KR20000057739A - Subband echo canceller and method therefor - Google Patents

Abstract

PURPOSE: A subband echo canceler and a method therefor are provided to decrease a complexity of an echo eraser by adapting subbands having a low echo residual energy during a short efficient sampling instant. CONSTITUTION: An analyzing filter(220) splits a full-band far-end signal on a bus(101) into the M number of adjacent subband far-end signals of output terminals(221-223). An analyzing filter(242) splits a full-band near-end signal on a bus(103) into the M number of adjacent subband near-end signals of output terminals(210-212). A synthesizing filter(230) combines the M number of subband echo-removed signals of output terminals(260-262) as a full-band echo-removed signal on a bus(105). The M number of adaptive filters(201-203) remove an echo signal from the M number of adjacent subband near-end signals of output terminals(210-212). An adaptive speed controller(240) outputs the M number of adaptive speed control signals of output terminals(241-243) to the M number of adaptive filters(201-203).

Description

Subband echo canceller and its method {SUBBAND ECHO CANCELLER AND METHOD THEREFOR}

The present invention relates generally to two-way communication systems, and more particularly to subband echo cancellers thereof.

The need for echo cancellation is emerging in many full duplex communication systems. One particular situation in which there is a need for reliable echo cancellation is full-duplex hands-free operation of cellular radiotelephone devices and teleconferencing devices. During hands-free operation of these devices, signals from the speaker are fed back to the microphone through various acoustic paths and delayed before reaching the original speaker. These feedback signals are recognized by the far-end user as echo signals. An echo signal, often called an acoustic echo, is very inconvenient and difficult to remove for participants participating in two-way communication.

One of the most effective solutions for removing echo signals is to use an echo canceller with an adaptive filter. Least Mean Squared (LMS) adaptive filters are the most common form of filters used. LMS filters are finite impulse response (FIR) filters that model echo paths through adaptive adjustment coefficients. The coefficients of the filter are adaptively trained using the far-end signal that drives the loudspeaker of the hands-free communication device and the near-end signal output from the microphone of the hands-free communication device. In a hands-free device, the adaptive filter adaptively synthesizes a replica of the acoustic echo from the far-end signal, which is subtracted from the near-end signal output by the microphone at the near-end. The result is a substantial echo-free signal transmitted further to the far end.

Subband echo cancellation circuits have been used in environments with long echo paths. In this circuit, the far-end signal is segmented into a set of signals, each representing a portion of the far-end signal in one subband. The near-end signal is also divided in the same way as the far-end signal. There are various ways to divide the far-end signal into subbands. The basic method is to use a filter bank to split the entire band of the transmitted signal into adjacent segments. The output of each filter is the signal from its subband. Thus, for each subband, a pair of signals corresponding to the far-end and near-end signals are obtained.

The subband echo canceller includes an echo canceller for each subband. Each echo canceller is equipped with a respective adaptive filter. Echo cancellation is performed in the time-domain or the frequency-domain. The echo-cleared signal generated by each subband echo canceller is combined into a full-band near-end signal using a synthesis filter. Adaptive subband filtering provides an efficient way to realize both performance enhancement and complexity reduction.

Stephen L. U.S. Patent No. 5,001,701, entitled "SUBBAND ECHO CANCELLER INCLUDING REAL-TIME ALLOCATION AMONG THE SUBBANDS," to Steven L. Gay, proposed a method for improving the performance of such a subband echo canceller. In this patent, it is assumed that the total computational resources available for echo cancellation are not fixed. Echo cancellers associated with subbands with larger adaptive filter misalignment are more per second than echo cancellers in subbands with smaller adaptive filter misalignment to realize the fastest convergence of the entire subband echo canceller. Iteration is provided. The misalignment of the adaptive filter is a weighted norm of the difference between the current and past filter coefficients.

Although this subband echo cancellation technique reduces the computational complexity of the acoustic echo canceller, in some applications such computational complexity is still very high. Further reduction in the computational complexity of the subband acoustic echo canceller is necessary to meet the requirements of many applications. In addition, it is desirable to reduce the computational complexity as much as possible while maintaining the overall performance recognized by the user.

The present invention is directed to determining echo residual energy in each of a plurality of subbands, assigning each adaptation rate to an adaptive filter associated with each subband as a function of echo residual energy in each subband, and And updating the coefficients of the adaptive filter at the respective adaptive rates selected in the subbands.

