of its response to a sinewave as the
system phase is varied, then every step must be taken to ensure that a strictly
sinusoidal input is applied. We cannot expect the measurement system to follow
the cos R law if the signal is grossly distorted; it follows that, for precision
measurements, the distortion of the signal waveform must be correspondingly
small.
The response of a phase-sensitive detector to a general periodic signal was
considered in Section 3.5. From the results obtained there, it is obvious that the
calculation of phase errors due to distortion components is a complicated
business, suited more to numerical computation then to general analysis. Also to
concentrate solely on the signal might distract attention from the reference which
should similarly be free from distortion if misleading results are to be avoided.
To see why, let us turn to Fig. 4.11 which shows synchronous signal and
reference voltages subjected to arbitrary harmonic distortion (greatly exaggerated
for emphasis). The effect of the harmonic components is to shift the zero
crossings of the reference voltage relative to those of its fundamental component.
As a result, the zero-phase switching waveform triggered from the positive zero
crossings of the reference waveform exists in an arbitrary phase relationship with
the fundamental component of either signal or reference.

Fig. 4.11 Phase error due to waveform distortion

There is a common misconception that a fundamental only responding lock-in

system automatically compensates for harmonic distortion in phase
measurements, but this is clearly not the case. It is true that, if the system is able
to suppress harmonic responses, then the null-shift procedures given in Section
3.5 can be applied to bring the reference switching waveform directly into phase
with the fundamental component of the signal. However, at this point the
reference phase indicated by the phase-control dial will differ from zero degrees
despite the precise synchronization of the externally applied voltages. The system
will indicate an apparent phase error e, which depends on the level of distortion,
and the output will vary according to cos (R + e), where R is the set phase of
the reference channel.

Chapter 420
When the lock-in system has harmonic responses, the null-shift procedures do not
generally apply to distorted signals and the cosine dependence is no longer
obtained. To arrive at a bound on the level of distortion which can be tolerated,
we can take the special case of second-harmonic distortion where the positive
zero crossings of a signal or reference are shifted by up to D2/D1 radians relative
to those of the fundamental component. Here, D1 and D2 are the magnitudes of
the fundamental and distortion components, with D2 < < D1.
Second harmonic distortion at a level of 1% can therefore contribute an apparent
phase error of up to 0.01 rad or 0.5. On this basis, and taking additional
experimental evidence into account, it appears that a total harmonic distortion of
0.1% represents a suitable target for the signal/reference oscillator. Failure to
achieve this target means that the oscillator contribution to phase error cannot be
entirely ruled out: unfortunately this level of distortion is far below that of typical
laboratory oscillators. Oscillators of the function generator variety often have
harmonic distortion in excess of 2% or 3% and are usually quite unsuited to
applications where phase accuracy is a critical factor.
4.6.5 Phase noise, phase drift and phase-sensitive detector
instability
When the output of a phase-sensitive detector is maximized by adjusting phase,
the response is first-order independent of phase variations due to noise and drift
in the reference channel. In this condition, the main sources of error in precision
measurements will be offset and drift in the phase-sensitive detector, which must
be minimized by using the detection system in a high stability mode as defined
earlier.
In most lock-in amplifiers an adjustment is provided to trim the offset of the
phase-sensitive detector before measurements begin, leaving the drift component
as the principal limitation on precision. In a good-quality system this can be as
low as 5 to 10 p.p.m./K, equivalent to 10 V/K or less in a 10 V output,
independent of the a.c. gain provided before phase-sensitive detection.
Instability in the reference channel has its worst-case effect when measurements
are carried out close to the point where the phase is adjusted for an output null.
The ability of the system to maintain a nulled output is then affected jointly by
phase drift in the reference channel and output drift in the phase-sensitive
detector.
If we suppose the system is first brought to a precise null condition, the
subsequent variation of the output for a temperature change of 1 K can be
expressed in the form
vo = Vs(VF/SF)T /180 + VF
where T is the phase drift (degrees/K) and is the fractional output stability of
the phase-sensitive detector operating with a maximum ouput voltage VF at full-
scale sensitivity SF.
It was noted earlier that the contribution due to phase drift depends on signal
magnitude and system sensitivity while the error due to phase-sensitive detector
instability will be fixed for a given level of output gain. We shall assume that the
latter is switched to its lowest possible value. The effect of phase-sensitive
detector drift will then be negligible compared with the phase-drift contribution,
provided:
Vs(VF/SF)T /180 >> VF

Chapter 421
In most phase measurements the ratio Vs/SF is either close to unity or deliberately
increased to the order of 10 or even 100 in order to enhance the phase-sensitivity
of the system. Let us suppose that Vs/SF is made equal to 10, giving a net phase
sensitivity of 10 VF volts/radian, while the phase drift is specified as 0.03/K. If
the drift of the phase-sensitive detector is not to dominate the final measurement
it must satisfy the condition
<< 0.0052 ( 5000 p.p.m./K)
We have seen that this criterion is easily satisfied with modern lock-in
equipment. The conclusion is that phase drift in the reference channel is likely to
make a much larger contribution to errors in phase measurement than phase-
sensitive detector instability.
So far we have ignored the effect of short-ternm fluctuations due to phase noise.
The time scale of these fluctuations is such that their contribution to the final
output can always be reduced by increasing the output time constant. Usually the
residual noise has minimal affect when setting the phase for a null output, its
worst effect being to obscure the system response to small phase variations.
At reasonably large time constants the r.m.s. value of the output fluctuation due
to phase noise is usually much less than the long-term deviation due to
temperature related phase drift. In this case, the smallest incremental phase-shift
which could be measured over a period of time would be comparable with the
uncertainty due to phase-drift in the reference channel.

4.7 References
Discussion on the systems aspects of lock-in amplifiers is confined, for the most
part to manufacturers data sheets and application notes. See, for example:
1 'Specifying lock-in amplifiers. Technical Note 116, Princeton Applied
Research Corp., Princeton, NJ.
2 MUNROE, D.M. (1973): The heterodyning lock-in amplifier Technical
Bulletin, Ithaca Corp., Ithaca, NY.

Chapter 422
CHAPTER 5

Two-phase lock-in amplifiers

5.1 Introduction

Fig. 5.1 Two-phase lock in amplifier

Two-phase lock-in amplifiers incorporate a pair of phase-sensitive detectors

operated with quadrature reference waveforms as shown in Fig. 5.1. They were
originally envisaged as a means of measuring the in-phase and quadrature
components of a synchronous sinewave signal, a typical application being in
network analysis as depicted in Fig. 5.2.

Fig. 5.2 Two-phase lock-in amplifier used in network analysis. The vector
computer converts the in-phase and quadrature outputs to
polar form

For a lock-in amplifier with full-scale sensitivity SF and output voltage swing VF
the output voltages VA and VB following the low-pass filters are:
VA = Vs(VF/SF) cos
VB = Vs(VF/SF) sin

Chapter 51
where Vs is the r.m.s. value of the signal and the ratio VF/SF gives the scaling
factor of the lock-in amplifier.
The vector computer indicated in Fig. 5.2 has become a standard feature of
commercial two-phase lock-in amplifiers. This is an electronic circuit which
operates on the two output voltages VA and VB to produce voltages proportional
to signal amplitude VS and the relative phase , giving:
Vo = Vs(VF/SF) = (VA2 + VB 2),
= tan1 VB/VA
Two-phase lock-in amplifiers can thus display their outputs in either cartesian or
polar form and are generally associated with measurements in two main areas.
The first includes applications which make use of the quadrature components of a
signal such as a.c. bridge measurements, Nyquist plotting and general impedance
measurements. The second involves applications where signal magnitude is to be
measured in the face of large phase variations and the polar form of the output is
particularly valuable. Experiments using a swept-frequency signal and reference
usually fall into this category.
The most powerful systems currently available are those which are capable of
suppressing harmonic responses and so behave as if the signal is multiplied by a
sinewave. These more advanced systems to be described further in Chapters 8
and 9 will respond only to the fundamental component of a periodic non-
sinusoidal signal. In the case of two-phase systems with fundamental-only
response, the amplitude of the fundamental component of a signal and its phase
shift relative to the reference can be measured in a true vector mode without
ambiguity.
When the lock-in amplifier is of the conventional type and subject to the
harmonic responses of the phase-sensitive detector, the application of a vector
computer is only meaningful when the signal is of sinewave form. Also, we shall
find that the vector computer falls short of the ideal when signals are very noisy,
resulting in a severe limitation on dynamic range. There is, therefore, a good case
for investigating two-phase techniques which extend the benefits of phase-
tracking to non-sinusoidal signals while retaining the noise rejection inherent in
the synchronous detection process.
It is significant that the majority of two-phase systems are catalogued as lock-in
analysers by their manufacturers, thereby emphasising their role in the analysis
of both signals and systems. In support of this we shall be reviewing the
operation of two-phase systems as wave analysers and spectrum analysers and in
other applications where noise is not identified as a particular problem. This is in
addition to a brief survey of some classic two-phase applications. Features
common to most of these applications are the need for wide dynamic range and
the ability to operate with extremely high resolution in the frequency domain. We
should also add cost- effectiveness to this list. The relatively low cost of lock-in
systems allied to their versatility and the possibility of computer-control (Chapter
10) increases their appeal as general-purpose measurement tools.

5.2 Examples of classic two-phase applications

5.2.1 A.C. bridge balancing
The ability of two-phase lock-in amplifiers to simultaneously measure the in-
phase and quadrature components of a low-level signal perturbed by noise makes
them ideally suited to null detection on a.c. bridges. A typical experimental
arrangement with a four-arm bridge is depicted in Figure 5.3. The bridge output

Chapter 52
is fed to the differential inputs of a preamplifier or directly to the lock-in
amplifier if this has a differential input stage. The notes on signal connections
and on the disruption of ground loops given in Appendix 6 are highly relevant to
this particular application.

Fig. 5.3 A.C. bridge balancing using a two-phase lock-in amplifier

The lock-in amplifier is referenced directly from the bridge excitation source with
zero phase shift in the reference channel. The in-phase and quadrature
components of the bridge output voltage are measured with respect to the
reference voltage and displayed directly on the twin outputs of the lock-in
amplifier.
Using the notation of Fig. 5.3 the residual output voltage from the bridge is:
Z4 Z3
Vo = Vi
Z1 + Z 4 Z 2 + Z 3
Where Z1 etc. represent the complex impedances of the bridge arms. The balance
condition for a null output is now:
Z1Z3 = Z2Z4
And balance is only achieved when the real and imaginary parts of Z1Z3 and Z2Z4
are separately equal.
Putting the impedances in the form:
Z = R + jX
we obtain the balance conditions:
Re {Z1Z3}= Re {Z2Z4}
R1R3 X1X3 = R2R4 X2X4
and
Im {Z1Z3} = Im {Z2Z4}
R1X3 + X1R3 = R2X4 + X2R4
We must avoid the pitfall of assuming that the in-phase and quadrature
components of the bridge output can be separately and uniquely identified with
each of the balance conditions. This has been responsible for abortive attempts to
null one component independently of the other. Even worse is the incorrect
assumption that the in-phase component can be nulled by balancing resistive

Chapter 53
elements and the quadrature component nulled by balancing reactances. * In
general, we must expect that balance is obtained when the in-phase and
quadrature components are both zero and that each component will only be zero
when each of the null conditions is satisfied.
This rather pessimistic conclusion is tempered in practice by the relative ease of
use of the two-phase lock-in amplifier in this application. No adjustments are
required except to the sensitivity control which can be switched to register deeper
and deeper nulls as balancing proceeds. This should be contrasted to
measurement with a single-phase lock-in amplifier which would require continual
switching between in-phase and quadrature components to cheek the outcome of
the smallest adjustment.
Null detection makes full use of the noise-rejection properties of the lock-in
system, but is subject to a source of error resulting from distortion on the bridge
excitation oscillator. If the bridge is frequency selective it should be possible to
achieve a null at the fundamental excitation frequency, leaving a residual voltage
in the output made up from the distortion components of the oscillator. These are,
of course, harmonically related to the excitation and hence the reference
frequency. If these unwanted components are allowed to contribute to the output
of the lock-in amplifier, an error will result in the determination of the null point.
Even when the excitation oscillator is of exceptionally high purity it is probable
that non-linearities in the bridge elements will contribute measurable harmonic
components in the bridge output. This applies particularly to bridges containing
cored inductors and includes some important examples as the Maxwell, Owen
and Carey-Foster bridge configurations.
The only solution is to use a lock-in amplifier with fundamental-only response,
and this has become the preferred type of system in bridge balancing
applications.
The extremely high sensitivity of lock-in systems, obtained through the use of
external preamplifiers, means that in critical applications the level of excitation
can be drastically reduced to avoid excessive dissipation in the bridge elements.
The entire detection system including all connections to the bridge should be
designed very carefully to avoid ground-loop problems, particularly if detection
is envisaged down to a level of tens of nanovolts.
The scaling factor of lock-in amplifiers is such that a bridge offset voltage of this
order could be made to provide a response of up to 10 V from the phase-sensitive
detector. Such an output could then be used to provide a feedback signal for a
control system; for example a temperature control system where the bridge
balances about a 'set' point corresponding to a particular temperature.
5.2.2 A. C. impedance measurements
We shall use the example of semiconductor capacitance measurement which uses
an experimental set-up similar to that shown in Fig. 5.4
The voltage source at the frequency of interest is arranged to have a very small
output impedance by virtue of the attenuator and is coupled to the device under
test by a large capacitor CB. Application of a d.c. voltage VDC causes CB to
charge and develop a ramp voltage VBIAS which is applied to the device under test
together with the a.c. excitation vs. The device under test is terminated at the

*
This misconception probably arises because of confusion with the method for a.c. impedance
determination described in the next section.

Chapter 54
virtual earth input of the current amplifier, and so sustains the full input voltage
vs + VBIAS.
The signal current is now:
Is = vs [Gx + jBx], Bx = Cx
Which comprises components in phase and in quadrature with the excitation
source.
When the phase of the reference channel is set correctly, the in-phase output of
the lock-in amplifier becomes proportional to Gx and the quadrature output
becomes proportional to Bx.

Fig. 5.4 Semiconductor capacitance measurement. Provision is made to

apply a ramp bias voltage to the device under test

The system can be set up for phase and sensitivity as follows. First of all the
output from the excitation source is adjusted to give as large an a.c. voltage as is
allowable across the device under test. A suitable full-scale meter reading for Cx
is chosen and a capacitor of this value is put in the test point. The reference phase
is then adjusted to give no output from the in-phase phase-sensitive detector and
the overall sensitivity is adjusted to give a full-scale reading on the quadrature
output. The phasing can be recheeked by inserting a non-reactive resistor at the
test point when it should be observed that the quadrature output is zero.

Fig. 5.5 Measurement of low impedances. The series resistor Rs is chosen

so that Rs >>| Zx|

An alternative approach is often used when the impedance to be measured has

very low magnitude in the frequency range of interest. This is to supply the
device under test from a source of relatively high impedance and to use a voltage
amplifier as shown in Fig. 5.5. When the series resistor Rs >> | Zx| the device

Chapter 55
under test is supplied with substantially constant current vi/Rs which is in-phase
with the signal source. The voltage developed across Zx is then, to a good
approximation:
vs = vi (Rx + jXx)/Rs
This method is most appropriate when accurate results are to he obtained with
very low power dissipation in the device under test. The notes on amplifier noise
matching given in Appendix 5 are relevant when the output voltages are
particularly low since it is difficult to maintain a good noise figure when
amplifying from a source of very low impedance. Fortunately, lock-in amplifiers
are usually supported by a range of preamplifiers and matching transformers
which enable a near-optimum noise match to be obtained under a wide range of
signal and source conditions.
5.2.3 Phase measurements
We can identify three main types of application where the relative phase of a
signal is to be measured, each of which involves the lock-in amplifier in a
different set of measuring procedures.
The first is where the phase-shift of a sinewave signal is to be measured with
respect to a reference voltage to a high order of precision. This can be achieved
using the null-shift procedure introduced in Section 3.5, carried out with regard to
the precautionary measures given at the end of Chapter 4. The null-shift
procedure brings the signal and reference in phase at the phase-sensitive detector,
whereupon the phase-shift of the signal can be read directly from the reference
channel phase settings. Either a single-phase or a two-phase lock-in amplifier can
be used in this type of measurement.
The second type of measurement involves the detection of small phase variations
or small phase increments on a signal. The lock-in amplifier is then set up to
operate as a linear phase detector by defining a null condition with the signal
and reference in quadrature at the phase-sensitive detector. The two-phase lock-in
amplifier comes into its own in this type of application because the amplitude of
the signal can be continuously monitored on the in-phase phase-sensitive detector
while the phase variations of interest are monitored on the quadrature channel.
The accuracy to which a two-phase system can maintain quadrature between the
two phase- sensitive detectors is usually of a very high order, typically 0.1 at
mid-band. In practice, any departure from true quadrature can be compensated
when nulling the output from the quadrature channel, leaving the system to
register small phase increments to a high level of precision. Sources of error are
the phase noise and phase drift of the reference channel and the output drift of the
phase-sensitive detector. These must be taken into account as described in
Section 4.6.5.
The general rule is to operate with the maximum possible value of time constant
in order to minimize fluctuations due to phase noise and to operate at the best
achievable output stability, that is with the minimum value of expand gain
following the phase-sensitive detector. The overall sensitivity to phase variations
can then be enhanced by increasing the a.c. gain of the system. This is a
legitimate step to take in a system with large dynamic reserve; even though the
output from the in-phase channel may now be greater than full-scale, the
quadrature output will maintain a linear response to small phase variations
provided the allowable input swing to the phase-sensitive detector is not
exceeded. As demonstrated in Section 4.6.5, it is usually possible to overcome
the drift of the phase-sensitive detector to give a measurement accuracy limited
only by the phase drift of the reference channel. In extreme cases, entire lock-in

Chapter 56
systems have been enclosed in environmental chambers to stabilize operating
conditions and minimize temperature-dependent phase errors.
The final type of measurement is the exclusive preserve of two-phase systems,
where continuously changing phase angles in the range 0 to 360 are to be
monitored without adjusting the controls of the lock-in amplifier. A two-phase
lock-in amplifier with a vector computer is normally used in this situation;
however, when the signal is particularly noisy, a vector tracking system of the
type described in Section 5.4 often provides a more accurate measure of phase
shift.

5.3 Noise limitations of the vector computer

The ability to measure the amplitude of signals independently of phase setting
has proved to be so attractive that commercial systems have become available in
which the vector magnitude provides the sole output. Aimed specifically at the
spectroscopy market, these systems offer ease of use with the phase control
entirely eliminated, and have become known as phase-insensitive detectors or
p.i.ds. There are, in addition, many research workers operating with conventional
two-phase systems who habitually make use of the vector computer output in
routine signal recovery applications as opposed to precision vector analysis. In
view of this, it is appropriate to investigate the behaviour of the vector computer
when the signal is noisy and subject to large variations in amplitude - as might be
expected in a typical spectroscopy application. We, therefore, begin with an input
to the lock-in amplifier of the form:
v(t) = 2 Vs sin (st + s) + n(t)
where n(t) represents wideband random noise which accompanies the signal.
The response of a synchronous detection system to random noise components is
discussed in Appendix 3 and the effect of the harmonic responses of a phase-
sensitive detector have been reviewed in Section 3.3. The net result is that a two-
phase system produces output voltages
VA = [Vs cos + nA(t)] VF/SF
VB = [Vs sin + nB(t)] VF/SF

where nA(t) and nB(t) are uncorrelated noise voltages derived from the
components of n(t) which originate close to the reference frequency and its odd
harmonics.
The prime advantage of synchronous detection methods over competing
techniques is that the residual noise voltages which filter to the final output
appear with zero average value and so make no net d.c. contribution to the overall
response. As explained in Chapter 2, this is a result of the essential linearity of
the synchronous detection process. Bearing this in mind, let us now consider the
effects of subjecting the output voltages VA and VB to a stage of non-linear
processing, using a vector computer. This provides an output voltage:
Vo = (VA2 + VB2)
The most noticeable effect of the vector computer is observed under zero or low
signal conditions. The system output then becomes:
Vo = [nA2(t) + nB2(t)] VF/SF
The squaring operation ensures that Vo can take only positive values and is thus
equivalent to a stage of rectification. The zero signal output is then a fluctuating

Chapter 57
unipolar waveform having a non-zero average value which we shall denote by the
symbol :
= Ave. value [nA2 (t) + nB2(t)]1/2 VF/SF
It is not necessary to enter into detailed calculations to predict the general effect
of this noise offset in the output. If we assume that the input noise is fixed while
the signal is allowed to vary, then the input signal-to-noise ratio must be greater
than some threshold value before the response to the signal begins to register in
the final output. The system will be subject to gross measurement errors for
signals close to the noise threshold; calculations then show that the average
output voltage due to the signal must be greater than 5 before the vector
computer is in error by less than 21/2%. The error decreases rapidly as the signal
increases beyond this point, falling as the square of the signal.
When operating a spectrometer under noisy conditions the effect of a vector
computer will be to introduce distortions on small features close to the baseline.
An estimate of the noise offset, , should always be made at the outset, therefore,
to predict the range over which a linear response can be obtained. Fortunately, in
many cases, this is easily achieved by temporarily shutting off the signal. If this
cannot be done without losing the noise, an alternative method of estimating the
noise offset is given by first of all estimating the fluctuation on either of the
output voltages VA or VB. When operating with large time constants and
correspondingly long response times this can be measured in terms of the peak-
to-peak fluctuation of the output meter; otherwise it will be necessary to observe
the fluctuation by connecting VA or VB directly to an oscilloscope. The value of
the noise offset is then given to a good approximation by:
= Vp-p/4
where Vp-p is the observed peak-to-peak fluctuation in volts. If the resulting value
of is judged to be too high compared with the smallest value of output voltage
then the most obvious step is to reduce the fluctuations nA and nB by increasing
the time constants on both phase-sensitive detectors. It will be recalled that the
r.m.s. fluctuation is reduced in proportion to the square root of the time constant;
hence increasing the time constant by a factor of 10 brings a reduction of 10 in
.
In cases where the maximum allowable time constant is established by
considerations such as spectrometer scan time, it is common practice to accept a
relatively poor output signal-to-noise ratio and then to introduce a stage of post-
detector averaging. In most cases this consists of crude averaging by eye of
output chart records, while in others, use is made of signal averagers and
waveform eductors which store and average the results of several successive
scans to a high degree of precision. Unfortunately, even these relatively
sophisticated techniques cannot restore the loss of dynamic range incurred
through the use of a vector computer. In general, the success of any averaging
technique depends upon the noise having zero average value. It follows that the
best place for such equipment is directly on the output of the phase-sensitive
detector. In fact, this should present no problem in spectroscopic measurements
where signal and reference usually exist in a fixed phase relationship at all times.

5.4 Vector tracking

When operating a broadband harmonically responding lock-in amplifier the use
of a vector computer is restricted to sinusoidal signals. Also, if the signal is very
noisy and the maximum allowable time constant is restricted by external factors

Chapter 58
(such as spectrometer scan time), the vector computer can impose a limitation on
dynamic range.
We shall now investigate a mode of operation known as vector tracking which
overcomes the noise limitations of the vector computer while retaining the ability
to monitor the magnitude of non-sinusoidal signals independently of reference
phase setting. The method is, in effect, an automated version of the null-shift
procedures discussed in Section 3.5 and uses a two-phase system in the
configuration illustrated in Fig. 5.6.

Fig. 5.6 Vector-track configuration

The arrangement operates as a closed-loop control system in which the output of

phase-sensitive detector B is integrated and fed back to control a voltage-
controlled phase-shifter in the reference channel. The control conditions are such
that the net input to the integrator is zero at all times, which implies that the
phase-shift is automatically adjusted to enforce a null condition at the output of
phase-sensitive detector B. The reference phase at phase-sensitive detector A is
displaced by 90 and will always yield a maximum response for signals which are
symmetrical in the sense defined in Section 3.5. Otherwise, the response will
not necessarily be a maximum. * The best that we can say in the general case is
that the phase is controlled to a well-defined condition which ensures that a
consistent measure of signal magnitude can be obtained under adverse conditions,
irrespective of the initial phase setting of the lock-in amplifier.
Vector tracking systems can accommodate variations in signal phase-shift up to a
limit set by the voltage-controlled phase-shifter, usually 100. When the phase-
shifter follows a linear law over its full range, the internal feedback voltage
becomes directly proportional to the phase offset of the reference channel and can
be scaled and displayed on a meter calibrated in degrees. If the signal requires a
phase offset greater than the range of the phase-shifter it will be necessary to
change quadrants by switching increments of 90 from the front-panel phase
controls. The phase offset can then be obtained by adding the set phase of the
reference channel to the phase indicated by the output meter.
The capacitive feedback of the control-loop integrator is usually switched from
the time-constant control of phase-sensitive detector B which should now be
treated as a means of adjusting the transient response of the overall system. The
control loop contains one integrating element which is readily identifiable, but in

*
Unless of course the system is fundamental-only responding. In this case, a control loop based on
the null-shift procedure will always maximize the response to the fundamental component of a non-
sinusoidal periodic waveform.

Chapter 59
examining the system response we must also take into account the behaviour of
the voltage-controlled phase-shifter. This cannot respond instantaneously to
changes in the feedback voltage and will be characterized by a time constant
which enters the control loop equations. As a result, the overall system has a
response which is at least second-order and liable to exhibit oscillatory transients
following abrupt changes in signal or reference-channel phase-shift. The
erstwhile time-constant control on channel B is important in this respect as it can
be used to control the damping of the system and adjusted to give a response free
from excessive over- shoot, increased damping being equivalent to a longer time-
constant selection.
The amplitude of the signal also influences the behaviour of the control loop
since any change in amplitude is equivalent to adjusting the internal gain of the
system. The effect is to bring about an increase in damping coupled to an increase
in response time as the signal is reduced. If the transient response of the system is
important in a particular application, the recommended procedure is to adjust the
damping at maximum signal strength knowing that the damping can only
increase* when the amplitude is reduced in the course of an experiment.
A vector tracking lock-in amplifier can thus ensure that the detection system is
brought to a well-defined condition for all types of periodic signal. This is
achieved with minimal effort on the part of the user and without generating
spurious offsets at the output due to rectified noise components. Vector tracking
can be advantageous under many circumstances, but it must be remembered that
if a non-sinusoidal signal suffers large phase variations in the experiment under
investigation then it is possible that the waveform will also change and that the
overall calibration of the measurement system will vary accordingly. The only re-
course here is to use a lock-in system which responds only to the fundamental
component of a synchronous signal. This would include systems using a signal
channel with a tuned filter which were discussed in Section 4.4. In this case the
vector track mode serves to compensate phase variations caused by signal
frequency drift relative to the centre frequency of the tuned filter.

