Interfacing a Microprocessor to the Analog World

In many systems, the embedded processor must interface to

the non-digital, analog world.

The issues involved in such interfacing are complex, and go

well beyond simple A/D and D/A conversion.

? ?
A/D CPU D/A

Two questions:

1. How do we represent information about the analog world in a

digital microprocessor?
2. How do we use a microprocessor to act on the analog world?

We shall explore each of these questions in detail, both

conceptually in the lectures, and practically in the laboratory
exercises.

EECS461, Lecture 2, updated September 6, 2015 1

 Sensors

• Used to measure physical quantities such as

- position
- velocity
- temperature
- sound
- light
• Two basic types:
- sensors that measure an (analog) physical quantity and
generate an analog signal, such as a voltage or current

physical analog
voltage digital
quantity
sensor A/D

* tachometer
* potentiometer
- sensors that directly generate a digital value

physical
digital
quantity
sensor

* digital camera
* position encoders

EECS461, Lecture 2, updated September 6, 2015 2

 Sensor Interfacing Issues

• Shall focus on issues that involve

- loss of information
- distortion of information

• Such issues include

- quantization
- sampling
- noise

• Fundamental difference between quantization and sampling

errors:
- Quantization errors affect the precision with which we can
represent a single analog value in digital form.
- Sampling errors affect how well we can represent an entire
analog waveform (or time function) digitally.

EECS461, Lecture 2, updated September 6, 2015 3

 Quantization

Digital representation of an analog number [2, 3, 6]

• Issue:
- an analog voltage can take a continuum of values
- a binary number can take only finitely many values

• Binary representation of (unsigned or signed) real number

- unipolar coding
- unipolar coding with centering
- offset binary coding
- two’s complement

• Resolution [2, 3]
- Idea: two analog numbers whose values differ by < 1/2n
may yield the same digital representation
- an n-bit A/D converter has a resolution equal to 2−n times
the input voltage range, v ∈ [0, Vmax]
- least significant bit (LSB) represents Vmax/2n

EECS461, Lecture 2, updated September 6, 2015 4

 Quantization: Example

• Suppose we quantize an analog voltage in the range (0, Vmax)

using a two bit binary number.
• The LSB thus represents Vmax/4

11

10

01

00
0 Vmax Vmax 3Vmax Vmax
4 2 4

input voltage

• Quantization error: from 0 to 1 LSB (e.g., 01 represents any

voltage from Vmax/4 to Vmax/2)
• For 11 to uniquely represent Vmax, divide voltage range into
2n − 1 intervals. LSB = Vmax/(2n − 1)
• Centering: 01 represents Vmax/8 to 3Vmax/8

11

10

01

00
0 Vmax Vmax 3Vmax Vmax
4 2 4

input voltage

• Quantization error: ± 12 LSB

EECS461, Lecture 2, updated September 6, 2015 5

 A/D Conversion

analog n-bit binary

voltage number
A/D

• Types of A/D converters [2]:

- flash
- successive approximation (MPC555 and MPC5643L)
- single-slope (or dual-slope) integration
- sigma-delta converters
- redundant signed digit (RSD) [5] (MPC5553)

• Design issues
- precision
- accuracy
- speed
- cost
- relative amount of analog and digital circuitry

• Performance Metrics [4]

- quantization error
- offset and gain error
- differential nonlinearity
- monotonicity
- missing codes
- integral nonlinearity

EECS461, Lecture 2, updated September 6, 2015 6

 Successive Approximation A/D Converter

• Used on the Freescale MPC5643L

• Bits set in succession, from most to least significant
Vs
+ n-bit binary
control
comparator logic
-

MSB
D/A
converter
LSB

• Control logic [1]

start

set all bits to 0

start at MSB

set bit = 1

is D/A output yes

go to next reset bit = 0
> Vs ?
lower bit
no

no all bits
checked?

yes

conversion done

end

• Issues
- timing (bits set one at a time)
- signal to noise ratio (lower bits based on small signals)
- cannot correct for wrong decisions on a given bit

EECS461, Lecture 2, updated September 6, 2015 7

 Sampling

• Convert an analog function of time into a sequence of binary

numbers
analog digital
physical waveform sequence
signal o
o o o
o o o
sensor sampler A/D

• sampler [3]
switch

analog -
input output
+

• Information loss in representing an analog function as a discrete

sequence [2, 3, 8, 6]

time

EECS461, Lecture 2, updated September 6, 2015 8

 Information Loss in Sampling

• How to describe information loss?