The present invention also includes a plurality of adaptive filters associated with a plurality of subbands, and an adaptive speed controller coupled to the plurality of adaptive filters, wherein the adaptive speed controller measures echo residual energy in each subband, A subband echo canceller characterized by assigning a normal adaptation rate to subbands with higher echo residual energy and a slow adaptation rate to subbands with lower echo residual energy.

1 is a circuit schematic diagram illustrating a full-duplex communication apparatus using a subband echo canceller.

2 is a circuit schematic showing a subband echo canceller for full-duplex communication.

3 is a schematic diagram illustrating an adaptive speed control apparatus for a subband echo canceller.

4 is a circuit schematic showing a residual energy calculator in an adaptive speed control device.

5 is a circuit schematic diagram illustrating an adaptive filter for one subband.

6 is a schematic diagram showing an adaptive filter control device.

<Description of the symbols for the main parts of the drawings.

101: A / D converter

102: speaker

103: output stage

104: microphone

105: output stage

106, 112: A / D converter

108, 110: D / A converter

114: Transceiver

116: antenna

By using adaptive filtering in the individual subbands, the subband echo cancellation technique provides an opportunity to use the computational resources of the device more efficiently. An efficient adaptation method reduces as much computational complexity as possible while maintaining the performance perceived by the entire user. This method identifies subbands with high echo residual energy. For subbands with high echo residual energy, adaptability occurs at normal speed. The other subbands adapt at lower speeds. This adaptability provides echo cancellation across all subbands, but performs more efficiently than conventional echo cancellers. Those skilled in the art will appreciate that in the described embodiments, time-domain adaptive filtering is used because of its structural simplicity and conceptual clarity, but other forms of adaptive filtering may be used.

Since a person recognizes echo disturbances from a high energy subband signal, computational complexity is reduced based on the rules that the user recognizes. It has been found that some subbands will produce an echo residual signal with dominant energy. Adaptive filter coefficients in other subbands with lower echo residual energy adapt at lower speeds, while adaptive filter coefficients in other subbands with higher echo residual energy adapt at normal speeds. Since only some subbands have high echo residual energy, most subbands have low echo residual energy to adapt at slower speeds.

The echo canceller is very efficient as the coefficient adaptation of the adaptive filter is normal speed only in some subbands, which greatly reduces the computational complexity of the entire subband echo canceller. The present invention is advantageous for the efficiency obtained by using different adaptation rates for adaptive filters in other subbands to uniformly suppress echoes across the subbands. The most efficient way to perform echo cancellation is to suppress the echo uniformly in all subbands.

Subband acoustic echo canceller 120 (FIG. 1) is used in bidirectional communication device 100 to improve hands-free operation. Full-duplex communication device 100 is shown as a hands-free device, a hands-free radiotelephone, hands-free teleconference device, hands-free satellite phone, hands-free radiotelephone, personal computer (PC) multimedia communication Device, or any other stable communication device. The communication device 100 includes a transceiver 114 that receives a remote signal (far-end signal) from a remote location and transmits a local signal (near end signal) to the remote site.

The transceiver may be any transceiver that is stable for cable, optical, wireless, wired or satellite communications, the operation of which is well known to those skilled in the art and no further details are described herein for brevity. In the described embodiment, the transceiver 114 is coupled to the antenna 116 for wireless communication in a cellular system. The transceiver 114 transmits the echo-erased near-end signal output by the D / A converter 110 to the far end communication device via the antenna 116 and receives the signal received from the remote device detected by the antenna 116. Input to A / D converter 112.

The receive path includes an analog-to-digital (A / D) converter 112 and a digital-to-analog (D / A) converter 108 connected to the loudspeaker 102. The transceiver 114 output is converted into a digital signal at the A / D converter 112 for the subband echo canceller 120. The digital-to-analog (D / A) converter 108 converts the digitized far end signal into an analog format that drives the local loudspeaker 102.

The transmission path includes a microphone 104 that converts the local acoustic signal into an electrical signal for transmission to a remote location, and the A / D converter 106 supplies the microphone electrical signal to the subband acoustic echo canceller 120. Convert to a digital signal. The echo-erased signal output from the subband acoustic echo canceller 120 is converted to an analog format by the D / A converter 110 for transmission via the transceiver 104.

