APPLICATION NOTE

ADAPTIVE FILTERING APPLICATIONS EXPLAINED

DESPITE THE DIVERSITY AND COMPLEXITY OF ADAPTIVE FILTERING, A SIMPLE

CLASSIFICATION DOES EMERGES AND PRACTICAL APPLICATIONS ARE DEMONSTRATED.

INTRODUCTION
Although well-known and widely used, adaptive filtering
applications are not easily understood, and their principles
are not easily simplified. Currently, adaptive filtering is
applied in such diverse fields as communications, radar,
sonar, seismology, and biomedical engineering. Although
these various applications are very different in nature, one
common feature can be noted: an input vector and a
desired response are used to compute an estimation error,
which is used, in turn, to control the values of a set of
adjustable filter coefficients. The adjustable coefficients
may take the form of tap weights, reflection coefficients, or
rotation parameters, depending on the filter structure
employed.
Despite the diversity and complexity, a simple classification
of adaptive filtering does emerge and practical applications
can be demonstrated. This app. note begins by describing
four basic classes of adaptive filtering applications and
follows with sections that detail various fundamentals,
techniques, and algorithms of several selected adaptive
applications (refer to Table 1).

Table 1. Adaptive Filtering Applications

Adaptive
Filtering
Class
Identification

Application
System Identification
Layered Earth Modeling
Inverse Modeling Predictive Deconvolution
Adaptive Equalization
Prediction
Linear Predictive Coding
Adaptive Differential PCM
Auto-Regressive Spectrum Analysis
Signal Detection
Interference
Adaptive Noise Canceling
Canceling
Echo Cancellation
Radar Polarimetry
Adaptive Beamforming
The following notations are used in Figures 14:
u = input applied to the adaptive filter

Classifying Adaptive Filtering Applications

y = output of the adaptive filter

Various applications of adaptive filtering differ in the

manner in which the desired response is extracted. In this
context, we may distinguish four basic classes of adaptive
filtering applications (depicted in Figures 1 to 4, which
follow):

d = desired response

Identification

Inverse Modeling

Prediction

Interference Canceling

AP96DSP0200

e = d -y = estimation error
The functions of the four basic classes of adaptive filtering
applications follow.

AN008001-0301

Adaptive Filtering Applications Explained

I. Identification (Figure 1). The notion of a mathematical
model is fundamental to sciences and engineering. In the
class of applications dealing with identification, an
adaptive filter is used to provide a linear model that
represents the best fit to an unknown plant.The plant and
the adaptive filter are driven by the same input. The plant
output supplies the desired response for the adaptive filter.
If the plant is dynamic in nature, the model will be time
varying.

values of the signal supply the input applied to the adaptive

filter. Depending on the application of interest, the
adaptive filter output or the estimation error may service as
the system output. In the first case, the system operates as
a predictor; in the latter case, it operates as a prediction
error filter.

System
Output 2

Random
Signal
u

Adaptive
Filter

Delay

+ d
y
e System
Adaptive

Filter
Output 1

Figure 3. Prediction

+
d

System
Input

System
Output

Plant

Figure 1. Identification

II. Inverse Modeling (Figure 2). In this second class of

applications, the adaptive filter provides an inverse model
representing the best fit to an unknown noisy plant. Ideally,
the inverse model has a transfer function equal to the
reciprocal of the plant's transfer function. A delayed
version of the plant input constitutes the desired response
for the adaptive filter. In some applications, the plant input
is used without delay as the desired response.

IV. Interference Cancelling (Figure 4). In this final class

of applications, the adaptive filter is used to cancel
unknown interference contained in a primary signal, with
the cancellation being optimized in some sense. The
primary signal serves as the desired response for the
adaptive filter. A reference signal is employed as the input
to the adaptive filter. The reference signal is derived from
a sensor or set of sensors located in relation to the
sensor(s) supplying the primary signal in such a way that
the information-bearing signal component is weak or
essentially undetectable.

