 |
Claims  |
|
|
The invention claimed is:
1. An echo canceller comprising:
a receive path and a send path each having an input port and an output
port;
a non-adaptive filter with first filter coefficients and an input port
coupled to the receive path;
an adaptive filter with second filter coefficients and an input port
coupled to the receive path;
a first subtracter with an addend input port coupled to the input port, a
subtrahend input port coupled to an output port of the non-adaptive
filter, and an output port coupled to the send output port;
a second subtracter with an addend input port coupled to the send output
port, a subtrahend input port coupled to an output port of the adaptive
filter, and an output port coupled to a feedback input port of the
adaptive filter; and
a controller coupled to the output ports of the first and second
subtracters, and coupled to the filters, for accumulating the second
filter coefficients on the first filter coefficients and for resetting the
second filter coefficients in response to signal levels at the output
ports of the first and second subtracters.
2. The echo canceller of claim 1 in which the first filter coefficients are
modified solely by accumulation with the second filter coefficients.
3. The echo canceller of claim 1 in which the resetting includes clearing
the second filter coefficients.
4. The echo canceller of claim 1 in which after accumulating the second
filter coefficients on the first filter coefficients and then resetting
the second filter coefficients, the signal level at the output port of the
first subtracter is essentially identical to what the signal level was at
the output port of the second subtracter before the accumulating and
resetting occurred.
5. The echo canceller of claim 1 in which the controller further includes a
spectral detector coupled to the receive path, wherein the accumulating is
dependent on a frequency distribution of a signal level at the receive
path being within a spectral flatness criterion.
6. An echo canceller with a receive path between a receive input port and a
receive output port, and a send path between a send input port and a send
output port, the echo canceller being used for cancelling an echo signal
at the send input port which occurs in response to a receive input signal,
the echo canceller comprising:
a first non-adaptive filter with first filter coefficients for generating a
first replica signal, which is an estimate of the echo signal, in response
to the receive input signal;
a second adaptive filter with second filter coefficients for generating a
second replica signal, which is an estimate of a send output signal, in
response to the receive input signal;
a first subtracter for generating the send output signal representing a
difference between a send input signal and the first replica signal;
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
and for applying the error signal to the second filter; and
a controller for replacing the first filter coefficients by a sum of the
first and second filter coefficients in corresponding tap positions and
for resetting the second filter coefficients in response to a calculation
that includes a first quantity associated with the send output signal and
a second quantity associated with the error signal.
7. The echo canceller of claim 6 in which the first filter coefficients are
modified solely by replacing the first filter coefficients by the sum of
the first and second filter coefficients.
8. The echo canceller of claim 6 in which the first filter has an impulse
response which is an estimate of an impulse response of an echo path
between the receive output port and the send input port.
9. The echo canceller of claim 6 in which the second filter includes an
adaptation processor for adaptively modifying the second filter
coefficients in response to the receive input signal and the error signal.
10. The echo canceller of claim 6 in which the first and second filters are
digital filters.
11. The echo canceller of claim 6 in which the controller is also for
resetting the second filter coefficients without replacing the first
filter coefficients when a ratio of the second quantity, representing a
power of the error signal, to the first quantity, representing a power of
the send output signal, exceeds a threshold.
12. The echo canceller of claim 6 in which the first filter coefficients
are implemented in single-precision words and the second filter
coefficients are implemented in double-precision words.
13. The echo canceller of claim 6, as part of an integrated circuit chip.
14. The echo canceller of claim 6, as part of a loudspeaking telephone set.
15. An echo canceller with a receive path between a receive input port and
a receive output port, and a send path between a send input port and a
send output port, the echo canceller being used for cancelling an echo
signal at the send input port which occurs in response to a receive input
signal, the echo canceller comprising:
a first non-adaptive filter with first filter coefficients for generating a
first replica signal, which is an estimate of the echo signal, in response
to the receive input signal, the first filter including a first
programmable filter coefficient memory for storing the first filter
coefficients;
a second adaptive filter with second filter coefficients for generating a
second replica signal, .which is an estimate of a send output signal, in
response to the receive input signal, the second filter including a second
programmable filter coefficient memory for storing the second filter
coefficients;
a first subtracter for generating the send output signal representing a
difference between a send input signal and the first replica signal;
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
and for applying the error signal to the second filter; and
a controller for replacing the first filter coefficients by a sum of the
first and second filter coefficients in corresponding tap positions and
for resetting the second filter coefficients in response to a calculation
that includes a first quantity associated with the send output signal and
a second quantity associated with the error signal.
