ECHO CANCELLER
BACKGROUND OF THE INVENTION
Field of the invention
The present invention generally relates t o an echo canceller for a digital
telephone signal. It is peculiarly useful when a digital telephone terminal is
connected to a "complex" telecommunication network. In the following
description, "complex network" means network comprising at least one digital
network connected t o at least one analog network. Due t o the analog network,
there is a so-called hybrid echo. Hybrid echo is generated by the reflection of
electrical energy by a line device called a hybrid (hence the term hybrid echo).
Most telephone local loops are two-wire circuits while transmission facilities are
four-wire circuits. A hybrid is used as an interface between a two-wire circuit and
a four-wire circuit. Each hybrid produces echoes in both directions, though the far
end echo is usually a greater problem for voice band.
A caller that uses a digital terminal t o call a callee via an analog network can
perceive the hybrid echo of his/her own background noise, during the ringing
phase, i . e. before the callee picks up the handset of his/her telephone terminal.
For instance, the connection goes via a media gateway, an analog trunk, and an
analog public switched telephone network. The network resources are opened
before the callee picks up the handset. In some situations, the analog trunk causes
an echo. So the noise of caller's environment returns back to the caller's
terminal and, depending on the echo level and the signal level amplification
applied in the caller's terminal, the echo of this noise can be disturbing.
Figure 1 is a block diagram showing an exemplary complex network, and
telephone terminals wherein an echo can be audible:
- An Internet protocol telephone terminal T 1 is linked t o a local area or a wide
area network N1.
- A gateway G1 is linking the digital network N1 t o a classical public telephone
network N2 by an analog line AL .
- An analog telephone terminal T2 is linked t o the classical public telephone
network N2 by an analog line AL2.
For instance, the telephone terminal T 1 is used by user A, in a room R, to call a
distant telephone terminal T2. In this room R, or in a neighboring room, other
people are speaking (for example they are working in an open space office). The
brouhaha done by these people is low; user A may not pay attention t o it.
However the brouhaha is picked up by the micro of terminal T . An echo E of this
brouhaha is sent back by the analog network N2 t o the terminal T1, via the
gateway Gl . The echo of the brouhaha is amplified by the terminal T 1 if it is set
up to provide some amplification to any received signal. User A can perceive the
brouhaha in the handset of terminal T 1. In some cases user A can listen to the
speech of other people, in the handset of terminal T1, even if user A can not
distinctly perceive this speech without the terminal T 1. This induces some
disturbances for user A: Even if the echo is low, the caller being focused on the
cal lee's awaited response is disturbed by the echo. This disturbance lasts until
the callee picks up the handset of the terminal T2.
In addition, this echo may generate some problem of confidentiality (for example
in a hospital) because the background conversations are amplified.
For example, the transmitted signal corresponding t o the brouhaha has a level of -
50 dB and the caller's terminal is set to amplify the received signal with +15 dB.
) During the call set up, there is a short period with no ring back tone. The echo
attenuation is -6dB even if a classical echo canceller is inserted on the link
between the caller and the analog network (because the level of the received
signal is not enough t o allow the activation the adaptive filter of an echo
canceller). Then the caller clearly perceives the brouhaha, because the level of
the received signal is -41 dB. As the caller's attention is peculiarly concentrated
on the awaited response, the caller may perceive some confidential information in
this brouhaha.
2) During the ringing phase, the echo attenuation is -6dB (because the level of the
received signal is still not enough to allow the activation the adaptive filter of an
echo canceller). Then the caller clearly perceives the brouhaha between the ring
back tone periods, because the level of the received signal -41 dB; and, during the
ring back tone periods, the caller perceives ring back tones that are downgraded
by the echo of the brouhaha at the level of -41 dB.
3) As soon as the callee answers the call, the caller's attention is now directed t o
the callee's response, so the caller is less sensitive t o the brouhaha. In addition,
the echo canceller is now adjusting its adaptive filter to attenuate the echo. The
echo is now attenuated by -20 dB. So the caller does not perceive the brouhaha
anymore.