Primary
Signal
d

+
Reference u
Signal
System

Plant

Adaptive

Adaptive
Filter

System
Output

System
y

Figure 4. Interference Canceling

Delay

Figure 2. Inverse Modeling

III. Prediction (Figure 3). In this example, the adaptive
filter provides the best prediction of the present value of a
random signal. The present value of the signal serves the
purpose of a desired response for the adaptive filter. Past
2

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

SELECTED ADAPTIVE FILTERING APPLICATIONS

Prediction
The coders used for the digital representation of speech
signals fall into two broad classes: source coders and
waveform coders. Source coders are model dependent,
which means that they use a priori knowledge about how
the speech signal is generated at the source. Source
coders for speech are generally referred to as vocoders.
Vocoders can operate at low coding rates; however, they
provide a synthetic quality, with the speech signal having
lost substantial naturalness.
Waveform coders, on the other hand, strive for facsimile
reproduction of the speech waveform. In principle, these
coders are signal independent. They may be designed to
provide telephone-toll quality for speech at relatively high
coding rates.
In the context of speech, linear predictive coding (LPC)
strives to produce digitized voice data al low bit rates (2.4
to 4.8 Kbps) with two important motivations in mind:
1. The use of linear predictive coding permits the
transmission of digitized voice over a narrow-band
channel.

2. The realization of a low-bit rate makes the encryption

of voice signals easier and more reliable than would
otherwise be the case.
Figure 5 shows a simplified block diagram of the classical
model for the speech production process. (In this particular
example, the sound-generating mechanism is linearly
separable from the intelligence-modulating, vocal-tract
filter.) The precise form of the excitation depends on
whether the speech sound is voiced or unvoiced.
Voiced speech sound is generated from quasi-periodic,
vocal-cord sound. In the speech model, the impulse-train
generator produces a sequence of impulses, which are
spaced by a fundamental period equal to the pitch period.
This signal, in turn, excites a linear filter whose impulse
response equals the vocal-cord sound pulse.
An unvoiced speech sound is generated from random
sound produced by turbulent air flow. In this case the
excitation consists simply of a white noise source. The
probability distribution of the noise samples does not
appear to be critical.

Pitch
Period

Impulse Train
Generator

Vocal-Cord
Sound Pulse

Voiced/Unvoiced
Switch

White-Noise
Generator

Vocal-Tract
Filter

Synthesized
Speech

Vocal-Tract
Parameters

Figure 5. Block Diagram of a Simplified Model for the Speech Production Process

AP96DSP0200

AN008001-0301

Adaptive Filtering Applications Explained

Figure 6 shows the block diagram of an LPC vocoder,
consisting of a transmitter and a receiver. The transmitter
first applies a window to the input speech signal, thereby
identifying a block of speech samples for processing. This
window is short enough for the vocal-tract shape to be
nearly stationary, so the parameters of the speechproduction model may be treated as essentially constant
for the duration of the window. The transmitter then
analyzes the input speech signal in an adaptive manner
block by blockby performing a linear prediction and pitch
detection. Finally, it codes the parameters made up of the

Speech
Signal

Window

set of predictor coefficients, the pitch period, the gain

parameter, and the voiced-unvoiced parameter, for
transmission over the channel. The receiver performs the
inverse operations by first decoding the incoming
parameters. In particular, it computes the values of the
predictor coefficients, the pitch period, and the gain
parameter, and determines whether the segment of
interest represents voiced or unvoiced sound. Finally, the
receiver uses these parameters to synthesize the speech
signal by utilizing the model of Figure 5.

LPC
Analyzer

Coder

Pitch
Detector
(a) Transmitter

Speech
Synthesizer

Decoder

Reproduction of
Speech Signal

(b) Receiver

Figure 6. Block Diagram of LPC Vocoder: (a) Transmitter, (b) Receiver

Adaptive Equalization
During the past three decades, a considerable effort has
been devoted to the study of data-transmission systems
that utilize the available channel bandwidth efficiently. The
objective here is to design the system to accommodate the
highest possible rate of data transmission, subject to a
specified reliability that is usually measured in terms of the
error rate or average probability of symbol error.

The transmission of digital data through a linear

communication channel is limited by two factors: 1)
intersymbol interference, and 2) additive thermal noise.