• Idea: Try to reconstruct the analog signal from its digital
representation. This may be done by a D/A converter.
reconstructed
analog digital analog signal
physical waveform sequence o
o o o o
signal o o o o o
o o o o
sensor sampler A/D ZOH

• “Staircase” output1: sin(2πt) sampled with sampling period

T = 0.05 seconds (sampling frequency f = 20 Hz)
sin(2π t), sin(2 π kT), T = 0.05 sec, and ZOH output
2
analog input
samples
ZOH output
1.5

0.5

−0.5

−1

−1.5

−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time, sec

• a staircase approximation of the input delayed by T /2 seconds

1 created with MATLAB files staircase approx.m and simulate ZOH.mdl

EECS461, Lecture 2, updated September 6, 2015 9

 Fast Sampling
• Compare fast and slow sampling2
• Input: a 1 Hz sinusoid, sin(2πt), sampled at 20 Hz
sample period, T = 0.05 sec, sample frequency, 1/T = 20 Hz
2

1.5

0.5

−0.5

−1

−1.5

−2
0 1 2 3 4 5 6 7 8 9 10
time, sec

• Input: a 1 Hz sinusoid, sin(2πt), sampled at 10 Hz

sample period, T = 0.1 sec, sample frequency, 1/T = 10 Hz
2

1.5

0.5

−0.5

−1

−1.5

−2
0 1 2 3 4 5 6 7 8 9 10
time, sec

2 MATLAB m-file fast slow sampling.m

EECS461, Lecture 2, updated September 6, 2015 10

 Slow Sampling

• Input: a 1 Hz sinusoid, sin(2πt), sampled at 2 Hz

sample period, T = 0.5 sec, sample frequency, 1/T = 2 Hz
2

1.5

0.5

−0.5

−1

−1.5

−2
0 1 2 3 4 5 6 7 8 9 10
time, sec

• Input: a 1 Hz sinusoid, sin(2πt), sampled at 1.11 Hz

sample period, T = 0.9 sec, sample frequency, 1/T = 1.1111 Hz
2

1.5

0.5

−0.5

−1

−1.5

−2
0 1 2 3 4 5 6 7 8 9 10
time, sec

EECS461, Lecture 2, updated September 6, 2015 11

 Observations on Sampling

• If sampling period is fast with respect to period of the signal,

then the reproduced signal approximates the original signal.
- slight staircase effect
- slight time delay

• If sampling period is relatively slow, then there are the

reproduced signal may differ significantly from the original
signal.
- It may equal zero!
- It may look like a periodic signal of equal amplitude but
longer period.

• Other issues[6]
- irregular sampling interval
- synchronizing sampling with the signal

EECS461, Lecture 2, updated September 6, 2015 12

 Aliasing

• Suppose we have two analog signals whose values are identical

at the sample points. Then their digital representations will
also be identical.
- Impossible to reconstruct original signal from its digital
representation.
- Any algorithm on the CPU will be unable to distinguish
between signals.
- Especially problematic when sampling noisy analog signals.
- sin(2πt) sampled at 0, 0.5, 1, 1.5, ... seconds is
indistinguishable from zero!
- cos(t) and cos((1 + π)t) are identical at 0, 2, 4, ...
seconds3!
sample period, T = 2
1

0.8

0.6

0.4

0.2

−0.2

−0.4
cos(t)
cos((1+ 1π) t)
−0.6 cos(2k)
cos((1+ 1π)2k)

−0.8

−1
0 1 2 3 4 5 6
time, seconds

• A higher frequency signal that “masquerades” as a low

frequency signal after sampling is said to be aliased.
3 MATLAB m-file aliasing.m

EECS461, Lecture 2, updated September 6, 2015 13

 Effects of Aliasing

• Aliasing is a type of information distortion that results from

undersampling.
• Questions:
1. How fast must one sample an arbitrary signal to avoid
aliasing?
2. When is aliasing likely to be a problem in sensor interfacing?
3. How does one minimize the effects of aliasing?
• We shall return to these questions after an example and a
review of some ideas from signals and systems.

Example: Consider a video of a rotating wheel marked with an

arrow, and made with a camcorder at a rate of 30 frames/second
[8]...

EECS461, Lecture 2, updated September 6, 2015 14

 Aliasing and the Wheel,I

• The effects of aliasing can be striking...

• Consider a wheel rotating counterclockwise (CCW) at R
rev/seconds.
• Suppose we
- View the wheel with a strobe light every T seconds, or
- Use a camcorder to make a video with one frame every T
seconds.

