 |
Description  |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates to an echo canceller comprising a receive path and a
send path, the two having each an input and an output, the output of the
receive path being connected to the input of an adaptive filter comprising
adapting means, which filter has an impulse response which is an estimate
of the impulse response of an echo path between the output of the receive
path and the input of the send path, the input of the send path being
coupled to a first input of a subtractor circuit and the output of the
adaptive filter being coupled to a second input of the subtractor circuit,
and the output of the subtractor circuit being coupled to the output of
the send path; this echo canceller also including detecting means which
generates a control signal for blocking the adapting means if the
detecting means detect a locally generated signal at the input of the send
path.
2. Description of the Related Art
An echo canceller of this type is known from Dutch Patent Application
8701633 which corresponds to U.S. Pat. No. 4,903,247.
In full-duplex signal transmission the presence of an undesired echo path
between the output of the receive path and the input of the send path, and
the presence of such an echo path at the far-end may cause a closed signal
loop to develop. Such a situation may occur, for example, in telephony
with the aid of loudspeaking telephone sets.
Because amplifiers are included in the send path, loop gain for a specific
frequency may be greater than 1. Consequently, a type of oscillation
occurs also denoted acoustic feedback when speech transmission is
concerned.
If the acoustic feedback is smaller than 1, oscillation will not occur, it
is true, but after a certain delay an echo of the signal applied to the
input of the send path will appear at the output of the receive path via
the far-end echo path. In telephony this means that a speaker hears his
own voice delayed by a specific period of time. This phenomenon is
experienced as extremely annoying especially in case of long delays.
In order to suppress these echoes, controllable attenuators included in the
send and receive paths are utilized in known fashion. In the case where
only a near-end signal generated by a near-end speaker is present, the
attenuator in the send path has a high transfer factor A.sub.max and the
attenuator in the receive path a low transfer factor A.sub.min. In the
case where only a signal coming from the far end is available, the
attenuator in the receive path has a high transfer factor A.sub.max and
the attenuator in the send path a low transfer factor A.sub.min. In the
case where there is a near-end generated signal as well as a far-end
generated signal available, the two attenuators are set to an equal
transfer value which equals .sqroot.A.sub.max *A.sub.min. In all cases the
loop gain is now reduced to such an extent that the problems caused by
echo signals are strongly reduced. A problem with this type of echo
cancelling is that the beginning of a speech signal is either not
transferred or is transferred in a strongly attenuated manner, because
after the detection of the speech signal a specific amount of time is
required before the attenuators can be adjusted to a correct transfer
factor. In addition, when doubletalk occurs, the transfer factor will be
lower than necessary because the two loudspeaking telephone sets introduce
an attenuation both in the send and receive paths. Consequently, when
doubletalk occurs, the signal received from the far end will be so
attenuated that the two speakers almost cannot hear each other any more.
A better method of transferring the beginning of a speech signal or speech
signals in case of doubletalk, without affecting the echo cancelling, is
to use an echo canceller either alone or in combination with adjustable
attenuators inserted in the send and/or receive path. This echo canceller
comprises an adaptive filter having an impulse response which is an
estimate of the impulse response of the echo path between the output of
the receive path and the input of the send path. Feeding the output signal
of the receive path to the input of the adaptive filter causes an estimate
of the echo signal to be available at the output of the adaptive filter.
By subtracting the output signal of the adaptive filter from the input
signal of the send path, a difference signal is obtained which comprises
only the near-end generated input signal of the send path in addition to
the remainder of the echo signal.
The desired impulse response of the adaptive filter is calculated by the
adapting means on the basis of the difference signal found. This is only
feasible if the difference signal comprises only the remainder of the echo
signal. If a near-end generated signal is present at the input of the send
path, this near-end generated signal will also be present in the
difference signal. Because the adapting means in the adaptive filter
cannot make any distinction between a remainder of the echo signal and a
near-end generated signal, a proper adaptation of the impulse response of
the adaptive filter will be disturbed. In order to avoid this, the
adapting means will be blocked if the detection means detect a near-end
generated signal.