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

Intersymbol Interference (ISI). Caused by dispersion in
the transmit filter, the transmission medium, and the
receive filter.
Additive Thermal Noise. Generated by the receiver at its
front end. For bandwidth-limited channels, intersymbol
interference seems to be the chief determining factor in the
design of high-data-rate transmission systems. Figure 7
shows the equivalent baseband model of a binary pulseamplitude (PAM) modulation system. The signal applied to
the input of the transmitter part of the system consists of a

binary data sequence, in which the binary symbol consists

of 1 or 0. This sequence is applied to a pulse generator, the
output of which is filtered first in the transmitter, then by the
medium, and finally in the receiver. Let u(k) denote the
sampled output of the receive filter in Figure 7; the sampling
is performed in synchronism with the pulse generator in the
transmitter. This output is compared to a threshold by means
of a decision device. If the threshold is exceeded, the
receiver makes a decision in favor of symbol 1. Otherwise,
it decides in favor of symbol 0.

Synchronization
Input
Binary
Data

Pulse
Generator

Transmit
Filter

Medium

Receive
Filter

Noise

Transmitter

()
Sampler

Decision
Device

Output
Binary
Data

Receiver

Figure 7. Block Diagram of a Baseband Data Transmission System (Without Equalization)

On a physical channel there is always intersymbol

interference. To overcome intersymbol interference,
control of the time-sampled function is required. In
principle, if the characteristics of the transmission medium
are known precisely, then it is virtually always possible to
design a pair of transmit and receive filters that will make
the effect of intersymbol interference arbitrarily small.

In practice, however, with the adaptive equalizer located in

the receiver, the equalizer is physically separated from the
origin of its ideal desired response.

For adequately reducing the intersymbol interference, an

adaptive equalizer will provide precise control over the
time response of the channel.

Training method. In this method, a replica of the desired

response is stored in the receiver. Naturally, the generator
of this stored reference must be electronically
synchronized with the known transmitted sequence.

There are two methods in which a replica of the desired

response may be generated locally in the receiver:
1) training method, and 2) decision-directed method.

An adaptive filtering algorithm requires knowledge of the

desired response to form the error signal needed for the
adaptive process to function. In theory, the transmitted
sequence is the desired response for adaptive
equalization.

AP96DSP0200

AN008001-0301

Adaptive Filtering Applications Explained

Decision-directed method: Under normal operating
conditions, a good replica of the transmitted sequence is
being produced at the output of the decision device in the
receiver. Accordingly, if this output was the correct
transmitted sequence, it may be used as the desired
response for the purpose of adaptive equalization. Such a
method of learning is said to be decision directed
because the receiver attempts to learn by employing its
own decisions.
A final comment pertaining to performance evaluation: A
popular experimental technique for assessing the
performance of a data transmission system involves the
use of an eye pattern. This pattern is obtained by applying
the received wave to the vertical deflection plates of an
oscilloscope, and a saw-tooth wave at the transmitted
symbol rate to the horizontal deflection plates. The
resulting display is called an eye pattern because of its
resemblance to the human eye for binary data. Thus, in a
system using adaptive equalization, the equalizer attempts
to correct for intersymbol interference in the system and
thereby open the eye pattern as far as possible.

Adaptive Differential Pulse-Code Modulation

In Pulse-Code Modulation (PCM), which is the standard
technique for waveform coding, three basic operations are
performed on the speech signal: 1) sampling, 2)
quantization, and 3) coding. The operations of sampling
and quantization are designed to preserve the shape of the
speech signal. As for coding, it is merely a method of
translating the discrete sequence of sample values into a
more appropriate form of signal representation. The
rationale for sampling follows from a basic property of all
speech signals: They are band limited. This means that a
speech signal can be sampled in time at a finite rate in
accordance with the sampling theorem. For example,
commercial telephone networks designed to transmit
speech signals occupy a bandwidth from 200 to 3200 Hz.
To satisfy the sampling theorem, a conservative sampling
rate of 8 kHz is commonly used in practice.