5.5 Asynchronous operation

5.5.1 Introduction
Up to this point we have assumed that the two-phase lock-in amplifier is
operating with the signal and reference precisely synchronized. In this section we
shall be relaxing this restriction and looking at the possibilities of using a two-
phase system fitted with a vector computer in applications where the signal and
reference are asynchronous. This will provide a background to the numerous
applications where two-phase lock-in amplifiers are used as wave analysers and
for high-resolution spectrum analysis.

5.5.2 Operation as a wave analyser

It has been established that the response of a synchronous detection system to an
asynchronous sinewave signal has the form of a beat component at the
difference frequency of the signal and reference. In practice, the only beat
components which survive to perturb the final output are those which correspond
to a frequency difference less than the bandwidth of the output low-pass filter.

*
These observations on response should be contrasted with these given in chapter 7 in relation to
phase-locked loops. We will find there that damping is reduced along with the signal level. The
difference is due to a modification of the integrator in phase-locked operation which gives
independent control of damping and response time (loop bandwidth). This feature is not usually
included in standard vector systems as supplied by manufacturers.

Chapter 510
Fig. 5.8 (a) High resolution spectrum obtained with a two-phase lock-in
amplifier; (b) line distortion resulting from a rapid scan

When deciding upon measurement bandwidth in a lock-in amplifier based system

the maximum acceptable value should be taken, corresponding to the smallest
acceptable value of output time-constant on the two phase-sensitive detectors.
This is because the allowable sweep rate, R (Hz/s) has an inverse dependence on
the square of the time constant (subject, of course, to the slew rate limitation of
the reference channel not being exceeded). For a distortion-free display the sweep
rate R should satisfy the criteria:
R = 1/(50 To2), 6 dB/octave filter
R = 1/(100 To2), 12 dB/octave filter
Note that the better frequency resolution of a 12 dB/octave filter demands a
correspondingly longer sweep rate. To see the practical significance of these
criteria suppose we wish to measure the frequency components of a signal
appearing in the range 950Hz to 1050 Hz at a resolution of 10 Hz using a
6 dB/octave filter. The required value of time constant is:
To = 1/(10) 30 ms
The recommended sweep rate for a distortion-free display is now:
R = 1/(50 To2)

= 22.2 Hz/s
The total frequency scan required is 100 Hz; hence the minimum scan time must
be 4.5 s. If, in a subsequent measurement, the frequency resolution was required
to be 1 Hz, the scan time would have to be increased to 450 s, a figure which
emphasizes the need to work with the maximum possible bandwidth.
To conclude, it is worth giving another set of criteria which can speed up the
measurement process in some circumstances. These give the sweep rate required
to display a spectral line with an error of less than 2% in amplitude, accompanied
by a small shift in the response peak but with negligible distortion otherwise. The
criteria are much less stringent than before:
R 1/(4 To2), 6 dB/octave filter

Chapter 513
 R 1/(8 To2), 12 dB/octave filter
Under these conditions, the shift of the line peak will be less than the system
resolution, giving an overall accuracy quite adequate for preliminary
measurements.

5.6 References
References and further two-phase applications:
1 The automatic measurement of semiconductor junction capacitance.
Application Report 5, Brookdeal Electronics, Bracknell, England
2 Bridge balancing and the two-phase lock-in amplifier. Application Note
104B; and Engineering applications of the lock-in amplifier. Application
Note 146. Princeton Applied Research Corp., Princeton, NJ

Chapter 514
CHAPTER 6

Limitations of conventional lock-in systems

6.1 Introduction
This short chapter provides an interlude where we can review the main
characteristics of conventional lock-in amplifiers before going on to consider
more complex system configurations. Of particular interest are the aspects of
performance which limit the effectiveness of conventional systems in different
applications. If a list of these performance limitations were to be drawn up it
would almost certainly include the following items:
(1) Dependence on the availability of a synchronous reference voltage
(ii) Trade-off between key specifications; for example, dynamic-range/linearity,
dynamic-reserve/output-stability
(iii) Harmonic responses of the phase-sensitive detector appearing in the overall
response of a lock-in amplifier
(iv) Slew-rate limitations in the reference channel
This list is not intended to be comprehensive, nor is it meant to imply that items
(i) to (iv) are identified as shortcomings in every application. What we can say is
that shortcomings in these areas have prevented conventional lock-in amplifiers
from being adopted as general-purpose measurement tools and from performing
tasks beyond their traditional role in signal recovery.
The absolute dependence of synchronous detection systems on a reference
voltage would appear to be a prime candidate in this respect, yet, in practice,
most research workers seem able to devise experiments where a local reference
is made available. There are, nevertheless, several examples of experiments
where the signal source is remote from the detection system and a local reference
must be generated by phase-locking an oscillator to the incoming signal.
An example in this category, the reception of satellite beacon signals, is
described in the next chapter. It should not be thought, however, that all phase-
lock applications are of this remote type. For example, in Fourier-transform
photometry the traditional light chopper is often replaced by a rotating optical
grating. This may be so fine-ruled that the usual method of generating a reference
by means of an auxiliary light source and phototransistor is impossible to apply.
In this case a reference can be generated by phase-locking to the signal in the
output of the experiment using the general arrangement illustrated in Fig. 6.1.
The problem of phase-locking to noisy signals merits a fairly extensive
discussion in the next chapter where we shall see how phase-locked systems can
be bust up using standard lock-in amplifiers and off-the-shelf modules.
Regarding the second item, it must be expected that some sort of trade-off will be
encountered when any electronic instrument is operated at its performance limits.
The trade-offs referred to in (ii) are inevitable when a conventional lock-in
amplifier is operated with a broadband signal channel, but we have seen that the

Chapter 61
trade-offs can be improved when filters are used to eliminate unwanted
components before detection. For example, the increase in signal-to-noise ratio

Fig. 6.1 General arrangement for generating a local reference in a lock-in

recovery system. The phase-locked loop normally incorporates a
second lock-in amplifier or phase-sensitive detector

obtained by filtering can be made equivalent to an increase in dynamic range; the

noise-handling capacity of the system is improved without incurring a penalty
due to increased offset and drift in the output. Alternatively, the filters could be
used to reduce the total level of signal and noise at the input to the phase-
sensitive detector with a possible increase in dynamic range sacrificed in favour
of an improvement in linearity.
In either case, the improvements attributable to the use of filters are obtained at
the expense of wideband capability. Furthermore, the use of filters offers no
solution to restoring the loss of dynamic range which is always observed when a
phase-sensitive detector is operated towards the upper limit of its recommended
frequency range. In general applications, therefore, there may come a point where
the trade-offs can only be improved by moving to systems with inherently better
performance. This might mean purchasing a conventional lock-in amplifier from
a later generation than existing equipment. From the point of view of a designer
working at the limits of available technology, it might be more appropriate to
investigate techniques which enhance the dynamic range of state of the art
phase-sensitive detectors. Methods of achieving this last objective, by means of
synchronous heterodyning, will be described in Chapter 8.
The remaining limitations arising from harmonic responses and slew rate can be
summarized with reference to specific measurement difficulties that have come to
light in earlier chapters. Some solutions will be found in Chapters 8 and 9 which
deal with more complex lock-in systems operating on the heterodyne and pulse-
width-modulation principles. Unlike the phase-locked systems discussed in the
next chapter, heterodyning and p.w.m. lock-in amplifiers are almost invariably
supplied and used as self-contained units and cannot, in general, be built up as
modular systems. It will be found that some drawbacks of these more advanced
systems are highlighted along with their inherent advantages. Certainly no single
lock-in amplifier will address all the limitations identified at the beginning of this
chapter; it is up to the user to define his own requirements and select the system
most suited to his needs.

Chapter 62
6.2 Limitations arising from harmonic responses
6.2.1 Susceptibility to interference
The response of a switching phase-sensitive detector is characterized by a set of
transmission windows centered on the reference frequency and its odd harmonics.
As discussed in Chapter 3, the higher-order windows have negligible effect on
the final noise output when the signal is accompanied by broadband white noise,
but can have the most serious effect when the signal is accompanied by
narrowband noise, or by discrete interference components.
The situation is by far the worst when the lock-in amplifier is synchronized to a
low-frequency reference signal at 100 Hz or less. For example, a reference
frequency of 100 Hz gives rise to 499 transmission windows between 100 Hz and
100 kHz at, or near to which, the lock-in amplifier can respond to interference
voltages. Moreover, at 100 kHz, where the relative magnitude of the harmonic
windows is of the order 1/1000, the window frequencies are only 0.2% apart,
leaving a very small margin for adjusting the reference frequency if large-scale
high-frequency interference components are to be avoided.
The problem is particularly acute when operating photometric equipment
incorporating infra-red detectors. Such detectors often exhibit considerable
thermal inertia and so limit the usable chopping frequency to a rnaximum of
10 Hz or so. The measurement difficulties are often aggravated by the need for a
matching transformer in the signal path to ensure a good noise match to the lock-
in amplifier (Appendix 5). This renders the system susceptible to inductive pick-
up and can result in relatively high-level interference at mains frequency.
Because the chopping frequency is limited to about 10 Hz it is necessary to
choose a value of this which does not have an odd harmonic within a few hertz of
mains frequency.* Otherwise, measurements will be perturbed by a difference-
frequency beat component in the final output. Although the beat can always be
reduced by using a larger output time-constant, this may not always be acceptable
in view of the increased response time of the measurement system.
The standard technique of suppressing harmonic responses by using a tuned filter
in the signal channel of a lock-in amplifier was described in Section 4.4.2. From
the discussion given there it is evident that a tuned filter offers only a partial
solution to the present problem unless the stability of the chopping system
(usually mechanical) is of a very high order. The use of a mains-frequency notch
filter would usually provide a better solution to measurement difficulties in this
case.
6.2.2 Ambiguity due to the harmonic responses
In a harmonically responding. system the output due to a sinusoidal signal at
frequency fR is indistinguishable from one with three times the amplitude and the
correct relative phase at frequency 3fR. This ambiguity in the response is not so
serious when the requirement is for signal recovery at a fixed frequency with
synchronous signal and reference, but can give rise to misleading results
elsewhere. For example, the problems with signal distortion in a.c. bridge
measurements have been described in Chapter 5. Also in that chapter were
examples of operation with asynchronous signals, where the lock-in amplifier is
used for wideband spectrum analysis. In this case the harmonic responses. will
give rise to spurious lines in the output spectrum which occur whenever an odd
harmonic of the reference coincides with a frequency component of the signal

*
9.3 Hz and 11 Hz are popular choices of chopping frequency in infra-red spectroscopy for this
reason.

Chapter 63
under investigation. Clearly, the use of a tuned filter for harmonic suppression
has no.relevance to this type of measurement where the frequency is changing
continually.
A further example relates to the discussion on phase-locked loops in Chapter 7. If
the phase detector in such a loop is provided by a phase-sensitive detector or
lock-in amplifier with harmonic responses it is possible for the loop to lock
securely at an odd sub-harmonic of the incoming signal frequency. If, in addition,
the signal has squarewave rather than sinewave form, there will be a large
number of additional frequencies where lock could be acquired. The effect here
will be at its worst when the phase-locked loop is required to operate
automatically. When the locking signal is very noisy it is recommended that the
initial locking conditions are brought under manual control. This would normally
be sufficient to ensure that spurious locking to harmonic components was
avoided in signal-recovery applications.
6.2.3 Detection of non-sinusoidal signals
The limitation here refers to the problem of obtaining a maximum response to
non-sinusoidal signals in the presence of noise. It was shown in Section 3.5 that
there is no problem in principle with so-called symmetrical signals. Here, the
null-shift procedures can always be applied to produce a maximum response
which is first-order independent of errors in the reference-channel phase setting.
Otherwise, the response will be less than maximum, resulting in a loss of output
signal.to-noise ratio and an increased susceptibility to phase changes in the
reference channel and in the applied signal.
The null-shift procedures can be applied with confidence to all types of periodic
signal when the lock-in amplifier has fundamental-only response. The detection
system is always brought to a condition where the fundamental components of
the signal and reference are in phase at the phase-sensitive detector. These
systems will also ensure that this condition is reached by the automatic vector
tracking system described in Chapter 5 in relation to two-phase lock-in
amplifiers, and when the lock-in amplifier is brought under computer control as
described in Chapter 10.
In practice, the loss of sensitivity to asymmetrical signals is often marginal in
conventional systems and the null-shift procedures offer the ultimate advantage in
giving a phase setting which can be reproduced under the noisiest conditions. The
problem of defining the phase-shift to give maximum response to an
asymmetrical signal remains, however, and serves as another example of the
measurement difficulties associated with harmonically responding systems.

6.3 Slew rate limitations

The specification of reference-channel slew rate was dealt with in Chapter 4. The
loss of accuracy, or indeed loss of lock, that results from a rapidly changing
reference frequency must be considered when setting up any experiment
involving a frequency sweep at the reference input.
In most cases, when using time-constant settings of a few hundred milliseconds
or greater, the primary limitation on sweep speed comes from the response of the
output filter of the lock-in amplifier. This aspect was discussed in some detail in
Chapter 5 when outlining the behaviour of a two-phase system used as a high-
resolution spectrum analyser. When operating at lower resolution, however, say
with resolution bandwidths of several hundred hertz, and with a relatively wide
frequency sweep (covering, for example, the entire audio frequency range), the

Chapter 64
maximum allowable sweep speed will often be determined by the slew rate of the
reference channel.
To take an example: suppose we operate with a 12 dB/octave output filter and a
time constant setting of 1 ms. This gives a 6 dB frequency resolution of
f = 1/(To) 300 Hz
The maximum allowed sweep-rate to avoid errors greater than 2% in the output
filter has been given in Chapter 5. We obtain
Rmax = 1/(8To2) = 125 kHz/s
If it was required to sweep from 2 kHz to 20 kHz, the system would allow a
sweep time
Ts = 18.103/125.103 = 144 ms
This short sweep time would allow the spectrum-analyser output to be displayed
in conventional fashion on an oscilloscope. The problem is, of course, that a
sweep time of 144 ms over the range 2 kHz to 20 kHz is equivalent to
72 ms/decade, a figure which is well beyond the slew rate capability of most
conventional lock-in amplifiers. The reference processing circuits incorporated in
pulse-width-modulated systems offer a solution to the slew rate problem as will
be shown in chapter 9.

Chapter 65
 from the phase detector, which, in turn, supplies the feedback essential to loop
control.
It is usual to express this result in the form
VD = KD sin
Where KD (volts/radian) measures the sensitivity of the phase detector. Since the
amplitude of the v.c.o. is usually fixed we find that the phase detector sensitivity
is proportional to the amplitude of the locking signal. In most of the standard
analysis which follows it will be assumed that KD is fixed, but we must
eventually consider the effect of variations in signal amplitude.

7.2.2 Static phase error

The notion of phase error must arise in a general description of any phase-control
system. To calculate its magnitude we must first consider the operation of the
v.c.o.
A linear model is to be used in which the frequency offset from the v.c.o.
free-running frequency, o, is proportional to the applied control voltage vc. This
gives
= K ov c
where Ko is the v.c.o. gain factor with dimensions radians/V-s.
Under the essentially static conditions assumed here, when the amplitude and
phase of the locking signal are free from variations, the v.c.o. offset frequency
will be
= i o
which is maintained by a control voltage
vc = KDF(0) sin
Here, F(0) is the magnitude of the zero frequency response of the loop filter. The
static frequency offset of the v.c.o. from its free-running frequency is therefore
= KoKDF(0) sin
The loop is usually designed to make the static phase error very small. In this
case we can make the approximation sin and so obtain

=
K o K D F(0)
The static phase error can accordingly be reduced by increasing the d.c. gain of
the loop, KoKDF(0), and by initially setting the free-running frequency of the
v.c.o. as close as possible to the incoming frequency, thus minimizing .

7.2.3 Hold-in range

If we now suppose that the incoming frequency is subject to a slow drift, the loop
will track and accommodate the frequency change by a change in the v.c.o. offset
frequency. However, the results of the last section show that the maximum
possible value of the frequency offset is
||max = K0KDF(0)
corresponding to a phase error magnitude of /2.

Chapter 74
 lag - lead filter (R0 / R2 >>1)
F( j) = F()(1+ jT2 ) / jT2 , T2 = R2C
imperfect integrator
where in both cases, the high-frequency gain of the filter, F() is a real ratio,
R2/R1.

If the loop-frequency response function is now derived, we find, for both loop
filters:
N = (K0KDF()/T2)
and
b = NT 2
The expression for b is exact for the imperfect integrator and an excellent
approximation for the lag-lead filter, for all reasonable values of damping
(b2 >> R2/R0).
Gardner1 has shown that for this type of filter the loop gain is correctly given by
the quantity K0KDF().
The noise bandwidth of the loop becomes very important when we deal with the
problems of phase-locking in noise. This can be expressed in terms of N and b
as:
N 1
BL = b +
2 4b
For any value of N the noise bandwidth is minimized when b = , and is then
given by N/2. Note that, following convention, the noise bandwidth is given in
hertz, while N is expressed in radians/s.
When using a lag-lead filter or imperfect integrator the natural frequency of the
loop and the loop damping ratio exhibit a square-root dependence on loop gain.
The practical significance of this becomes evident when the amplitude of the
locking signal is allowed to vary in the course of a phase-locked experiment. We
saw in section 7.2.1 that the phase-detector constant KD is strictly proportional to
Vs. If all the other loop components are held at a fixed value we obtain the
following functional dependence on Vs for N and b:
N = c 1V s
b = c 2V s
where c1 and c2 are suitably defined constants. The variation of the noise
bandwidth now takes the form:
c1
BL =
8c 2
[1+ 4c 22Vs ]
showing that BL increases in direct proportion to the signal amplitude.

7.5 Noise and phase-locked loops

When the locking signal is accompanied by a random noise disturbance, noise
enters the phase-locked loop via the phase detector and gives rise to a random
phase error on the v.c.o. output. When the phase error is small it appears as phase

Chapter 79
 SNRI = Vs2/(2BIWN)
while the signal-to-noise ratio measured within the effective bandwidth of the
loop is
SNRL = Vs2/(4BLWN)
We shall not attempt to attach any physical significant to SNRL which is often
loosely called the loop signal-to-noise ratio. It can be shown nevertheless that the
probability of cycle slipping will be very small provided that this noise measure
is greater than about 6 dB (see, for example, the results reviewed by Gardner1 and
Blanchard2). We shall err on the side of safety when calculating the required
value of BL and use the criterion:
SNRL = Vs2/(4BIWN) 10
In this condition the system is amenable to a small-signal analysis. The results in
the literature1 give a good estimate of the mean-square phase error* on the v.c.o.
output, namely:
1
N2 =
2SNRL
Let us now assume that the signal-to-noise ratio of the locking signal has been
estimated and that the input bandwidth is known. We have:
SNRL = SNRI BI /(2BL)
and, putting SNRL 10, we obtain the condition:
BL SNRI BI /20
to ensure that lock is maintained in the presence of noise on the locking signal.
With this condition satisfied, the v.c.o. has a mean-square phase error

N2 0.05 radian2
The next step is to derive optimization procedures which guarantee that these
conditions obtain even when the signal amplitude is allowed to vary over a wide
range. This will represent a significant improvement over the standard treatment
where the signal and hence the loop parameters are fixed, and changes in
signal-to-noise ratio are arranged by allowing the noise intensity to vary.
It was noted earlier that we have taken no account of phase noise which is
inherent in the v.c.o. itself. In designing phase-locked loops for high precision it
is normally arranged that the loop bandwidth is wide enough to accommodate the
bulk of these variations, and it can be shown that their effect is reduced in
proportion to the loop gain of the system. This requirement is obviously
incompatible with choosing a narrow loop bandwidth to combat noise of external
origin. Also, it will be shown that having determined the bandwidth of the loop,
consistent with a reasonably high damping ratio, the choice of loop gain is
restricted. It turns out, however, that when operating in the audio-frequency range
with loop components of reasonable stability, any residual jitter due to
imperfections on the v.c.o. will be masked by noise on the locking signal,
provided that the loop bandwidth is not made unnecessarily small. This is also an
important consideration when the loop is expected to track slow frequency
variations on the locking signal as will be shown in Section 7.7.

*
The mean square phase error is calculated on the assumption that the static phase error of the loop
(see Section 7.2.2) has been trimmed to zero.

Chapter 711
7.6 Optimization procedure
The optimization procedure will be derived for a classic signal recovery example,
where the locking signal amplitude is varying over a wide range against a noise
background with uniform spectral characteristics. It is necessary for us to have an
estimate of the minimum value of the locking signal, corresponding to the worst-
case signal-to-noise ratio at the input to the loop, (SNRI)MIN.
When the loop incorporates a lag-lead filter or imperfect integrator, the noise
bandwidth will be dependent upon signal level as described in Section 7.4. We
accordingly design the loop to have its minimum noise bandwidth at minimum
signal level. Using the result of Section 7.5, the minimum noise bandwidth is
chosen to satisfy:
(BL)MIN (SNRI)MIN BI /20
The behaviour of the loop as the signal increases from its minimum value can be
predicted as follows.
First of all we note that the loop signal-to-noise ratio is given by:
SNRL = Vs2/(4BLWN)
Since BL increases, at most, linearly with signal we find a steady improvement in
SNRL as the locking signal increases, causing a proportional reduction in the
mean-square phase error (= (2SNRL)1) due to external noise. The rise in noise
bandwidth also makes the loop more effective in reducing the effect of phase
noise inherent in the v.c.o. It is shown in section 7.4 that the loop damping will
also rise with signal level. Fortunately, this rise is accompanied by an increase in
the natural frequency N. This joint behaviour results in a response which is not
too sluggish as would be the case if b, alone, were to increase.
Turning now to the question of loop damping: this would normally be chosen to
have a minimum value of about to ensure that the loop transient response is not
marred by excessive overshoot and ringing. It also turns out that choosing this
particular minimum value greatly simplifies the optimization procedures. The
damping ratio falls with locking signal amplitude; we therefore arrange for a
minimum damping ratio of at the minimum anticipated signal level.
The general expressions for noise bandwidth and damping ratio are:
N 1
BL = b +
2 4b
b = NT 2
Hence, if we put b = bMIN = and decide on a minimum value for BL we find

that T2 is given immediately:

T2 = 1/(2BL)MIN
while the minimum value of N is given by
(N)MIN = 1/T2 = 2(BL)MIN
Using the value of the phase detector constant appropriate to minimum signal
level, together with the required value of T2, the loop can now be designed to
give N = (N)MIN at b = bMIN. From the results given in Section 7.4 this implies
that the minimum value of loop gain is given by
K0(KD)MIN F() = 1/T2

Chapter 712
T2 has already been determined, so we cannot improve the minimum loop gain.
We have already noted that residual phase noise on the v.c.o. will make a
contribution to phase jitter under conditions of low loop bandwidth and low loop
gain. The effect of these incidental variations must be checked in a trial run using
a noise-free signal. This is, in any case, a useful first step in setting up a phase-
locked loop of even moderate complexity.

7.7 Notes on acquisition and tracking

Acquisition is a general term which is used to describe the extremely complex
processes by which a phase-locked loop picks up an incoming signal and moves
towards lock.
When the loop is in an unlocked condition the output from the phase detector will
initially be a beat waveform which contains the difference frequency between
the locking signal and the instantaneous frequency of the v.c.o. The maximum
possible peak-to-peak value of the beat waveform is KD volts which is
subsequently attenuated by the effect of the loop filter. If the resulting voltage
swing at the v.c.o. is sufficient to make the v.c.o. and input frequencies coincide,
the system moves smoothly into lock without slipping cycles.
In a second-order system using a lag-lead filter the attenuation at high beat
frequencies (corresponding to a high initial frequency difference between the
locking signal and the v.c.o.) has a constant value F(). In this case the
maximum available swing of the v.c.o. frequency is K0KDF() which defines
the capture range of the loop.
For initial frequency offsets within the capture range, locking is assured and fast.
However, an important consequence of using an integrating loop filter is that the
loop will eventually pull-in to a locking signal which is at a frequency far
removed from the v.c.o. frequency. The reason is that the beat waveform is
highly asymmetrical and contains a d.c. component. This can build up in the
integrator and gradually drive the v.c.o. towards the signal. Pull-in can be a
cumbersome and time-consuming process which is greatly affected by noise on
the signal. It is usually overcome by manual tuning of the v.c.o. to bring the
frequency difference within the capture range of the loop. Locking is then, for
practical purposes, instantaneous.
It must be assumed that acquisition is always assisted when locking to very noisy
signals and that the loop is finally trimmed to minimize the offset frequency of
the v.c.o. As indicated in Chapter 6, manual assistance of acquisition serves also
to avoid ambiguities in locking when the phase detector has responses to odd
harmonics of the v.c.o. frequency.
Finally, let us look briefly at the case where the locking signal frequency is
subject to a slow variation or drift. It can be shown that a loop containing an
imperfect integrator is capable of tracking a locking signal with a changing
frequency, but that the phase of the v.c.o. suffers a slip relative to the phase of
the locking signal. This is analogous to the phase slip observed in the reference
channel of a lock-in amplifier in response to a changing reference frequency, as
discussed in Section 4.5. In the case of the phase-locked loop the rate of change
of locking frequency R is related to the phase slip by 1,2:
2R = N2
where R is expressed in Hz/s. Note that the phase slip is exclusive of any transient
phase error that may occur following the application of a frequency sweep.
When designing a loop for recovery from noise it has been suggested that the
natural frequency has a minimum value
Chapter 713
7.8.2 Identifying the loop constants
When used as a phase detector with a time-constant filter, the observed sensitivity
of a lock-in amplifier to static phase errors is, from Section 4.1:
VF(Vs/SF) volts radian
This observed phase sensitivity is inclusive of the d.c. gain of the time-constant
filter, R0/R1. In phase-lock operation, the phase detector sensitivity KD that
appears in the loop equations is exclusive of the gain provided by the filter block.
We thus obtain:
VsVF
KD = (R1/R0)
SF
In practice, the ratio R0/R1 accounts for the bulk of the phase-sensitive detector
dynamic reserve. The phase detector constant to be used in the loop equations is
therefore given approximately by the overall phase sensitivity reduced by a factor

equal to the dynamic reserve. For either a lag-lead filter or imperfect integrator
we have F() = R2/R1. The loop gain is accordingly

Vs VF R1 R2
K0KDF() = K0
SF R0 R1
Putting T2 = R2C, T0 = R0C we obtain

Vs VF T2
K0KDF()
= K0
SF T0
Here, T0 is the time-constant setting on the front panel of the lock-in amplifier. In
view of our earlier remarks about assigning a value to R2, T2 might have to be
identified from
a circuit diagram or by inspection of the time constant switch.
Using this value of loop gain, the natural frequency of the loop can now be put in
the form
K 0VsVF 1/ 2
N =
SFT0
In the usual arrangement, T0/T2 appear in a fixed ratio:
T0/T2 = R0/R2 = r

The damping ratio is therefore:

b = NT2 = NT0/r
Provided that r can be identified, the loop parameters N and b are now given in
terms of Vs and the lock-in amplifier settings; time constant T0 and sensitivity SF,
for an output voltage swing VF.
7.8.3 Optimization procedures for lock-in amplifiers
The specification for the locking signal gives us a minimum anticipated r.m.s.
signal level VMIN, an input noise bandwidth BI and a worst-case signal-to-noise
ratio (SNRI)MIN. This enables us to choose a minimum value of loop noise
bandwidth (BL)MIN commensurate with the input conditions:
(BL)MIN (SNRI)MIN BI /20

Chapter 715
First of all, the noise bandwidth of the loop: at minimum signal level this must
satisfy
(BL)MIN (SNRI)MIN BI /20
101 5 103 /20 Hz or 25 Hz
The smallest time constant consistent with this value is:
T0 = r/(2 25)
We have r = 1000; hence
T0 = 20 s
If a time constant of 20 s is not available, the next largest should be selected. The
optimization procedures will ensure that the damping ratio will have a minimum
of as required.
Finally:
K0VMIN/SF = r2/(VFT0)
K0 has been given. Using VF = 10 V we obtain:
VMIN/SF = 0.079
The minimum anticipated signal level is 1 mV r.m.s. The lock-in amplifier should
therefore be set to a full-scale sensitivity
SF = VMIN / 0.079 = 12.6 millivolts
In practice, a full-scale sensitivity of 10 mV will be indistinguishable from the
optimum setting. Note that the locking signal may subsequently take larger
values which exceed the full-scale sensitivity of the lock-in amplifier without
incurring a fault condition, provided the total allowable swing on the signal input
is not exceeded.*
If an inconveniently low or high ratio VMIN/SF is predicted, there may be scope to
change the value of K0. It should not be overlooked that many v.c.o.s found in
measurement laboratories have overlapping decade ranges. By judicious choice
of operating frequency it may be possible to change K0 by a factor of 10 or even
100, depending on range selection.
When very narrow loop bandwidths are required, the ability to track signals of
changing frequency (Section 7.7) must be taken into consideration. Also, a trial
run using a noise-free signal should be carried out to assess the residual phase
jitter in the loop arising from incidental phase- and frequency-modulation on the
v.c.o. Here again, the ability to switch between overlapping ranges might prove
useful and enable the v.c.o. to be operated in a region where its self-noise is
lower.