16. An echo canceller with a receive path between a receive input port and
a receive output port, and a send path between a send input port and a
send output port, the echo canceller being used for cancelling an echo
signal at the send input port which occurs in response to a receive input
signal, the echo canceller comprising:
a first non-adaptive filter with first filter coefficients for generating a
first replica signal, which is an estimate of the echo signal, in response
to the receive input signal;
a second adaptive filter with second filter coefficients for generating a
second replica signal, which is an estimate of a send output signal, in
response to the receive input signal;
a first subtracter for generating the send output signal representing a
difference between a send input signal and the first replica signal;
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
and for applying the error signal to the second filter; and
a controller for replacing the first filter coefficients by a sum of the
first and second filter coefficients in corresponding tap positions and
for resetting the second filter coefficients in response to a calculation
that includes a first quantity associated with the send output signal and
a second quantity associated with the error signal, the calculation
includes determining whether a ratio of the first quantity, representing a
power of the send output signal, to the second quantity, representing a
power of the error signal, exceeds a threshold.
17. The echo canceller of claim 16 in which the powers of the send output
signal and the error signal are instantaneous signal powers of these
signals averaged over a specific period of time.
18. An echo canceller with a receive path between a receive input port and
a receive output port, and a send path between a send input port and a
send output port, the echo canceller being used for cancelling an echo
signal at the send input port which occurs in response to a receive input
signal, the echo canceller comprising:
a first non-adaptive filter with first filter coefficients for generating a
first replica signal, which is an estimate of the echo signal, in response
to the receive input signal;
a second adaptive filter with second filter coefficients for generating a
second replica signal, which is an estimate of a send output signal, in
response to the receive input signal;
a first subtracter for generating the send output signal representing a
difference between a send input signal and the first replica signal;
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
and for applying the error signal to the second filter; and
a controller for replacing the first filter coefficients by a sum of the
first and second filter coefficients in corresponding tap positions and
for resetting the second filter coefficients in response to a calculation
that includes a first quantity associated with the send output signal and
a second quantity associated with the error signal, in which resetting in
the second filter coefficients includes clearing the second filter
coefficients after replacing the first filter coefficients.
19. The echo canceller of claim 18 in which the first filter provides
essentially identical echo cancellation to that provided by the first and
second filters combined before the replacing and clearing occurs.
20. An echo canceller with a receive path between a receive input port and
a receive output port, and a send path between a send input port and a
send output port, the echo canceller being used for cancelling an echo
signal at the send input port which occurs in response to a receive input
signal, the echo canceller comprising:
a first non-adaptive filter with first filter coefficients for generating a
first replica signal, which is an estimate of the echo signal, in response
to the receive input signal;
a second adaptive filter with second filter coefficients for generating a
second replica signal, which is an estimate of a send output signal, in
response to the receive input signal;
a first subtracter for generating the send output signal representing a
difference between a send input signal and the first replica signal;
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
and for applying the error signal to the second filter; and
a controller for replacing the first filter coefficients by a sum of the
first and second filter coefficients in corresponding tap positions and
for resetting the second filter coefficients in response to a calculation
that includes a first quantity associated with the send output signal and
a second quantity associated with the error signal, the controller
including a spectral detector for determining whether a frequency
distribution of the receive input signal is within a spectral flatness
criterion.
21. The echo canceller of claim 20 in which the spectral detector prevents
the replacing after the receive input signal is a narrowband signal for a
time period.
22. The echo canceller of claim 21 in which the narrowband signal includes
single-tones and double-tones.
23. The echo canceller of claim 22 in which the narrowband signal is a
single-tone.