In PCM, as used in telephony, the speech signal is

sampled at the rate of 8 kHz, nonlinearly quantized, and
the coded into 8-bit words, as shown in Figure 8(a). The
result is a good signal-to-quantization noise ratio over a
wide dynamic range of input signal levels. This method
requires a bit rate of 64 Kbps.
Differential Pulse-Code Modulation (DPCM), another
example of waveform coding, involves the use of a
predictor as shown in Figure 5(b). The predictor is
designed to exploit the correlation that exists between
adjacent samples of the speech signal in order to realize a
reduction in the number of bits required for the
transmission of each sample of the speech signal and yet
maintain a prescribed quality of performance. This is
achieved by quantizing and then coding the prediction
error that results from the subtraction of the predictor
output from the input. If the prediction is optimized, the
variance of the prediction error will be significantly smaller
than that of the input signal, so a quantizer with a given
number of levels can be adjusted to produce a quantizing
error with a smaller variance than would be possible if the
input signal were quantized directly as in a standard PCM
system. Equivalently, for a quantizing error of prescribed
variance, DPCM requires a smaller number of quantizing
levels than PCM. Differential PCM uses a fixed quantizer
and a fixed predictor. A further reduction in the
transmission rate can be achieved by using an adaptive
quantizer together with an adaptive predictor of sufficiently
high order.
Adaptive Differential Pulse-Code Modulation (ADPCM)
can digitize speech with toll (8-bit PCM) quality at 32 Kbps
(see Figure 8[c]).

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

Sampled
Speech
Input

Nonuniform
Quantizer

PCM
Wave

(a)
+

Sampled
Speech
Input

DPCM
Wave

Quantizer

Predictor

(b)

Adaptive
Algorithm

Sampled
Speech
Input

ADPCM
Wave

Quantizer

Predictor

Adaptive
Algorithm
(c)

Figure 8. Waveform Coders: (a) PCM, (b) DPCM, (c) ADPCM

AP96DSP0200

AN008001-0301

Adaptive Filtering Applications Explained

Adaptive Noise Canceling

As the name implies, adaptive noise cancelling relies on
the use of noise cancelling by subtracting noise from a
received signal, an operation controlled in an adaptive
manner for the purpose of improved signal-to-noise ratio.
Ordinarily, it is inadvisable to subtract noise from a
received signal because such an operation could produce

disastrous results by causing an increase in the average

power of the output noise. However, when proper
provisions are made, and filtering and subtraction are
controlled by an adaptive process, it is possible to achieve
a superior system performance compared to direct filtering
of the received signal.

Primary
Sensor

Output

Signal
Source

Reference
Sensor
Noise
Source

Adaptive
Filter

Estimate of
Noise

Adaptive Noise Canceller

Figure 9. Adaptive Noise Cancellation

Basically, an adaptive noise canceller is a dual-input,

closed-loop adaptive control system as illustrated in Figure
9 and Figure 1. The two inputs of the system are derived
from a pair of sensors: a primary sensor and a reference
sensor.

The primary sensor receives an information-bearing

signal s(n) corrupted by additive noise v0(n).

The signal and the noise are not correlated with each
other. The reference sensor receives a noise v1(n)
that is not correlated with the signal s(n) but correlated
with the noise v0(n) in the primary sensor output in an
unknown way:

E[s(n),v1(n-k)]=0, for all k

E[v0(n)v1(n-k)]=p(k)
where, as before, the signals are real valued and p(k) is an
unknown cross correlation for lag k.
The reference signal v1(n) is processed by an adaptive
filter to produce the output signal y(n). The filter output is
subtracted from the primary signal d(n), serving as the
desired response for the adaptive filter. The error signal is
defined by:
e(n) = d(n)-y(n)

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

The error signal is used, in tern, to adjust the tap weights
of the adaptive filter, and the control loop around the
operations of filtering and subtraction is thereby closed.
Note that the information bearing signal s(n) is indeed part
of the error signal e(n). Now, the adaptive filter attempts to
minimize the mean-square value (average power) of the
error signal e(n). The information bearing signal s(n) is
essentially unaffected by the adaptive noise canceller.
Hence, minimizing the mean-square value of the error
signal e(n) is equivalent to minimizing the mean-square
value of the output noise v0(n)-y(n). With the signal s(n)
remaining essentially constant, it follows that the
minimization of the mean-square value of the error signal
is indeed the same as the maximization of the output
signal to noise ratio of the system.
The effective use of adaptive noise cancelling therefore
requires that the reference sensor be placed in the noise
field of the primary sensor with two specific objectives in
mind:
1. The information-bearing signal component of the
primary sensor output is undetectable in the reference
sensor output.

2. The reference sensor output is highly correlated with

the noise component of the primary sensor output.
Moreover, the adaptation of the adjustable filter
coefficients must be near optimum.