7.9 The final measurement

Let us finally return to the measurement system proposed at the beginning of this
chapter, where the v.c.o. output is used as a reference voltage for the detection of
the locking signal in a second lock-in amplifier.
Two-phase lock-in amplifiers modified for phase-locking are ideal for this type of
measurement; quadrature channel B is used for phase-locking, leaving the

*
This comment is in line with the procedures for increasing the phase sensitivity of lock-in
amplifiers that were given in Sections 4.7.5. and 5.2.3.

Chapter 717
signal to be measured in in-phase channel A. When the dual phase-sensitive
detectors are fitted with independently switched expand amplifiers it is possible
to operate the two channels at sensitivities differing by a factor of 10 or even 100.
This is usually sufficient to ensure that the sensitivities of the two phase-sensitive
detectors can be separately optimized for phase locking and for signal detection.
When the signal is noisy, the v.c.o. output inevitably exhibits phase jitter relative
to the locking signal. Fortunately, as a result of the cos dependence, this phase
jitter has only a second-order effect on the measurement at the second phase-
sensitive detector. It turns out in practice that if the phase jitter in the loop is
sufficiently small to ensure long-term locked operation without slipping cycles
(as we have assumed throughout), then the phase jitter will have minimal effect
in the final measurement. In the worst case, at minimum signal level, its effect
can be reduced by increasing the output time constant in the second lock-in
channel beyond the value normally used at a given signal-to-noise ratio.

7.10 References
1 GARDNER, F.M. (1979): Phase lock Techniques (John Wiley, New York)
2 BLANCHARD, A. (1976): 'Phase-locked loops: Application to coherent
receiver design (John Wiley, New York)
3 LINDSEY, W.C. (1972): Synchronization systems in communication and
control (Prentice-Hall, N.J.)
4 HOGG, D.C. and CHU, T.S. (1975): The role of rain in satellite
communications, Proc. IEEE, 63, pp. 1308-1331
5 BAYLISS, A. (1974): A guide for orbital test satellite experiments. Proc.
European Conf. on Electrotechnics, Eurocon 74, Amsterdam

Chapter 718
CHAPTER 8

Heterodyne lock-in amplifiers

8.1 Introduction
In a heterodyne lock-in amplifier, phase-sensitive detection is carried out at a
relatively high, fixed, frequency following a stage of frequency translation of the
applied signal. Since the phase-sensitive detector operates at a frequency that
bears no harmonic relationship to the applied signal, the harmonic responses that
characterize a conventional lock-in system are suppressed. Practical lock-in
amplifiers operating on the heterodyne principle thus conform very closely to
ideal fundamental-only responding systems. The first of these, offering relative
freedom from harmonic responses over a moderate frequency range, was the
Ithaco Dynatrac lock-in amplifier. This was introduced in both single- and two-
phase versions in the early 1970s.
In the early system the benefits of fundamental-only response were obtained to
the detriment of performance in other areas. For example, the true frequency
range was only about one decade and coverage of the audio-frequency range
required a total of five sets of plug-in circuit cards. Also, the phase accuracy left
much to be desired, particularly at the extremes of the individual ranges. Even
under the most favourable conditions, this heterodyne system suffered by
comparison with the high precision of conventional lock-in amplifiers. The
system nevertheless enjoyed considerable success and served to focus attention
on the limitations associated with harmonic responses which were reviewed in
Chapter 6.
The block diagram of the Ithaco Dynatrac begs comparison with that of a
superhet radio, and so the system attracted the "heterodyne" label from the time
of its first introduction. This description is now applied more or less
indiscriminately to all lock-in amplifiers that incorporate one or more stages of
frequency translation. It can be argued that these systems bear only a superficial
resemblance to classical heterodyne systems and that the use of expressions such
as "intermediate frequency" and "i.f. filter" in relation to lock-in amplifiers is
likely to cause confusion, especially with those who have a clear understanding
of the conventional usage of these terms. It is therefore necessary to introduce a
note of caution about the terminology employed in this chapter which reflects the
usage that is now prevalent among lock-in amplifier manufacturers and appears
in data sheets and publicity material.
The Ithaco Dynatrac has since been matched by alternative and improved
heterodyne lock-in amplifiers from the EG&G companies, PAR and Brookdeal.
These lock-in amplifiers have greatly benefited from developments in technology
relating to both reference channel and phase-sensitive detector design. In this
chapter we shall be taking note of these developments and aiming to highlight
areas of specification which are peculiar to heterodyne lock-in amplifiers.
Of particular interest in this respect will be the problem of identifying the
spurious responses which occur when the frequency of an asynchronous signal
lies close to a "critical" frequency. In a conventional system these critical
frequencies correspond to the odd harmonics of the reference frequency. We shall
show that in principle heterodyne systems can be designed to be inherently free
Chapter 81
of these harmonic responses. There are nevertheless a number of additional
critical frequencies, each with its related transmission window, which must be
taken into account when a heterodyne model is adopted. A major objective of
heterodyne system design is therefore to achieve a high degree of suppression of
all unwanted responses, consistent with maintaining wideband, wide dynamic
range performance.
Unfortunately, as we shall see, this last requirement cannot be achieved without
sacrificing the total rejection of harmonic responses, which is inherent in an
"ideal" heterodyne system. In our discussions we must therefore make a clear
distinction between principles and practice and be prepared to examine the trade-
offs which are necessary to produce heterodyne systems with good all-round
performance.
Also included in this chapter is an appraisal of the synchronous heterodyne
technique. Synchronous heterodyning offers a means of improving the dynamic
range of phase-sensitive detectors and lock-in amplifiers. The technique can also
be used to counteract the loss in dynamic range which occurs when an otherwise
conventional phase-sensitive detector is operated towards the upper limit of its
frequency range.
This last approach is used to obtain a competitive dynamic range specification in
the EG&G Brookdeal heterodyne lock-in amplifiers where phase-sensitive
detection is carried out at a fixed high frequency. We shall also be giving
consideration to the spurious responses associated with synchronous
heterodyning. The treatment given falls short of a full analysis, but it is shown
how some of the major contributory factors can be overcome in practical systems.

8.2 Principles of heterodyne operation

The principles of heterodyne lock-in amplifiers can be established in terms of the
idealized system shown in Fig. 8.1.
The frequency translator has inputs from the signal channel amplifier, from the
applied reference signal and from an internal oscillator. The latter operates at
frequency fI which we shall call the intermediate frequency. The purpose of the
frequency translator is to produce an output with magnitude proportional to the
signal input but with its frequency shifted from fs to a new and higher value
fI fR fs . The translated signal is then applied to a phase-sensitive detector
which we assume is referenced and phase-shifted at the intermediate frequency.

Fig. 8.1 Basic heterodyne lock-in amplifier, showing the frequencies at

different points in the system

When the signal and reference are fully synchronous, fs fR . The translated
output then appears at frequency fI and yields a classic phase-sensitive response
at the phase-sensitive detector.
Chapter 82
For signals with frequencies close to fR the response will be an alternating output
at frequency fR fs , which is attenuated in the usual way by the output low-pass
filter. However, in this system there is no scope for unwanted responses when the
signal frequency is coincident with an odd harmonic of fI . The latter are included
on the assumption that a switching phase-sensitive detector is used with its
associated harmonic transmission windows.
The critical signal frequencies are therefore those which satisfy the relationship:
fI fR fs KfI

where K is an odd integer. Solving for fs , we obtain the critical frequencies:

fs 1 K fI fR

and,
fs 1 K fI fR

For each value of K there are therefore two frequencies where an interference
component will be able to excite a harmonic response in the phase-sensitive
detector. The relative sensitivity of the system to inputs at these critical
frequencies is denoted by S K and is given by
SK 1/ K

Fig. 8.2 Location and relative magnitude of transmission windows in a

heterodyne system, for K = 1, 3, 5, 7 and 9. fI = 10fR

The transmission windows of the overall detection system are thus derived from
the phase-sensitive detector windows and are located at frequencies determined
jointly by the intermediate and reference frequencies. The transmission windows
corresponding to K in the range 1 to 9 are drawn schematically in Fig 8.2 for the
relationship, fI 10 fR . The weighting factor assigned to each window is shown in
each case and gives the relative sensitivity of the detection system to inputs at the
appropriate critical frequency.
Fig. 8.2 reminds us that there must be two transmission windows, corresponding
to K 1 , where the relative sensitivity is unity. These windows correspond to the
"primary" response at fs fR , and the "image" response given by signals with
frequency fs 2 fI fR . An image response occurs in all systems operating on the
heterodyne principle. In this case, a signal for which fs 2 fI fR yields a
translator output close to the intermediate frequency and the resulting behaviour
of the detection system cannot then be distinguished from the "true" response

Chapter 83
when fs fR . The only satisfactory way to deal with the image response is to
eliminate signal components at the image frequency by filtering. The most
convenient arrangement is to use a low-pass filter in the signal channel with a
sharp cut-off defined at a frequency below 2 fI fR .
This filter must be introduced before the frequency translator and should have a
cut-off frequency greater than fRMAX, the maximum anticipated value of the
reference frequency. An image filter is an essential component in a heterodyne
lock-in amplifier, irrespective of the precise system configuration.
It turns out that in the present, ideal, case a properly designed image suppression
filter would be effective in suppressing inputs at all other critical frequencies.
This ideal heterodyne system is thus inherently free from responses at harmonics
of the reference frequency and can be made relatively immune to the incidence of
spurious responses at other, non-related, frequencies.

8.3 Practical considerations

8.3.1 Frequency translation
When discussing principles of operation it was assumed that the frequency
translator had the characteristics of a single-sideband generator; a single
component at frequency fs gave rise to a translated output at a single frequency
fI fR fs . Single-sideband generators can be devised and constructed to operate
with the required degree of precision. Unfortunately, the implementation of such
a scheme to operate over a wide frequency range is both complex and expensive.
A more cost-effective solution is to use double-sideband generation in which the
required translation is achieved with a signal channel mixer and frequency
synthesizer as shown in Fig. 8.3.

Fig. 8.3 Heterodyne system with double-sideband frequency translation

The frequency synthesizer provides an output at a precisely defined frequency

fI fR which yields mixer products at frequencies fI fR fs . When the signal
and reference are fully synchronous, the mixer output will consist of two
components, one at frequency fI and the other at fI 2 fR . The first of these will
give rise to a phase-sensitive response as before. The second gives rise to an
alternating output at frequency 2 fR . This is no more serious than the "ripple"
component which is expected in conventional phase-sensitive detector operation.
As is usually the case, the ripple component is rejected by the low-pass filter in
the phase-sensitive detector output.
Turning now to asynchronous signals, we find an important consequence of using
a double-sideband generator. This is a reduction in the allowable voltage swing
due to asynchronous components at the input to the signal channel. Each of these
Chapter 84
gives rise to two mixer products of equal amplitude. The input voltage swing at a
given sensitivity is thus limited to half the value which could be sustained when
using the phase-sensitive with a single-sideband frequency translator. For a
phase-sensitive detector of given characteristics the achievable dynamic reserve
is consequently reduced by 6 dB.
Further complications arise in a practical implementation of the frequency
translator. The signal channel mixer, like the phase-sensitive detector, is almost
invariably a switching multiplier which uses a squarewave drive at frequency
fI fR . This is another example where linearity and dynamic range are obtained
through the adoption of a switching operation. The result is that the output of the
signal mixer consists of a large number of components at frequencies
N fI fR fs , N odd

The amplitude of each of these components is weighed by a factor 1/N, reflecting

the reduction in the harmonics of the switching waveform with increasing order.
As we shall see, the use of a switching mixer leads to a reappearance of the
harmonic responses which were so successfully rejected by the ideal heterodyne
scheme.
8.3.2 Formulation of spurious responses
Spurious responses will be generated when one of the mixer products has a
frequency coincident with the intermediate frequency or one of its odd
harmonics. Denoting the odd harmonics by KfI as before, we obtain the critical
frequencies:
fs N K fI NfR

and
fs N K fI NfR

To calculate the relative sensitivity of the detection system to inputs at these

critical frequencies, we must take the following factors into account: First of all, a
factor 1/N resulting from the use of a squarewave drive to the mixer; secondly a
factor 1/K to allow for the reduction of the phase-sensitive detector transmission
windows with increasing harmonic order. The relative sensitivity at frequencies
corresponding to given values of N and K is therefore:
SN,K 1/ NK
For example, when N = K = 1 we obtain the primary response corresponding to
fs = fR and the image response corresponding to fs = 2fI + fR. The relative
sensitivity of the system to inputs at these frequencies is unity in both cases.
Let us now take N K 1 . In this case, the system will be sensitive to inputs at
fs = NfR and 2NfI NfR . The response to the Nth harmonic of fR is thus reinstated
at a relative sensitivity of 1/N2, compared to the figure of 1/N which would be
obtained in a conventional system.
As noted in section 8.2 an image suppression filter would be essential in any
practical system. Unfortunately this has no influence on signal components at
frequencies NfR that fall within the filter bandwidth. The resulting harmonic
responses can only be satisfactorily suppressed by using a second, tuned, filter in
front of the phase-sensitive detector.

Chapter 85
8.3.3 Suppression of the spurious responses
We envisage a system such as that shown in Fig. 8.4 which includes an image
suppression filter in the signal channel and a tuned filter following the signal
mixer.

Fig. 8.4 Incorporation of tuned filter in a heterodyne system

The image suppression filter is a low-pass filter which exhibits a "flat" amplitude
response for all signal frequencies up to the maximum value of the reference
frequency as shown in Fig. 8.5.

Fig. 8.5 General transmission characteristics of an image-suppression

filter

The filter is designed for certain minimum attenuation at the lowest expected
image frequency and gives progressively higher attenuation for frequencies
beyond the image frequency. An attenuation of the order 60 to 80 dB at the image
frequency would be typical of practical systems. We thus find that asynchronous
signals characterized by frequencies:
fs N K fI NfR ; N, K 1

all lie beyond the cut-off of the image filter and are effectively eliminated before
the signal mixer.
The only asynchronous signals which are likely to yield discernable responses
must therefore have frequencies which satisfy the condition:
fs N K fI NfR

The nature of the responses corresponding to different values of N and K are

summarized in Table 8.1.

Chapter 86
Table 8.1 Catalogue of responses for asynchronous signals with frequencies
|(N K)fI + NfR|

N, K Comments

N K 1 fs fR : primary respoonse of system

N>K Asynchronous signals always have frequencies image

frequency

N>K1 Spurious responses whenever asynchronous signal has

frequency fs NfR up to bandwidth of image filter

K>N Miscellaneous spurious responses for asynchronous signals

lying within the bandwidth of image filter

The first of the categories listed in Table 8.1 corresponds to the "wanted"
response of the system while signals in the second category are suppressed by the
image filter. In the absence of a tuned filter, the system would display a relative
sensitivity SN,K 1/ NK at all critical frequencies in the final two categories. The
effect of a tuned filter will be to introduce additional attenuation of these
responses giving a relative sensitivity:
SN,K H jKI / NK

Here, H j is the frequency response of the tuned filter normalized to a

magnitude of unity at the intermediate frequency.
We note that the largest spurious response will be obtained when N K 3 ,
corresponding to an asynchronous signal with frequency 3 fR *. The tuned filter
should therefore be set up to reduce this response to an acceptably small value.
8.3.4 Tuned filter requirements
In all commercial systems the tuned filter is of the low-pass type described in
Appendix 4. The use of a tuned filter for suppressing the harmonic responses of a
phase-sensitive detector was discussed in Section 4.5.2. For a low-pass filter
tuned to the intermediate frequency fI , the attenuation at frequency KfI in the
high-Q approximation is:
1
H jK I 2 , K 3, 5, 7, K
(K 1)Q
The relative sensitivity of the heterodyne system, SN,K , can now be written as
1
SN,K
NK(K 2 1)Q
The value of the Q-factor required for a given level of suppression can thus be
calculated.
For example, suppose the system is required to have a relative sensitivity of 10-4
(80 dB) to a signal with frequency close to 3 fR . From Table 8.1 we put
N K 3 and calculate the required Q-factor of the tuned filter:

*
Note that the response due to K = 3, N = 1 corresponds to an asynchronous signal at frequency
2fI fR.This would normally be heavily attenuated by the image filter.

Chapter 87
 Q 10 4 /(3 3 8) 139
From the discussion given in Section 4.5.1 it is clear that operation with such a
high value of Q-factor places severe demands on the system with regard to
alignment and maintaining good phase accuracy. This is all the more troublesome
in a purpose-built system where the filter is not usually accessible for routine
realignment by the user.
An alternative approach which greatly eases alignment problems and which
places less demands on filter performance has been used in the EG&G Brookdeal
heterodyne system. This is to use two filters of relatively low Q-factor in cascade.
The attenuation introduced by the filter stage then becomes:
1
H jKI
(K 1) 2 Q 2
2

In this case, a Q-factor of 5 is sufficient to give a relative sensitivity of 83 dB to

signals with frequency fs 3 fR .
Of course, these figures for harmonic rejection are strictly theoretical. There is a
considerable technical hurdle to be overcome in order to realize these figures in
practice and commercial systems usually specify suppression factors of around
60 dB (1/1000). In any specific case, it is always worth checking whether the
figure given refers solely to third-harmonic rejection or to the maximum level of
spurious responses arising from all possible sources.
8.3.5 Phase-shifting
It has been assumed so far that the user-controlled phase-shifter which is essential
for phase-sensitive detection is introduced at the intermediate frequency, on the
"reference" side of the phase-sensitive detector. This has a distinct advantage in
that the phase-shifter can be designed to operate at a fixed frequency rather than
over a wide range of frequencies as is usually the case.
In principle, there is no reason why phase-shifting should not be carried out at the
original reference frequency or, indeed, on the output of the frequency
synthesizer, at frequency fI fR . These are all found to be equivalent when
system operation is analysed. Phase-shifting at the reference frequency requires
similar broadband circuitry to that found in a conventional lock-in amplifier,
whereas a phase-shifter placed at the output of the frequency synthesizer,
assuming fI fRMAX , would require relatively narrowband capability.

8.4 Practical limitations

8.4.1 The frequency synthesizer
The generation of a waveform at a precisely defined frequency fI fR in response
to an external signal applied at frequency fR is the most demanding task facing the
designers of heterodyne lock-in systems. The ultimate objective is to produce a
system which:
(i) Maximises the range of reference frequencies which can be handled at a
particular intermediate frequency.
(ii) Has an acceptably low level of both random and discrete phase-noise over the
specified frequency range.
(iii) Minimizes phase errors between the absolute phases of the synthesized
waveform and the applied reference waveform.
(iv) Has an acceptable value of reference slew rate and a small acquisition time.

Chapter 88
A shortfall in any of these areas would be noticed by any user whose interest in
lock-in amplifiers extended beyond the detection of noisy signals at a fixed
frequency. The designer's difficulty is to reconcile these requirements and make
the correct trade-offs to produce a system with all-round acceptable performance.
The problems encountered in reaching an acceptable compromise in synthesizer
performance can be highlighted by referring to the system illustrated in Fig. 8.6.

Fig. 8.6 Heterodyning phase-locked loop

This example of a "heterodyning" phase-locked loop received considerable

publicity when the Ithaco Dynatrac lock-in amplifiers were first introduced1.The
phase detector compares the phase of its two inputs and its output is amplified
and smoothed by the loop filter. The v.c.o. is controlled from the loop filter
output and produces an output at frequency fV , which is mixed with a signal at
the intermediate frequency fI provided from a stable sinewave oscillator. The
purpose of the low-pass filter is to transmit only the low-frequency mixer product
at frequency fV fI to the phase detector. The difference frequency fV fI is thus
phase-locked to the incoming reference signal at frequency fR , making fV equal
to fI fR .
When a strictly conventional approach is taken to designing the phase-locked
loop, following the treatment given in Chapter 7, the designer is faced with a
difficult decision in choosing the loop characteristics. The loop bandwidth should
be small enough to suppress discrete phase modulations resulting from unwanted
mixer components at the output of the low-pass filter, and yet wide enough to
accommodate the phase noise inherent in the v.c.o. and to give adequate slew rate
performance. In addition, severe phase errors might be incurred when the low-
pass filter achieves the desired separation of mixer products by virtue of a sharp
cut-off. Almost inevitably, the trade-offs are such that acceptable performance
can only be achieved over a relatively small range of reference frequencies.
Limitations are observed in terms of poor phase tracking and excessive phase
noise at the extremes of the operating frequency range.
In the loop used by EG&G Brookdeal, switching waveforms derived from the
applied reference and from the intermediate-frequency oscillator are combined in
a multiplexer. The output is a switching waveform in which the positive
transitions occur an average rate ( fI fR ) / 4 . This entire waveform is then used
to phase-lock a v.c.o. in a phase-locked loop which incorporates a 4 counter. In
a conventional phase-locked loop, the output from the v.c.o would be a
squarewave at frequency fI fR subjected to a high level of discrete phase-
modulation which could only be suppressed by using a loop of small bandwidth.
In the EG&G Brookdeal system the unwanted modulation is well defined and can
be suppressed to a high order by adding a compensation signal to the output of
the phase detector. Provided the compensation signal is accurately generated and
controlled, the constraints on loop bandwidth are considerably relaxed. The loop
characteristics can then be optimized with respect to a low incidence of phase
noise and obtaining acceptable handling characteristics over a wide range of
reference frequencies. The system, in fact, proves capable of operating with

Chapter 89
reference frequencies from less than 1 Hz up to about one fifth of the
intermediate frequency. The quoted figures of phase noise and phase drift for the
overall lock-in amplifier are similar to those given in Chapter 4 for conventional
systems. Also, because the method of generation does not rely on the use of
filters to separate mixer products, the frequency synthesizer itself does not
contribute a reference frequency-dependent phase error of major significance.
8.4.2 The image filter
It has been noted that the image filter should have a flat response up to signal
frequencies corresponding to fRMAX. The rate of cut-off beyond this point must
then be extremely large in order to introduce adequate attenuation at the image
frequency. In most commercial systems the ratio fI / fRMAX is less than 10, so that
a filter of high order, 4 to 6 pole, is required to achieve the necessary roll-off.
Like all filters used for signal conditioning, the image filter introduces phase
errors into the measurement system. When the filter is of high order, the phase
error can be in excess of 100 at about one half of the cut-off frequency. A
fortunate consequence of using high-order filters is that the amplitude response
can be made uniformly flat up to frequencies very close to cut off. In addition, the
phase shift within this range can be made proportional to frequency. When the
filter approximates to such a linear-phase model, the phase-shift of the signal
channel can be compensated by introducing a suitable time delay in the reference
channel. In the Ithaco Dynatrac, the image-filter phase characteristic was actually
compensated by the characteristic of the low-pass filter in the synthesizer phase-
locked loop. In other systems, such as that from EG&G Brookdeal, the
synthesizer has inherently low phase error and the compensating time delay need
be no more complicated than a monostable circuit operating at a fixed pulse
width.
When we take matching constraints into account and add uncompensated phase
errors accrued in the intermediate-frequency filter and in the synthesizer, we
conclude that the overall phase precision of heterodyne systems must fall short of
that obtainable in a conventional lock-in amplifier operating in the same
reference frequency range.
8.4.3 The signal mixer
The most serious performance limitations associated with the signal mixer are
due to non-linearity and "feedthrough". It has been stressed elsewhere that the
synchronous demodulation process should be supported by linearity in all
preceding stages; hence the linearity of the signal mixer should be at least of the
same order as the linearity of the phase-sensitive detector. This requirement is
eased in practice because the two components, being similar in concept, draw
upon similar technologies.
"Feedthrough" is a phenomenon associated with voltage offsets and capacitive
coupling in the signal mixer, whereby components at the switching frequency,
fI fR , appear at the mixer output in the absence of signal, and independently of
signal channel gain selection.
Feedthrough is not normally specified explicitly but its effect becomes evident
when operating a heterodyne lock-in amplifier at high dynamic reserve. It
appears as an alternating component at frequency fR in the output of the system
and appears at its worst when operating at low reference frequencies with a
relatively short time constant selected.
A level of feedthrough of 105 (100dB) represents a reasonable target for a
signal channel mixer operating with an intermediate frequency in excess of
100 kHz. When used in conjunction with a phase-sensitive detector having a

Chapter 810
dynamic reserve of 104 (80 dB) the output a.c. variation at frequency fR could
have a maximum value of 1/10 (20 dB) of full scale. Unless the intermediate-
frequency tuned filter has unusually high Q-factor - and correspondingly narrow
bandwidth - it is unlikely to have much influence on the level of feedthrough
reaching the phase-sensitive detector.
8.4.4 The phase-sensitive detector
It was noted in Chapter 3 that the dynamic range of switching phase-sensitive
detectors is reduced at high reference frequencies owing to the appearance of the
so-called "h.f. offset". The mechanisms at work here are very similar to those
which give rise to feedthrough in switching mixers. It is claimed, quite
reasonably, that in heterodyne lock-in amplifiers operating at a fixed value of
intermediate frequency, the dynamic range is constant over the entire range of
signal and reference frequencies. It should also be said, however, that the phase-
sensitive detector is fated always to operate at a relatively high frequency where
its dynamic range is less than optimum. In the Ithaco system, operation in a given
frequency range involved selecting plug-in circuit cards to provide a suitably high
value of intermediate frequency. Not surprisingly, the figures for dynamic range
showed a marked deterioration as the system was configured to operate at higher
reference frequencies and correspondingly higher values of intermediate
frequency.
In the EG&G Bookdeal heterodyne lock-in amplifier the maximum intermediate
frequency is 1 MHz, which allows operation with reference frequencies over the
entire audio-frequency range up to 200 kHz. In this case, the dynamic range of
the phase-sensitive detector is maintained at a competitively high value
(>120 dB) by a technique known as synchronous heterodyning. It is shown in
Section 8.8 that this approach can be used to improve the dynamic range of
phase-sensitive detectors operating in either a conventional lock-in amplifier or in
a heterodyne system. Synchronous heterodyning is not an alternative to the
heterodyne mode of operation described so far; rather it is a supplementary
technique applied with the objective of improving dynamic range.