Description of the prior art
- The caller may decrease the volume, in particular i f he/she has previously
increased the volume.
- An echo canceller may be inserted on the connection path, either in the
terminal or in the digital network. It works fine when the received signal has a
high level. One may think of stetting up the echo cancellation filter i n order t o
reach a good echo rejection even when there is no speech signal. But it would
lead t o instability and divergence, especially when there is background noise in
the environment of the distant user. This instability would cause echo. For this
reason, in a classical echo canceller, the adaptive filtering is not applied on low
level signals such as noise, to avoid instability. So a classical echo canceller
doesn't reduce the echo of its own background.
Figure 2 is the block diagram of a classical echo canceller. It is placed between a
terminal and a complex network not represented. It comprises:
- An input receiving a signal TXi sent by the terminal.
- An output directly linked to the input receiving the signal TXi, and forwarding, t o
the network, a signal TXo identical to the signal TXi received from the terminal.
- An input receiving a signal RXi supplied by the network and containing an echo
signal E that is a part of the signal TXo sent to the network, which is sent back by
the network.
- An output supplying, t o the terminal, a signal RXo obtained by processing the
signal XRi in order t o attenuate the echo signal E.
- A linear processor, also called adaptive filter, AF1 , having: a control input, a
signal input coupled t o the input receiving the signal TXi, and an output. It aims t o
generate a signal identical to the echo signal E and t o be subtracted from the
received signal RXi for cancelling the echo signal E.
- A subtractor S1 having: a first input linked t o the input receiving the signal RXi; a
second input linked t o the output of the adaptive filter AF1; and an output
supplying a signal RXr that contains a residual part of the echo signal E.
- A non linear processor NLP1 having: a signal input linked to the output of the
subtractor S1 , a control input, a first output linked to the output supplying the
signal RXo.
- A single talk detector STD and a double talk detector DTD for detecting:
-- when the near end user is talking,
-- when both the near end and the far end user are talking,
-- when none is talking.
- A control unit CU having inputs for receiving signals from the single talk detector
STD and the double talk detector DTD, and having outputs for supplying control
signals t o the non linear processor NLP1 and t o the adaptive filter AF1
The transmitted signal TXi is processed through the adaptive filter AF1 i n order t o
produce an echo estimate, when the control unit CU activates the adaptive filter
AF1 . This echo estimate is then subtracted from the received signal RXi. The
difference between the two signals (i. e. the residual signal RXr) is then processed
by the non linear processor NLP1 to remove the residual echo signal as far as
possible, i . e. to attenuate the components that could not be removed by the
linear filter AF1 and the subtractor S1 alone.
In current narrowband VoIP solutions, echo canceller performances must be
compliant with the ITU-T recommendation G.168. This recommendation describes
the characteristics of an echo canceller including the requirements for control
mechanisms. It also describes a number of laboratory tests that should be
performed on an echo canceller t o assess its performance under conditions likely
to be experienced n the network.
G.168 recommendation is applicable t o the design of echo cancellers using digital
techniques, and intended for use in circuits where the delay exceeds the limits
specified by recommendations ITU-T G. 4 and ITU-T G.1 3 1 . In particular, packet
networks based on IP transport protocol (i.e. VoIP) are known to introduce delays
that exceed these limits.
The problem of echo during the ringing phase is not handled by the ITU G.168
recommendation in its present state. So even if an echo cancellation module is
located in a terminal, or i n a media gateway, and is compliant with the current
G.1 8 recommendation, users can perceive an echo during the ringing phase.
Thus, there is a need to provide an improved echo canceller.