24. An integrated circuit including an echo canceller with a receive path
between a receive input port and a receive output port, and a send path
between a send input port and a send output port, the echo canceller being
used for canceling an echo signal at the send input port which occurs
along an echo path between the receive output port and the send input port
in response to a receive input signal at the receive input port, the echo
canceller comprising:
a first programmable digital filter, with a first filter coefficient memory
for storing first filter coefficients, for generating a first replica
signal which is an estimate of the echo signal in response to the receive
input signal, such that the first filter has as impulse response which is
an estimate of an impulse response of the echo path, wherein the first
filter devoid of an adaptation processor;
a second adaptive digital filter, with a second filter coefficient memory
for storing second filter coefficients, for generating a second replica
signal which is an estimate of the echo signal in response to the receive
input signal, the second filter further including an adaptation processor
for adaptively modifying the second filter coefficients in response to the
receive input signal and an error signal;
a first subtracter for generating a send output signal at the send output
port representing a difference between a send input signal at the send
input port and the first replica signal;
a second subtracter for generating the error signal representing a
difference between the send output signal and the second replica signal;
and
a controller for performing an update operation when both (a) a ratio of a
first quantity, representing a power of the send output signal, to a
second quantity, representing a power of the error signal, exceeds a first
threshold, and (b) a frequency distribution of the receive input signal is
within a spectral flatness criterion,
the controller also for performing a reset operation when a second ratio of
the second quantity to the first quantity exceeds a second threshold,
wherein the update operation includes replacing the first filter
coefficients by a sum of the first and second filter coefficients in
corresponding tap positions and clearing the second filter coefficients,
thereby updating the echo canceller, and
the reset operation includes clearing the second filter coefficients
without changing the first filter coefficients, thereby removing drifting
second filter coefficients.
25. A method of operating an echo canceller with a receive path between a
receive input port and a receive output port, a send path between a send
input port and a send output port, a first non-adaptive filter with first
filter coefficients for generating a first replica signal, a second
adaptive filter with second filter coefficients coupled to an adaptation
processor for generating a second replica signal, a first subtracter for
generating a send output signal at the send output port representing a
difference between the send input signal and the first replica signal, and
a second subtracter for generating an error signal representing a
difference between the send output signal and the second replica signal,
the method comprising the steps of:
modifying the second filter coefficients by using the adaptation processor
in response to the receive input signal and the error signal; and
modifying the first filter coefficients by accumulating the second filter
coefficients on the first filter coefficients in corresponding tap
positions in response to signal levels of the send output signal and the
error signal, thereby updating the echo canceller.
26. The method of claim 25, further comprising clearing the second filter
coefficients after modifying the first filter coefficients so that the
first replica signal is essentially identical to what the second replica
signal was before the modifying and clearing occurred.
27. The method of claim 25, further comprising detecting a frequency
distribution of the receive input signal, and preventing the modifying of
the first filter coefficients after the receive input signal has been a
narrowband signal for a time period. |
|
 |
|
 |
Claims |
|
|
|
Description |
|
|
FIELD OF THE INVENTION
The invention relates generally to echo cancellers, and more particularly
to echo cancellers with at least two filters.
BACKGROUND OF THE INVENTION
Speech typically results in reflected waves. When the reflected wave
arrives a very short time after the direct sound, it is perceived as a
spectral distortion or reverberation. However, when the reflection arrives
a few tens of milliseconds after the direct sound, it is heard as a
distinct echo. Such echoes may be annoying, and under extreme conditions
can completely disrupt a conversation.
Line echoes (i.e., electrical echoes) occur in telecommunications networks
due to impedance mismatches at hybrid transformers that couple two-wire
local customer loops to four-wire long-distance trunks. Ideally, the
hybrid passes the far-end signal at the four-wire receive port through to
the two-wire transmit port without allowing leakage into the four-wire
transmit port. However, this would require exact knowledge of the
impedance seen at the two-wire ports, which in practice varies widely and
can only be estimated. As a result, the leaking signal returns to the
far-end talker as an echo. The situation can be further complicated by the
presence of two-wire toll switches, allowing intermediate four-two-four
wire conversions internal to the network. In telephone connections using
satellite links with round-trip delays on the order of 600 ms, line echoes
can become particularly disruptive.
Acoustic echoes occur in telecommunications networks due to acoustic
coupling between a loudspeaker and a microphone (e.g., in a speakerphone).
During a teleconference, where two or more parties are connected by a
full-duplex link, an acoustic reflection of the far-end talker through the
near-end conference room is returned to the far-end talker as an echo.