8.5 Overload capability of heterodyne systems

It was the advent of the Ithaco Dynatrac that focused popular attention on the
specification of overload capability in lock-in amplifiers. Clearly, a simple
statement of phase-sensitive detector dynamic reserve is not sufficient when a
lock-in system is so heavily supported by filters.
From the point of view of overload capability, the image suppression filter in any
heterodyne lock-in amplifier can be regarded as a low-pass signal conditioning
filter. The rate of cut-off of the image filter is extremely high and this leads to a
dramatic improvement in overload capability for asynchronous signals having
frequencies greater than about one half the intermediate frequency.
The selectivity of the tuned intermediate-frequency filter is chosen solely on the
grounds of suppressing harmonic responses. Claims that the tuned filter is
responsible for a significant increase in overload capability should therefore be
treated with caution, particularly when the reference frequency is very low
compared with the intermediate frequency. In a heterodyne system the overload
characteristics are similar to those of a conventional lock-in amplifier using a
tuned filter at the reference frequency with a bandwidth equal to that of the
intermediate frequency filter, fI /Q . The effective Q-factor is thus dependent on
the reference frequency and is given by Qeff Q fR / fI . As a result, the filter is
likely to have a significant effect on overload capability only when the reference
frequency approaches its maximum value; even then the maximum reference

Chapter 811
frequency fRMAX should represent a substantial fraction of fI . Moreover, the Q-
factor of the tuned filter itself should also be high; if sharp cut-off is obtained by
cascading low-Q sections as described in Section 8.3.4. the overload
characteristics will be broadly independent of frequency within the bandwidth of
the image-suppression filter.
The interference rejection characteristics plotted in Fig. 8.7 serve to emphasize
these points. The graphs are drawn for a system in which fI 5 fRMAX and the
intermediate-frequency tuned filter has a Q-factor of 20. These conditions
approximate to those it the Ithaco Dynatrac lock-in amplifier. In a system such as

Fig. 8.7 Interference rejection in a heterodyne lock-in amplifier

(a) fR = fRMAX/10; (b) fR = fRMAX/2; (c) fR = fRMAX

the EG&G Brookdeal heterodyne, the reference frequency is allowed to take

values up to 6 decades below the intermediate frequency. Rejection
characteristics of type (a) in Fig. 8.7 are thus applicable over most of the
operating frequency range.

8.6 Double heterodyne lock-in amplifiers

In the heterodyne system described so far, the final detection has been carried out
at the intermediate frequency. In a double heterodyne detection is carried out at
the original reference frequency as shown in Fig. 8.8.

Fig. 8.8 Double-heterodyne lock-in amplifier showing the principal

frequencies of interest

Chapter 812
The intermediate frequency stage incorporates a filter of extremely high
selectivity. In the case of the EG&G PAR Crystal-Het lock-in amplifier a two-
section crystal filter is used to define a Q-factor of 50 000 at the intermediate
frequency of 1MHz. The resulting 20 Hz bandwidth is significantly smaller than
the reference frequency over much of the 100kHz frequency range, and thus has a
profound effect in removing harmonically related components and noise before
detection.

Fig 8.9 Effective bandpass response at different values of reference

frequency

The system can be thought of in terms of a tracking filter with a fixed bandwidth
of 20 Hz, giving an effective Q-factor, Qeff 50 000 fR / fI . The effectiveness of
the system in rejecting interference components is illustrated in Fig. 8.9 which is
drawn for different reference frequencies.
As might be expected, the phase-tracking of such a narrowband system is greatly
inferior to that of a conventional lock-in amplifier. The level of spurious
responses is nevertheless very low at midband. This is evident from the results
given in Fig. 8.10 for operation at a reference frequency of 1kHz. The largest
spurious response in this case occurs at a level of 70dB relative to the "primary"
response at 1kHz.

Fig. 8.10 Spurious responses of the EG&G PAR Crystal-Het at a reference

frequency of 1kHz

Chapter 813
8.7 Brief comparison of single and double
heterodyne systems
It is evident that the role of the intermediate frequency tuned filter is quite
different in the two cases. In the single heterodyne system, which has attracted
most of our attention, the tuned filter is essential in order to maintain relative
freedom from unwanted responses, particularly those occurring at the odd
harmonics of the applied reference frequency. The system then approximates to a
fundamental-only responding lock-in amplifier over the entire operating
frequency range. As we have seen, however, the tuned filter might have only a
marginal influence on overload characteristics, becoming effective in a limited
way towards the high end of the operating frequency range.
In a double heterodyne lock-in amplifier the tuned signal has similar status to a
tuned filter incorporated in the signal channel of a conventional lock-in amplifier.
The considerations of Section 4.4 thus apply provided that allowance is made for
the variation of the effective Q-factor with operating frequency. The essential
difference between the two heterodyne schemes is that, in the double heterodyne
system, harmonic rejection depends on the effective Q-factor; in a single
hetrodyne, harmonic rejection depends only on the actual Q-factor of the tuned
filter measured at the intermediate frequency. We can conclude from this that the
double heterodyne approximates to a fundamental-only responding system only
for reference frequencies greater than a specified value. In the case of the EG&G
PAR Crystal-Het system referred to earlier, this minimum value of reference
frequency must certainly be greater than 20 Hz, which limits the system's
effectiveness in the critical low-frequency regime identified in Chapter 6.
It is also evident that alignment problems in a double heterodyne system are far
worse than those encountered in a single heterodyne operating at the same
intermediate frequency and covering the same range of reference frequencies. As
we have seen, the double heterodyne lock-in amplifier requires a tuned filter with
a Q-factor several orders of magnitude higher than that required by a single
heterodyne system. In the case of the Crystal-Het system the crystal filter and the
crystal-controlled intermediate-frequency oscillator are based on carefully
matched components and a high level of performance is maintained by matching
temperature coefficients of drift and holding the devices at a controlled
temperature. The benefits are obtained in the form of greatly improved overload
capability and a much lower level of spurious responses compared with a single
heterodyne lock-in amplifier.

8.8 Synchronous heterodyning

8.8.1 Introduction
In the mid 1970s EG&G PAR introduced a lock-in amplifier known as the
Syncro-Het which offered a considerable improvement in dynamic range over
conventional lock-in amplifiers. The system enjoyed success in its own right, but
attracted little in the way of direct competition. It appears that no attempt was
made to refine the basic technique or to develop further commercial instruments
operating on the synchronous heterodyne principle. This was the situation until
recently when synchronous heterodyning was revived in the context of high-
frequency lock-in systems. The objective was stated in Section 8.4.4; namely to
improve the dynamic range of phase-sensitive detectors operating at a high
frequency.
It was remarked in Section 3.6.5 that the d.c. response to a synchronous signal
can be separated from offset voltages at the output of a phase-sensitive detector
by introducing a phase reversal of 180 in the reference channel. This approach to
Chapter 814
signal detection forms the basis of the synchronous heterodyne system illustrated
in Fig. 8.11. Here, the phase reversal is introduced by systematically switching
the signal with a squarewave taking values +1 and 1. The switching is carried
out at a frequency fSYN. Final detection takes place in the phase-sensitive detector
referenced at fSYN. The output, shown in Fig. 8.12(f), is smoothed by the action
of the output low-pass filter. The system thus yields a conventional phase-
sensitive response as the relative phase-shift of the signal and reference inputs is
varied at frequency fR .

Fig. 8.11 Principles of synchronous heterodyning

Fig. 8.12 Waveforms in the system of Fig. 8.11

In this mode of operation we find that the phase-sensitive detector can be

constrained to operate in a frequency range well below the actual value of the
reference frequency fR . For example, phase-sensitive detection of a signal at
100 kHz could be achieved with a "syncrohet" frequency, fSYN, of 100 Hz. The
overall dynamic range would then be equivalent to that obtainable when
operating the phase-sensitive detector in a conventional fashion at 100 Hz rather
than the higher value of 100 kHz. It is this aspect of operation that has made the
synchronous heterodyne approach so attractive in heterodyne systems where the
phase-sensitive detector is referenced at a high value of intermediate frequency.
8.8.2 Dynamic range improvement
The synchronous heterodyne system described so far offers no inherent
improvement in dynamic range at low reference frequencies. This was obtained

Chapter 815
in the EG&G PAR Syncro-Het system by imposing a filter between the final
mixer and the phase-sensitive detector as shown in Fig. 8.13.

Fig 8.13 Improvement of dynamic range by using an interstage filter in a

synchronous heterodyne lock-in amplifier. The filter is of the
"synchronous" or "rotating capacitor" variety switched at
frequency fSYN

The filter used in the Syncro-Het system is a so-called "synchronous" or "rotating

capacitor" filter2 which is supplied with a reference at the final detection
frequency fSYN. The synchronous filter has the correct response characteristics to
transmit the components of a squarewave at frequency fSYN, while attenuating all
asynchronous components due to noise, signal related "ripple", and drift in the
mixer stages. The filter output thus conforms closely to a squarewave with a
amplitude proportional to the in-phase component of the synchronous signal. The
squarewave is, moreover, relatively free from noise and can be subjected to a
further stage of a.c. amplification before detection. To restore the sensitivity, the
output d.c. gain can then be reduced in proportion to the extra a.c. gain supplied
in the signal path. The result is a system which offers improved output stability at
a given level of dynamic reserve. In the light of the definitions given in
Chapter 3, this is equivalent to a system with improved dynamic range.
The Syncro-Het lock-in amplifier operated with fSYN = 11 Hz and proved capable
of 3000 (70 dB) dynamic reserve, consistent with an output stability of
10 p.p.m. This is equivalent to an input dynamic range of a massive 3.108
(170 dB). Unfortunately, the overall performance was marred by a series of
spurious low-frequency outputs which occurred for critical combinations of the
applied reference frequency and the internal "syncrohet" frequency fSYN.
Mixer feedthrough, giving rise to a residual mixer output in the absence of signal,
is a major cause of these low-frequency "beat" products. In general, the residual
output from the second mixer will comprise components at combinations of fR,
fSYN and their odd harmonics, characterized by frequencies:
MfSYN NfR ; M , N odd
The phase-sensitive detector is referenced at frequency fSYN and is responsive to
inputs at frequencies K fSYN where K is an odd integer. Spurious responses will
occur whenever fR takes values such that:
MfSYN NfR KfSYN

It should be stressed that the effect of mixer feedthrough is observed in the

absence of signal and that spurious responses are obtained for critical values of
the applied reference frequency, given by:

Chapter 816
 M K
fR fSYN
N
and
M K
fR fSYN
N
Since M, N and K are odd we find that spurious responses can be avoided (at least
with respect to mixer feedthrough) provided that fSYN is constrained to be an odd
submultiple of fR . If this condition is established, the worse-case effect will be a
beat product from the phase-sensitive detector at frequency fSYN. This response
will not be discernible provided fSYN is chosen to be greater than the maximum
bandwidth of the low-pass filter in the output of the phase-sensitive detector.
Clearly, such a well-defined relationship is lacking in the Syncro-Het lock-in
amplifier, where fSYN is fixed and fR is allowed to take values over a wide range.
As we shall see, however, the constraint is not a serious one when synchronous
heterodyning is applied to a heterodyne lock-in amplifier operating at a fixed
intermediate frequency.
8.8.3 Application to heterodyne lock-in amplifiers
The implementation of synchronous heterodyning in the EG&G Brookdeal
heterodyne lock-in amplifiers follows the block diagram shown in Fig. 8.14(a).
The phase sensitive detector operates at a frequency fSYN which is well below the
highest intermediate frequency of 1MHz. The complexities of interstage filtering
are avoided, and further reduction in complexity is achieved by using only one
mixer as opposed to the double mixer stage assumed earlier.

Fig. 8.14 (a) Incorporation of synchronous heterodyning in a heterodyne

lock-in amplifier. Reference processing and provision for phase-
shifting are omitted for clarity
(b) Generation of the switching input to the syncrohet mixer
fSYN is obtained by odd-integer frequency division from fI to
minimize spurious responses. In practice, fSYN will be several
orders of magnitude less than the highest value of intermediate
frequency

Chapter 817
In our original description of synchronous heterodyning, the signal was mixed
successively with signals at frequencies fSYN and fR. For a signal s(t), the output
from the second mixer has the form:
s(t) R1(t) R2(t)
where R1(t) and R2(t) are squarewaves at fSYN and fR .
The alternative approach, using only one mixer, is to multiply the signal directly
by a two-state waveform:
R3(t) = R1(t) R2(t)
In practice, R3(t) can be formed by combining the individual switching
waveforms R1(t) and R2(t) in an exclusive-NOR circuit. The resulting waveforms
are derived in Fig. 8.14(b). In the EG&G Brookdeal heterodyne lock-in amplifier,
fSYN is chosen to be about 1 kHz, to lie beyond the bandwidth of the output filter,
and is generated by odd-integer frequency division from the intermediate
frequency oscillator. As explained in the last section this minimises the incidence
of spurious outputs resulting from feedthrough in the syncrohet mixer. The
method of generation avoids the coincidence of transitions in the switching
waveforms R1(t) and R3(t) which is, in itself, a source of spurious outputs in
systems where fSYN takes arbitrary values.

8.9 Conclusions
This chapter has shown how a simple idea, that of frequency-shifting to avoid
harmonic responses, has been progressively modified to allow for practical
limitations in the performance of the various subsystems which comprise a
heterodyne lock-in amplifier. The result is an instrument operating at a level of
complexity greatly in excess of a conventional broadband lock-in amplifier and
which falls short of conventional systems in several important specification areas,
notably dynamic range and phase accuracy.
These drawbacks, to judge from the popularity of heterodyne systems, are more
than offset by the advantages in operating with fundamental-only response allied
to a low level of spurious responses. In terms of all-round performance, therefore,
heterodyne systems confirm most closely with the characteristics demanded of a
"general purpose" measurement tool capable of making unambiguous
measurements with a variety of signal types over a wide range of operating
frequencies. The availability of these instruments with a comprehensive facility
for digital control adds to their appeal in a wide range of applications. Some of
the characteristics of these systems are reviewed in Chapter 10, while a
comparison with p.w.m. systems, which offer an alternative approach to
achieving fundamental-only response, is given at the end of Chapter 9.

8.10 References
1 MUNROE, D.M. (1973): "The heterodyning lock-in amplifier". Technical
Bulletin, Ithaco Corp., Ithaca, NY.
2 FRANKS, L.E. and SANDBERG, I.W. (1960): "An alternative approach to
the realization of network transfer functions: The N-path filter", Bell Syst.
Tech. J., 39, pp. 1321-1350.

Chapter 818
CHAPTER 9

P.W.M systems

9.1 Introduction
Single and double heterodyne systems were developed to overcome the problem
of harmonic responses in lock-in detection while retaining the ability to track
signals over a wide frequency range. In this chapter, we shall be looking at an
alternative method of suppressing harmonic responses, by using pulse-width
modulation in the reference channel.
It will become apparent that heterodyne and pulse-width modulation systems
have rather different status. Whereas heterodyne lock-in amplifiers have a system
configuration which is far removed from the "conventional" arrangement, the
pulse-width modulation approach requires only a relatively simple modification
to an otherwise basic system. As a result, the pulse-width modulation, or p.w.m.,
circuitry can be supplied as an option to a conventional lock-in amplifier, leaving
the user to select either conventional response or fundamental-only response as
required. As in the case of heterodyne systems, the suppression of harmonic
responses is achieved at the expense of spurious responses at apparently arbitrary
frequencies and is attended by a loss of dynamic range. We shall therefore be
examining some of the trade-offs which must be made to achieve a detection
system with overall acceptable characteristics.
P.W.M. lock-in amplifiers are characterized by phase accuracy of a very high
order and represent a solution to suppressing harmonic responses which is both
cost-effective and flexible. Other benefits, such as the potential to operate with
greatly improved slew rate and the ability to operate the phase sensitive detector
as an "ideal" multiplier, will also be noted.
A brief comparison of p.w.m. systems with heterodyne lock-in amplifiers is given
at the end of this chapter.

9.2 Principles of operation

Fig. 9.1 Effect of reference waveform symmetry on the average gain of a

phase-sensitive detector shown for the signal at a fixed level, V1
volts. The average gain is (a) zero; (b) +m; (c) m

The harmonic responses of a phase-sensitive detector result from the abrupt

change of gain between +1 and 1 which occurs whenever the reference

Chapter 91
switching waveform changes polarity. If we were to measure the average gain of
the phase-sensitive detector over many consecutive reference cycles the result
would of course be zero, provided that the reference switching waveform spent
an equal amount of time in its two states. As shown in Fig. 9.1(a), (b) and (c) the
effect of changing the symmetry of the switching waveform is to change the
average gain of the phase-sensitive detector to a value somewhere between the
extremes of +1 and -1.

Fig. 9.2 (a) Modulation voltage; (b) pulse-width-modulated switching

waveform

Suppose now that the waveform symmetry is subject to a long-term variation

imposed by modulating the switching waveform as shown in Fig.9.2. The
average gain of the phase-sensitive detector (Measured over a time which is long
compared with a switching cycle but short compared with a modulation period) is
now subject to a continuous variation which is free from discontinuities. The
variation in gain, moreover, reproduces the modulation waveform exactly.
This approach, the pulse-width modulated reference channel, provides a means of
achieving a sinusoidal gain variation in the signal path while retaining the
dynamic range benefits of a switching phase-sensitive detector. A sinusoidal
variation has been chosen because this clearly brings us a step closer to a system
with fundamental-only response. To take a broader view: the system described
approximates to an ideal multiplier model in which the reference input could be
of any waveform, supplied in the form of a modulation voltage. Fig. 9.3 shows
this configuration of a lock-in system working on this principle.

Fig. 9.3 Lock-in amplifier operating on the p.w.m. principle. The output
is a low-pass filtered version of the product of the signal s(t) and
a general reference waveform r(t)

Chapter 92
 9.3 Frequency composition of the p.w.m.
waveform
P.W.M. waveforms can be generated in a number of different ways. The
waveform shown in Fig. 9.2 is just one particular type of p.w.m. waveform which
has been used commercially by EG&G Brookdeal. Analysis of these waveforms
is generally a protracted business even for a sinusoidal modulation. For the
moment it will be sufficient to identify some characteristics which are shared by a
number of p.w.m. schemes that are suited to incorporation in lock-in systems.
The amplitude spectrum of these modulated switching waveforms can be divided
into two distinct frequency regions as shown in Fig 9.4. Most important from the
point of view of signal recovery is the low-frequency region which incorporates
the complete unmodified spectrum of the modulation waveform. The high
frequency region contains Fourier components related to the switching frequency
f0 and its harmonics, each of which carries sidebands derived from the Fourier
components of the modulation.
The switching waveform thus combines the Fourier components of the
modulation voltage with other, non-harmonically related, components at higher
frequencies. The latter account for the switching characteristics of the waveform
and each is associated with a transmission window where the detection system is
susceptible to interference components.

Fig 9.4 Amplitude spectra of: (a) modulation voltage; (b) a typical
p.w.m. switching waveform

For fundamental-only response, the modulation will be sinusoidal at frequency fR

which is usually much less than f0. We then identify the principal components of
the spectrum at frequencies:
fR , Kf 0 Lf R ; K odd, L 0, 1, 2, 3 ...
Each component in the spectrum, other than at fR, is located at a "critical"
frequency near which spurious responses will be obtained. As in the case of the
heterodyne systems discussed in Chapter 8, the ability to suppress responses at
odd harmonics of fR is offset to some extent by the appearance of responses at
unrelated frequencies.

Chapter 93
9.4 Basic design considerations
9.4.1 Dynamic range
In practical p.w.m. systems giving fundamental-only response the magnitude of
the component at frequency fR is found to vary linearly with the amplitude of the
modulation voltage which, in turn, determines the depth of modulation in the
p.w.m. waveform. The component at fR can take reasonably large values - of the
same order of magnitude as the dominant high-frequency components - without
causing over-modulation. This turns out to be very important in phase-sensitive
detection where the "primary" response at fs = fR should be as large as possible.
Failure to achieve this means that the overall detection system will suffer a
significant loss of sensitivity compared to operation with a conventional
squarewave reference. Some loss of sensitivity is inevitable in a p.w.m. system:
unfortunately this can be equated to a reduction in dynamic range. The reason is
that the input overload level to the phase-sensitive detector and the output
stability remain substantially unchanged as the modulation depth, and hence the
sensitivity, is varied. At a given modulation depth, the sensitivity can be restored
by using a larger gain factor either in the signal channel or in the post-detection
"expand" amplifier. From the considerations given in Chapter 4, we find that the
first approach results in a loss of dynamic reserve, leaving the output stability
unaffected, while the second option maintains dynamic reserve at the expense of
output stability. In either case, the net dynamic range of the system is lowered by
the same factor; the implication is that the p.w.m. waveform should be modulated
to the maximum possible depth in the interests of maintaining dynamic range.
This brings us to an important design trade-off. Increasing the depth of
modulation on the p.w.m. waveform reduces the loss in dynamic range but is
accompanied by a rise in the sideband components centred on the switching
frequencies and its harmonics. Any attempt to recoup dynamic range using this
approach is consequently matched by an increase in the general level of spurious
responses. In addition, high-order sidebands which had negligible magnitude will
now introduce transmission windows in a frequency range closer to the reference
frequency, and so assume greater practical importance.
9.4.2 Spurious responses
In general, the magnitude and extent of the sideband arrays centred on f0 and its
harmonics vary non-linearly with modulation depth and can only be determined
for a given modulation scheme by exact analysis. The scheme exploited by
EG&G Brookdeal in their Sinetrac systems is particularly difficult to analyse
because, in addition to modulating the mark/space ratio of the switching
waveform, the modulating signal also causes a shift of the carrier frequency. We
shall therefore restrict ourselves to a general review of system behaviour.
In the Sinetrac system, the sinewave modulation voltage at frequency fR is
specified at a standard level of 1 V r.m.s. The modulation depth corresponding to
this level of modulation signal gives a loss of 10 dB in the sensitivity of the
phase-detector. The sensitivity is regained by introducing an extra gain stage of
10 dB in the signal channel, with a consequent loss of 10 dB in the system
dynamic reserve.
When the modulation is at a reference frequency fR which is much lower than f0,
the sidebands centred on the carrier frequency are closely spaced at frequencies
f 0 Lf R but the extent of the sideband array is strictly limited as shown

Chapter 94
schematically in Fig. 9.5. The total width, 2f , is typically 20 kHz and
substantially independent of modulation frequency when fR takes sufficiently low
values, giving a spectrum similar to that of a frequency-modulated carrier under
conditions of large-index modulation.

Fig. 9.5 General form of the p.w.m. waveform spectrum at low reference
frequencies

This general behaviour, giving rise to a p.w.m. spectrum with sensibly fixed
characteristics, corresponds to operation with fR less than about f0/30. For
reference frequencies in this range, a clear separation is maintained between the
low- and high-frequency regions of the spectrum. The lowest critical frequency,
where a spurious response of significant magnitude could be obtained, is now in
the region of f0 f ,which is well removed from fR.
When fR is increased, keeping the depth of modulation constant, the separation of
the sidebands becomes correspondingly larger. In order to predict the critical

frequencies and the magnitude of their associated transmission windows, it now
becomes necessary to give individual attention to each of the sidebands. In
practice it will be the sidebands below the carrier frequency which prove to be
the most troublesome, located at frequencies f0 LfR.
It turns out that for sufficiently high reference frequencies, the major contribution
to spurious responses comes from the low-order sidebands for which L 4 . This
is illustrated by Fig. 9.6, drawn for the specific case where fR = f0/10. Here, the
separation between the low- and high-frequency regions of the p.w.m. spectrum
is much less well defined. Also, it is evident that, if fR exceeds a certain value,
one of the sidebands will enter the frequency range below fR.

Fig. 9.6 Major sideband components for fR = f0/10 , shown relative to the
"primary" component at fR

Note that, in the Sinetrac system, application of the modulation voltage shifts the mean carrier
frequency by about 10% from its "free-running" value. We shall ignore this effect in the following
discussion.
Chapter 95
 To avoid the possibility of a transmission window appearing at an apparently
arbitrary frequency below the reference, we must place an upper bound on the
reference frequency. Assuming that sidebands corresponding to L 4 have
negligible magnitude, this gives us the condition
f0 4 fR fR

which limits the reference frequency to the range
fR f0 /5

If responses at harmonics of fR are to be heavily suppressed, the sinewave

modulation voltage must have very low distortion. Otherwise the distortion
components will appear in the low-frequency region of the p.w.m. spectrum and
give rise to a set of harmonically-related transmission windows. In practice it
must be expected that the modulation voltage has some residual distortion and
that the modulation process is not perfectly linear, either in principle or in its
practical implementation. Such deviations from the ideal will inevitably result in
the occurrence of harmonic responses, albeit at low level.
9.4.3 Choice of switching frequency
For maximum separation between the low- and high-frequency regions of the
p.w.m. spectrum, the switching frequency f0 should be chosen to be much higher
than the maximum anticipated reference frequency. The limitation on f0 is
decided ultimately, by the dynamic range of the phase-sensitive detector which
deteriorates at high switching frequencies. This deterioration is compounded by
the additional loss in dynamic range inherent in the p.w.m. approach. It has been
observed that a p.w.m. reference channel can be configured as an option to an
otherwise conventional lock-in amplifier. If this is the case, the phase-sensitive
detector will be optimized over a range of frequencies rather than at a fixed high
frequency. The p.w.m. switching frequency must then be chosen to be
comparable with the highest frequency envisaged in conventional operation.
Inevitably, the highest permitted value of fR in p.w.m. operation must then be
significantly less than this value. The EG&G Brookdeal Sinetrac lock-in
amplifiers are subject to such a constraint; here the maximum recommended
reference frequency in p.w.m. operation is 25 kHz in a system which can operate
to frequencies above 100 kHz in conventional mode.