SUMMARY OF THE INVENTION
The object of the invention is an echo canceller comprising:
- an adaptive filter for receiving a signal affected by an echo, and for supplying a
filtered signal that is an estimate of the echo;
- a subtractor for subtracting this estimate from the received signal and supplying
a residual signal;
characterized in that it further comprises:
- means for detecting a ring back tone i n said residual signal, comprising:
-- means for comparing the difference between the energy of the received
signal and the energy of a transmitted signal with a first threshold, and then
concluding that a ring back tone is detected when this difference is greater
than this first threshold,
-- and means for comparing the energy of the transmitted signal to a second
threshold, and comparing the cumulated time during which this energy is
greater than the second threshold with a fourth threshold, then assuming
that the ringing phase is elapsed when this cumulated time is greater than
this fourth threshold;
- means for blocking the received signal and replacing i t by a locally generated
ring back tone if a ring back tone is detected in the received signal.
According t o a preferred embodiment, the proposed echo canceller further
comprises:
- a timer t o determine a time period;
- and means for, during said time period, replacing the residual signal by some
synthetic comfort noise when there is no ring back tone.
According to a peculiar embodiment, the means for replacing the residual signal
by some synthetic comfort noise comprise means for replacing the received signal
by some synthetic comfort noise only if the difference between the energy of the
transmitted signal and the energy of the residual signal is greater than the current
value of a threshold.
According t o a peculiar embodiment, the means for means for replacing the
residual signal by some synthetic comfort noise, comprise means for incrementing
the current value of said threshold with a predetermined increment when the
difference between the energy of the transmitted signal and the energy of the
residual signal is greater the sum of the current value of the threshold and said
increment.
Other features and advantages of the present invention will become more
apparent from the following detailed description of embodiments of the present
invention, when taken in conjunction with the accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
In order to illustrate in detail features and advantages of embodiments of the
present invention, the following description will be with reference to the
accompanying drawings. If possible, like or similar reference numerals designate
the same or similar components throughout the figures thereof and description, i n
which:
- Figure 1 (above described) is a block diagram showing an exemplary complex
network, and telephone terminals wherein echo can be audible.
- Figure 2 (above described) is the block diagram of a classical echo canceller.
- Figure 3 is the block diagram of a first embodiment of the echo canceller
according to the invention.
- Figure 4 is the block diagram of a second embodiment of the echo canceller
according to the invention.
- Figure 5 is the block diagram of a part of this second embodiment.
- Figure 6 represents a flow chart of the process run i n this second embodiment of
the echo canceller according t o the invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
Figure 3 is the block diagram of a first embodiment of the echo canceller
according to the invention, which comprises the same components as the classical
echo canceller represented on figure 2, plus a preprocessing module PM1 inserted
between the input receiving the signal RXi from the network, and the first input of
the subtractor S . This echo canceller i s embedded in a gateway for instance.
The preprocessing module PM1 applies a specific non linear processing from the
call set up request until the callee has picked up the phone (i. e. the end of the
ringing phase) or until the caller speaks, which causes the convergence of the
echo canceller. This non linear processing module PM1 cancels the echo during
this period.
In a preferred embodiment, when there is no ring back tone, it replaces the
removed signal by a synthesized comfort noise with the same statistical
characteristics as the suppressed back ground noise.
When there are ring back tones, i t replaces the removed signal by synthesized ring
back tones.
After the callee has picked up the phone, or after the caller has begun t o speak,
the control unit CU de-activates the preprocessing module PM1, so that the
classical echo cancelling process is applied.
Figure 4 is the block diagram of a second embodiment of the echo canceller
according to the invention, which comprises the same components as the classical
echo canceller represented on figure 2, except that the non linear processor NLP1
is replaced by an improved non linear processor NLP2 integrating a preprocessing
module PM2. A block diagram of this improved non linear processor NLP2 is
represented on figure 5. It comprises:
- A classical non linear processor CNLP similar to the non linear processor NLP1 . It
receives a signal supplied by the enhanced non linear processor EN, and i t supplies
the processed received signal RXo to the near end terminal.
- The preprocessing module PM2 comprising a signal qualifier SQ for detecting a
ring back tone, and an enhanced non linear processor EN.
The signal qualifier SQ receives the residual signal RXr provided by the output of
the subtractor S2. It also receives a so called "Adaptation Signal" AD that is a
control signal supplied by the control unit UC2 and that can take three values:
■ 0 when control unit CU2 detects that the echo canceller EC3 cannot be i n
an "adaptation phase" i . e. the adaptive filter AF2 cannot be adapted
because the received signal RXi signal is too low to enable a convergence
to a stable filtering.