Noise Canceling Applications

Now, let us consider two useful applications of the
adaptive noise cancelling operation.
Canceling 60-Hz Interference in Electrocardiography
(ECG). In ECG, commonly used to monitor heart patients,
an electrical discharge radiates energy through a human
tissue, and the resulting output is received by an electrode.
The electrode is usually positioned in such a way that the
received energy is maximized. Typically, however, the
electrical discharge involves very low potentials. Hence
extra must be exercised in minimizing signal degradation
due to external interference. By far, the strongest form of
interference is that of a 60-Hz periodic waveform picked up
by the receiving electrode from nearby electrical
equipment. Figure 10 shows a block diagram of the
adaptive noise canceller used to reduce the harmonics.

ECG
Preamplifier
Output

Adaptive Noise Canceller

Primary
Signal

^ (n)
w
0
+
Reference
Signal

ECG
Recorder

Attenuator

^ (n)
w
1

90Phase
Shifter

LMS
Algorithm

Figure 10. Adaptive Noise Canceller for Suppressing 60-Hz Interference

in Electrocardiography. (After Widrow, and others [1975b].)

AP96DSP0200

AN008001-0301

Adaptive Filtering Applications Explained

Reduction of Acoustic Noise in Speech. At a noisy site,
for example, the cockpit of a military aircraft, voice
communication is effected by the presence of acoustic
noise. This is particularly serious when linear predictive
coding (LPC) is used for the digital representation of voice
signals at low-bit rates.

time delay involved. If the delay between the speech and

the echo is short, the echo is not noticeable but perceived
as a form of spectral distortion or reverberation. If, on the
other hand, the delay exceeds a few tens of milliseconds,
the echo is distinctly noticeable.
To see how echoes occur, consider a long-distance
telephone circuit depicted in Figure 11. Every telephone is
connected to a central office by a two-wire line called the
customer loop. The two-wire line serves the need for
communications in either direction. However, for circuits
longer than 35 miles, a separate path is necessary for
each direction of transmission.

The noise corrupted speech is used as the primary signal.

To provide the reference signal, a reference microphone is
placed in a location where there is sufficient isolation from
the source of speech.

Echo Cancellation
Almost all conversations are conducted in the presence of
echoes. An echo may not be distinct, depending on the

Echo of As Speech

L1

Hybrid
A

Hybrid
B

Speaker
A

Speaker
B

L2

Echo of Bs Speech

The boxes marked N are balancing impedances.

Figure 11. Long-Distance Telephone Circuit

10

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

Accordingly, there must be provision for connecting the
two-wire circuit to the four-wire circuit. This connection is
accomplished by means of a hybrid transformer,
commonly referred to as a hybrid.
Basically, a hybrid is a bridge circuit with three ports. If the
bridge is not perfectly balanced, the in port becomes
coupled to the out port, thereby giving rise to an echo
(refer to Figure 12).

In

Balancing
Network

Speaker

The basic principle of echo cancellation is to synthesize a

replica of the echo and subtract it from the returned signal.
This principle is illustrated in Figure 13 for only one
direction of transmission. The adaptive canceller is placed
in the four-wire path near the origin of the echo. The
synthetic echo is generated by passing the speech signal
from speaker A through an adaptive filter that ideally
matches the transfer function of the echo path. The
reference signal, passing through the hybrid, results in the
echo signal. This echo, together with a near-end talker
signal x, constitutes the desired response for the adaptive
canceller. The synthetic echo is subtracted from the
desired response to yield the canceller error signal. In any
event, the error signal is used to control the adjustments
made in the coefficiencies of the adaptive filter. For the
adaptive echo cancellation circuit to operate satisfactorily,
the impulse response of the adaptive filter should have a
length greater than the longest echo path that needs to be
accommodated.

Out

Figure 12. Hybrid Circuit

Adaptive
Filter

^r (n)

e(n)

Speaker
B

Hybrid
Speaker As
Echo r(n)

Speaker Bs
Signal x(n)

+
r(n) + x(n)

Figure 13. Signal Definitions for Echo Cancellation

AP96DSP0200

AN008001-0301

11

Adaptive Filtering Applications Explained

LMS Algorithm
The well-known least-mean-square (LMS) algorithm is an
important member of the family of stochastic gradientbased algorithms. A significant feature of the LMS
algorithm is its simplicity. It does not require
measurements of the pertinent correlation functions, nor
does it require matrix inversion. Indeed, it is the simplicity
of the LMS algorithm that has made it the standard against
which other adaptive filtering algorithms are benchmarked.
The operation of the LMS algorithm is descriptive of a
feedback control system. Basically, it consists of a
combination of two basic processes:

u(n)