9.5 Reference phase-shifting

As described so far, p.w.m. systems depend on the provision of a sinewave
reference voltage in order to achieve fundamental-only response. This would
obviously place a severe restriction on the utility of such systems compared with,
say, heterodyne lock-up amplifiers which are able to operate with a wide range of
externally applied reference waveforms.
Also, to be of practical value, a p.w.m. system must be supported by a reference
phase-shift network to enable the phase of the sinewave modulation voltage to be
adjusted relative to a synchronous signal.
Clearly, these two drawbacks can only be overcome by adding to the complexity
of the reference channel. In deciding on a suitable processing system, the
following factors must be taken into account:
(i) Although we have identified an upper limit on fR, to avoid low-frequency
spurious responses, there is no fundamental limit on the lowest value of fR
which might be used. This implies that to exploit the p.w.m. technique to the
full, the reference processing circuits should be capable of operating over a
wide range of frequencies, amounting to several decades.
Chapter 96
(ii) The sensitivity of the phase-sensitive detector, and hence the calibration of
the overall system, depends directly on the amplitude of the sinusoidal
modulation voltage, which should therefore have constant value over the
entire operating range.
(iii) The sinewave applied to the pulse-width modulator should have a very low
level of distortion if the detection system is to reject responses at harmonics
of the reference frequency.
These requirements are fully met by the reference channel used in the EG&G
Brookdeal Sinetrac series of lock-in amplifiers. The configuration is shown in
Fig. 9.7. The reference input stage and the broadband phase-shift network are
those of a strictly conventional lock-in amplifier, capable of operating with high
precision over a frequency range in excess of five decades. The phase-shifted
output of the reference channel is a closely controlled squarewave which is
subsequently converted, first to a triangle and then to a sinewave. The circuits
used for squarewave-to-triangle conversion have been described by Carter and
Faulkner1; the final conversion to sinewave form is given by a piecewise linear
network adjusted for a low level of distortion over the full frequency range.

Fig. 9.7 Broadband lock-in amplifier with facility for fundamental-only

operation in p.w.m. mode

Fig. 9.7 also shows the relatively simple arrangement of switches required to
convert a harmonically-responding, conventional, lock-in amplifier to a system
with fundamental-only response. An additional signal-channel gain stage of
10 dB is used to maintain the overall sensitivity of the system in p.w.m. mode.
The diagram serves to emphasise that the p.w.m. configuration does not rely on
the use of filters, and is consequently free from phase and amplitude errors due to
filter misalignment. The phase accuracy and the residual phase noise are
therefore comparable with those of the basic conventional system and
significantly better than that of a heterodyne lock-in amplifier operating in the
same reference frequency range.

9.6 Two-phase systems

The extension to two-phase systems requires an additional squarewave-to-
sinewave convertor operating on the squarewave output of the quadrature
reference channel. A second pulse-width modulator, operating independently of
the first, is used to supply the reference input to the quadrature phase-sensitive
detector.

Chapter 97
It should be noted that p.w.m. systems do not depend on a carefully adjusted
carrier frequency. In practice, the carrier frequencies of the two pulse-width
modulators used in a two-phase lock-in amplifier need to be no more than
nominally equal.

9.7 Analogue correlation

"Analogue correlation" is a term used in the context of p.w.m. lock-in amplifiers
to cover the various modes of operation made possible when an external
reference waveform is applied directly to the pulse-width modulator. This opens
up numerous possibilities some of which are reviewed in the following sections.
9.7.1 Matched detection
The idea of a "matched" detector was mentioned in Section 3.5.5 in relation to
the measurement of a squarewave signal using a conventional squarewave
reference. If we now suppose that a periodic but non-sinusoidal waveform was
used as the modulation input to a p.w.m. lock-in amplifier, the resulting detection
system would be characterized by a set of transmission windows which could be
exactly matched in amplitude, frequency and phase to the Fourier components of
the signal. Such a system proves capable of yielding the best possible output
signal-to-noise ratio for signals obscured by white noise. Unfortunately, signal
recovery problems are usually associated with noise spectra far more complicated
than this, so that the benefits of matched detection (which are, in many cases,
marginal) are difficult to realize in practice. It usually turns out that the ability to
operate in a fundamental only response mode with relative freedom from
transmission windows close to the reference frequency gives a far greater
advantage when measuring non-sinusoidal signals in noise. As noted in Section
3.5.5 and in Chapter 6, the null-shift procedures can be applied when the
detection system has fundamental-only response, to quickly bring the reference
phase-shift to an optimum setting under very noisy conditions.
9.7.2 Two-frequency lock-in analysis
In a two-phase lock-in amplifier operating on the p.w.m. principle, the pulse-
width modulators associated with each phase-sensitive detector operate
independently and have separate inputs. These can be supplied with external
reference waveforms having different fundamental frequencies and different
waveforms if so required. The ability to use independent reference inputs means
that different spectral components of the signal can be separately measured on
each of the phase-sensitive detectors.
Since direct connection of external references to the modulator inputs bypasses
the reference phase-shifting networks, this facility is likely to be most useful in
such applications as optical spectroscopy where there is minimal phase-shift
between the reference and signal. An example of spectrometer operating with two
chopper frequencies is shown in Fig. 9.8. The light paths from the two samples
are combined at the cathode of single photomultiplier. A two-phase p.w.m. lock-
in amplifier is subsequently used to measure the two chopped signals separately
and simultaneously, using "analogue" reference inputs derived from the drives to
the optical choppers. A ratiometer can be a useful accessory in this type of
measurement; however, a digital interface of the type described in Chapter 10
provides a more flexible approach to processing the outputs from the twin
channels.

Chapter 98
Fig. 9.8 A two-frequency lock-in amplifier application. The outputs VA
and VB are proportional to the signals transmitted by samples 1
and 2 respectively.

9.7.3 High slew rate applications

In principle, the limited slew rate capability of conventional lock-in amplifiers
could be overcome by bypassing the reference phase-shifting networks and
applying a swept-frequency switching waveform directly to the phase-sensitive
detector. Such a procedure is valid when the signal and reference remain sensibly
in phase over the desired frequency sweep. In practice, the procedure can be
applied to conventional lock-in systems only when the phase-sensitive detector is
accessible in the form of a modular unit.

Fig. 9.9 Swept-frequency spectrum analysis using a quadrature oscillator

A difficulty arises when high slew rate is required in association with

fundamental-only response. In this case, a p.w.m. lock-in amplifier operated with
direct access to the pulse-width modulator provides a workable solution. P.W.M.
systems operating with carrier frequencies of the order of 100 kHz are able to
respond to rapid frequency sweeps applied to the modulation input. Provided a
swept-frequency sinusoid is available at the standard level required for proper
calibration, it is possible to obtain fundamental-only response consistent with
slew rates far in excess of most practical requirements.
Wide-band swept spectrum analysis was cited in Chapter 6 as a lock-in amplifier
application where fundamental-only response and high slew rate were essential
joint requirements. In order to exploit the characteristics of a two-phase p.w.m.
Chapter 99
system in this application, it is necessary that the two pulse-width modulators are
operated in strict quadrature at the swept reference frequency. This highlights the
need for a swept-frequency oscillator providing quadrature sinewave outputs at a
standard level. Such oscillators have been made available as accessories to two-
phase p.w.m. lock-in amplifiers for use in the configuration shown in Fig. 9.9. In
wideband applications using high frequency resolution, the sweep-rate limitation
in this type of system lies with the output filters as explained in Section 5.5.3.

9.8 Interference rejection filters

In an earlier version of the Sinetrac lock-in amplifier, the signal channel was
fitted with a 2-pole low-pass filter cutting off above the maximum reference
frequency of 25 kHz but well below the 100 kHz switching frequency. In
addition, a notch filter was used to enhance the suppression of signal components
in the region of 100 kHz. The object was to overcome the major spurious
responses associated with p.w.m. operation.

Fig 9.10 (a) Combined frequency response of signal channel filters. (b), (c)
Amplitude spectrum of PWM reference at "low" and "high"
reference frequencies

In practice, such a combination of filters is likely to prove desirable only at

comparatively low reference frequencies for the following reasons. First of all,
we have seen that at low reference frequencies the transmission windows are
concentrated in the spectral regions close to the switching frequency and its odd
harmonics. Spurious responses are therefore associated with high-frequency
interference and the filters are most effective in suppressing these in advance of
detection. Secondly, at low frequencies the phase-shift introduced by the filters
will be relatively small, enabling the overall system to operate with good phase
precision.
Conversely, at high reference frequencies, the signal channel filters will introduce
large phase errors into the measurement system while the transmission windows
move to much lower frequencies by reason of the wider sideband separation. The
contrasting situations are illustrated in Fig. 9.10. We find that at higher reference

Chapter 910
frequencies there will be a number of critical frequencies, for example f1 and f2
which fall within the bandwidth of the signal channel filter. Interference
components close to these frequencies would suffer minimal attenuation in the
filter and be able to excite spurious responses. In some cases, therefore, the
inclusion of the filter acts more to the detriment of the system, introducing large
phase errors while having only partial success in overcoming the problem of
spurious responses.
The decision to eliminate interference rejection filters with fixed characteristics
in later versions of the Sinetrac system resulted in a lock-in amplifier capable of
fundamental-only response, consistent with excellent phase accuracy, over many
decades of frequency. The system is, nevertheless, associated with large spurious
responses at relatively high frequencies. It was shown in Chapter 6 that
fundamental-only responding systems have considerable application in
experiments operating with low reference frequencies. In this regime, signal
conditioning filters can be used to suppress interference components at
frequencies comparable with the high-frequency switching components of the
p.w.m. reference. If a range of optional filters is made available, this can be
achieved without sacrificing phase accuracy when the system is required to
operate with fundamental only response up to the highest possible value of
reference frequency: for example, when measuring the swept-frequency response
of networks over the full audio-frequency range.

9.9 Comparison of p.w.m. systems with

heterodyne lock-in amplifiers
P.W.M. and heterodyne lock-in amplifiers were devised with the same objective
in mind; to give a synchronous detection system with wide dynamic range and
relative freedom from harmonic responses over a wide range of reference
frequencies. As we have seen, the two approaches to this problem lead to vastly
different solutions, both of which involve system designers in a number of trade-
offs and compromises.
On balance, modern heterodyne lock-in amplifiers appear to offer the widest
frequency range consistent with the lowest level of spurious responses, whereas
commercial versions of the p.w.m. system operate up to a maximum frequency of
about 25 kHz and have a number of large transmission windows accessible in the
frequency range immediately beyond this value.
The overall phase accuracy of p.w.m. systems is superior to that of heterodyne
lock-in amplifiers, which are susceptible to alignment errors in a number of sub-
systems and generally require a far more complex configuration. The difference
in complexity is reflected in system cost, since p.w.m. lock-in amplifiers usually
offer a cheaper means of obtaining fundamental-only response than their
heterodyne counterparts. Also p.w.m. systems can usually be converted by
pushbutton selection to operate as conventional lock-in amplifiers, giving an
extension of the frequency range and allowing the fundamental-only response to
be traded for greater dynamic range. This flexibility in choosing the response of
the system extends to choosing an arbitrary response given by the Fourier
components of the applied reference waveform.
A feature that both heterodyne and p.w.m lock-in amplifiers have in common is
that fundamental-only response at low frequencies is obtained by operating the
phase-sensitive detector at a relatively high frequency. The result in both cases is
a system with a dynamic range independent of reference frequency but less than
that which might be achieved if the phase-sensitive detector was operated in
conventional fashion. It was shown in Chapter 8 that the dynamic range of

Chapter 911
heterodyne systems can be recouped by using synchronous heterodyning as a
supplementary technique. Unfortunately, it is not obvious how this could be
applied to phase-sensitive detectors operating with a p.w.m. reference without
incurring additional spurious responses in the low-frequency region. As a result,
the dynamic range of a heterodyne lock-in amplifier can be comparable with that
of a p.w.m. system where the phase-sensitive detector is operating at much lower
frequency.

9.10 Reference
CARTER, S.F., and FAULKNER, E.A. (1977): "Accurate broadband square-to-
triangle converter", Electron, Lett., 3, pp. 381-382.

Chapter 912
CHAPTER 10

Computer-controlled lock-in amplifiers

10.1 Introduction
The advent of the microprocessor and the increasing availabity of desk-top
computing power have provided a challenge to both designers and users of
electronic measuring equipment. Instrument designers are faced with a demand
for intelligent instruments capable of performing programmed tasks or able to
communicate with other instruments via a computer controller. As for the
instrument user; he is concerned with using these new instruments to the best
effect and with devising measurement procedures that take advantage of the latest
developments in instrument technology.

Fig. 10.1 Bus-controlled measurement system

This general situation is reflected in the specific case of instrumentation for

signal recovery where increasing emphasis is being placed on computer control in
its widest sense. Developments in this area have been greatly influenced by the
widespread adoption of the IEEE-488 bus system for providing two-way digital
communication between measuring instruments and a computer-controller. This
has resulted in the availability of a large range of compatible instruments that can
be, literally, plugged together to produce a computer-controlled measurement
system. Many of the applications listed in Appendix 1 could benefit from such an
approach; for example, Fig. 10.1 shows a bus-compatible lock-in amplifier
operated in conjunction with a number of other controllable devices to provide an
automatic system for frequency-response measurement.
The IEEE-488 bus protocol is rigidly defined; thus, at any time, only one device
is permitted to talk, that is send data or commands over the bus, while several
devices may listen in order to receive data or commands. In the example shown,
the X-Y plotter and signal generator would probably operate as listeners while
the lock-in amplifier would both talk and listen, to transmit data to the
computer and receive commands. Overall control comes from the computer,
which is programmed to change the frequency of the signal generator in discrete
steps, to manage the take-up of data from the lock-in amplifier, and to process
data for presentation on the X-Y plotter.
In the computer-controlled lock-in amplifiers to be discussed in this chapter, the
phase-sensitive detector remains intact at the heart of the system, supported by
signal and reference channels having characteristics similar to those described in

Chapter 101
earlier chapters. The incorporation of digital control lines to switch the sensitivity
and the internal configuration of the lock-in amplifier can be achieved without
compromising key specifications such as input dynamic range and operating
frequency range. In practice, therefore, the only serious limitation incurred in
operating with a computer interface is with regard to output dynamic range. This
is now limited by the use of an analogue-to-digital convertor on the phase-
sensitive detector output. An attempt to match the 100 dB dynamic range of a
typical analogue output would require a 17-bit conversion and would be difficult
to justify on grounds of cost in a general-purpose measurement system. The usual
provision is for a 31/2 digit conversion, giving a resolution of 10 mV in a 10 V
output with 100% over-range. If this is inadequate for a particular application, the
analogue output is available on its usual socket and can be separately converted
to high precision if so required.*
The handling characteristics of the lock-in amplifier being relatively unchanged,
the main problem in digital control is to create programs which reproduce the
measurement routines and setting-up routines that are associated with the
detection of noisy signals. In giving consideration to these routines it will be
convenient to distinguish between the two main types of controllable lock-in
amplifier in general use. These are programmable lock-in amplifiers where the
software control routines are resident in an external computer controller, and
microprocessor-based systems so called intelligent lock-in amplifiers. The
latter feature a number of stored software routines that can be initiated by front-
panel switch selection or by a command transmitted on the interface bus.

10.2 Programmable lock-in amplifiers

In early lock-in amplifiers, the pushbutton and switch selectors controlling the
overall system configuration were heavily interlocked and interlinked and
required front panel assemblies that were both complex and labour-intensive in
production. At a later stage, f.e.t. switches, controlled by the application of
standard logic levels, became widely used for both gain selection and mode
selection and there was a move to transfer hard-wired switching logic to
integrated circuits mounted on the printed circuit board. This change leads to a
dramatic simplification in switch design. For example, the sensitivity switch of a
typical lock-in amplifier is reduced from a multi-wafer assembly to a single-pole
selector as illustrated in Fig. 10.2.

Fig. 10.2 Simplification of switching operations by using a combinational

logic circuit

*
Note that to exploit the full output dynamic range of these instruments generally places severe
demands on peripheral equipment, both analogue and digital.
Chapter 102
The sensitivity lines S1 to Sn provide an address input to a combinational logic
block which is tabled to produce the appropriate output combination on the f.e.t.
control lines F1 to FN. These are then used to switch the gain of the amplifier
stages in the lock-in amplifier signal channel. In a comprehensive detection
system, giving a choice of high stability or high reserve operation, the address
lines might be augmented by an additional input which is either HIGH or LOW
depending on mode selection. The combinational logic circuit is then arranged to
control the f.e.t. switches for any combination of mode and sensitivity, without
adding to the complexity of the mechanical switching assembly.
The transition from a standard instrument designed along these lines, to one
where the internal switches can be controlled by the application of logic levels
from an external source is relatively straightforward. In a fully programmable
lock-in amplifier, all the functions that are normally switched from the front
panel, such as sensitivity, time constant, phase quadrants and expand, can be
controlled from logic levels applied to a digital input. In lock-in amplifiers
having a voltage-controlled phase-shifter, the provision of a digital-to-analogue
convertor enables the reference phase to be added to the list of controllable
parameters. In its simplest form, the digital input might be a multi-way socket
connected to a set of remotely operated switches. In modern instruments,
however, the digital input is more likely to be a port having access to a standard
bus system, such as the IEEE-488 bus referred to earlier. When the lock-in
amplifier output is provided with an analogue-to-digital convertor, a properly
defined bus system enables data to be transferred to the computer controller and
to other instruments connected to, and controlled from, the bus.
In order to exploit this type of system effectively, the computer-controller must
be provided with programs sufficiently powerful to undertake the management of
the lock-in amplifier under a wide range of signal and noise conditions. We thus
envisage a control program that defines an overall measurement procedure and
contains a number of subroutines for sensitivity and phase selection. The
specification of these subroutines requires a certain familiarity on the part of the
user with the handling characteristics of lock-in amplifiers and would normally
involve several stages of refinement before an acceptable solution was found.
Some essential features of these routines are identified and discussed in Sections
10.4 and 10.5.

10.3 Microprocessor-based systems

The incorporation of a microprocessor to monitor and supervise the switching
functions of a lock-in amplifier represents a significant advance in system
concept and design. The result is a self-contained lock-in amplifier with the
ability to undertake sequential switching operations controlled by software
associated with the microprocessor. The control system takes additional data
from
(i) front panel switch arrays
(ii) a digital interface to an external keyboard or controller
(iii) the outputs of the phase-sensitive detectors.
A lock-in system with this overall capability greatly cases demands on the user
who no longer requires such detailed familiarity either with lock-in techniques in
particular or signal recovery in general. For example, the resident software
routines could enable the lock-in amplifier to adjust sensitivity and phase
automatically to maximize the output for a given signal. As far as the user is
concerned, the lock-in amplifier now operates like a special type of a.c.
microvoltmeter which can read the amplitude and phase of a signal in response to

Chapter 103
a single key stroke or bus command. This reduced level of operational
complexity is reflected in the amount of programming effort required to control
the lock-in system when it forms a component part of a larger bus-controlled
system.
There are other aspects of operation which benefit non-specialist users that apply
to almost any type of microprocessor-based instrument. For example, the digital
output can be scaled to reflect the input signal level, taking all factors such as
sensitivity multipliers and amplifier gains into account. The system is then much
less prone to operator error than a mechanically switched system fitted with a
pointer scale where the danger of overlooking a scaling factor is always present.
Also, since all switching operations from the front panel are supervised by the
microprocessor, the system is able to inhibit or give warning of undesirable or
unorthodox combinations of front-panel controls, supported by a display or print-
out of the appropriate error message.
In an instrument such as a lock-in amplifier which is subject to frequency-
dependent errors in the signal and reference channel circuits, there is ample scope
for using the microprocessor in an automatic calibration routine. This would
measure and store calibration errors over the frequency range of the instrument
with the object of providing corrected results in the final measurement. In
principle, the calibration routine could be extended to correct phase-sensitive
detector offsets and to compensate the amplitude and phase characteristics of
signal conditioning filters introduced into the signal channel.
Clearly, the incorporation of a microprocessor has progressively greater impact as
the complexity of the lock-in system is increased and should, ideally, enable a
greater number of facilities to be offered without sacrificing case and clarity of
operation. This objective generally requires a fresh approach to front panel
design. For example, the familiar phase dial of a lock-in amplifier might be
replaced by a counter that can be incremented or decremented using a pushbutton
switch. The reference-channel phase-shifter is then controlled from a digital-to-
analogue convertor taking its input from the microprocessor data bus. The
problem of monitoring the status of the instrument is overcome by displaying the
phase setting on a digital panel meter which also serves to display error codes and
fault conditions when the system is operated.
As an additional constraint on the system designers, experienced users would
normally require that the system is able to revert to full manual control where
various combinations of front-panel settings could be tried without being
restricted to operate from a menu of stored routines. This constraint would
certainly apply to any microprocessor-based lock-in amplifier that was offered as
a general purpose measurement tool rather than as a special-purpose instrument,
rigidly programmed to perform a specific range of tasks.
The selection of routines available on commercial instruments is limited but
carefully chosen to enhance the handling characteristics of the lock-in amplifier
in a wide range of applications. In addition to the software routines for sensitivity
and phase selection referred to earlier, there is usually the possibility to offset
data by a fixed amount and to normalize data, providing an output expressed as a
fraction of percentage of some predetermined level. Routines of this type are
therefore applied after detection and serve as a first stage of output processing. If
more complex processing is required, this would normally be carried out by a
computer interfaced to the lock-in amplifier, programmed to suit the needs of a
specific experiment.
Management of this interface by a microprocessor resident in the lock-in
amplifier offers several advantages over a hardware only design. Thus,
transmitted data can be presented in an easily understood format, only relevant
Chapter 104
data need be transmitted, and received commands can be less complex and more
meaningful. The impact of the microprocessor on the design of an IEEE-488
compatible lock-in amplifier is brought out in Table 10.1. Comparison is made
with a notional design based on classical techniques, and improvements are
attributed to specific characteristics of the microprocessor.
Regarding the routines for sensitivity and phase adjustment; it is essential here
that the criteria for range and phase switching are clearly stated if the lock-in
amplifier is to behave predictably under the worst conditions of signal and noise.
The following sections give further discussion on these routines and apply
equally to a detection system under software control from a microprocessor or
from an external controller linked by a data bus.

Table 10.1

Instrument Classical design Improvement in Microprocessor

characteristic techniques microprocessor- characteristic
based design leading to
improvement
Digital display of Output in range 0 Direct scaling Multiplication
output to 10V scaled reflecting input program
by reference to level
gain setting and
1, 2, 5
multipliers
Digital display of 0 to 99.9 0 to 359.9 Addition
phase Quadrant Direct indication capability
information on
+90and +180
switches
Autorange mode Hardware design Software-only Digital
requiring physical design: no comparison
links with additional Data manipulation
sensitivity and hardware required Program storage
time-constant to implement
switches mode
Zero offset Manual operation Manual-automatic Program storage
operation Data manipulation
Subtraction
capability
Initial Set-up None Automatic Program storage
operation Data manipulation
Digital
comparison
Mathematical
capability
Normalize Manual operation Manual/automatic Program storage
operation Data manipulation
mathematical
capability

Chapter 105
Table 10.1 (continued)

Variable phase 3-digit d.a.c. or 10-bit d.a.c. B.C.D. to binary

and zero-offset 10-bit binary program storage
hardware d.a.c. plus b.c.d.
to binary
conversion
hardware
Output conversion 31/2 digit a.d.c. 12-bit binary Binary to b.c.d.
a.d.c. program storage
IEEE 488 Fixed format. Flexible format Data manipulation
Transmitted data Interpretation No interpretation Storage of format
required for phase required styles
and output data No redundant data Mathematical
need be capability
transmitted Read/write
storage capability
IEEE 488 Fixed format: Flexible format Data manipulation
Received hardware Meaningful Data storage
commands determined commands Storage of format
styles
Front panel Rotary, Pushbutton Data manipulation
control pushbutton and switches only. Program storage
toggle switches. Ergonomic
Hardware improvements.
determined Group of controls
not constrained by
internal design of
instrument

10.4 Automatic sensitivity selection

When a lock-in amplifier is operated manually, the response to a synchronous
signal is usually adjusted by switching sensitivity to obtain an output as close as
possible to a full-scale reading.
In a single-phase lock-in amplifier, the response will not necessarily be maximum
and may even take negative values unless the phase of the reference channel has
been correctly adjusted. When using a two-phase system, the sensitivity is usually
adjusted to maximize the output of the phase-sensitive detector giving the largest
response. Alternatively, when the use of a vector computer is appropriate, the
sensitivity can be adjusted by observing the magnitude output of the vector
computer.
If the residual noise output of the lock-in amplifier is sufficiently large, it will be
necessary to ensure that fluctuations in the output do not carry the indication due
to the signal beyond full-scale. This would normally require an observation time
amounting to several time constants and may result in the system being switched
to lower sensitivity. Most systems offer a 1:2:5 or 1:3:10 switching sequence, so
this final step can usually be achieved without a significant loss of output voltage.
The object of a sensitivity routine or autorange routine is to bring this sequence
of operations under automatic control by comparing the magnitude of the lock-in
amplifier output with predetermined threshold levels. Some of the difficulties
encountered when autoranging with a noisy signal are demonstrated by the
following examples.
Chapter 106
Fig. 10.3 Definition of switching thresholds in a decade autoranging
system.