In the echo canceller known from afore-mentioned Dutch Patent Application,
this detection is performed by detecting means which calculates the ratio
of the signal power of the difference signal to the signal power of the
input signal of the send path and comparing this ratio with a given
threshold. If this ratio exceeds the threshold, a near-end generated
signal is assumed to be present. This is because once the adaptive filter
has been adjusted the ratio of the signal power of the difference signal
(which signal is then equal to the residual signal) to the signal power of
the input signal of the send path will be relatively small (order of
magnitude: 0.01-0.001) when a locally generated signal is absent.
A general problem for echo cancellers is that the adaptive filter is unable
to adapt its own impulse response sufficiently rapidly when there is a
fast change of the impulse response of the echo path. Consequently, the
ratio of the signal power of the difference signal to the signal power of
the input signal of the send path strongly increases and the detection
means could erroneously detect the presence of a near-end generated signal
even when it is not actually present.
SUMMARY OF THE INVENTION
It is an object of the invention to provide an echo canceller of the type
mentioned in the opening paragraph in which the probability of the
detection means erroneously detecting the presence of a near-end generated
signal is considerably reduced compared with an echo canceller according
to the state of the art.
For this purpose, the echo canceller according to the invention is
characterized in that the detection means detect a near-end generated
signal based on whether the ratio of the power of the input signal in the
send path to the power of the output signal of the adaptive filter exceeds
a threshold value.
The invention is based on the understanding that the power of the output
signal of the adaptive filter for specific types of echo paths (for
example, acoustic echo paths) is a good measurement of the power of the
echo signal at the input of the send path, even if the impulse response of
the echo path as such changes relatively fast. This property will be
discussed in further detail hereinbelow.
If the output signal of the receive path may be considered white noise, the
following holds according to a known theorem from the stochastic signal
theory:
##EQU1##
In (1) P.sub.iz is the signal power of the input signal of the send path,
P.sub.uo the signal power of the output signal of the receive path and
h(.THETA.) is the impulse response of the echo path at instant .THETA..
For certain types of echo paths it holds that the impulse response of the
echo path may change considerably, but that the integral according to
formula (1) remains substantially constant.
For example, in telephony by means of laudspeaking telephone sets, the mean
value of the squared impulse response h(.THETA.) of the echo path over a
specific time .delta..THETA.(.delta..THETA.<<.THETA.) at any instant is
mainly determined by the ratio of the instant .THETA. to the reverberation
time of the room in which the telephone set is located. This property is
known from the book entitled "Principles and Applications of Room
Acoustics", vol. 1, L. Cremer & H. Muller, published by Applied Science
Publisher. If the impulse response of the echo path changes, for example,
because a speaker walks up and down, this will hardly affect the
reverberation time of the room and the square value of the impulse
response of the echo path at a given instant will remain substantially
equal. The integral according to formula (1) will, for that matter, be
substantially independent of changes in the echo path.
Simulations and experiments have shown that in addition to white noise it
also holds for speech signals that the signal power of the output signal
of the adaptive filter is a proper measurement for the signal power of the
part of the input signal of the send path coming from the receive path.
A remaining problem is that it is possible that if the adaptive filter has
not yet been properly adjusted and if signal peaks appear in the output
signal of the receive path, these signal peaks appear at the input of the
send path at an earlier instant than at the output of the adaptive filter.
Consequently, the signal power of the output signal of the adaptive filter
is then much smaller than the signal power of the input signal of the send
path. This may lead to an erroneous detection of a near-end generated
signal being present at the input of the receive path.
The number of erroneous detections of the presence of near-end generated
speech signals caused by the occurrence of signal peaks may now be further
reduced in that the echo canceller is characterized in that the detection
means determine the signal power of both said signals on the basis of the
mean value of the instantaneous signal powers of these signals over a
specific period of time.
Since the signal power is equal to the mean value of the instantaneous
signal powers over a specific period of time, the effect of signal peaks
on the signal power is reduced, reducing the chance of an erroneous
detection of a near-end generated signal.
Generally, adaptive filters are arranged as transversal filters.