Acoustic echo cancellation tends to be more difficult than line echo
cancellation since the duration of the acoustic path is usually several
times longer (100-400 ms) than typical electrical line paths (20 ms), and
the acoustic path may change rapidly at any time due to opening doors,
moving persons, changing temperatures, etc.
Echo suppressors have been developed to control line echoes in
telecommunications networks. Echo suppressors decouple the four-wire
transmit port when signal detectors determine that there is a far-end
signal at the four-wire receive port without any near-end signal at the
two-wire receive port. Echo suppressors, however, are generally
ineffective during double-tang when speakers at both ends are talking
simultaneously. During double-talk, the four-wire transmit port carries
both the near-end signal and the far-end echo signal. Furthermore, echo
suppressors tend to produce speech clipping, especially during long delays
caused by satellite links.
Echo cancellers have been developed to overcome the shortcomings of echo
suppressors. Echo cancellers include an adaptive filter and a subtracter.
The adaptive filter attempts to model the echo path. The incoming signal
is applied to the adaptive filter which generates a replica signal. The
replica signal and the echo signal are applied to the subtracter. The
subtracter subtracts the replica signal from the echo signal to produce an
error signal. The error signal is fed back to the adaptive filter, which
adjusts its filter coefficients (or taps) in order to minimize the error
signal. In this manner, the filter coefficients converge toward values
that optimize the replica signal in order to cancel (i.e., at least
partially offset) the echo signal. Echo cancellers offer the advantage of
not disrupting the signal path. Economic considerations place limits on
the fineness of sampling times and quantization levels in digital adaptive
filters, but technological improvements are relaxing these limits. Echo
cancellers were first deployed in the U.S. telephone network in 1979, and
currently are virtually ubiquitous in long-distance telephone circuits.
See generally Messerschmitt, "Echo Cancellation in Speech and Data
Transmission", IEEE Journal on Selected Areas in Communications, Vol.
SAC-2, No. 2, March 1984, pp. 283-298; and Tao et al., "A Cascadable VLSI
Echo Canceller", IEEE Journal on Selected Areas in Communications, Vol.
SAC-2, No. 2, March 1984, pp. 298-303.
In order for the adaptive filter to correctly model the echo path, the
output signal of the echo path must originate solely from its input
signal. During double-talking, speech at the near-end that acts as
uncorrelated noise causes the filter coefficients to diverge (or drift).
In open-loop paths, coefficient drift is usually not catastrophic although
a brief echo may be heard until convergence is established again. In
closed-loop paths (which typically include acoustic echo paths), however,
coefficient drift may lead to an unstable system which causes howling and
makes convergence difficult. To alleviate this problem, double-talk
detectors are commonly used for disabling the adaptation during
double-talking. Double-talk detectors may employ, for instance, Geigel's
test to detect double-talking. Unfortunately, double-talk detectors fail
to indicate the presence of double-talking for a time period (e.g., a
whole syllable) after double-talking begins. During this time period, the
coefficients may drift and lead to howling as mentioned above.
Furthermore, double-talking becomes increasingly difficult to detect as an
acoustic echo becomes large in comparison to the near-end signal.
An adaptive echo canceller arranged for overcoming the double-talking
problem is proposed in Ochiai et al., "Echo Canceler with Two Echo Path
Models", IEEE Transactions On Communications, Vol. COM-25, No. 6, June
1977, pp. 589-595. Ochiai et al. discloses a parallel filter arrangement
with a programmable foreground filter and an adaptive background filter.
Each of the filters generates a replica of the echo signal. The filter
coefficients of the foreground filter are replaced by those of the
background filter when the replica signal from the background filter
provides a better estimate of the echo signal than the replica signal of
the foreground filter. Therefore, during uncorrelated double-talking, the
foreground filter is relatively immune from coefficient drift in the
background filter. There are, however, drawbacks to this approach. First,
in the event the filter coefficients of the background filter drift into
an unstable state and are subsequently cleared, there may be a relatively
long delay before the background filter works back to providing a better
replica signal than the foreground filter. As a result, the convergence
time for the foreground filter may be significantly delayed. Secondly, the
filter coefficients of the background filter may converge to provide good
cancellation for narrowband signals (i.e., a narrow range of frequencies),
but poor cancellation outside this frequency range. In this instance, the
background filter may provide the foreground filter with an undesirable
update.