Those skilled in the art will appreciate that A / D converter 112 and D / A converter 110 are used in analog systems. Alternatively, in a digital implementation the A / D converter 112 is replaced by a speech decoder and the D / A converter 110 is replaced by a speech encoder. For example, transducers 112 and 110 may be speech decoders and encoders of digital interfaces in a communications device for Global System for Mobile Communications (GSM) or Integrated Services Digital Network (ISDN). It should also be appreciated that the combination of antenna 116 and transceiver 114 may be replaced by a network interface device in some applications, such as for wired or optical communication systems.

In operation, the far-end voice signal, x (n), in digital form at the output 101 of the A / D converter 112, is to the D / A converter 108 which generates an analog signal to drive the loudspeaker 102. Is entered. A portion of the far-end signal output by the speaker 102 is detected by the microphone 104 as a near-end audible signal and converted into a digital format by the D / A converter 106. The final near-end signal y (n) at output 103 is the input to subband acoustic echo canceller 120. The digital-type echo-cleared speech signal r (n) is output to the output terminal 105 of the subband acoustic echo canceller 120 and input to the D / A converter 110.

Subband acoustic echo canceller 120 may be implemented in a digital signal processor (DSP), a microprocessor, a programmable logic device, or the like. Subband acoustic echo canceller 120 transmits the full-band far-end signal x (n) on bus 101 to the M adjacent subband far-end signals x ₁ (n) to x _M (n) at output stages 221-223. Analytical filter 220 (FIG. 2), which decomposes the full-band near-end signal y (n) on bus 103 into M adjacent subband near-end signals y ₁ (n) to y _M of outputs 210-212. an analysis filter 242 that decomposes into (n), and the M subband echo-erased signals r ₁ (n) to r _M (n) of the output stages 260-262 are full-band echoes on the bus 105. A synthesis filter 230 which combines the erased signal r (n). M adaptive filters 201-203 are used to cancel echo signals at the M adjacent subband near-end signals y ₁ (n) to y _M (n) of outputs 210-212, and adaptive speed controller 240 is used. Outputs M adaptive speed control signals s ₁ (n) to s _M (n) of the output terminals 241-243 to the M adaptive filters 201 to 203. Note that the time index n for the subband signal may or may not be the same as the time index n for the full band signal. For example, it should be noted that the adaptation occurs more often in the subband echo canceller because the time index (sampling interval) for the subband signal may be a smaller interval than the time index for the fullband signal.

Ideally, analysis filters 220 and 242 output M subband signals where each subband signal does not overlap with an adjacent subband in the frequency-domain. However, since absolute isolation between adjacent subbands is very difficult, some small overlap is allowed. This will not affect the performance of the acoustic echo canceller as long as the overlap is small.

In addition, the operation of synthesis filter 230, which is completely opposite to the operation of analysis filters 220 and 242, will be difficult or very expensive to implement in practice. After all, ideally the output signal of the synthesis filter may be a delayed version of the signal input to the synthesis filter, but in practice a small amount of magnitude and phase distortion is allowed to simplify the design and reduce computational complexity.

There are various ways of implementing the analysis filter 220, 242 and the synthesis filter 230. Such implementations are well known to those skilled in the art and are not described herein any further for the sake of brevity.

The adaptive filter 201 (FIG. 2) is used to cancel the echo signal of the near-end signal y ₁ (n) of the output terminal 210 in subband 1, and the adaptive filter 202 is the near end of the output terminal 211 in subband 2. An adaptive filter 203 is used to cancel an echo signal of the signal y ₂ (n), and an adaptive filter 203 is used to cancel an echo signal of the near end signal y _M (n) of the output terminal 212 in M subbands. Adaptive filters 201 to 203 may be, for example, LMS or RLS algorithms. The analysis filters 220, 242, the synthesis filter 230, and the adaptive filters 201-203 may be implemented in the time-domain or the frequency-domain.

This operation is described for a time-domain system using the LMS algorithm for each adaptive filter 201-203, and the number M of subbands is 32 (three adaptive filters 201-203 are shown in FIG. 2). Although shown, those skilled in the art will appreciate that an additional 29 adaptive filters are provided for the other 29 subbands in an embodiment with 32 subbands). LMS adaptive filters are widely used because of their simplicity and powerful performance. The M adaptive filters may be different or the same depending on the implementation. By using the same structure for all M adaptive filters, the system may be advantageous for one circuit implementation for program reuse of DSP processors or for M adaptive filter implementations using different hardware. Each adaptive filter 201-203 includes a finite impulse response (FIR) filter having an adjustable coefficient for generating an echo estimate, a subtractor for generating an echo-erased signal, and a control device for updating the coefficient.