Transversal filters comprise a series combination of delay elements, the
output signals of the various delay elements being added together by means
of adjustable weight factors. The use of transversal filters in adaptive
filters presents the advantage that undesired spontaneous oscillation will
be absent for all the possible weight factors due to the absence of
feedback paths. In addition, when these filters are used, the adaptation
process leads to an optimum estimate of the impulse response to be
simulated. These properties of transversal filters are described in the
book entitled "Discrete-Time Signal Processing", by A. van der Enden & N.
Verhoeckx, published by Prentice Hall.
Another property of these filters is that their impulse response is of a
limited duration T.sub.f, which duration is equal to the sum of the delays
of the individual delay elements. In most cases, however, the impulse
response of the echo path, on the other hand, will be infinitely long.
Therefore, if the adaptive filter is adjusted, the impulse response of the
adaptive filter and the impulse response of the echo path at instants
situated between 0 and T.sub.f will be substantially equal. For instants
greater than T.sub.f the impulse response of the adaptive filter is equal
to 0, whereas the impulse response of the echo path will then be unequal
to 0. The integral in formula (1) will then always be greater for the
impulse response of the echo path than the integral for the impulse
response of the adaptive filter, because the integral for the impulse
response of the echo path comprises an extra contribution of the impulse
response of the echo path at instants greater than T. Consequently, the
signal power of the output signal of the adaptive filter is always smaller
than the power of the echo signal. This difference in signal powers could
reduce the reliability of the detection of a near-end generated signal.
In order to further enhance the reliability of the detection of the
presence of a near-end generated signal, the echo canceller is
characterized in that the detection means add a correction term to the
signal power valid at a current instant of the adaptive filter output
signal, before the ratio is determined by the detection means.
The addition of a suitable correction term to the signal power of the
adaptive filter output signal, prior to this signal power being compared
with a signal power of the input signal of the send path, enhances the
reliability of this comparison.
Another consequence of the finite duration of the impulse response of the
transversal filter is that the input signal of the send path is gradually
attenuated to zero at the end of a speech signal received from the far
end, whereas the output signal of the transversal filter abruptly becomes
equal to zero at time interval T.sub.f after the speech signal has ended.
Consequently, the signal power of the input signal of the send path is
suddenly much greater than the signal power of the output signal of the
transversal filter, so that it may be erroneously inferred that a near-end
generated signal is available at the input of the send path.
An embodiment of an echo canceller according to the invention which reduces
this effect is characterized in that the correction term is a sum weighted
with a weight factor of the signal powers of the adaptive filter output
signal situated at a number of measuring instants in a time interval
ending at the current instant, reduced by the duration of the impulse
response of the adaptive filter. The weight factor is a monotonously
decreasing exponential function of the difference between the current
instant and the measuring instant.
Due to these measures a correction term is added to the signal power of the
echo canteller output signal, which correction term is representative of
the error in this signal power due to the limited duration of the impulse
response of the transversal filter. This correction term is equal to the
weighted sum of signal powers of the output signal of the transversal
filter at instants t-.tau., t is the current instant and where .tau.>T. As
a result, the corrected signal power also depends on the signal powers at
instants situated before t-T, so that the influence of the limited
duration of the impulse response of the transversal filter is reduced. The
weight factor is equal to the squared estimate of the impulse response of
the echo path at an instant .tau. because the squared impulse response
denotes how much the signal power is attenuated with time.
In telephony, when loudspeaking telephone sets are used the impulse
response of the echo path is determined by the room in which the telephone
set is located. From the afore-mentioned text "Principles and Applications
of Room Acoustics" it is known that the square of the impulse response of
such a room may be estimated by an exponential function of .tau. so that
the correction term contains such a function.
By adding the correction term to the signal power of the output signal of
the transversal filter, the instant at which the corrected signal power of
the output signal of the adaptive filter suddenly becomes zero is
postponed, so that the signal power of the input signal of the send path
has been reduced even further by the time the corrected signal power
becomes equal to 0. Consequently, the chance of erroneously detecting a
nearend generated signal is reduced.