■ 1 when control unit CU2 detects that the echo canceller EC3 is i n its
"adaptation phase".
" 2 when the control unit CU2 detects that the adaptive filter AF2 has just
finished its "adaption phase", i . e. it is the beginning of a phone
conversation.
The signal qualifier SQ provides the enhanced non linear processor EN with a
binary signal SN indicating whether the residual signal RXr that i t receives
contains, or not, a ring back tone.
This improved non linear processor NLP2 fulfills the same functions as those
described above for the combination of the preprocessing module PM1 and the non
linear processor NLP1 .
The preprocessing module PM1 uses two kinds of threshold:
o An energy threshold: Eth
o A convergence status and time threshold: CTth.
If the difference of the energies of the signals respectively present on
the inputs TXi and RXi is greater than the threshold Eth, during the ringing phase,
- and i f it is during an interval wherein there is no ring back tone, then
the preprocessing module PM1 completely cancels the received signal and
replaces i t by a locally generated comfort noise;
- and i f i t is during a ring back tone, then the preprocessing module PM1
completely cancels the received signal and replaces i t by a locally generated ring
back tone.
The threshold CTth is used by the signal qualifier S t o detect the ringing
phase. The energy of the transmitted signal on the input TXi is measured and is
compared t o a threshold MinEnergy. If the cumulated time, during which this
energy i s greater than the threshold MinEnergy is higher than the threshold CTth,
then the ringing phase is assumed to be elapsed. Then the preprocessing module
PM1 is disabled to avoid clipping the received signal RXi during speech
transmission. So, as soon as the callee has picked up the handset of terminal T2,
the preprocessing module PM1 stops the specific non linear processing.
Figure 6 represents a flow chart of the process run i n the enhanced non linear
processor EN represented on figure 5. A similar process is run in the preprocessing
module PM1 of figure 3:
Step 1: Initialization of the processing:
■ resetting the value of a timer Tadapt that will be used for stopping the
filter adaptation process when the adaptation phase is supposed t o be
over, i . e. some time after the phone conversation has started, in order to
let some time after the end of the ringing phase, for the convergence of
the adaptive filter;
■ and resetting the value of a variable threshold Tth that will be used for
comparing the energy Etx of the transmitted signal TXo and the energy Erx
of the received signal RXi. The initialization value of Tth is 6 dB in this
example.
Step 2: Starting the processing for a new received signal frame.
Step 3: Checking the value of the "Adaptation signal" AD. If the value is 0, it
means that the echo canceller is not in the "adaptation phase"; the processing
continues with step 7. If the value is not 0, it means that the echo canceller is in
the "adaptation phase" (if value i s 1) or has recently finished this adaption phase
(if value is 2); the processing continues with step 4.
Step 4: Discriminating whether the value of the signal "Adaptation" is 1 or 2. If
the value is 1, it means that the echo canceller is in the "adaptation phase"; the
processing continues with step 5. If the value is 2, it means that the echo
canceller has recently finished the "adaptation phase"; the processing continues
with step 6.
Step 5: The timer Tadapt is incremented with a frame period
(Tframe = 10 ms in this example). The processing continues with step 7.
Step 6: The timer Tadapt is set t o the maximal value Tth ( 1 second in this
example). This value Tth is used as a threshold value t o stop the adaptation
process in a later step. So the processing will be stopped by this later step when
the timer will have been incremented 100 times while processing 100 signal
frames. The processing continues with step 7.
Step 7: Detecting a ring back tone in the residual signal RXr received by the
enhanced non linear processor EN.
Step 8 : Checking the signal SN supplied by the signal qualifier S to the enhanced
non linear processor EN. If the signal SN indicates that a ring back tone is
detected in the received signal RXr, then the processing continues with step 9.
Otherwise it continues with step .
Step 9: The received signal RXr is replaced by locally generated ring back tones so
that the user will clearly hear ring back tones without any echo.