The described embodiment may be any other number of subbands and / or groups, but assumes M (number of subbands) is 32 and N (number of subband groups) is eight. Adaptive speed control device 240 includes 32 subband echo residual signals {r _i (n) on ports 260, 261, 302-306, 262; 32 echo residual energies {e _i (n) of the output stages 320 to 327 for i = 1, 2, ..., 32}, 32 to find i = 1, 2, ..., 32} Energy calculators 310-317 (FIG. 3) (only eight are shown). Eight adders 330-331 (two are shown) are each used to sum some M / N subband echo residual energy. The maximum group selector 340 selects the maximum value of eight group energies {C _i (n): i = 1, 2, ..., 8} at the ports 332-333. The output speed control generator 350 outputs the 32 adaptive speed control signals s ₁ (n), s ₂ (n), and s _M (n) of the output stages 241-243 to the 32 adaptive filters 201 -203. Generation (Fig. 2).

In the described embodiment, the 32 subband far-end signals {x _i (n): i = 1, 2, ..., 32} (FIG. 2) generated in the analysis filter 220 are output stages 221-223. Is output from This output of analysis filter 220 is input to adaptive filters 201-203. Near-end signals {y _i (n): i = 1, 2, ..., 32} of 32 subbands at the outputs 210-212 of the analysis filter 242 are also 32 adaptive filters 201 -203. ) Is entered. The thirty-two subband adaptive filters 201-203 comprise thirty-two subband echo- canceled signals {r _i (n) at the output stages 260-262; i = 1, 2, ..., 32}. This subband echo canceled signal is input to a synthesis filter 230 which outputs a full-band echo- canceled signal r (n) at the output stage 105.

Subband acoustic echo canceller 120 utilizes adaptive speed controller 240. The rate at which the adaptive filters 201-203 adapt is set by this adaptive speed controller 240. The 32 adaptive speed control signals {s _i (n): i = 1, 2, ..., 32} at the output stages 241-243 of the adaptive speed controller 240 are 32 adaptive filters 201 -203. Is entered. The signals x _i (n), y _i (n), r _i (n), and s _i (n) are used for one subband i.

The adaptive speed controller 240 assigns an adaptive speed to each of the M adaptive filters 201 to 203. Adaptive speed controller 240 includes energy calculators 310-317. Energy calculators 310-317 are connected to receive echo residual signals at ports 262, 302-306, 261, 260. Each energy calculator includes a square or absolute value generator circuit 410 (FIG. 4) to generate a positive value. The output of circuit 410 is filtered at low pass filter 412 to generate an average value. The 32 energy calculators 310 to 317 calculate the echo residual energy {e _i (n): i = 1, 2, ..., 32} at the ports 320 to 327 according to the following equation (1).

e _i (n) = (1-g) e _i (n) + gr _i (n) r _i (n) i = 1, 2, ..., 32

Where g is a scalar value between 0 and 1, for example may be selected from 0.01. The 32 subband echo residual energies {e _i (n): i = 1, 2, ..., 32} at the outputs 320 to 327 are divided into eight groups of four adjacent subbands as follows. . In other words,

Group 1: {e1 (n), e2 (n), e3 (n), e4 (n)}

Group 2: {e5 (n), e6 (n), e7 (n), e8 (n)}

Group 3: {e9 (n), e10 (n), e11 (n), e12 (n)}

Group 4: {e13 (n), e14 (n), e15 (n), e16 (n)}

Group 5: {e17 (n), e18 (n), e19 (n), e20 (n)}

Group 6: {e21 (n), e22 (n), e23 (n), e24 (n)}

Group 7: {e25 (n), e26 (n), e27 (n), e28 (n)}

Group 8: {e29 (n), e30 (n), e31 (n), e32 (n)}

For each group, four adjacent subband echo residual energies are added together to generate one group energy. Thus, eight adders 330 to 331 generate eight energies {C _i (n): i = 1, 2, ..., 8} at output stages 332 to 333. In other words,

C _i (n) = e _4i-3 (n) + e _4i-2 (n) + e _4i-1 (n) + e _4i (n) i = 1, 2, ..., 8

The maximum group energy signal Cmax (n) at the output stage 341 is the maximum selector 340 in response to the input {C _i (n): i = 1, 2, ..., 8} from the output stages 332 and 333. Is generated by Cmax (n) = max {C ₁ (n), C ₂ (n), ..., C ₈ (n)}.