Based on the foregoing, there is a need for an echo canceller which
protects against double-talking, provides rapid convergence, and has
robustness to narrowband signals.
SUMMARY OF THE INVENTION
An embodiment of the invention is an echo canceller which provides
excellent echo cancellation performance with rapid convergence, robustness
to double-talking and channel noise disturbances, and/or robustness to
narrowband signals.
In accordance with one aspect of the invention, an echo canceller includes
receive and send paths, with each including an input and output port. The
echo canceller is adapted to cancel an echo between the receive output
port and the send input port. The echo canceller also includes a first
non-adaptive filter with first filter coefficients for generating a first
replica signal in response to a receive input signal, a second adaptive
filter with second filter coefficients for generating a second replica
signal in response to the receive input signal, a first subtracter for
generating a send output signal representing a difference between a send
input signal and the first replica signal, a second subtracter for
generating an error signal representing a difference between the send
output signal and the second replica signal, and a controller for
replacing the first filter coefficients by the sum of the first and second
filter coefficients and for resetting the second filter coefficients in
response to a calculation that includes a first quantity associated with
the send output signal and a second quantity associated with the error
signal. Preferably, the first filter coefficients are modified solely by
replacing the first filter coefficients by the sum of the first and second
filter coefficients.
In accordance with another aspect of the invention, the controller includes
a spectral detector for determining when the receive input signal is a
narrowband signal in order to prevent updating the first filter
coefficients in response to a narrowband signal.
BRIEF DESCRIPTION OF THE DRAWING
The invention, together with its various features and advantages, can be
readily understood from the following detailed description taken in
conjunction with the accompanying drawing, in which:
FIG. 1 is a schematic diagram illustrating the manner in which an echo
canceller is generally used as part of a loudspeaking telephone set;
FIG. 2 is a schematic diagram illustrating an embodiment of an echo
canceller according to the invention, and
FIG. 3 is a schematic diagram illustrating an embodiment of the echo
canceller of FIG. 2.
DETAILED DESCRIPTION
FIG. 1 shows a simplified schematic diagram illustrating the manner in
which an echo canceller is generally used as part of a loudspeaking
telephone set. Echo canceller 10 includes a receive path 12 with receive
input port 14 and receive output port 16, and send path 18 with send input
port 20 and send output port 22. A receive input signal x(t), representing
a far-end signal, is applied to receive input port 14. Receive input
signal x(t) is coupled via path 14 to receive output port 16, which
couples signal x(t) to loudspeaker 24. Acoustic echo path 26 with transfer
characteristic H[t] is located between loudspeaker 24 and microphone 28.
Echo path 26 causes signal x(t) at loudspeaker 24 to appear as echo signal
H[x(t)] at microphone 28. Microphone 28 also generates near-end signal
u(t) due to speech at the near-end. Therefore, microphone 28 generates
send input signal s(t) consisting of near-end signal u(t) added to echo
signal H[x(t)] by way of superposition. Echo canceller 10 further
comprises adaptive digital transversal filter 30 and subtracter 32.
Adaptive filter 30 has a finite impulse response (FIR), and its filter
coefficients are adaptively updated to model the transfer characteristic
H[t] at sample intervals. Adaptive filter 30 synthesizes a replica signal
y(t) as an estimate of the undesired echo signal H[x(t)] in response to
receive input signal x(t). Subtracter 32 subtracts replica signal y(t)
from send input signal s(t) to form error signal e(t). Error signal e(t)
is coupled to send output port 22 to provide the send output signal. Error
signal e(t) is also fed back to adaptive filter 30. Error signal e(t) can
be described as follows:
e(t)=u(t)+[H[x(t)]-y(t)]
From this expression, it is clear that signal e(t) represents near-end
signal u(t) when replica signal y(t) is a reliable estimate of echo signal
H[x(t)].