Step 10: Determining the respective energies Etx and Erx of the transmitted signal
TXo and of the residual signal RXr.
Step 11: Comparing the timer value Tadapt with the threshold
Tth = 1 second. If Tadapt is not lower than Tth, the processing, in preprocessing
module PM2, comes t o an end with step 12. If Tadapt is lower than Tth, the
processing continues with step 13.
Step 12: The preprocessing module PM2 is bypassed, the echo canceller works
now in the classical way. The output signal RXo is the received signal RXi
classically filtered by the adaptive filter AF2 and the classical non linear processor
CNLP. The processing continues with the same step 1 for the next frame.
Step 13: The difference between the energy Etx of the transmitted signal TXo and
the energy Erx of the residual signal RXr is compared with the current value of the
threshold Eth. If this difference i s lower than, or equal to, Eth, the processing is
ending with step 1 because the level of the residual signal RXr is high enough so
that an echo cannot be disturbing anymore. Otherwise the processing continues
with step 15 because the residual signal RXr is much lower than the transmitted
signal, so one can suppose that it is mostly constituted by an echo.
Step 14: The preprocessing module PM2 is bypassed, the echo canceller works
now i n the classical way. The output signal RXo is the received signal RXi
classically filtered by the adaptive filter AF2 and the classical non linear processor
CNLP. The processing continues with step 2 for the next frame.
Step 15: The difference between the energy Etx of the transmitted signal TXo and
the energy Erx of the residual signal RXr is compared with the sum of the current
value of the threshold Eth and an increment Ethmarg. If this difference is lower
than, or equal to, Eth, the processing is ending with step 17. Otherwise the
processing continues with step 16.
Step 16: The current value of the threshold Eth is incremented with an increment
Ethmarg that is 1 dB in this example. Then the processing is ending with step 17.
Step 17: The residual signal RXr is blocked. The restituted signal RXo supplied to
the near end terminal is a synthesized comfort noise replacing the residual signal
RXr. Then the processing continues with step 2 for the next frame.
The echo canceller according to the invention is peculiarly advantageous for
telephone terminals that are used for full duplex hands-free modes. It may be
embedded in:
- Laptop/desktop PC-based soft phones,
- hardware IP phones, Smart phones, PDA phones,
- VoIP Media gateways,
- IP PBX.
THERE IS CLAIMED:
1) An echo canceller comprising:
- an adaptive filter for receiving a signal affected by an echo, and for supplying a
filtered signal that is an estimate of the echo;
- a subtracter for subtracting this estimate from the received signal and supplying
a residual signal;
- means for detecting a ring back tone in said residual signal, comprising:
-- means for comparing the difference between the energy of the
received signal and the energy of a transmitted signal with a first
threshold, and then concluding that a ring back tone is detected
when this difference is greater than this first threshold,
-- and means for comparing the energy of the transmitted signal t o
a second threshold, and comparing the cumulated time during
which this energy is greater than the second threshold with a fourth
threshold, then assuming that the ringing phase is elapsed when this
cumulated time is greater than this fourth threshold;
- means for blocking the received signal and replacing it by a locally generated
ring back tone when a ring back tone has been detected in the received signal,
and until the ringing phase is assumed t o be elapsed.
2) An echo canceller according to claim 1, further comprising:
- a timer to determine a time period;
- and means for, during said time period, replacing the residual signal by some
synthetic comfort noise when there is no ring back tone.
3) An echo canceller according t o claim 1, wherein said means for replacing the
residual signal by some synthetic comfort noise, comprise means for replacing the
received signal by some synthetic comfort noise only if the difference between
the energy of the transmitted signal and the energy of the residual signal is
greater than the current value of a threshold.
4) An echo canceller according to claim 1, wherein said means for replacing the
residual signal by some synthetic comfort noise, comprise means for incrementing
the current value of said threshold with a predetermined increment when the
difference between the energy of the transmitted signal and the energy of the
residual signal is greater than the sum of the current value of the threshold and
said increment.