BRIEF DESCRIPTION OF THE DRAWING
The operation of the invention will now be further explained with reference
to the drawing Figures, in which:
FIG. 1 shows a block diagram of an echo canceller to be used in a
loudspeaking telephone set;
FIG. 2 shows a state diagram of an echo canceller shown in FIG. 1,
representing the activity of two speakers;
FIG. 3 shows a flow chart of a program for a programmable processor, for
calculating a number of parameters necessary for adjusting the attenuators
of the echo canceller shown in FIG. 1 to a proper value;
FIG. 4 shows a flow chart of a function for calculating the background
noise power of the received and sent signal to be utilized in a program
according to FIG. 3;
FIG. 5 shows a flow chart of block 38 in the program shown in FIG. 3 for
calculating the value of the adaptive filter echo transfer at a given
instant; and
FIG. 6 shows a flow chart of block 39 in the program as shown in FIG. 2 for
calculating the acoustic attenuation of the echo path.
DETAILED DESCRIPTION OF THE REFERENCED EMBODIMENTS
In the echo canceller in FIG. 1 the input of the receive path is connected
to the input of an analog-to-digital converter 1. The output of the
analog-to-digital converter 1 is an output signal x'(k) which is supplied
to the input of an attenuator 2 having a transfer factor A.sub.1s, the
transfer factor and being adjustable by means of a control signal. The
output of attenuator 2 is an output signal x(k) which is supplied to the
input of a frequency-domain adaptive filter 8, which filter comprises
adapting means, to the input of a time-domain programmable filter 9, and
to the input of a digital-to-analog converter 3. The output of the
digital-to-analog converter 3, which constitutes the output of the receive
path, is connected to a loudspeaker 4.
The output of a microphone 5 is connected to the input of an
analog-to-digital converter 6, which input likewise forms the input of the
send path. The output of the analog-to-digital converter 6 is an output
signal z(k) which is supplied to a first input of a subtractor circuit 7
and also to a first input of another subtractor circuit 10. A second input
of the time-domain subtractor circuit 7 is connected to an output of the
programmable filter 9, which produces an output signal y(k). A second
input of the subtractor circuit 10 is connected to an output of the
adaptive filter 8, which filter comprises impulse response adapting means.
The output of the subtractor circuit 10 is an output signal r.sub.a (k)
which is supplied to a residual signal input of the adaptive filter 8.
The output of the subtractor circuit 7 is an output signal r(k) which is
supplied to the input of an attenuator 11 having a transfer factor
A.sub.mic which is adjustable by a control signal. The output of the
attenuator 11 is an output signal r"(k) which is supplied to the input of
a threshold circuit 12 which can be adjusted by a control signal. The
output of the threshold circuit 12 is an output signal r"(k) which is
supplied to the input of a digital-to-analog converter 13. The output of
the digital-to-analog converter 13 forms the output of the send path.
A first output of a control unit 14 is connected to the control input of
the attenuator 2. A second output of the control unit 14, providing a
control signal to block the adapting means in adaptive flter 8, is
connected to the control input of the attenuator 11. A third output of the
control unit 14 is connected to a control input of the adaptive filter 8,
whereas a fourth output of the control unit 14 is connected to the control
input of the threshold circuit 12. The adaptive filter 8 has an output bus
for transferring parameters to the programmable filter 9 for programming
the impulse response thereof.
In FIG. 1 the input signal of the receive path is applied to the
analog-to-digital converter for further processing of this signal in
digital form. The function of the complete echo canceller may be achieved
by a digital signal processor of the DSP56001 type (Motorola). The output
signal of the analog-to-digital converter 1 is fed via the adjustable
attenuator 2 to the digital-to-analog converter 3 which converts the
signal to an analog signal suitable for the loudspeaker 4.
The microphone 5 in the send path receives the local speech signal together
with an echo of the signal reproduced by the loudspeaker. The output
signal of the microphone 5 is converted to a digital signal z(k) by the
analog-to-digital converter 6. The subtractor circuit 7 subtracts an
estimate of the echo signal y(k) from the signal z(k), so that a signal
r(k) is obtained which substantially solely consists of the local speech
signal. This signal r(k) is fed via the attenuator 11 and the threshold
circuit 12 to the digital-to-analog converter 13 which converts the signal
r"(k) to a signal suitable to feed to the telephone line.