Suppose we have a lock-in amplifier where the sensitivity control is divided into
a series of decade ranges, and that the response to a synchronous signal has been
brought to a level corresponding to about half-scale output as shown in
Fig. 10.3(a).
The lock-in amplifier controller is programmed to switch to a lower sensitivity
(range-down) when the signal increases to a level corresponding to full scale
output, and to switch to higher sensitivity (range-up) when the signal falls to 10%
of full-scale. To be realistic we must allow for a small amount of residual noise
appearing with the signal as shown in Fig. 10.3(b); hence the first down
transition will occur when the total output due to signal plus noise exceeds the
100% level. Fig. 10.3(c) shows the new situation which applies after the
sensitivity has been switched. Signal and noise appear in the same relation as
before, so it is only a matter of time before the total output falls below the 10%
threshold level, causing the system to range-up to its original sensitivity. The
situation illustrated in Fig. 10.3(b) is thus restored. If the signal remains at a
constant value the system will attempt to switch gain alternately up and down
with the result that a stable condition is never attained. To overcome this
difficulty it is necessary to redefine the switching thresholds, for example by
increasing the upper threshold to greater than 100%, or by reducing the lower
threshold to less than 10%. In this way, the system can be made to tolerate
residual noise on the output, at least up to a certain peak-to-peak level, and will
be able to switch sensitivity to reach a well-defined condition.
Let us now look at the behaviour of the system under conditions of very low
signal. If, at some point, the output signal-to-noise ratio falls drastically, or if the
signal is removed, leaving only noise in the output, the sensitivity controller will
attempt to switch gain to the maximum achievable value. The system will then
remain in a stable condition at maximum sensitivity, provided the output noise
peaks do not exceed the upper threshold level. If this level is exceeded, the
controller will switch the sensitivity to a lower value. Unfortunately, the noise is
bipolar and so repeatedly takes values close to zero voltage. The controller will
thus restore the sensitivity to its maximum value at the first opportunity and
subsequently make random transitions between the two most sensitive range
positions.
These considerations suggest that the maximum usable sensitivity in autorange
operation is where the peak output noise is just less than the full-scale output.
This maximum sensitivity can be determined by experiment; the autorange
program should then be arranged to inhibit the selection of sensitivities beyond
this maximum permitted value, to avoid the control system becoming unstable
with low signal inputs.
Next, let us consider the effect of the lock-in amplifier time constant on an auto-
ranging system. It should be remembered that time constants up to about 100 ms,
which would be considered fast by a human observer, are relatively slow

Chapter 107
compared with the response time of a digital controller. Decisions to switch range
should therefore be deferred for an appropriate time while the output of the lock-
in settles to a new value following a switching operation. In commercial systems
supplied with an autorange facility, a settling time of four or five time constants
is usually allowed between successive switching operations. In a fully integrated
system, the micro-processor will be provided with the time-constant setting as a
matter of course. In a bus-controlled system, the time-constant setting may have
to be read by the controller via the bus interface and entered into the autorange
subroutine. The subroutine would usually feature a number of WAIT instructions
to ensure that the program runs at an appropriate rate for the particular time-
constant selection.
The final point to be taken into consideration concerns the role of expand
selection in determining the sensitivity in autorange operation. It is shown in
Chapter 4 that, in some lock-in amplifiers, a given sensitivity can he obtained for
two or more combinations of a.c. and d.c. gain in the system and that the choice
of combination influences the dynamic performance of the lock-in amplifier. It
follows that, if the lock-in amplifier is required to autorange in a high reserve
mode, the switching program should be arranged to give a gain combination that
uses the maximum possible value of expand gain. Conversely, a high stability
switching program would be biased in favour of using high a.c. gain in order to
achieve the best possible output stability for precision measurements.
In a microprocessor-based system, these factors could be taken into account
automatically, depending on the mode of operation selected by the user. Other
facilities that would normally be made available include a procedure for entering
the maximum autorange sensitivity (in the interests of system stability as
described above) and a procedure for entering the threshold switching points,
usually expressed as a percentage of full-scale deflection. In some cases, the
upper threshold is fixed at 110% of full-scale output while the autorange routine
covers the 1:2:5 range sequence of the lock-in amplifier. A system with these
characteristics would switch range until the output indication lay somewhere
between 40% and 110% of full-scale with the overall switching time determined
by the time-constant selected on the lock-in amplifier.
It should be acknowledged that autorange switching routines are, at best,
systematic and, at worst, cumbersome. At a time constant of 1 second, a typical
autorange routine would take about 1 minute to switch from minimum to
maximum sensitivity in a 1:2:5 sequence. When operating with a very wide range
of signal levels, using a programmable system with a choice of programs, there is
a possibility to include a trial routine confined to decade switching. The idea is
to obtain an order-of-magnitude estimate of signal level; this estimate can then be
improved using the 1:2:5 switching sequence in a final iteration.

10.5 Automatic phase selection

We can identify two basic routines for phase adjustment in lock-in amplifier
measurements. The first is used in signal recovery work where the phase of the
reference channel is adjusted to maximize the output from the phase-sensitive
detector. The second is associated with precision phase measurement where the
phase adjustment is made with the objective of nulling the output of the phase-
sensitive detector.
Phase measurements might be made with either a single- or two-phase lock-in
amplifier. If the latter is used, it is normally the output of the quadrature phase-
sensitive detector which is to be brought to a null condition. Since phase
measurements are usually made with noise-free signals, the accuracy to which the
null can be set will depend on the ability of the analogue-to-digital convertor

Chapter 108
associated with the phase-sensitive detector to resolve small output changes. The
resolution of the phase-shift control will be similarly limited in a digitally
controlled system. In commercial systems the phase can usually be advanced in
increments as small as 0.1, which is comparable with the resolution of a
conventional phase dial giving a continuous adjustment.
The phase-null routine can be initiated by subtracting increments of about 30
until the output changes sign. Smaller increments, say 5, are then progressively
added to the set phase until the output changes sign yet again. The procedure is
repeated with successively smaller increments until the null is achieved to within
the resolution capability of the system, or to within some specified limit.
Regarding signal recovery applications using a single-phase lock-in amplifier, an
alternative approach to setting the phase is defined as follows, starting from an
arbitrary initial phase condition:
(i) read the in-phase value of the signal, VA
(ii) add 90 to the set phase of the reference channel
(iii) read the quadrature value of the signal, VB.
(iv) compute = tan1 VB/VA: reduce set phase by 90
(v) add to the set phase.
This routine could be accomplished in a time equivalent to about 10 lock-in
amplifier time-constants. When the lock-in amplifier has fundamental-only
response, the resulting response will always be maximized and first-order
independent of errors accrued in the measurement and in the computation of the
signal phase. This procedure is perfectly adequate for use in general signal
recovery applications and, more importantly, can be used to extend the usefulness
of single-phase lock-in amplifiers in tasks which are normally reserved for two-
phase systems.
Of course, if a two-phase system is available, the problem of setting phase need
not arise in signal recovery work. All that is required is an autoranging control
system to bring the vector magnitude to a suitable on-scale value. In the case of
a single-phase lock-in amplifier, an autoranging routine would normally be
executed prior to setting the phase, followed by a final autorange routine to bring
the maximized response within range.

Chapter 109
APPENDIX 1

Principal applications

The following list catalogues some of the principal applications of lock-in

amplifiers and phase-sensitive detectors. The compilation was made by the staff
at EG & G Brookdeal and covers a wide range of disciplines in applied science
and technology. It is inevitable that a number of applications appear under more
than one title. To compensate for this, many readers will doubtless find a number
of important applications that have been overlooked!

Absorption spectroscopy Filter calibration

A.C. bridges Fluorescence spectroscopy
Antenna Patterns Frequency-response
Astronomical spectroscopy measurements
Atomic absorption Frequency-shift measurements
Audio amplifier frequency
response Hall effect: single frequency
Audiometry Hall effect: double frequency
Auger spectroscopy
Infra-red (near and far)
Biomedical stimuli response spectroscopy
measurements Interferometry
Bode plots
Klystron stabilization
Cochlea microphonics
Common mode rejection Laser research
measurements Line ripple measurement in
Complex impedance amplifier power supplies
measurements
Contact potential measurements Magnetic-field measurements
Crosstalk in cables, amplifiers, Magnetometry
etc. Magnetoresistance studies
C-V plotting Marx gauging
Cube interferometry Mass spectroscopy
Microphone calibration
De Haas Van Alphen effect Microwave reflections,
Densitometry attenuation
Detectivity compensation Microwave spectroscopy
Displacement measurements Moisture content measurement
Doppler measurements (C-G)
Dual-beam optical Molecular-beam spectroscopy
measurements
N.M.R. spectroscopy
Eddy-current flaw testing N.O.R. spectroscopy
Edge shift in GaAs Nyquist plots
Electrochemistry
Electroluminescence Operational amplifier gain
Emission spectroscopy measurement
E.P.R./e.s.r. spectroscopy Optical derivative
measurements
Appendix 11
Photometry Stress-strain measurements
Plasma-physics research
Pyrometry Temperature control
Temperature measurement
Radiometry Torque measurements
Raman spectroscopy
Ratiometric measurements Ultra-violet spectroscopy
Resistance thermometry
R.F. measurements Visible spectroscopy

Second sound Whistler signal measurements

Seismic measurements Work function measurements
Semiconductor research
Source compensation Youngs modulus
Spectrophotometry
Strain gauging Zeeman effect

Appendix 12
APPENDIX 2

Selected topics on signals and noise

A2.1 Introduction
Fig. A2.1 shows the frequency ranges which can be assigned to a number of
noise sources of practical importance. As explained in Chapter 2, we usually
make a distinction between interference sources of external origin and noise
which is inherent in the measurement system. The means to combat external
interference sources are many and varied but inevitably involve the use of
screening and attention being paid to cable runs. In many cases, susceptibility to
mains-frequency pick-up can be reduced by physical re-orientation of circuits and
components; factors giving rise to ground loops whereby mains-borne
interference is introduced along with the signal are treated in Appendix 6.

Fig. A2.1 Spectrum of noise and interference

In dealing with the spectrum of noise and interference it is usual to treat sources
of discrete interference separately from random noise sources. The former can
sometimes be estimated and presented in the form of an amplitude spectrum,
showing the magnitude of the various interference components relative to that of
the signal. However, it is often more relevant to estimate the actual peak-to-peak
values of the interference components; these components often give rise to
saturation in amplifiers, which is most conveniently expressed in peak-to-peak
terms.
Of course, the fundamental system noise cannot be treated in terms of an
amplitude spectrum. The noise manifests itself as a fluctuating voltage in the
output, which is the resultant of components distributed over a wide frequency
range. It is characteristic of well-behaved noise sources, however, that these
essentially random fluctuations deliver a consistent average power into an
external load circuit. In view of this, it is appropriate to express the frequency

Appendix 21
distribution of the random noise components in terms of a power spectrum or,
more exactly, a power density spectrum, P(f). P(f) is usually a continuous
function of frequency and has dimensions of watts/Hz. By definition of a density
spectrum the power delivered from a small frequency range f centered at a
frequency f is simply P(f)f. Therefore, when P(f) is specified, we can calculate
the total power in any desired frequency range f1 to f2 from the integral

f2
PTOT P( f )df
f1

This is shown graphically in Fig. A2.2 for a white noise spectrum with constant
spectral density P0 and for a more typical spectrum which might be encountered
in practice. In both cases the total noise power in the frequency range of interest
is given by the area under the spectral density plot.

Fig. A2.2 Power density spectra for (a) white noise; (b) typical
experimental noise
The total noise power in the frequency range f1 to f2 is given by
the area of the shaded region in each case

A2.2 Voltage noise and current noise spectra

The integral of a power spectrum is given practical significance when a bandpass
filter is used to reject all noise components except for those lying in a selected
frequency range. The value of the integral then gives a measure of the noise
power which might be measured in the filter output.
In practice, noise power measurements are usually reserved for v.h.f. and other
systems operating with well-defined impedance levels. Elsewhere, it is generally
more convenient to measure the filter output in terms of its mean-square value.
Since we are concerned here with electronic systems, the output from the filter
will either be a voltage fluctuation or attributable to a current fluctuation: so we
express the noise intensity in a given frequency range as a total mean-square
voltage or current as appropriate.
This change of emphasis leads us to define mean-square voltage and current
noise spectra, WV(f) and WI(f), expressed in units V2/Hz and A2/Hz respectively.
In a given frequency range we measure a voltage signal or a current signal in
association with a total mean-square fluctuation:

f2
v2 WV ( f )df
f1

or

f2
i2 WI ( f )df
f1

A further point remains to be considered. In manufacturers data sheets, the noise

inherent in transducers and amplifiers is commonly given as an r.m.s. fluctuation

Appendix 22
measured in a specified bandwidth. If a noise voltage has a spectral density
expressed in V2/Hz, then the r.m.s. spectral density must have dimensions V/Hz.
Thus, doubling the measurement bandwidth for a white-noise spectrum doubles
the measured intensity while the r.m.s. value increases by only 2.

A2.3 Signal spectra

It is clear that signals have a different status to noise in that a fairly precise
description can often be given of their time-domain behaviour. Indeed, as has
often been remarked, we must have at least an outline description for a signal
before we can embark on the process of signal recovery.
In the restricted view of signal recovery which includes lock-in techniques, the
signal is usually an amplitude- or phase-modulated carrier described by one of
the general forms:
s(t) = A0 [1 + m(t)] cos0t
or
s(t) = A0 cos [0t + m(t)]
In each case, the carrier frequency 0 is known within fairly close limits and the
information or modulation signal, m(t), is to be determined. Just as we have
general information about the form of the signal, it is likely that the broad
characteristics of m(t) will also be known; the experiment itself will set limits to
the maximum amplitude range of m(t) and to its maximum rate of change.
In many instances, the signal is of very simple form and can be separated into
sinewave components by the use of trigonometric identities, or expanded as a
Fourier series. The frequency composition of the signal and its representation as
an amplitude spectrum can then be inferred directly from the time domain
description.
Elsewhere, a knowledge of Fourier transforms plays a role in deducing the form
of the signal spectrum, but a rigorous approach is not necessarily the most
beneficial. In most cases, it is usually sufficient to know: (i) the location of the
spectrum, (ii) the width of the spectrum and (iii) the mean-square value of the
signal. Even when the modulation m(t) is only broadly specified, these three
points can usually be answered. For example, in the case of amplitude-modulated
carrier, we have:
s(t) = A0 [cos0t + m(t)cos0t]
We suppose that m(t) is a slowly varying function compared with cos0t. If the
Fourier transform of m(t) exists, and is given by M(j), then we can use the
narrowband transformation:
m(t) cos 0 t

= M(j + j0) + M(j j)

The Fourier transform of the signal is thus
A0 A A A
S(j) = ( 0 ) 0 ( 0 ) + 0 M ( j j 0 ) 0 M ( j j 0 )
2 2 2 2
An example is shown in Fig. A.2.3 with the modulation strictly limited to a
bandwidth Bm. A knowledge of this bandwidth is sufficient to estimate the width
of the spectrum located about the carrier frequency, although the precise shape of
the spectrum may not be known. In this way points (i) and (ii) raised above can
be answered. Regarding the mean-square value of the signal we have:

Appendix 23
 A02
s 2 (t) [1 m 2 (t)]
2
If the frequency range of the signal can be estimated together with its mean-
square value we can avoid the need for a formal definition of intensity spectra for
periodic and other deterministic signals. Also, we see, that provided m(t) < 1,
we can estimate the signal intensity to within a factor of 2 even when m(t) itself
might not be precisely specified. If this seems inadequate in the light of the
conventional approach to modulation systems it is no worse than the degree of
approximation used in estimating the level of the background noise!

Fig. A2.3 Spectrum of an amplitude-modulated carrier

In the case of phase modulation: when m(t) << 1, corresponding to low-index

modulation, the signal can be expressed as:
s(t) = A0 [cos0t cosm(t) sin0t sinm(t)]
A0 [cos0t m(t)sin0t]
The amplitude spectrum and the effective bandwidth of s(t) is thus the same as
for amplitude modulation by the same information signal m(t). Note, however,
that the phase relationships are different in the two cases.
At the other extreme, when the index of modulation is large, Carsons rule can be
used to estimate the spread of the spectrum from its centre frequency. This
gives an estimate of the signal bandwidth:
Bs = 2(m(t)max + 1)Bm
where Bm is the bandwidth of the modulation or information signal.
When calculating the mean-square value of phase modulated signals, we find that
the phase terms make no contribution to the final result. This is simply

s 2 (t) A02 /2
In general we might expect that the signal carries both amplitude and phase
modulation. While this is a possibility we should also note that, in the vast
majority of cases, the modulations can be very slowly varying functions, often
limited to a bandwidth of a few hertz. Thus, very often, the signals of interest are
extremely narrowband, occupying a relative bandwidth of no more than a few per
cent.

Appendix 24
A2.4 Thermal noise and shot noise
There are a number of well defined mechanisms which give rise to broadband*
noise in experimental systems. Some of these, for example, noise due to the
generation and recombination of charge carriers in semiconductors (g.r. noise),
are associated with a range of time constants and the spectrum is limited to an
upper cut-off frequency. In the majority of cases, however, the fundamental noise
mechanisms can be traced to thermal noise or shot noise, both of which generally
occupy a frequency range far in excess of the signal frequencies of interest.
In electrical systems, thermal noise (Johnson noise) is generated by the random
motion of electrons in resistive material at a finite temperature. Shot noise is
attributed to the passage of discrete charge carriers when current flows through
electronic devices. In both cases the noise can be modelled by a white-noise
spectrum over all frequencies of practical importance.
First of all, regarding thermal noise: this is a fundamental source of fluctuation in
all physical systems in a state of thermodynamic equilibrium. We can be sure,
therefore, that it will be found in all linear, passive devices, irrespective of their
form. It is often the case that such devices and for that matter entire
experimental systems can be reduced to a simple description in terms of an
equivalent electrical analogue circuit. The most common is either the Thvenin
or the Norton form shown in Fig. A2.4.

Fig. A2.4 (a) Thvenin and (b) Norton source equivalent circuits including
resistance noise generators

The equivalent source resistor is defined in terms of the physical characteristics

of the linear device and the thermal noise associated with the source resistance is
included in the form of a voltage or current noise generator. When Rs is specified
and the temperature is known we can immediately obtain the spectral density
functions of these noise generators:
Wv(f) = 4 kTRs V2/Hz
WI(f) = 4 kT/Rs A2/Hz
Here, k is Boltzmanns constant (1.38 10-23 joules/K), Rs is the equivalent
source resistance in ohms and T is the absolute temperature.
In calculations involving resistive sources at laboratory temperature it is
convenient to remember that a resistor of x kilohm is associated with a random
voltage generator of 4x nanovolts/Hz or a random current generator of 4x
picoamperes/Hz.

*
Broadband noise is recognised by its having a spectrum that is generally free from local peaks
and extends to zero frequency.
Appendix 25
Shot noise, unlike thermal noise, is always associated with current flow. The
y

random passage of charge carriers in vacuum tubes and semiconductors gives rise
to a fluctuation which depends on the average current. The spectrum is that of
white noise which extends over a wide frequency range limited only by transit-
time effects in the electronic device. We have
WI(f) = 2qI0 A2/Hz
where q is the electronic charge (1.6 1019 coulombs) and I0 is the average
current. The r.m.s. spectral density of the noise caused by a 1 nA current flow is
therefore about 1.8 1014 A/Hz.
Shot noise will be present in all semiconductor devices operating with finite bias
current, and is usually the dominant source of broadband noise in optical
detectors. Here, a periodic current variation due to a chopped light beam must
often be measured against a more or less steady bias current which flows in
response to a much greater background illumination due to light leakage or
sample fluorescence. Many such detectors conform closely to an ideal current
source, and the output can be measured by connecting the detector to an external
load resistor RL. Fig. A2.5 gives the noise equivalent circuit of this arrangement,
which shows that the signal current is appears in competition with the shot noise
of the bias current IDC and the thermal noise of the load resistor. To ensure that
the signal-to-noise ratio inherent in the detector is not degraded further by the
thermal noise in RL we investigate the condition:
2qIDC 4 kT/RL
which gives
RL 2kT/(qIDC)
The quantity 2kT/q is approximately equal to 50 mV at laboratory temperatures.
Thus, for a bias current of 1 mA, the source will be dominated by shot noise
provided that RL is in excess of 50 . In fact, the usual tendency is to choose very
large values of RL to increase the output voltage due to the signal current. In this
case, the shot-noise contribution is usually the dominant one even at low bias
currents. This topic is discussed further in Appendix 5 in relation to amplifier
selection and the use of current amplifiers.

Fig. A2.5 Noise equivalent circuit for a detector terminated in a resistor

RL. RL is much less than the output resistance of the detector
i sh2 2q I DCf ; i NR
2
4kTf / RL

A2.5 Noise bandwidth

The overall bandwidth of the noise appearing with a signal of interest will always
be limited to some finite value, if only because of the effect of stray reactance.
More usually, however, the bandwidth is fixed at a well defined value owing to
the low-pass filter effect of the transducer and output amplifier.

Appendix 26
When the noise inherent in the experimental process is broadband in nature, the
combined frequency response of the transducer/amplifier combination is often
responsible for the spectral characteristics of the noise observed in the final
measurement. For example, suppose the experimental noise has constant spectral
density WN over a wide frequency range, and that we can identify a frequency-
response function, H(j), with comparatively narrow bandwidth. The spectrum of
the output noise can then be approximated by
W(f) = WN H(j)2, = 2f
Conversely, an observed noise spectrum can often be modelled by assuming that
it originates from the passage of broadband white-noise through a filter with
appropriate frequency-response characteristics. Broadband spectra and
narrowband spectra such as those shown in Fig. A2.6 are examples where this
approach is often successful.
In order to calculate the total fluctuation due to the noise we integrate its
spectrum over all frequencies to obtain the mean-square value:

N0 = WN 0 H(j)2d /2

Since the integral depends only on the filter transfer function we can simplify all
subsequent discussions by defining the noise equivalent bandwidth of the filter.
This gives the bandwidth of the rectangular filter shown in fig. A2.6 which
transmits the same fluctuation as the actual filter of interest.
The noise output of the noise-equivalent rectangular filter with bandwidth BN is
N0 = WN HMAX 2BN
hence, equating this with the integrated noise in the filter characteristic H(j), we
obtain the filter noise-equivalent bandwidth in terms of the integral:

H( j ) 2


BN = 2
d /2
0
H MAX

Fig. A2.6 (a) Broadband and (b) narrowband noise spectra obtained by
filtering white noise. BN denotes the noise equivalent bandwidth

The noise-equivalent bandwidth (or, simply, noise bandwidth) of a practical filter

is somewhat greater than its 3 dB bandwidth, but becomes closer for filters of
higher order. This reflects the sharper cut-off of high-order filters, which
approximate more closely to filters with ideal cut-off characteristics.
In principle, therefore, the output fluctuation of any filter in response to a white-
noise input can be calculated once the maximum gain of the filter and its noise

Appendix 27
bandwidth are known. The latter is usually obtained from a catalogue of noise
bandwidths such as that given in Appendix 4.
Finally, it should be noted that noise bandwidth is conventionally expressed in
hertz and not radians/second.

A2.6 Signal-to-noise-ratio improvement by

filtering
We envisage the situation shown in Fig. A2.7, where a signal, s(t), appears
against a background of white noise having noise bandwidth BI and spectral
density WN.
The signal-to-noise ratio, measured on a mean-square basis, is simply

s 2 t
SNRI =
WN BI
A signal-conditioning filter is now used to attenuate all noise components except
those lying within the frequency band occupied by the signal. The filter
bandwidth is sufficiently wide to transmit the signal without distortion, but
significantly smaller than the input noise bandwidth BI. In this case, the signal-
to-noise ratio at the filter output is given to a good approximation by

s 2 t
SNR0
WN B0
where B0 is the noise bandwidth of the filter. Note that the gain modulus of the
filter, HMAX, does not appear in the expression for signal-to-noise ratio.
Dividing the two signal-to-noise ratios we obtain the signal-to-noise ratio
improvement factor
SNR0/SNRI = BI /B0
This is the classic improvement factor for the recovery of signals from white
noise by filtering.

Fig. A2.7 Spectrum of signal and noise shown with the transmission
characteristics of a noise reduction filter

Appendix 28
A2.7 Low-frequency noise
Practical noise spectra almost invariably display a steady rise in spectral density
as lower and lower frequencies are taken into account. This is the so-called
flicker-noise region where the spectral density follows a law
W(f) = W0/f x
Here, W0 is a constant and x takes values, typically, in the range 0.8 to 1.4. The
term 1/f noise is also used to describe spectra of this general type.
Flicker-noise is associated with a wide range of physical processes. Although its
origins are obscure, its spectral characteristics are usually well-defined for a
given experimental set-up.

Fig. A2.8 Spectral model for broadband noise with a low-frequency noise
component

Fig. A2.8 gives an example where flicker noise has added a low-frequency tail
to a broadband noise spectrum. This is typical of the noise-spectrum
characteristics of a low-noise voltage amplifier where the corner frequency, fc,
marks the transition between the low-frequency and white-noise regions. We can
use the corner frequency to provide the following description of the overall
spectrum:
W(f) = WN [1 + fc/f x], f < BN
where WN is the constant spectral density in the white-noise zone and BN is the
overall noise bandwidth.
The widespread incidence of flicker-noise accounts for the equally widespread
use of a.c. excitation in experimental work, the object being to bring the signal of
interest into the spectral region above the corner frequency. If a clear separation
is not achieved, then it may sometimes be necessary to calculate the total
fluctuation from the frequency interval f1 to f2 shown in Fig. A2.8 which includes
the corner frequency. For the purpose of calculation it is usual to assume that
x = 1. To do otherwise implies that the spectrum of low-frequency noise has been
characterised very carefully.
Our spectral model gives a total mean-square fluctuation:

f2
WN [1 + fc/f] df = WN (f1 f2) + WN fc 1n (f2/f1)
f1

We thus find that the white-noise component gives rise to a mean-square

fluctuation proportional to the measurement bandwidth, (f2 f1), while the
flicker-noise contribution depends on the frequency ratio. We conclude from this
that the flicker-noise fluctuation measured per octave or per decade is constant
over all frequencies.

Appendix 29
A2.8 More about narrowband noise
The term narrowband noise is generally used to describe noise that has zero
spectral density in the vicinity of f = 0. It was remarked in Section A2.5 that
narrowband noise can often be modelled by supposing that white noise has been
transmitted by a filter with appropriate frequency-response characteristics. In the
example shown in Fig. A2.9, the filter is highly selective and the output noise has
a bandwidth much less than the centre frequency, f0. Under these circumstances
we find that the noise has the appearance of a noisy sinewave since all
components except those in the immediate vicinity of f0 have been suppressed by
the filter.