R rev/sec

• Depending upon the relative values of T and R, the wheel

may appear to be
- rotating CCW – as we expect to see
- stationary – not moving!
- rotating clockwise (CW) – backwards!

EECS461, Lecture 2, updated September 6, 2015 15

 Aliasing and the Wheel, II

• We visually determine the direction of motion by noting the

difference between consecutive measurements of the position
of the arrow.
• If T is fast with respect to the speed of rotation, then motion
appears to be CCW:

actual
3T
2T
T perceived

• If T is slow with respect to the speed of rotation, then motion

appears to be CW:

actual

2T
perceived
3T

EECS461, Lecture 2, updated September 6, 2015 16

 Aliasing and the Wheel, III

• At an even slower value of T , wheel appears to be stationary:

actual

T, 2T, 3T,... perceived

• At an intermediate value of T , we are only confused:

actual

T, 3T,...

2T, 4T,...

EECS461, Lecture 2, updated September 6, 2015 17

 Aliasing and the Wheel, IV

• Suppose the wheel rotates CCW at a fixed rate R rev/sec.

Can we determine the maximum value of T so that the wheel
always seems to be rotating (and rotating CCW)?
• Terminology
- sample period, T seconds
- sampling frequency, fs = 1/T Hz or ωs = 2π/T rad/sec
- rotation rate, R rev/sec, or 2πR rad/sec
• Position of wheel in (x, y) coordinates is given by

θ
x

x(t) = cos(2πRt)
y(t) = sin(2πRt)

⇒ Taking a picture of the wheel every T seconds is equivalent

to “sampling” a sine wave every T seconds

EECS461, Lecture 2, updated September 6, 2015 18

 Aliasing and the Wheel, V

• It takes 1/R seconds for the wheel to make a complete

revolution.
• Suppose that initially θ(0) = 0◦. Hence if we sample at
T = 1/R samples/second, then the position coordinates at
the sample times kT, k = 1, 2, 3... satisfy

x(kT ) = cos(2πk) = x(0) = 1

y(kT ) = sin(2πk) = y(0) = 0

⇒ the wheel looks as though it were stationary

• To determine the correct direction of rotation, we need to take
at least one sample before it reaches the halfway point:
y

T
θ 0

• The wheel reaches θ = 180◦ in 1/2R seconds, hence we

require
- sample period T < 1/2R sec
- sample frequency ωs > 4πR rad/sec (fs > 2R Hz)
• Later we shall rederive this result from the Shannon sampling
theorem [6, 8]

EECS461, Lecture 2, updated September 6, 2015 19

 Fourier Series

• Consider a periodic time signal f (t), t ≥ 0, with period T :

f (t) = f (t + kT ), k = 0, 1, 2, . . ..
• Examples:
- sine wave
- square wave
- sawtooth wave
• Then f (t) may be expressed as a sum of (possibly infinitely
many) sines and cosines.
• Terminology
- T : period of signal
- ω0 = 2π/T : frequency in rad/sec
- f = 1/T : frequency in Hz
• Then
∞
X
f (t) = a0 + (an cos(nω0t) + bn sin(nω0t))
n=1

• More terminology
- Fourier coefficients: ai, bi
- DC term: a0
- fundamental: n = 1, sinusoids of frequency ω0
- harmonics: n > 1, sinusoids of frequency > ω0

EECS461, Lecture 2, updated September 6, 2015 20

 Examples of Fourier Series

• Example: A sine wave with period T

 
2π
f (t) = sin t
T

is its own Fourier series expansion

• Unit amplitude square wave with period T has Fourier

expansion
∞
X 4
f (t) = sin(nω0t)
nπ
n=1
n,odd
where ω0 = 2π/T is the frequency of the square wave in
rad/sec (f = 1/T is the frequency in Hz)

4
- Fundamental: n = 1, π sin(ω0t)
4
- 1st harmonic: n = 3, 3π sin(3ω0t)
4
- 2nd harmonic: n = 5, 5π sin(5ω0t)

EECS461, Lecture 2, updated September 6, 2015 21

 More Terms ⇒ Better Approximation
Fourier series of a square wave with period T = 2 seconds4.
1.5
square
fundamental
1st harmonic
2nd harmonic
1 fund+1st+2nd