The complete adaptive filter comprises a combination of frequencydomain
adaptive filter 8 and a time-domain programmable filter 9. The impulse
response of the complete adaptive filter is adapted to the impulse
response of the echo path by the adapting means in the frequency-domain
adaptive filter 8. The estimate y(k) of the echo signal, which estimate is
subtracted from the signal z(k), is calculated by the time-domain
programmable filter 9, which for that purprose receives parameters for
adjusting the desired impulse response from the frequency-domain adaptive
filter.
When a local speech signal is present at the input of the send path it
disturbs the adaptation of the impulse response of the adaptive filter 8.
In order to avoid that this disturbed adaptation will lead to an erroneous
estimate of the impulse response of the echo path, the two filters 8 and 9
are used. If the signal power of the residual signal r.sub.a (k) is
greater than the signal power of the residual signal r(k), or if a
near-end generated speech signal is detected, the adaptation of the
adaptive filter 8 will be disturbed and so the detection means in the
filter 8 will block the transfer of parameters therefrom to the
programmable filter. Consequently, the disturbance of the adaptation of
the adaptive filter 8 will no longer have any effect on the estimate y(k)
of the echo signal.
The adaptive filter 8 is realised in the frequency domain because a
timedomain adaptive filter for the necessary length of the impulse
response is of much greater complexity than a frequency-domain adaptive
filter. In addition, a time-domain adaptive filter has the disadvantage
that adapting its impulse response to the impulse response of the echo
path for signals having a strong autocorrelation takes much longer than
for a frequency-domain adaptive filter for which a decorrelation that is
simple to implement is used. Introducing such decorrelation in a
time-domain adaptive filter would involve much greater complexity.
The programmable filter 9 is realised in the time domain because the signal
in such a filter is not subjected t dditional delay in contradistinction
to a frequency-domain adaptive filter in which certain additional delay is
unavoidable. The structure and operation of these filters is further
discussed in the afore-mentioned Dutch Patent Application.
In order to assume that the power of the undesired echo signals, even if
the adaptive filter has not yet been properly adjusted, remains below a
certain level, the sum signal power transfer between the input of the
receive path and the output of the send path is maintained at a constant
value A.sub.tot by means of the attenuators 2 and 11. For calculating the
correct value for the transfer factor of the attenuators, the value of the
echo transfer ERLE of the adaptive filter and the value of the acoustic
attenuation A.sub.xz are calculated on the basis of the mean signal powers
P.sub.x, P.sub.z and P.sub.r over a specific period of time. From the
values of A.sub.tot, ERLE and A.sub.xz the product A.sub.vsw of the
transfer factors A.sub.1s and A.sub.mic can be calculated with the aid of:
##EQU2##
Distribution of the product A.sub.1s and A.sub.mic over the two attenuators
2 and 11 depends on the activity of the speakers, which is measured by
detection means present in the control unit. If only the speaker at the
far end speaks, A.sub.1s is set at 1 and A.sub.mic at A.sub.vsw. If only
the speaker at the near end speaks, A.sub.1s is set at Avs.sub.w and
A.sub.mic at 1. If the two speakers are active, the transfer factor
A.sub.vsw is uniformly distributed over the two attenuators 2 and 11. The
transfer factors A.sub.1s and A.sub.mic are not adapted suddenly but
gradually in order to avoid strong clicking sounds in the loudspeaker
signal.
The threshold circuit 12 is used for blocking a small residual echo,
whereas a local speech signal is to be transferred substantially
completely. For this purpose, the threshold circuit 12 produces an output
signal which is equal to zero if the input signal remains below a
threshold, and an output signal which linearly depends on the input signal
if the input signal exceeds the threshold. The threshold is only present
in cases where only the far-end speaker is active. In other cases the
transfer of the threshold circuit is equal to 1.
The signal powers P.sub.x, P.sub.z and P.sub.r are calculated by averaging
the value of the instantaneous powers of these signals over blocks having
a time duration T. At the beginning of each block the activity of the
speakers is determined. The values of the various transfer factors ERLE,
A.sub.xz and A.sub.vsw, used for adjusting the attenuators 2 and 11, are
determined only every fourth block. This is done because these
calculations require averaging of the various signal powers over a rather
long period of time (4.T) to reduce the effect of signal peaks.