Generally, the transfer characteristic H[t] of echo path 26 will be
time-varying. Echo signal H[x(t)] approximates the linear convolution of
signal x(t) with the impulse response h(t) of echo path 26. Therefore,
adaptive filter 30 adjusts its impulse response w(t) as best it can to
match impulse response h(t). The adaptive adjustment of filter 32 is
controlled by error signal e(t). This adaptive adjustment is continued as
long as there is a correlation between error signal e(t) and receive input
signal x(t). When receive input signal x(t) is present and near-end signal
u(t) is absent (i.e., far-end speech without near-end speech), adaptive
filter 30 generates replica signal y(t) as a reliable estimate of echo
signal H[x(t)]. When, however, both x(t) and u(t) are present (i.e.,
double-talking), adaptive filter 30 can become grossly misadjusted due to
near-end signal u(t) as a disturbing factor in error signal e(t). This
misadjustment prevents replica signal y(t) from providing a reliable
estimate of H[x(t)], in which case H[x(t)] is improperly or inadequately
canceled.
Further discussion of echo cancellers appears in U.S. Pat. Nos. 5,390,250;
5,371,789; 5,263,020; 5,146,494; 4,903,247; 4,564,934; and in the articles
by Ochiai et at., Messerschmitt, and Tao et al., supra, which are all
incorporated herein by reference.
Since the present invention is best performed by a digital echo canceller,
the following description will utilize discrete-time modeling.
Discrete-time modeling can be obtained by assuming in FIG. 1 that signals
x(t) and s(t) are applied to analog-to-digital converters before being
applied to input ports 14 and 20, respectively, and likewise that signals
x(t) and e(t) are received from digital-to-analog converters coupled to
output ports 16 and 22, respectively, and further that all the relevant
signals in echo canceller 10 are digital signals. These digital signals
are denoted in a conventional manner so that, for example, x(k) denotes a
quantized sample of continuous-time sample x(t) at instant t=kT, where 1/T
is the sampling frequency at the appropriate (Nyquist) rate.
Digital-to-analog and analog-to-digital converters may be employed for
connecting the echo canceller 10 with analog channels.
FIG. 2 shows a schematic diagram of one embodiment of the invention. Echo
canceller 110 includes a receive path 112 with receive input port 114 and
receive output port 116, and send path 118 with send input port 120 and
send output port 122. Echo path 124 with transfer characteristic H[z] is
disposed between receive output port 116 and send input port 120. A first
non-adaptive filter 126 with first filter coefficients has an input port
coupled to receive path 112, and a second adaptive filter 128 with second
filter coefficients has an input port coupled to receive path 112. A first
subtracter 130 has an addend input port coupled to send input port 120, a
subtrahend input port coupled to an output port of filter 126, and an
output port coupled to send output port 122. A second subtracter 132 has
an addend input port coupled to send output port 122, a subtrahend input
port coupled to an output port of filter 128, and an output port coupled
to a feedback input port of filter 128. Thus, filters 126 and 128 are
arranged in cascade with filter 126 in the foreground and filter 128 in
the background. Filter 126 provides the actual echo cancellation for echo
canceller 110 at output port 122. Filter 128 attempts to cancel whatever
echo is not canceled by filter 126 in order to subsequently improve the
performance of filter 126.
Controller 134 has input ports coupled to the output ports of subtracters
130 and 132. Controller 134 also has both input and output ports coupled
to filters 126 and 128 for accessing the first filter coefficients of
filter 126 and the second filter coefficients of filter 128. Controller
134 is used for replacing the first filter coefficients with the sum of
the first and second filter coefficients when the combination of filter
126 and filter 128 provides better echo cancellation than filter 126
alone. Controller 134 performs this update operation by accumulating the
second filter coefficients on related first filter coefficients, and then
resetting the second filter coefficients. Controller 134 performs the
update operation in response to a function determined at least by the
levels of the outputs of the first and second subtracters. That is, when
second subtracter 132 generates a smaller output signal than that of first
subtracter 130, then the combination of filters 126 and 128 provides
better echo cancellation than filter 126 alone, and the update operation
improves the echo canceller's performance. After updating occurs, if the
second filter coefficients are cleared, then echo cancellation provided by
filter 126 is essentially identical to that provided by the combination of
filters 126 and 128 before updating occurs. In addition, after updating
occurs, adaptive filter 128 resumes its adaptive processing. As a result,
the updating operation may be repeated as often as necessary to
incrementally improve the echo cancellation at send output port 122.