Fig. A2.9 Generation of narrowband noise

We shall find that the structure of narrowband noise lends itself to a time-domain
description which proves to be very useful when considering the response of
synchronous detectors to noise inputs. To provide a time-domain model we
suppose that we start with a clean sinewave at frequency f0 and then impose
random variations on its instantaneous amplitude and phase. The result is a
voltage:
n(t) = R(t) cos [0t +(t)]
R(t) and (t) are random modulations that vary very slowly in comparison with
cos0t. (t) is a simple phase modulation while Fig. A2.9 shows that we can
interpret R(t) as the envelope* of the noise. Because R(t) is a relatively slow
variation, we find that there is no dramatic change in the envelope over several
cycles at frequency f0.
We now expand n(t) into its constituent components to obtain:
n(t) = R(t) cos (t) cos0t R(t) sin (t) sin0t
and then define:
ni(t) = R(t) cos (t)
nq(t) = R(t) sin (t)
Thus:
n(t) = ni(t) cost nq(t) sin0t
where
[ni2 (t) + nq2(t)]1/2 = R(t)
and
nq(t)/ni(t) = tan(t)

*
An ideal rectifier, incorporating an output low-pass filter, would deliver an output voltage which
varied in response to the envelope function, R(t).
Appendix 210
We can clarify some of the steps taken so far by supposing that the filter is
precisely tuned to a signal, cos0t, which appears in the filter output together
with the narrowband noise. In this case, it appears that the noise has a component
ni(t) that lies in-phase with the signal and a component nq(t) in quadrature with
the signal. This suggest a phasor representation for the noise as shown in Fig.
A2.10.

Fig. A2.10 Phasor representation of narrowband noise

Relationships between n(t), ni(t) and nq(t) and their spectra are derived in most
books on communication theory; for example, Taub and Schilling1, and Haykin2,
but for our purposes it will be sufficient to note the following properties:
(i) Mean value: We are, dealing with noise at the output of a bandpass filter
which implies that n(t), and hence ni(t) and nq(t), have zero mean value, that
is:
n(t) ni (t) n q (t) 0

(ii) Mean-square value: The phasor diagram shown in Fig. A2.10 is a snapshot
taken at a particular instant. The fact that ni and nq have zero mean values
implies that the noise phasor spends an equal amount of time on average
in all four quadrants. This symmetry suggests that ni and nq have equal mean
square values and it can be shown that this is indeed the case:

ni2 (t) nq2 (t)

In general, the noise processes ni(t) and nq(t) are uncorrelated. Hence, the total
mean-square fluctuation of the narrowband noise, n(t), is:

n 2 (t) ni2 (t) cos 2 0 t n q2 (t) sin 2 0 t 12 n i2 (t) 12 n q2 (t)

We thus obtain:

n 2 (t) ni2 (t) nq2 (t)

(iii) Spectral density of ni(t) and nq(t): If we denote the spectrum of the narrow-
band noise by Wn(f), then the spectra of ni(t) and nq(t) are identical, obtained
through the transformation1,2:
Wi(f) = Wq(f) = Wn(f f0) + Wn(f + f0)
where f0 is the centre frequency of the narrowband noise process.
Wi(f) and Wq(f) generally have the form of low-pass spectra. In the special case
where Wn(f) is symmetrical about f = f0, we obtain:

Appendix 211
 Wi(f) = Wq(f) = 2Wn (f + f0)

Fig. A2.11 (a) Band-limited white noise; (b) spectrum of nI(t) and nq(t)

For example, suppose the narrowband noise has the form of band-limited white
noise as shown in Fig. A2.11. We have:
W , f0 BI /2 f f0 BI /2
Wn f N
0, elsewhere
In this case, Wi(f) and Wq(f) take the form:
2WN , f BI/2
Wi(f) = Wq(f)
0, elsewhere
Integrating Wi(f), Wq(f) and Wn(f ) over all frequencies we confirm the results
given earlier, namely:

ni2 (t) nq2 (t) n 2 (t)

where, in this case:

n 2 (t) WN BI

A2.9 References
1 TAUB, H., and SCHILLING, D.L. (1971): Principles of communication
systems (New York, McGraw Hill)
2 HAYKIN, S. (1978): Communications systems (New York, John Wiley &
Sons)

Appendix 212
APPENDIX 3

Synchronous detection and noise

A3.1 Signal-to-noise-ratio improvement

Let us consider the response of a synchronous detector to an amplitude-
modulated signal perturbed by random noise, giving an input of the form:
v in t m tcos 0 t n t

We have seen that the operation of synchronous detectors does not necessarily
depend on the elimination of unwanted noise components by filtering in advance
of detection. However, it was shown in Appendix 2 that the structure of narrow-
band noise is particularly convenient when it comes to performing calculations in
the time domain. We shall therefore assume that the input noise is band-limited
as shown in Fig. A3.1 with a bandwidth BI much greater than the signal
bandwidth 2BM .

Fig. A3.1 Spectra of: (a) input noise and (b) amplitude-modulated signal

If we further assume that the noise spectrum is centered on the signal frequency,
we can use the results of Appendix A2.8 and write:
n t ni tcos 0 t n q tsin 0 t

The input voltage to the synchronous detector now has the form:
v in t m t n i tcos 0 t n q t sin 0 t

Following the arguments developed in Appendix A2.8, we identify ni t and nq t

as the components of the noise lying, respectively, in phase and in quadrature
with the signal.
A reference voltage, synchronous with the signal of interest, is now introduced at
the synchronous detector and the phase is adjusted to bring signal and reference
in phase. The reference voltage is:

Appendix 31
 v R t 2VR cos 0 t

giving multiplication products:

v in tv R t
VR
2
m t n i t 1 cos 2 0 t R nq t sin20 t
V
2
At this stage we shall assume that the output low-pass filter is used only to
eliminate components centered on frequency 2 0 without modifying the low-
frequency output of the multiplier. We obtain an output voltage:

v 0 t
VR
2
mt ni t
giving an output signal-to-noise ratio:

SNR0 m 2 t / n i t
2

This important result shows that the output signal-to-noise ratio is given in terms
of the noise components that lie in-phase with reference voltage. We thus
conclude that the quadrature noise components nq t are rejected at the point of
detection and so make no contribution to the low-frequency output.
The input signal-to-noise ratio is:

SNRI m 2 tcos 2 0 t / n 2 t 12 m 2
t / n 2 t

From Section A2.8 we have:

ni t n 2 t WN BI
2

Hence:
SNR0
2
SNRI
A signal-to-noise improvement factor of 2 is thus inherent in the operation of the
synchronous detector.

Fig. A3.2 (a), (b) Spectra of output noise and recovered modulation signal
in a synchronous detector; (c) frequency response of noise-
reduction filter in final output

Appendix 32
The spectra of the recovered modulation signal and the output noise are shown in
Fig. A3.2 for the case where VR 2 . We now suppose that the bandwidth of the
low-pass filter is greatly reduced in order to eliminate noise components from the
final output. The output noise bandwidth is accordingly set to a value B0 as
indicated in Fig. A3.2 where B0 is sufficiently wide to transmit the recovered
modulation signal without distortion.
When B0 << BI we can approximate the mean-square value of the noise following
the low-pass filter by:
N 0 2WN B0
and so obtain the output signal-to-noise ratio:

m 2 t
SNR0
2WN B0
compared with the input signal-to-noise ratio:

m 2 t
SNRI
2WN BI
The signal-to-noise improvement factor is therefore:
SNR0
BI / B0
SNRI
The "classic" signal-to-noise improvement factor derived in Appendix 2 for
linear filters is thus applicable to synchronous detectors. The noise bandwidth of
the detector is determined simply by B0 , the noise bandwidth of the low-pass
filter.

Fig. A3.3 Synchronous detector with arbitrary input noise spectrum

An alternative approach which helps to put these results into perspective involves
the idea of a transmission "window" which was first introduced in Section 2.4.
We have seen that the only asynchronous components which survive to perturb
the final output of a synchronous detector are those which are confined to a
transmission window centred on the reference frequency, having a noise
bandwidth equal to twice the noise bandwidth of the low-pass filter. Using the
spectral model shown in Fig. A3.3 for an arbitrary noise spectrum, the mean-
square fluctuation associated with the components within the transmission
windows is, approximately:
N 0 2B0WN
Appendix 33
However, from the results given above, it is evident that the synchronous detector
responds only to the components of the noise that lie in-phase with the reference.
Of the total mean-square fluctuation appearing within the transmission window
we can ascribe one half to noise components lying in phase with the reference
and one half to noise components in quadrature with the reference. If we now
suppose that a synchronous detector has a full-scale sensitivity SF and a full-scale
output VF, the noise appearing in the final output will have a mean-square value:

v N2 12 VF / SF N 0 VF / SF B0WN
2 2

The factor 1/2 accounts for the loss of the quadrature noise components at the
point of detection. Although the synchronous detector transmission window has a
noise bandwidth 2B0 , the rejection of the quadrature noise components results in
an effective noise bandwidth of B0 .
Note that if a two-phase lock-in amplifier is used with a noisy input, the residual
noise outputs from the two phase-sensitive detectors will originate respectively
from the in-phase and quadrature components of the noise. The fluctuations
observed at the two outputs will, in general, be uncorrelated but otherwise have
similar statistical properties.

A3.2 Noise measurements

Lock-in amplifiers are often used in a noise-measurement mode where the
reference frequency and low-pass filter are selected to define a narrow
measurement bandwidth centered on a spectral region of interest. A "noise
measurement" unit is then used to measure the r.m.s. noise output from the low-
pass filter. The result is a measure of the "spot" spectral density in the immediate
vicinity of the reference frequency. This is clearly an application for a
fundamental-only responding lock-in system: otherwise noise "leakage" from
harmonic transmission windows could seriously affect the outcome of a
measurement.
The noise bandwidth of a two-section RC low-pass filter is:
B0 1/8T0

where T0 is the selected time constant. If the noise has an r.m.s. spectral density
VN at the reference frequency, the noise appearing in the final output will have an
r.m.s.value:

VNVF 1
1/ 2

Vr.m.s
2SF 4T0
If Vr.m.s is measured with a noise measurement unit, the noise voltage spectral
density referred to input is:

VN 2ANVr.m.sSF 2T0 /VF

1/ 2

where AN is a scaling factor specified for the noise measurement unit.

Lock-in amplifiers give a unique mode of measurement whereby the noise can be
measured in the presence of a synchronous signal without errors due to
intermodulation. In practice, this means that noise-measurement units are almost
invariably a.c. coupled to ensure that d.c. components due to detected signal do
not affect the measurement of the r.m.s. value of the noise output.

Appendix 34
APPENDIX 4

Signal conditioning filters

A4.1 Low-pass Filters

A4.1.1First-order

0 1/ RC 1/T0

H j 1/ 1 2 / 02
1/ 2

tan 1 / 0
Noise bandwidth:
BN 0 / 4 1/4T0 , Hz
A4.1.2Second order

0 1 / RC 1 / T0

Appendix 41
 H j 1 / 1 2 / 02

2 tan 1 / 0
Noise bandwidth:
BN 0 /8 1/8T0 , Hz
A4.2 High-pass Filters
A4.2.1First order

0 1 / RC 1 / T0
/ 0
H j
1 2 / 02 1 / 2
tan 1 0 /
A4.2.2Second order

Appendix 42
 0 1 / RC 1 / T0
/ 02
H j
1 2 / 02
2tan 1 0 /

A4.3 Active tuned filters

A4.3.1Band pass

j0 /Q
H j
02 j0 /Q 2
3dB bandwidth: 0 / Q

Noise bandwidth (for Q 12 ) :

BN 0 /4Q, Hz
Approximations, Q 5 , 0 :
1. << 0 /Q :

H j 1/ 1 4Q 2 2 / 02
1/ 2

tan 1 2Q / 0
2. >> 0 /Q :

/0
H j
Q1 2 / 02

Appendix 43
A4.3.2Low-pass

02 /Q
H j
02 j 0 /Q 2
3dB bandwidth: 0 / Q

Noise bandwidth (for Q 12 ) :

BN 0 /4Q, Hz
Approximations, Q 5 , 0 :
1. << 0 /Q

H j 1/ 1 4Q 2 2 / 02
1/ 2

tan1 2Q / 0 /2
2. >> 0 /Q

1
H j
Q1 2 / 02

Appendix 44
A4.4 Active notch filter

02 / 2
H j
j0 /Q 2
2
0

Notch width at 3dB points:

0 / Q
Attenuation at 0 :

> 70 dB (typical)

Appendix 45
APPENDIX 5

Amplifier selection and noise matching

A5.1 Introduction
If operational problems are encountered with a lock-in recovery system there is a
strong likelihood that they can be traced to the critical interface between the
signal source and the equipment used for detection. If every effort is made to
optimize performance in the "front end" area, then lock-in recovery becomes a
relatively straightforward business. So often, however, the interface is given very
cursory treatment and scant attention is paid either to the choice of a suitable
preamplifier or to the layout of the equipment and cable runs. As a result the
signal-to-noise ratio encountered at source cannot be maintained through the
point of detection, so measurement becomes a time-consuming and painstaking
task.
The basic problems to be tacked in this Appendix are as follows. First of all, how
to decide on the type of amplifier to be used in a given application so as to ensure
that the signal is handled in the most effective and predictable way. Secondly,
how to ensure that the input signal-to-noise ratio is not unduly degraded in the
process of amplification, recognizing that even a "low-noise" amplifier can
generate a significant amount of noise in some circumstances.
It will be assumed throughout that the principal noise limitations arise from
thermal noise and shot noise. For an extension to more complicated noise models,
reference should be made to a paper by Faulkner1. Note that design aspects of
low-noise amplifiers is a topic excluded from the present treatment.

A5.2 What kind of amplifier?

We have assumed throughout our earlier discussions that the signal of interest
appears in the form of an electrical signal, usually at the output of an electrical
transducer. From now on we must regard this as a signal source and we shall find
it convenient to represent it in the form of a circuit model. This is a necessary step
if we are to consider the effect of making external connections.
The exact model may be more or less complex, but in many practical situations it
is sufficient to use the Thevenin or Norton forms shown in Fig. A5.1 These basic
forms remind us that signals always appear from a source of finite inpedance Zs.
In Fig. A5.1(a) the source is represented as a signal voltage generator vs and the
additional noise voltage generator vNs accounts for all the noise generated within
the experiment and within the transducer itself. In Fig. A5.1(b) the signal appears
in the form of a current is which is perturbed by a noise current represented here
as the generator iNs.

Appendix 51
Fig. A5.1 Signal source equivalent circuits

In either case, the purpose of a preamplifier is to provide an output voltage which

is proportional to the output of the signal source. The choice of an appropriately
"low-noise" amplifier will ensure that the signal-to-noise ratio encountered at
source is not significantly degraded in the process. It must be remembered that,
while we are free to choose from a range of amplifiers or to make external
connections to the signal source, the characteristics of the source must be
assumed to be fixed. The problem is to choose an amplifier to suit the source, not
the other way around!
In general, it is possible to define the characteristics of an amplifier to ensure that
the final output depends on vs or is alone, rather than on some combination of vs or
is with Zs. For example, with a voltage source such as Fig. A5.1(a) it is sufficient
to use an amplifier in which the input inpedance has magnitude Z in >> Z s . This
will normally be specified as a voltage amplifier having a well-defined voltage
gain over the frequency range of interest.
In the case of the current source of Fig. A5.1(b) we require an amplifier with a
very low input inpedance compared with Zs in order to measure the output current
independently of Zs. Such an amplifier would normally deliver an output voltage
which is proportional to is and so is often referred to as a transimpedance
amplifier. A common way to achieve a transimpedance amplifier is to use a
voltage amplifier with a low impedance ZL shunted across its input terminals as
shown in Fig. A5.2. Those familiar with operating photomultiplier tubes will
recognize this arrangement where ZL is replaced by RL, the anode load resistor.
One reason for the popularity of this configuration is that RL converts a hitherto
"unseen" current into a more readily observable voltage variation, isRL, which is
subsequently amplified to give an output AvisRL. Dividing output by input we
obtain a transimpedance AvRL although the overall operation is very rarely
thought of in this light. Thus, increasing the load resistor is almost invariably
looked upon as a means of increasing the signal voltage rather than as a means of
obtaining a larger transimpedance.

Fig. A5.2 Transimpedance amplifier. Cs represents stray reactance in the

input circuit. |Zin| >> |Zs|

Appendix 52
As we shall see, the noise performance of this arrangement deteriorates at low
values of RL. Furthermore, there are severe operational difficulties when using
high values of RL due to stray reactance and microphony. In practice, these
difficulties can be largely avoided by the use of purpose-built current amplifiers
in which high values of transimpedance consistent with low input inpedance are
obtained through the use of parallel feedback.
For a given noise performance with high-impedance sources such as
photomultipliers, current amplifiers can give much improved handling
characteristics in terms of gain stability and relative freedom from "cable"
effects. These characteristics are reviewed in Section A5.4, with special reference
to photometric measurements.

Fig. A5.3 (a) Application for a differential amplifier; (b) equivalent circuit
of source. ZA = Z2 // Z3; ZB = Z1 // Z4

Finally, let us look at another type of voltage source exemplified by the bridge
circuit shown in Fig. A5.3. Here, the signal of interest appears as the difference in
potential between two points in the bridge where neither point is at ground
potential. This gives us a typical application for a differential voltage amplifier
connected as in Fig. A5.3(a). The usual arrangement is that the impedance of
each amplifier input is much larger than the source inpedance presented by the
bridge. This can be found by applying Thevenin's theorem to obtain the source
equivalent circuit shown in Fig. A5.3(b). When the output inpedances ZA and ZB
are identical the bridge is said to behave as a balanced source (not to be confused
with a balanced bridge). In an unbalanced source there may be a large difference
between the impedances of the two arms; however, the output inpedances are
usually affected only slightly by the small adjustments which are made to the
bridge at its null point.
A differential amplifier has three input terminals: A, B and ground. A voltage
applied between terminals A and B is called a series or differential mode voltage.
The mean voltage of A and B with respect to ground is called the common-mode
voltage. A prime specification of a differential amplifier is its common-mode
rejection ratio (c.m.r.r.) which gives the ratio of the series-mode gain to the
common-mode gain. C.M.R.R. thus measures the ability of a differential
amplifier to reject a voltage applied equally to its inputs. For example, for an
amplifier with a series mode or differential gain of 100 (40 dB) and a c.m.r.r. of
105 (100 dB), a 1V common-mode voltage would produce an output of
1 V 100 10 5 1 mV .
C.M.R.R. is usually specified at a midband frequency, say 1 kHz, and will be in
excess of 120 dB for a good-quality amplifier. Rejection falls with increasing
frequency and a front-panel adjustment is often provided to maximize rejection at
a frequency of interest.

Appendix 53
Fig. A5.4 (a) Differential with single-ended source; (b) alternative
configuration using a single cable and an amplifier switched to
"A-diff" mode

Differential amplifiers have an important role to play in suppressing ground loops

in measurement systems. For this reason they are often specified for use with
single-ended voltage sources as shown in Fig. A5.4(a). This arrangement is
discussed further in Appendix 6. To facilitate connections with BNC
terminations, some amplifiers incorporate a switch position labelled "A-diff"
whereby the connections illustrated in Fig. A5.4(b) are made automatically. If the
differential amplifier has identical input impedances on inputs A and B the inputs
are said to be balanced. Other amplifiers offer a pseudo-differential or
unbalanced input where a single input connector is wired as shown in Fig.
A5.4(b) but the two input impedances are not the same. In this case, the common-
mode rejection ratio is not so spectacular (usually around 80 dB), but it is
possible to achieve moderate differential performance for ground-loop
suppression consistent with extremely low noise. "True" differential amplifiers
are generally 3 dB more noisy than their single-ended counterparts and are
normally used where ultra-low-noise performance is not required. In practice, the
increased immunity of a true differential stage to common-mode inputs may
outweigh its additional noise contribution, and this is usually the preferred
configuration for "general purpose" amplifiers.

A5.3 Noise in voltage amplifiers

A5.3.1Introduction

Fig. A5.5 Voltage source with amplifier

We begin with Fig. A5.5, which shows a voltage amplifier connected to a signal
source. The entire source noise is accounted for by the random voltage generator

Appendix 54
vNs which has spectral density W f 0 in the vicinity of the signal frequency f 0 .
The source signal-to-noise ratio measured in a small bandwidth f centered on
f 0 is therefore

v s2
SNRI
W f0 f

We shall now assume that the signal originates in a linear passive device and that
the source inpedance is resistive with value Rs over the frequency range of
interest. In this important case the limitation at source is due to thermal noise so
that the best possible value of source signal-to-noise ratio is

SNRI v s2 /4kTRs f

where k is Boltzmann's constant and T the equilibrium temperature.

The output of the amplifier would, ideally, be an amplified version of the total
input from the signal source. In practice, we must allow for noise generated
within the amplifier which gives rise to an additional output fluctuation with
mean-square value v A2 . The output signal-to-noise ratio is therefore less than
SNRI, and is given by:

Av2 v s2
SNR0
v A2 Av2 4kTRsf
where Av is the gain of the amplifier at the signal frequency.
We now define the noise figure of the amplifying system
best possible SNR0
F
actual SNR0
This ratio will always be greater than unity for any real combination of voltage
source and amplifier. In the present example, the noise figure takes the form:

v A2
F 1
Av2 4kTRsf
Let us now turn to the noise model shown in Fig. A5.6(a) which will enable us to
predict the behaviour of the amplifier under a wide range of operating conditions.

Fig. A5.6 (a) Noise model for a voltage amplifier (b) Typical spectral
densities of the noise generators vNA and iNA
Appendix 55
Here, the total noise of the amplifier is attributed to a pair of random-noise
generators connected at its input. As is usual in such models, the amplifier itself,
including the input impedance ZI, is assumed to be noise-free. In this
representation, it is clear that the signal is in competition with the amplifier noise
generators vNA and iNA. in addition to the noise associated with the source. A
fully documented voltage amplifier will have vNA and iNA specified in terms of
their r.m.s. spectral densities over the entire frequency range of the amplifier.
Fig A.5.6(b) gives an example which is typical of modern amplifiers using a
j.f.e.t. input stage.
An alternative presentation which proves to be extremely useful in practicc is
2 2
derived as follows. We note that the mean-square fluctuations v NA and i NA
appearing in a bandwidth f centred on any frequency of interest can always be
associated with equivalent noise resistances RNv and RNi defined by
2
v NA
RNv
4kTf

RNi 4kTf /i NA
2

From Fig. A5.6(b) we see that RNv and RNi will take more or less constant values
over several decades of frequency, deviating at the extremes of the operating
range of the amplifier.
In the case of j.f.e.t. amplifiers the current noise originates almost entirely with
the passive resistor which is used to define the amplifier input resistance. As a
result, we find a virtual one-to-one correspondence in manufacturer's catalogues
between the midband value of RNi and input resistance RI. The noise of the
amplifier measured with the input open-circuit is thus, for the most part, due to
the thermal noise amplified from its input termination resistor. At the other end of
the scale, the short-circuit noise gives a measure of the noise-voltage generator
VNA but here the results depend very much on the selection of input transistors
and on the circuit configuration. A common feature is that the voltage noise
exhibits a flicker-noise dependence at frequencies below the flicker-noise corner
frequency (see Appendix 2).
The corner frequency is usually below 1kHz, and is usually specified by
manufacturers. In this case we can account for the variation in spectral density
and find the appropriate value of RNv by using the approximate relationship:
2
v NA
RNv
4kTf
1 fc / f

Where v NA 2
/ f is the midband spectral density of the noise voltage generator and
fc is the corner frequency.
A5.3.2Noise-figure calculations
We shall use our amplifier noise model to calculate the output signal-to-noise
ratio when the source is limited by the thermal noise of the source resistor.
To do this it is helpful to transform the input circuit of Fig. A5.6(a) to obtain the
modified noise equivalent circuit of Fig.A5.7. This shows clearly how the
contribution of i NA depends on the source resistance Rs. We shall assume that
v NA and i NA are independent noise sources (that is they exhibit no correlation)
2 2
and that they give rise to mean-square fluctuations v NA and i NA in a frequency

Appendix 56
interval f centred on the signal frequency. The signal then appears in association
with a total fluctuation

v T2 4kTRs f v NA
2
Rs2 i NA
2

and the signal-to-noise ratio measured at the amplifier output becomes

SNR0 v s2 /[4kTRs f v NA
2
Rs2 i NA
2
]

Fig. A5.7 Transformed noise-equivalent circuit

An ideal amplifier would give an output signal-to-noise ratio equal to the value
measured at source, v s2 / 4kTRsf . The actual value is therefore worse by a factor
best possible SNR0
F
actual SNR0
2
v NA Rs2 i NA
2
1
4kTRs f
With our particular amplifier noise model, the expression for noise figure can be
put into much simpler form using the equivalent noise resistances defined in the
previous section. The result is
F 1 RNv / Rs Rs / RNi
which shows more clearly the dependence of F on the source resistance Rs. Also,
F is subject to the frequency dependence of RNv and RNi.
If noise figure is plotted versus source resistance, using values of RNv and RNi
appropriate to the frequency range of interest, the result is the parabolic curve of
Fig. A5.8 which exhibits a minimum value for a value of source resistance given
by Rs RNv RNi .

Fig. A5.8 Dependence of noise figure on source resistance.

Appendix 57
The rise in noise figure for values of Rs less than RNv is a reminder that we cannot
expect an amplifier to generate less noise than the thermal noise of an arbitrarily
small source resistance. The noise figure similarly increases as Rs exceeds RNi:
from the remarks made in the previous section this would normally correspond to
operating a voltage amplifier from a source resistance which is greater than the
amplifier input resistance.
It is evident from Fig. A5.8 that the noise-figure graph has a very broad minimum
when RNi >> RNv . Under this condition we obtain a noise figure of 3 dB (F = 2)
when Rs = RNv or when Rs = RNi, and a value of F 1 for Rs RNv RNi .
Although it is instructive to observe how the noise figure of a given amplifier
varies with source resistance, in practice we are usually faced with a fixed value
of source resistance. We must then choose an amplifier from a wide range of
competing devices which gives an acceptable noise figure. Any amplifier which
achieves this end with a noise figure of less than 3 dB can be said to be "low
noise" within the context of a given experiment, and an amplifier which is
capable of giving this performance over a wide range of source resistance is said
to have a high figure of merit, M, defined by
M RNi / RNv
Modern j.f.e.t. amplifiers have figures of merit in the range 100 to 1000 and they
can be roughly classified according to the lowest value of source resistance which
can be handled with an acceptable noise figure. Thus, an amplifier catalogued as
"low noise" would normally be useful for source resistances as low as 1 k
(voltage noise of 4 nV/ Hz at midband) while "ultra low-noise" units extend the
useful range to 40 or 50 (voltage noise of 800 pV/ Hz at midband).
A5.3.3Minimum noise figure and optimum source resistance
The value of source resistance which minimizes the noise figure is known as the
optimum source resistance:
Ropt RNv RNi

The minimum noise figure obtained with Rs Ropt is:

FMIN 1 2 RNv / RNi 1 2 / M

For practical purposes we can say that a noise figure of 1 dB (F = 1.26) is
indistinguishable from the best possible figure of 0 dB. Since figures of merit of
100 and greater are obtainable, the minimum noise figure is easily achieved when
Rs falls in the correct range. More to the point, however, we see that noise figures
indistinguishable from the ideal can be obtained from a wide range of source
resistances - even when Rs differs from Ropt by an order of magnitude - as shown
in the first example below.
Example 1
A low level signal is to be measured from a source of 10 k resistance at a
frequency of 5 kHz. An amplifier is available with the following specification:
r.m.s. noise voltage density (f 1 kHz):
4 nV / Hz, RNv 1 k
r.m.s. noise current density
14 fA / Hz, RNi 100 M
What is the noise figure that can be achieved with this combination?