In the state diagram shown in FIG. 2 the states and the transition
conditions have the following connotations:
______________________________________
NUM-
BER INSCRIPTION CONNOTATION
______________________________________
20 A Far-end speaker is active.
21 B Both speakers are active.
22 C Neither speaker is active.
23 D Near-end speaker is active.
24 1*NE & (4* FE)
Near-end speech signal in
previous block.
25 4* NE & (4* FE)
No near-end speech signal in four
preceding blocks.
26 4* FE & 1*NE No far-end speech signal in four
preceding blocks, but near-end
speech signal in previous block.
27 4* NE & 1*FE No near-end speech signal in four
preceding blocks, but far-end
speech signal in previous block.
28 4* FE & 1*NE No far-end speech signal in four
preceding blocks.
29 1*FE & 1* NE Far-end speech signal in previous
block.
30 1*NE & 1*FE Near-end speech signal and far-
end speech signal in previous
block.
31 4* NE & 4* FE Neither near-end speech signal
nor far-end speech signal in four
preceding blocks.
32 1*FE & (4* NE)
Far-end speech signal in previous
block.
33 4* FE & (4* NE)
No far-end speech signal in four
preceding blocks.
34 1*NE & 1* FE Near-end speech signal in
previous block.
35 4* NE & 1* FE No near-end speech signal in four
preceding blocks.
______________________________________
Because the various control signals for controlling the attenuators 2 and
11, the threshold circuit 12 and the adaptive filter 8 are derived from
the speakers' activity, the speakers' activity is translated to four sta
further shortened to FSM, whose state diagram is represented in FIG. 2.
The FSM is in state A if only the far-end speaker is active. The FSM is in
state B if both speakers are active. The FSM is in state C if neither
speaker is active and is in state D if only the near-end speaker is
active. The transitions between the various states are determined by the
two speakers becoming active or inactive. A certain speaker is assumed to
be active if in the previous block a speech signal was present in the
signal originating from that speaker. A speaker is assumed to remain
active until no further speech signal is present any more in four
successive blocks of the signal coming from that speaker. This assumption
is made to avoid the FSM changing state during pauses between words and
sentences.
A far-end speech signal is assumed to be present if the signal power
P.sub.x is greater than a background noise term N.sub.x. A near-end speech
signal according to the innovative idea is assumed to be present if the
following inequality holds:
.delta.[P.sub.z (i)-N.sub.z ]>P.sub.y (i)+.alpha..sup.-2 P.sub.y
(i-2)+.alpha..sup.-3 P.sub.y (i-3)
or equivalently:
[P.sub.z (i)-N.sub.z ]/[P.sub.y (i)+.alpha..sup.-2 P.sub.y
(i-2)+.alpha..sup.-3 P.sub.y (i-3)]>1/.delta.
In this inequality, .delta.or 1/8 is a threshold value, N.sub.z a
background noise term, P.sub.y (i) is the signal power of the signal y in
the current block and P.sub.y (i-1), P.sub.y (i-3)) are the signal powers
of the signal y in the two preceding blocks. The exponential terms
.alpha..sup.-2 and .alpha..sup.-3 together with the signal powers P.sub.y
(i-2) and P.sub.y (i-3) form the correction term according to the
innovative idea for cancelling the finite impulse response of the adaptive
filter which has an impulse response 2T in length.
The Table below represents which actions belong to the four states of the
FSM.