Preferably, the first filter coefficients are modified solely by
accumulation with the second filter coefficients. That is, filter 128
includes an adaptation processor, whereas filter 126 is devoid of an
adaptation processor. The elimination of a separate adaptation processor
for filter 126 reduces hardware requirements and provides for a more
compact, efficient implementation. Furthermore, filter 126 becomes less
susceptible to coefficient drift during uncorrelated double-talking.
In operation, receive input signal x(k), representing a far-end signal, is
applied to receive input port 114. Signal x(k) is applied to filters 126
and 128, and to receive output port 116. Send input port 120 receives send
input signal s(k) consisting of near-end signal u(k) added to echo signal
H[x(k)]. Filter 126 generates a first replica signal y.sub.1 (k), as an
estimate of echo signal H[x(k)], in response to signal x(k). Likewise,
filter 128 generates a second replica signal y.sub.2 (k), as an estimate
of echo signal e.sub.1 (k), in response to signal x(k). Subtracter 130
generates error signal e.sub.2 (k) , representing the difference between
signals y.sub.1 (k) and s(k). Signal e.sub.1 (k) is coupled to send output
port 122. Subtracter 132 generates error signal e.sub.2 (k), representing
the difference between signals y.sub.2 (k) and e.sub.1 (k). Signal e.sub.2
(k) is fed back to filter 128 for adaptive processing.
Controller 134 may employ any number of algorithms for determining whether
to perform the update operation. For instance, controller 134 may update
the first coefficients in response to a calculation that includes a first
quantity associated with e.sub.1 (k), and a second quantity associated
with e.sub.2 (k). The calculation, for example, may determine whether a
ratio of the first quantity, representing a power of e.sub.1 (k), to the
second quantity, representing a power of e.sub.2 (k), exceeds a
predetermined threshold. The powers can be instantaneous signal powers of
e.sub.1 (k) and e.sub.2 (k) averaged over a specific period of time.
FIG. 3 shows a schematic diagram of an embodiment of FIG. 2. As is seen,
non-adaptive filter 126 includes X register 202, convolution circuit 204,
and coefficient register 206. Adaptive filter 128 includes X register 212,
convolution circuit 214, coefficient register 216, and adaptation
processor 218. X registers 202 and 212 each store N samples x(k), x(k-1),
. . . x(k-N) of the input receive signal. The value of integer N is
directly related to the delay of echo path 124. That is, N is selected for
a sufficiently accurate representation of the impulse response of echo
path 124. Coefficient register 206 provides a memory element for storing N
first filter coefficients corresponding to the impulse response of filter
126, and coefficient register 216 provides a memory element for storing N
second filter coefficients corresponding to the impulse response of filter
128. Registers 202, 206, 212 and 216 are each recirculated shift
registers. Convolution circuit 204 generates replica signal y.sub.1 (k) by
performing linear convolution on the samples stored in X register 202 with
the first filter coefficients stored in coefficient register 206.
Likewise, convolution circuit 214 generates replica signal y.sub.2 (k) by
performing linear convolution on the samples stored in X register 212 with
the second filter coefficients stored in coefficient register 216. If
desired, X registers 202 and 212 can be combined into a single X register.
Adaptation processor 218 adaptively updates the second filter coefficients
stored in coefficient register 216, in response to signals x(k) and
e.sub.2 (k), in order to minimize the level of signal e.sub.2 (k). Various
adaptation algorithms for correcting the second filter coefficients are
well known in the art. For instance, the Least Mean Square (LMS) algorithm
is an iterative stochastic gradient algorithm that minimizes the expected
value of the squared error signal. Various modifications of the LMS
algorithm, such as the Normalized Least Mean Square (NLMS) algorithm are
also suitable. Alternatively, the Least Squares (LS) algorithm provides a
noniterative, block- oriented adaptation algorithm.
Coefficient register 206 is not coupled to adaptation processor 218.
Instead, the first filter coefficients in coefficient register 206 are
programmably updated by controller 134. During the update operation, the
first filter coefficients in register 206, and the second filter
coefficients in register 216 corresponding to the same tap positions as
the first filter coefficients, are added together in arithmetic unit 220,
and the sum is stored in register 206. In other words, the first tap in
| | |