Appendix 58
The noise figure is given by:
F 1 RNv / Rs Rs / RNi
1 0.1 10 5
= 1.1 (0.4 dB)
Let us now calculate the minimum noise figure which can be obtained using this
amplifier in the same frequency range but with the optimum value of source
resistance. This is
FMIN 1 2 RNv / RNi

= 1.006 (0.27 dB)

which is obtained at a value of source resistance given by
Ropt RNv RNi

316 k
If the signal is sufficiently strong that a degradation of 3 dB in signal-to-noise
ratio can be tolerated, this same amplifier will be suitable for sources with
resistances in the range 1 k to 100 M .
Example 2
A signal of 100 nV r.m.s. is to be measured from a source of resistance 100 in
a bandwidth of 1 kHz using the same amplifier as in example 1.
In this case the system will have a noise figure
F 1 10 10 6
= 11 (10.4 dB)
The r.m.s. noise voltage associated with a source resistance of x k at laboratory
temperature is (Appendix 2):
4 x nV / Hz
Hence the input signal-to-noise ratio is
(100 10 9 ) 2
SNRI , f 1 kHz
16.10 18 0.1 f

= 6.25 (8 dB)
which will be reduced to 6.25/11 when the signal is amplified.
There is a clear case for seeking an amplifier with better noise performance
Suppose now, however, that the signal appears at a level of 1 V r.m.s. The input
signal-to-noise ratio is now increased to
SNRI 625 (28 dB)
In this case an amplifier noise figure of 10 dB or so would reduce the output
signal-to-noise ratio to about 18 dB, which might be considered quite adequate if
the signal is to be measured in a recovery system. A decision to select an "ultra
low-noise" amplifier with RNv 100 or less may then be uneconomical
provided that 1 V r.m.s. represents the minimum value of the signal for all time.
If, at a later stage, the signal amplitude is likely to be reduced beyond this value,
then the question of amplifier noise will undoubtedly be raised again.

Appendix 59
Finally it should be noted that these examples are for operation at a fixed
frequency, using the values of RNv and RNi appropriate to that frequency. If
operation over a wider frequency range is envisaged, then more information is
required. Fortunately, this is usually available, as discussed in the following
section.
A5.3.4Noise-figure contours
Since noise figure depends upon frequency in a fairly complicated way, most
manufacturers elect to present their data graphically in the form of noise-figure
contours.
Figure A5.9 shows a single contour, the 3 dB contour, drawn against axes
labelled with source resistance and frequency on logarithmic scales. The shape of
the contour is derived as follows: first of all, the regions (i) and (ii). These define
the lower and upper limits of Rs required to give a noise figure of 3 dB at
midband. They thus coincide with the midband values of RNv and RNi . The rise in
the lower contour in region (iii) results from the rise in RNv at low frequencies due
to flicker-noise effects in the amplifier, and shows that the value of Rs required to
maintain a 3 dB noise figure becomes progressively larger as the operating
frequency is reduced.

Fig. A5.9 The 3 dB noise-figure contour for a low-noise voltage amplifier

The sloping characteristic in region (iv) indicates that RNi is, in fact, reduced at
high frequencies and that Rs must be reduced in proportion if the noise figure is
to be held at the midband value of 3 dB. This is one area where the contours are
particularly useful since the high-frequency cut-off depends on the "noise
capacitance" of the amplifier which may not otherwise be specified. We can be
sure, however, that if the amplifier input is heavily loaded with cable capacitance,
the turn-over in the upper contour will shift to lower frequencies as indicated by
the broken line appropriate to "system" operation.

Appendix 510
The area enclosed by the 3 dB contour is often called the "optimum performance"
region. Its vertical extent is maximized by selecting an amplifier with a high
figure of merit, but this should be consistent with a low value of RNv if optimum
performance is required from low source resistances. In many voltage-amplifier
applications where the source resistance is less than 1 M, the loss of
performance in region (iv) is not normally significant. However, the turn-over
frequency can be as low as 1 kHz for amplifiers with extremely high input
resistance, of the order of 1 G.
A complete set of contours for a general-purpose low-noise voltage amplifier is
shown in Fig. A5.10 exactly in the form which might be encountered in a
manufacturer's catalogue. The contours provide the means for a researcher to
compare competing amplifiers and find the one which offers the best noise
performance for a particular combination of source resistance and operating
frequency.

Fig. A5.10 Typical noise-figure contours

Alternatively, if an amplifier is available and there is sufficient latitude in, say,

the choice of experimental frequency, it may be possible to arrange to operate
within the optimum performance region of the amplifier. For example, if an
optical detector provides a signal from a source resistance of 10 k and the
amplifier contours are those given by Fig. A5.10, we find that changing the
optical chopping frequency from 10 Hz to 100 Hz brings an improvement in
noise figure from 3 dB to less than 1 dB even though both frequencies lie below
the flicker-noise corner frequency of the amplifier.
A5.3.5Cryogenic sources
The noise figure data supplied by manufacturers almost invariably refer to the
source and amplifier at normal laboratory temperatures. Let us now return to our
original definition of noise figure and rework the results for the more general case

Appendix 511
where the source is at a different temperature to the amplifier. We shall assume
that vNA and iNA are specified at laboratory temperature TL, say 290 K, and
represent the source temperature by Ts. The noise figure is now:

i NA
2 2 2
v NA Rs
F 1
4kTs Rsf
Using the equivalent noise resistances calculated at TL we obtain:
F 1 RNv / Rs Rs / RNi TL /TS

which indicates an inevitable deterioration in noise figure if Ts is less than TL.

If we denote the noise figure obtainable with the source at laboratory temperature
by FL, then

RNv / Rs Rs / RNi FL 1
and the general expression for the noise figure becomes:
F 1 FL 1 TL /TS

The next step is to identify the quantity FL 1 TL with the noise temperature,
Te, of the amplifier. The cryogenic noise figure is then
F 1 Te /Ts
Thus, we find that to give a noise figure of 2 (3 dB) with a cryogenic source, the
noise temperature of the amplifier must equal the temperature of the source.
Unfortunately, low-frequency voltage amplifiers are rarely specified directly in
terms of noise temperature. The following examples show the sort of calculation
which must usually be undertaken.
Example 1
An amplifier has a noise figure of 2 dB when operated with an optical detector at
room temperature (FL = 1.58). What is the achievable noise figure when
operating with a detector of the same resistance at a temperature of 77 K?
We have
FL 1.58
so the noise figure obtainable at a source temperature of 77 K is:
F 1 (1.58 1) 290 / 77
3.18 (5 dB)
Example 2
For a given value of source resistance, what noise figure must be achieved with a
source at laboratory temperature to ensure a cryogenic noise figure of 3 dB for
sources at (a) 77 K and (b) 4 K?
The amplifier should be capable of operating at a noise temperature of 77 K in
the first case and 4 K in the second. The maximum acceptable noise figure with
the source at laboratory temperature is then
FL 1 Te /290
This gives (a) FL = 1.26 (1 dB) and (b) FL = 1.014 (0.06 dB).
In this situation it is clear that the "optimum performance" region - which is
bounded by the 3dB noise-figure contour for operation with sources at laboratory
temperature - is now considerably reduced in area and effectively replaced by a
Appendix 512
smaller region bounded by the 1 dB or even the 0.05 dB contour. This places
restrictions on the choice of operating frequency and defines a much tighter
bound on the value of source resistance required to maintain an acceptable noise
figure. We can conclude that extremely low noise figures have rather more than
academic interest when cryogenic sources are involved. If the source resistance
differs widely from the optimum value appropriate to a given amplifier, it will be
necessary to introduce a stage of noise "matching" using a signal transformer.
A5.3.6Transformer noise matching
The voltage noise resistance RNv of even an "ultra low-noise" amplifier may, in
many cases, be too high to give an acceptable noise figure from sources either of
low resistance or at low temperature. In such cases it may be necessary to resort
to noise "matching" whereby the source resistance is transformed to a new value
which is much closer to the optimum sources resistance of a given amplifier. It
should be noted that noise matching is achieved in the interest of maximizing the
signal-to-noise ratio at the output of an amplifier, and is quite distinct from any
attempt made to maximize either the signal voltage or the signal power through
impedance matching.
As a first step we can disregard any attempt to "transform" the source resistance
by the addition of resistors between the source and the amplifier. Series resistors
merely add to RNv when performing noise calculations, and parallel resistors
cause a reduction in RNi and will always degrade the signal-to-noise ratio. Far
from "reducing signal and noise equally" an input attenuator will always
introduce noise at the expense of the signal.
The usual approach is to introduce a transformer of turns ratio nT as shown in
Fig. A5.11. We shall assume for the moment that the transformer is ideal with no
loss, wide bandwidth and infinite self-inductance. The transformer reflects a
voltage nTvs into its secondary circuit and a resistance nT2 Rs . The signal-to-noise
ratio at the transformer output is thus unchanged and remains at its "best
possible" value while the amplifier "sees" a source of resistance nT2 Rs . By suitable
choice of nT, therefore, we can arrange for the noise matching condition:

nT2 Rs Ropt

and so ensure that the overall system operates at its minimum noise figure.
Practical transformers can bring about a significant improvement in system
performance, but, nevertheless, fall short of the ideal in almost every respect.
Among the factors which must be taken into account are a reduced response
when the transformer is operated outside its recommended frequency range and
the effect of noise generated within the transformer itself. The latter includes the
effects of vibration and the susceptibility of the transformer to pick-up,
particularly at line-related frequencies.

Fig. A5.11 Noise matching using a transformer

Appendix 513
The useful frequency range of a transformer depends jointly on the source
resistance and, in those with multiple tappings, on the selected turns ratio. The
usual behaviour is a restriction on bandwidth when either Rs or nT is increased;
the information is most usefully presented in graphical form.
In commercial transformers the pick-up problem is reduced by packaging the
transformer in a heavily screened box, while the effects of vibration and
microphony are suppressed by the use of shock-absorbing mounting materials.
This leaves the resistance of the windings as the main source of internally
generated noise since the thermal noise of the primary coil plus the noise
reflected from the secondary effectively add to the applied signal.
The noise resistance of a transformer is given in terms of the primary and
secondary coil resistances R1 and R2 by:

RT R1 R2 / nT2
It is thus possible to define a noise figure for a transformer and - as in the case of
amplifiers - to present noise figure as a function of source resistance and
operating frequency. Most useful of all, however, are the noise-figure contours
plotted directly for a given combination of transformer and amplifier that are
made available by some manufacturers.

A5.4Noise in current amplifiers

Our interest in this section is with current amplifiers obtained by applying
parallel feedback to a high-gain, low-noise voltage amplifier. A typical
arrangement is shown in Fig. A5.12(a) in which the current amplifier is
connected to a high-impedance transducer modelled as an ideal current source.

Fig. A5.12 (a) Using a current amplifier; (b) noise equivalent circuit

The current signal is appears in association with a d.c. bias current IDC and the
current amplifier fulfils an essential requirement in that it provides a low-
impedance path to ground for this component. Amplifiers with this property are
said to be able to "sink" a d.c. current (which may be many orders of magnitude
greater than the signal current).
The dominant sources of noise are due to the shot noise of the current source and
thermal noise in the feedback resistor RF, which are included in the noise
equivalent circuit of Fig. A5.12(b). If the amplifier is not to degrade the signal-to-
noise ratio encountered at source, then the shot-noise contribution must exceed
the thermal-noise contribution. Following the arguments developed in Section
A2.4, we obtain the condition:

Although strictly a transimpedance amplifier, the arrangement in Fig. A5.12(a) is usually

catalogued as a current amplifier when the transimpedance is substantially real and constant over
the operating frequency range. The transimpedance is given in this case directly by the feedback
resistor, RF.

Appendix 514
 2kT
I DC , (2kT / q 50 mV at T 290 K)
qRF
This defines a lower limit on IDC to maintain an acceptable noise performance. At
large values of IDC we find a further limitation; this is given by the maximum
value of bias current which can be sunk without driving the amplifier into
saturation. We obtain:
I DC VOUT MAX / RF

and so arrive at the allowed range of IDC:

50 mV / RF I DC VOUT MAX / RF

Even if (VOUT)MAX were little more than 50 mV, there would always be a value
of RF which gave the required current sinking and which contributed noise not
greater than the input current shot noise. In practice (VOUT)MAX is usually of the
order 10 V. Thus, for any given value of RF, the smallest and largest currents
which can be accepted by the amplifier without, on the one hand, suffering
significant noise degradation, and, on the other hand, exceeding the amplifier's
current sinking capability, may be in the ratio 50 mV:10 V or 1:200.
One reason which is often cited for choosing "true" current amplifiers is their
relative immunity to "cable" effects. This refers to the effects of pick-up,
capacitive loading and microphony which can result when a high-impedance
source is terminated by a large load resistor and connections are made via long
cable lengths. When a current amplifier is used, the additional impedance
introduced by such a cable is shunted by the relatively low input impedance of
the amplifier given by:
Z I RF / A( j )
where A(j) is the open-loop voltage gain of the amplifier. It should be noted that
when A(j) has a single-lag response:
A( j) A0 /1 j / 1

with 1 equal to the open-loop bandwidth, the input impedance has an inductive
component and takes the general form:
Z I R jL
where
R RF / A0
and
L RF / 1 A0
Although current amplifiers are particularly immune to the effect of stray
capacitance on the input cable, this capacitance can have an adverse affect on
noise performance at sufficiently high frequencies. To see this, we must take into
account the input voltage noise generator of the amplifier vNA which gives rise to
an input noise current:
i Nv v NA 2fCs
Cs represents the total capacitance of the source and the amplifier input. Its effect
is illustrated graphically in Fig. A5.13 for the case where vNA has a midband
value of 4 nV/ Hz . The graph shows the value of Cs that causes an increase of

Appendix 515
3 dB in the amplifier noise contribution as a function of feedback resistor RF and
frequency.

Fig. A5.13 Maximum allowed value of Cs for optimum noise performance of

a current amplifier

A5.5 References and further reading

1 FAULKNER, E.A. (1975): "The principles of impedance optimisation and
noise matching", J. Phys. E: Sci. Instrum., 8, pp. 533-540.
2 Technical Note 101 (1077): "The use of current preamplifiers", (EG&G
Brookdeal, Bracknell, England).
3 Technical |Note 243 (1976): "Noise in amplifiers", (EG&G Princeton
Applied Research Corp., Princeton, NJ).
4 "Noise figure contours" (1969), (EG&G Princeton Applied Research Corp.,
Princeton, NJ).

Appendix 516
APPENDIX 6

Interference and ground loop suppression

A6.1 Introduction
Although consideration has been given to sources of fundamental noise in
experiments and to the noise contribution of amplifiers, it is noise injected by
pick-up from external sources that is usually the most troublesome in practice. In
this Appendix we shall therefore be dealing with some of the problems that are
met when small signals are handled in a typical laboratory environment.
A useful starting point is to list the principal mechanisms by which interference
couples to experiments:
(i) capacitive pick-up;
(ii) inductive pick-up;
(iii) electromagnetic interference (e.m.i.);
(iv) high-frequency interference superimposed on mains supplies;
(v) ground loops.
Problems associated with ground loops will be left until a later stage. Otherwise,
means of overcoming the first four sources of interference are generally well
known and have been thoroughly documented1,2. These can be summarized as
follows:
(a) Use screened cables to reduce capacitive "hum" pick-up between signal and
power lines to suppress crosstalk between adjacent signal cables. Reinforce
this approach by ensuring that low-level signal cables are routed separately
from mains cords and digital highways. Reduce point-to-point capacitive
pick-up within an experiment by the use of metal enclosures or fine mesh
screens.
(b) Arrange for a large separation of signal lines from sources of power-
frequency magnetic fields such as transformers and electric motors. Reduce
the susceptibility of circuits to stray magnetic fields by eliminating large
circuit loops.
Transmit signals via screened twisted pairs where spurious voltages induced
in successive small loops tend to cancel. These would normally be essential
first steps before resorting to expensive solutions involving high-permeability
screening. Note that lock-in amplifier construction, toroidal transformers
having low external fields are almost always used in order to reduce "hum"
pick-up within the instrument case.
(c) When laying out experiments it should be remembered that loops of wire act
as antennas at radio frequencies and that the nature and quality of a signal
ground is considerably obscured when the length of the ground path becomes
comparable with a wavelength at the interference frequency3.
The effect of electromagnetic interference in a long cable run can be suppressed
by techniques that raise the r.f. impedance of the cable. For example, it is worth

Appendix 61
investigating the effects of coiling the cable to form an r.f. choke. Other
transmission-line techniques involve the use of transmission-line transformers
and coaxializers. The most simple form consists of a few turns of screened cable
wound on a ferrite toroid as illustrated in Fig. A6.1. This arrangement has the
practical advantage of maintaining d.c. continuity throughout the length of the
cable and is sometimes known as a longitudinal choke1. The tight coupling
introduced by the winding results in a considerable attentuation of r.f. common-
mode voltage but presents a low impedance to signal current. This technique can
also be applied successfully when a differential signal is transmitted over a
screened twisted pair4.

Fig. A6.1 (a) A longitudinal choke; (b) equivalent circuit

In general, the only satisfactory approach to suppressing electromagnetic

interference is to systematically trace the interference source and to identify the
path by which it couples to the experiment. In severe environments it may be
necessary to erect a mesh screen forming a Faraday cage either around the
experiment or around the offending source when this is most convenient. The
screen should be earthed at one point only, using the most direct route possible.
(d) In many cases, the source of high-frequency interference can be traced to the
coupling of large transients to the powerlines from pulsed lasers, thyristor
controllers, laboratory ovens and other ancillary equipment used in
experimental work. The solution here may be to use plug-in r.f. filters on the
mains inputs of sensitive instruments to prevent transient interference
appearing on instrument power supplies.
Let us now turn to the last item in our original list: ground loops. The fact is that
even when detailed attention has been paid to screening and laying out cable
runs, experiments can still be plagued with interference. The reason is that
screened connections have finite inpedance, and so are able to support spurious
voltage drops. These can give rise to severe measurement difficulties unless
careful attention is paid to experiment design. Lock-in amplifier-based
experiments are particularly prone to earthing problems. The result can be a large
component of synchronous voltage appearing at the signal input that could be
much larger than the "true" synchronous signal of interest. This aspect is
discussed further in Section A6.4.

A6.2 Ground loops: single-ended amplifiers

Fig. A6.2 shows the connection of a transducer voltage source to a single-ended
amplifier via a screened cable. The cable screen is securely referenced to the
amplifier earthing point and is thus effective in shielding the sensitive inner
conductor against capacitive pick-up from external sources.

Appendix 62
Fig. A6.2 Connection of a "floating" source and grounded amplifier

Unfortunately, it is not always possible to maintain the screen at the same

potential along its entire length. When this cannot be achieved, the effectiveness
of the screen is reduced because variations in screen potential become
capacitively coupled to the inner conductor. Also, in some circumstances the
voltage developed across the cable screen is able to add directly to the signal
voltage.

Fig. A6.3 Ground loop established when items of "grounded" equipment

are interconnected

Consider, for example, the connections shown in Fig. A6.3(a). Here, the signal
source is located inside a screening enclosure. Following "good practice" for
optimum screening, the signal-source common and the cable screen have been
connected to a single earth point to ensure that no signal currents or earth currents

Appendix 63
flow through the enclosure. The main problem in this, or any similar,
arrangement, is to define a true "wet earth" for screen connections. The use of
chassis symbols with separate labels for "experiment" and "amplifier" reminds us
that in an extensive laboratory installation it is not unusual to find a.c. potential
differences of several hundred millivolts between adjacent chasses. Any attempt
to connect these via a cable screen of finite impedance then results in a ground
loop as indicated in Fig. A6.3(b). this loop is sensitive to any potential difference
between the two chasses and is additionally susceptible to inductive coupling to
stray magnetic fields. Both these effects are accounted for by the inclusion of a
ground loop generator in Fig. A6.3(b) which develops a voltage drop vg across
the cable screen. Because the amplifier senses the potential difference between
the screen and the inner conductor of the connecting cable, the so-called common
mode voltage, vg is effectively added to the signal.
This spurious input may dominate the measurement of the signal unless
appropriate steps are taken. In principle, the common-mode signal can be
eliminated by bringing the source and amplifier chasses to the same potential, but
attempts to achieve this are rarely successful in practice. Even when units are
brought into close proximity and bolted to a metal plate it is not unusual to find
large potential differences between "earth" points only a few inches apart. A far
better approach is therefore to investigate ways of "floating" either the source or
the amplifier with a view to breaking the ground loop completely.
In the case of this source this might be achieved by using insulating bolts and
washers to prevent direct contact between the transducer case and the
experimental chassis. Where this is not feasible, a battery-powered preamplifier
provides a reliable (and safe) way of isolating the amplifier input. An alternative
and usually more convenient approach is to use a semi-floating amplifier in
which a "float" resistor, typically in the range 50 to 1 , is used to provide a
degree of isolation which approaches actual breaking of the ground loop.
Fig. A6.4 shows how the float resistor is inserted between the cable screen and
the amplifier chassis; as a result, most of the ground loop voltage is dropped
across the float resistor, leaving a relatively small fraction across the cable
screen. In effect, the common-mode signal vg is reduced by the ratio of the screen
resistance (typically a few tens of milliohms) to the float resistance. Note that the
reduction of the spurious voltage across the screen also results in a corresponding
reduction in interference capacitively coupled from the screen to the inner
conductor of the connecting cable.

Fig. A6.4 Introduction of an amplifier "float" resistor, resulting in a much

reduced common-mode votlage vg

Appendix 64
A6.3 Ground loops: differential amplifiers
The ground-loop suppression afforded by a semi-floating single-ended amplifier
is impressive, but yet may be insufficient in some circumstances. When, for
reasons of safety or practicability, it is not possible to use a floating source,
residual connection problems can usually be overcome by using a differential
amplifier as shown in Fig. A6.5. As before, an amplifier float resistor is used to
bring about a large reduction in the spurious cable screen voltages which now
appear as equal common-mode voltages across the amplifier inputs. As explained
in Appendix 5, the extremely high common-mode rejection of the amplifier will
then ensure suppression of ground-loop interference at power frequencies.
In order to maximize suppression it is essential to provide identical routes
between source and amplifier for both cables to ensure that there is no differential
pick-up in the two screens. This problem can be overcome by arranging to
transmit differential signals over twisted pairs in a common screen, an approach
favoured by instrumentation engineers. The symmetry of a twisted-pair
connection also tends to equalize capacitive pick-up between the screen and the
two conductors, and is additionally effective in reducing inductive pick-up.
The common-mode rejection ratio of a differential amplifier falls at high
frequencies. Common-mode pick-up at radio frequencies should therefore be
reduced as far as possible using screening and the coaxilizer techniques referred
to earlier. Even at moderate interference frequencies, spurious phase shifts caused
by the distributed cable capacitances acting with unequal resistances in the two
paths may cause incomplete cancellation of common-mode voltages. An
improvement will usually be obtained when a fully balanced source such as an
a.c. bridge is used with a "true" differential amplifier with balanced inputs
(Appendix 5) and matched cable lengths.

Fig. A6.5 Using a differential amplifier to overcome grounding problems

A6.4 Ground loops and lock-in recovery

Careless connection of signal and reference cables to a lock-in amplifier can lead
to ground-loop problems over and above those described so far.
In most signal-recovery experiments, an oscillator is used as both an excitation
and reference source, while the signal is taken from a "single-ended" transducer
output. Fig. A6.6 gives an example where the oscillator and transducer output are

Appendix 65
strapped to the chassis of the experiment with direct connections made to the
reference and signal inputs of a lock-in amplifier. The impedance ZE represents
the load provided by the excitation circuit to the experiment. To take a specific
example, the oscillator could provide the drive to a vibration or "shaker" table,
while the signal output is derived from a vibration transducer mounted on the
device under test. In this case, the drive current could be several orders of
magnitude greater than the current flowing in either the reference or signal
circuits.

Fig. A6.6 Indiscriminate connections in a lock-in amplifier-based

experiment causing a synchronous loop

Problems will arise in practice with the arrangement shown because the return
path from ZE to the excitation oscillator is shunted by a second path formed by
the screen connections on the reference and signal connection cables. A fraction
of the drive current is thus able to circulate in this path and generate a
synchronous voltage drop across the signal screen in series with the signal
voltage of interest. Although the amplifier "float" resistance RA will attenuate the
voltage considerably, the spurious voltage could well be greater than the signal
voltage. The fact that spurious voltage is also synchronous with the applied
reference would severely restrict the range over which measurements could
usefully be performed.
Note that if the drive load were omitted completely, the signal screen would
continue to provide a return path for a fraction of the current flowing in the
reference circuit. Fortunately, the reference current is limited by the input

Fig A6.7 Improved experimental layout

Appendix 66
resistance of the reference circuit. However, the spurious, synchronous, voltage
set up in the signal input is often troublesome in low-level applications. A
floating differential input can provide the solution in most cases. This leads us to
the improved arrangement illustrated in Fig. A6.7, which also shows how to
overcome the effect of the drive current. Here, the drive circuit is completely
isolated from the sensitive signal circuit and contact is made to the experimental
chassis at a single point. It is certainly possible that this, or some equivalent,
arrangement could be arrived at by trial and error. It would be far better,
however, to give some preliminary thought to ground loop problems; the
alternative is almost invariable an unco-ordinated attempt to achieve "on-line"
solutions when a large investment has already been made in mounting equipment
and fixing cable runs.

A6.5 References
1. OTT, H.W (1976): "Noise reduction techniques in electronic systems" (John
Wiley, New York).
2. MORRISON, R. (1967): "Grounding and shielding techniques in
instrumentation" (John Wiley, New York).
3. Technical Note 117 (1978): "Grounding and shielding in electrochemical
instrumentation - Some basic considerations" (EG&G Princeton Applied
Research Corp., Princeton, NJ).
4. HOROWITZ, P., and HILL, W. (1980): "The art of electronics" (Cambridge
Univ. Press, Cambridge).

Appendix 67