______________________________________
State A.sub.mic A.sub.ls Threshold circuit
______________________________________
A A.sub.vsw 1 switched on
B .sqroot.A.sub.vsw
.sqroot.A.sub.vsw
switched off
C (T < T.sub.h)
previous value
previous value
switched off
C (T > T.sub.h)
.sqroot.A.sub.vsw
.sqroot.A.sub.vsw
switched off
D 1 A.sub.vsw switched off
______________________________________
In state A the attenuator 11 is set at A.sub.vsw and the attenuator 2 at
the transfer factor 1. The threshold circuit 12 is switched on. In state B
the two attenuators 2 and 11 are set at a value .sqroot.A.sub.vsw, while
the threshold circuit 12 is switched off. If the FSM remains in state C
shorter than T.sub.h, the transfer factors of the attenuators are
maintained at a same value because if state C is shorter than T.sub.h the
speaker who was speaking before state C was changed to will probably start
speaking again. If state C takes longer than T.sub.h, the probability for
the near-end speaker or the far-end speaker to become active becomes
equally great and the two attenuators are set at a value A.sub.vsw. In
state D only the near-end speaker is active and A.sub.mic will be set at 1
and A.sub.1s at A.sub.vsw.
In FIG. 3 the numbered instructions have the connotations as will be
described in the Table below.
__________________________________________________________________________
NUMBER
INSCRIPTION CONNOTATION
__________________________________________________________________________
36 START All variables are initialized.
37 NEXT 4 BLOCKS Waiting period until the next
parameter values are to be calculated.
38 N.sub.x : = MAX[NOISE(P.sub.xs), 10.sup.-6 ]
Background noise in signal x is
calculated.
39 N.sub.4 : = NOISE(P.sub.rs)
Background noise in signal r is
calculated.
40 N.sub.z : = MIN(NOISE(P.sub.zs), N.sub.r)
Background noise in signal z is
calculated.
41
##STR1## Mean signal power over four blocks in signal x is
calculated.
42
##STR2## The noise power in signal z is compared with the
sum signal power of signal z.
43
##STR3## The mean signal power of the signal z over four
blocks and corrected for background noise is
calculated.
44
##STR4## The mean signal power of the signal z over four
blocks is calculated.
45
##STR5## The noise power in signal r is compared with the
sum signal power in signal r.
46
##STR6## The mean signal power of the signal r over four
blocks and corrected for background noise is
calculated.
47
##STR7## The mean signal power over four blocks in signal r
is calculated.
48 CALCULATED ERLE The echo attenuation of the adaptive
filter is calculated.
49 CALCULATE A.sub.xz
The attenuation of the acoustic echo
path is determined.
50
##STR8## The overall transfer for the two attenuators 2 and
11 is determined.
__________________________________________________________________________
The programme of FIG. 3 is used for calculating the overall attenuation
A.sub.vsw in the attenuators 2 and 11 necessary for maintaining the
attenuation of the echo signal at a specific level.
At instruction 36 all constants and variables are initialized. At
instruction 37 there is a wait until a time interval corresponding to four
blocks has elapsed. At instruction 38 the background noise power N.sub.x
in the signal x is determined by calling a function of NOISE. The noise
value N.sub.x always gets a minimum value because the value N.sub.x is
also used as a threshold for detecting a far-end speech signal. At
instruction 39 the background noise power N.sub.r in the signal r is
determined and at instruction 40 the background noise power N.sub.z in the
signal z. If the noise power in signal x is large, the value of N.sub.z
will mainly be determined by the noise in the echo signal instead of the
near-end generated background noise signal. N.sub.r is then a better
estimate for the nearend generated background noise signal because the
noise signal originating from the loudspeaker is cancelled by the echo
canceller. At instruction 41 the signal power P.sub.x1 is determined on
the basis of the sum of the signal powers P.sub.x8 determined in the
preceding four blocks. This renders P.sub.x1 equal to 4.times. the mean
value of P.sub.x8. At instruction 42 the noise power in signal z is
compared with a sum signal power which is equal to the signal power of
signal z summed over 4 blocks. If 4 times the noise power exceeds this sum
signal power, the estimate of the noise power is not correct and at
instruction 44 the mean signal power over four blocks in signal z is not
corrected for noise. If the noise power is smaller than the sum signal
power, the noise-corrected signal power Pzl is determined at instruction
43. At instruction 45 the noise power in signal r is compared with a sum
signal power that is equal to the signal power summed over four blocks in
signal r. If 4 times the noise power exceeds this sum signal power, the
estimate of the noise power is not correct and the mean signal power over
four blocks in signal r is not corrected for noise at instruction 47. If
the noise power is smaller than the sum signal power, the signal power
P.sub.r1 corrected for noise is calculated at instruction 46. At
instruction 48 a sub-routine is called which calculates the echo transfer
ERLE of the adaptive filter on the basis of the signal powers P.sub.z1 and
P.sub.r1. At instruction 49 the acoustic attenuation A.sub.xz of the echo
path is calculated. At instruction 50 the necessary residual overall
transfer factor A.sub.vsw of the attenuators 2 and 11 is calculated on the
basis of the variables ERLE and A.sub.xz. Calculating the square root at
instruction 50 is necessary because A.sub.xz and ERLE are signal power
transfer factors, whereas A.sub.vsw is an amplitude transfer factor. Once
instruction 50 has been carried out, instruction 37 is reverted to where
the passage of 4 blocks is waited for.