0.5

−0.5

−1

−1.5
0 0.5 1 1.5 2 2.5 3 3.5 4
time, seconds

1.5 square
fundamental
fund+...+5th

0.5

−0.5

−1

−1.5
0 0.5 1 1.5 2 2.5 3 3.5 4
time, seconds

4 Matlab m-file sq wave.m

EECS461, Lecture 2, updated September 6, 2015 22

 Frequency of a Signal

• Consider a periodic signal, such as a square wave, that has

“sharp corners”.
• In general, many high frequency terms are needed to construct
such “sharp corners”. In fact, any signal with relatively abrupt
changes will contain high frequencies, even if the changes are
not discontinuous.
• It is useful to sketch the location, and relative amplitude, of
the various frequency components of a signal
• Example: unit amplitude square wave

4/π

4/3π
4/5π

ω0 2ω0 3ω0 4ω0 5ω0

EECS461, Lecture 2, updated September 6, 2015 23

 Fourier Transform
• Most signals are not periodic. Nevertheless, it is possible to
think of “almost any” signal as the sum of sines and cosines
of all frequencies.
• Fourier transform [7]: Under certain conditions, we can write
Z ∞
−jωt
F (ω) = f (t)e dt
−∞
Z ∞ Z ∞
= f (t) cos(ωt)dt − j f (t) sin(ωt)dt
−∞ −∞

• Inverse Fourier transform:

∞
1
Z
jωt
f (t) = F (ω)e dω
2π −∞

• We will not need any of the details of the Fourier transform.

However, it is important to remember that time signals may
be given a frequency representation.
• Can visualize the frequency content of a signal by plotting
F (ω) as a function of frequency:
|F(ω)|

EECS461, Lecture 2, updated September 6, 2015 24

 Fourier Transform of a Periodic Signal

R∞
• Recall: delta function: δ(ω) = 0, ω 6= 0; −∞ δ(ω)dω = 1
• The Fourier transform of a sinusoid of frequency f Hz consists
of two delta functions located at frequencies ±f Hz.
• The frequency response of a square wave consists of delta
functions corresponding to all frequency components of the
Fourier series expansion of the square wave.
• Example5: Square wave of period T = 2 seconds, f = 0.5
Hz has frequency components at ±f, ±3f, ±5f, . . .. The
Fourier transform of a square wave may be approximated using
algorithms from [7]

approx. Fourier transform of square wave with period T = 2 sec, f = 0.5 Hz

5

4.5

3.5

2.5

1.5

0.5

0
−5 −4 −3 −2 −1 0 1 2 3 4 5
frequency, Hz

5 MATLAB m-file sq wave.m

EECS461, Lecture 2, updated September 6, 2015 25

 Frequency Response in Embedded Systems
Applications

• Many embedded systems for control, communications, and

signal processing applications – and anything to do with audio
or video – require knowledge of frequency content of signals.
• an important class of embedded processors – DSP chips –
has a special architecture that allows rapid computation of
the frequency response of a signal using the Fast Fourier
Transform (FFT) algorithm.
• Knowledge of frequency content is needed to design the
interface electronics for an embedded system. For example,
circuits that implement lowpass filters to remove unwanted
high frequencies.
• Frequency response ideas arise in the study of sampling and
aliasing, and in the use of Pulse Width Modulation (PWM) to
drive a DC motor.

EECS461, Lecture 2, updated September 6, 2015 26

 Shannon Sampling Theorem
• Recall Question 1: How fast must we sample to avoid aliasing?
• Shannon’s Theorem [6, 8]
- Consider a signal f (t) with frequency response F (ω).
- Suppose we sample f (t) periodically, with period T sec, and
define the Nyquist frequency ωN = π/T radians/second
(fN = 1/2T Hz).
|F(ω)|

0
-ωΝ ωΝ ω

- If F (ω) = 0, for |ω| ≥ ωN , then it is possible to

reconstruct f (t) exactly from its samples f (kT ).
• Reconstruction requires an ideal lowpass filter :
f(t) f(kT)
f(t)
o
o o o
o o o
sampler A/D ideal LPF

• In practice, reconstruction can only be done approximately,

because perfect reconstruction requires all samples of the
signal, even those in the future! 
P∞ π(t−kT )
• f (t) = k=−∞ f (kT ) sinc T ; sinc(x) = sinx x
• Nevertheless, this result tells how fast we must, in principle,
sample to avoid aliasing: at least twice as fast as the highest
frequency in the signal!

EECS461, Lecture 2, updated September 6, 2015 27

 Aliasing and the Wheel, VI

• Suppose that the wheel rotates at R rev/sec, or 2πR rad/sec.