1. An adaptive process, which involves the automatic

adjustment of a set of tap weights.
2. A filtering process, which involves: (a) forming the
inner product of a set of tap inputs and the
corresponding set of tap weights emerging from the
adaptive process to produce an estimate of a desired
response, and (b) generating an estimation error by
comparing this estimate with the actual value of the
desired response. The estimation error is used, in turn,
to actuate the adaptive process, thereby closing the
feedback loop.
Correspondingly, we may identify two basic components in
the structural constitution of the LMS algorithm, as
illustrated in Figure 14.

^
d(n |Un)

Transversal Filter
w(n)

Adaptive Weight-Control
Mechanism

e(n)

+
d(n)

Figure 14. Block Diagram of Adaptive Transversal Filter

12

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

First we have a transversal filter, around which the LMS
algorithm is built. This component is responsible for
performing the filtering process. Second, we have a
mechanism for performing the adaptive control process on
the tap weights of the transversal filter.

u(n)

u(n 1)
z 1

w0 (n)

z 1

Details of the transversal filter component are presented in

Figure 15. The tap inputs from the elements of the M-by-1
tap input vector u(n), where M-1 is the number of delay
elements.

u(n M

u(n 2)
. . .

w1 (n)

w2 (n)

. . .

. . .

z 1

u(n M +

(n)
wM-2

wM-1 (n)

d(n

e(n)

+
d(n)

Figure 15. Detailed Structure of the Transversal Filter Component

AP96DSP0200

AN008001-0301

13

Adaptive Filtering Applications Explained

Figure 16 presents details of the adaptive weight-control
mechanism. Specifically, a scaled version of the inner
product of the estimation error and the tap input is
computed. The result obtained defines the correction

applied to the tap weight. The scaling factor used in this

computation is called the adaptation constant or step size
parameter.

w0 (n)
u(n)

w1 (n)
u(n 1)

w2 (n)
u(n 2)

.
.
.

.
.
.

u(n M + 2)

u(n M + 1)

wM2 (n)

wM1 (n)

Figure 16. Detailed Structure of the Adaptive Weight-Control Mechanism

The tap weight vector computed by the LMS algorithm

executes a random motion around the minimum point of
the error performance surface. This random motion gives
rise to two forms of convergence behavior for the LMS
algorithm: 1) Convergence in the mean, and 2)
Convergence in the mean square.

14

It is important to realize, however, that the misadjustments are under the designer's control. In
particular, the feedback loop acting around the tap weights
behaves like a low-pass filter, with an average time
constant that is inversely proportional to the step size
parameter u. Hence, by assigning a small value to u, the
adaptive process is made to progress slowly, and the
effects of gradient noise on the tap weights are largely
filtered out. This, in turn, has the effect of reducing the misadjustments.

AN008001-0301

AP96DSP0200

Adaptive Filtering Applications Explained

tap-weight vector, w(n). The iterative procedure is started
with the initial guess w(0)=0. In general, the LMS algorithm
requires only 2M+1 complex multiplications and 2M
complex additions per iteration, where M is the number of
tap weights used in the adaptive transversal filter.

LMS Algorithm Summary

1. FIR Filter output: y(n) = w'(n)u(n)
2. Estimation error: e(n) = d(n) - y(n)
3. Tap weight adaptation: w(n+1) = w(n) + uu(n)e(n)
u(n) .. tap-input vector: u(n), u(n-1), ... u(n-M+1)
w(n) .. tap-weight vector: w0(n), w1(n) ... wM-1(n)
Equation 1 and 2 define the estimation error e(n), the
computation of which is based on the current estimate of
the tap-weight vector, w(n). Note also that the second
term, uu(n)e(n), on the right side of Equation 3 represents
the correction that is applied to the current estimate of the

AP96DSP0200

References
1. Haykin, S. Adaptive Filter Theory. Prentice-Hall
International, Inc. 1991, pages 1720, 3140, 4955,
303.
2. Honig, M. and Messerschmitt, D. Adaptive Filters:
Structures, Algorithms, and Applications. Boston:
Kluwer Academic Publishers. 1984.

AN008001-0301

15