In FIG. 4 the numbered instructions have the connotations as described in
the Table below.
__________________________________________________________________________
NUMBER
INSCRIPTION CONNOTATION
__________________________________________________________________________
51 MIN: = MIN[P.sub.s (i)]
The minimum value of the signal
power in the preceding four blocks is
calculated.
52 MAX: = MAX[P.sub.s (i)]
The maximum value of the signal
power in the preceding four blocks is
calculated.
53 MIN < N The variable MIN is compared with
the previous value of the noise power.
54 MAX/MIN > .epsilon.
The ratio of the variable MAX to MIN
is compared with a constant .EPSILON..
55 NOISE: = MIN The result of the function of NOISE is
made equal to the variable MIN.
56 NOISE: = N The result of the function of NOISE is
made equal to the previous value of
the noise power.
57 AV: = .phi.*N + (1 - .phi.)*MIN
The variable AV is made equal to the
mean value of the previous noise
power and the variable MIN.
58 NOISE: = MAX(.THETA., MIN(AV, 1.5*N)
The result of the function of NOISE is
made equal to the maximum of the .THETA.
values and an estimate of the previous
noise level.
__________________________________________________________________________
In FIG. 4 first the minimum of the signal powers P.sub.s in the preceding 4
blocks is calculated at instruction 51. At instruction 52 the maximum of
the signal powers P.sub.s in the preceding 4 blocks is calculated. At
instruction 53 the variable MIN is compared with the previous noise value
N. If MIN is smaller than N, the noise level is assumed to have decreased
and the new value of the noise power is made equal to the variable MIN at
instruction 55 after which the function is left. If MIN is not smaller
than N, however, the ratio of MAX to MIN is compared with the constant
.epsilon. at instruction 54. If this ratio exceeds .epsilon., it is
assumed that the difference between MIN and MAX is relatively large, so
that the signals in the preceding four blocks are not stationary and thus
probably comprise speech. At instruction 56 the previous value of the
noise power is then taken as an estimate of the noise power and after this
the function is left. If the ratio of MIN to MAX is smaller than
.epsilon., it is assumed that the signal is stationary and that the
preceding four blocks probably only comprise noise signals. At instruction
57 the mean value is then calculated on the basis of the previous noise
value and the variable MIN. At instruction 58 the new noise value is made
equal to the maximum of .THETA. and the minimum of AV and 1.5* the
previous value of N. The comparison with 1.5*N is made to avoid the value
of the noise level increasing abruptly when strong stationary signals not
being noise suddenly appear. The comparison with .THETA. takes place to
avoid the value of the noise power becoming and remaining 0. After
completion of instruction 58, the function is left.
In FIG. 5 the numbered instructions of the connotations are as described in
the Table below:
__________________________________________________________________________
NUMBER
INSCRIPTION CONNOTATION
__________________________________________________________________________
60 ERLE.sub.-- P: = ERLE
The content of variable ERLE is stored
in variable ERLE.sub.-- p.
61 L: = 0, i: = 1 The counters L and i are set to 0 and 1
respectively.
62 P.sub.xs (i) < P.sub.xsil
The signal power in signal x is
compared with a threshold value.
63 L: = L + 1 The counter L is incremented by 1.
64 i: = i + 1 The counter i is incremented by 1.
65 i > 4 | | |