• Then position coordinates

x(t) = cos(2πRt)
y(t) = sin(2πRt)

are sinusoids with frequency ω0 = 2πR.

• Nyquist says that to avoid aliasing we sample fast enough that

π 1
ω0 < ωN = rad/sec ⇒ T < sec
T 2R

• Same result as we derived before!

EECS461, Lecture 2, updated September 6, 2015 28

 Nyquist and Embedded System Applications

• A frequency analysis is done of each analog signal that must

be measured with a sensor and represented in digital form.
• Although the signals will have energy at all frequencies, usually
the “information” lies in some low frequency range, say
ω < ω0 .
• If possible, set the sample period T so that the Nyquist and
sampling frequencies satisfy

π 2π
ωN = > ω0 ⇔ ωs = > 2ω0
T T
(usually, we set sampling frequency ωs > (5 − 10)ω0, twice
as fast is only the theoretical limit)

EECS461, Lecture 2, updated September 6, 2015 29

 Problems with Aliasing

• When is aliasing likely to be a problem?

• Almost all signals are corrupted by noise
- 60 Hz hum
- EMI from spark ignition
-
• Often the noise is at a higher frequency than the information
contained in the signal. If the noise is at a sufficiently high
frequency, it will get “aliased” to a lower frequency, and
corrupt the signal we are trying to measure.
• How to resolve?

EECS461, Lecture 2, updated September 6, 2015 30

 Frequency Response Functions

• a linear filter has a frequency response that determines how it

responds to periodic input signals
• Example: RC circuit

vi(t) + C vo(t)
-

- frequency response function

1
H(jω) =
1 + jωRC

- magnitude, or gain

1
|H(jω)| = √
1 + ω 2 R2 C 2

- phase
−1
∠H(jω) = − tan (ωRC)
• After transients die out, the steady state response of the filter
to a sinusoid is determined by it frequency response function:

vi(t) = sin(ω0t) ⇒ vo(t) → |H(jω0)| sin(jω0t+∠H(jω0))

EECS461, Lecture 2, updated September 6, 2015 31

 Gain and Phase Plots

• Bode plots: gain and phase vs frequency6

Bode plots of RC filters

−10

−20
Magnitude (dB)

−30

−40

−50

−60
0
RC = 0.01
RC = 0.1
RC = 1
RC = 10
Phase (deg)

−45

−90
−3 −2 −1 0 1 2 3 4
10 10 10 10 10 10 10 10
Frequency (rad/sec)

• Lowpass filter
- passes low frequencies
- attenuates high frequencies
- introduces phase lag
• Bandwidth of RC filter proportional to 1/RC

6 MATLAB m-file RC filter.m

EECS461, Lecture 2, updated September 6, 2015 32

 Anti-Aliasing Filters

• Potential solution to aliasing problem: “anti-aliasing filters”

that are inserted before the sampler to remove high frequencies
signal +
noise noise
digital
physical sequence
+ o
signal o o o
anti-alias o o o
sensor sampler A/D
filter

• Commercial devices often have an AA filter built in, but may

need to build another one to configure the frequency response
for the application.
• Practical AA filters are higher order.
• Potential problems:
- may not have frequency separation between signal and noise
- phase lag in control applications

EECS461, Lecture 2, updated September 6, 2015 33

References

[1] http://hyperphysics . phy-astr . gsu . edu /

hbase/electronic/adc . html#c3.
[2] D. Auslander and C. J. Kempf. Mechatronics: Mechanical
Systems Interfacing. Prentice-Hall, 1996.
[3] W. Bolton. Mechatronics: Electronic Control Systems in
Mechanical and Elecrical Engineering, 2nd ed. Longman,
1999.
[4] J. Feddeler and B. Lucas. ADC Definitions and Specifications.
Freescale Semiconductor, Application Note AN2438/D,
February 2003.
[5] M. Garrard and P. Ryan. Design, Accuracy, and Calibration
of Analog to Digital Converters on the MPC5500 Family.
Freescale Semiconductor, Application Note AN2989, July
2005.
[6] S. Heath. Embedded Systems Design. Newness, 1997.
[7] E. Kamen and B. Heck. Fundamentals of Signals and Systems
using MATLAB. Prentice Hall, 1997.
[8] J. H. McClellan, R. W. Schafer, and M. A. Yoder. DSP First:
A Multimedia Approach. Prentice-Hall, 1998.

EECS461, Lecture 2, updated September 6, 2015 34