The output speed control generator 350 outputs 32 adaptive speed control signals {s _i (n): i = 1, 2, ... in response to the output 341 of the maximum selector 340. , 32}. This adaptive speed control signal is input to the M adaptive filters 201 to 203 (Fig. 2). In operation, if C _i (n) from group i has the maximum group energy, then the adaptive speed control signal at output stages 241 to 243 is generated as follows.

s _4i-3 (n) = s _4i-2 (n) = s _4i-1 (n) = s _4i (n) = 1 (if adaptive filter associated with group i with maximum value)

sj (n) = K and j ≠ 4i, 4i-1, 4i-2, 4i-3 and K> 1 (if adaptive filter associated with other groups)

Where K is the number of samples in which one coefficient adaptation occurs. For example, if K = 2, the adaptation rate of the adaptive filter other than the selected group is 1/2 rate (update once every two samples), or if K = 4, the adaptation rate of the adaptive filter other than the selected group Is 1/4 rate (update once every 4 samples).

Thus, the output speed control generator 350 outputs a separate control signal {s _i (n) at the output stages 241 to 243 for each adaptive filter 201 to 203 indicating whether coefficient adaptation is performed at a particular sampling instant. generates i = 1, 2, ..., 32}. The normal adaptation rate at which the adaptation occurs at every valid sampling instant is provided to a group of subbands, or groups, with a high residual echo group energy, while the low adaptation rate at which the adaptation is skipped for some sampling intervals is a low group residual. Assigned to a group of subbands with echo energy. By delegating the allocated adaptive energy to the subband with the maximum echo residual energy, the subband echo canceller 120 can use less processing resources without significantly degrading the performance of the echo canceller.

The adaptive speed controller 240 may specify the control signals {s _i (n): i = 1, 2, ..., 32} at the output stages 241 to 243 in various alternative ways. Another way of specifying the control signal may use individual subband energy and select a band having a threshold level of residual echo energy above. Alternatively, the band with the largest residual energy can be selected regardless of the energy value in the adjacent band. The advantage of the group of selected subbands is that the processing is simplified and it is advantageous for adjacent subbands to have maximum residual echo energy.

5, an adaptive filter 201 is shown. Adaptive filter 201 includes FIR filter 500 and adaptive filter control device 502. Adaptive filter control device 502 generates an adaptive coefficient W ₁ (n) at output stage 504. FIR filter 500 generates a subband echo estimate z ₁ (n) at output 508 using coefficient W ₁ (n) at output 504. At the current sampling instant n, subband far-end speech sample x ₁ (n) is received at port 221 from analysis filter 220 and subband near-end speech sample y ₁ (n) is output of analysis filter 242. As received from port 210. Subband signals x ₁ (n) and y ₁ (n) at outputs 221 and 210 are synchronized as A / D converter 106 (FIG. 1) and D / A converter 108 use the same clock. do.

The echo estimate z ₁ (n) at the output stage 508 is synthesized by the FIR filter 500 based on Equation 4 below.

Superscript T means the binomial of the vector or matrix, L is the order of the FIR filter 500, and X ₁ (n) = [x ₁ (n) x ₁ (n-1) ... x ₁ (n -L + 1)] ^T is the L maximum current subband far-end speech sample for the subband.

Adaptive filter 201 (FIG. 5) subtracts the echo estimate z ₁ (n) at output stage 508 from the near-end signal y ₁ (n) at input stage 210 in the bilateral conversation state at output stage 260. A subtractor 506 for generating an echo- canceled signal r ₁ (n) (or an echo residual signal in a monologue state) of

r ₁ (n) = y ₁ (n)-z ₁ (n)

to be. Here, the monologue state is defined as no local speech and the near-end signal includes only the echo signal, while the quantum conversation state is defined as the presence of the near-end speech in addition to the echo signal. Coefficient adaptation should stop in quantum dialogue. Quantum conversation detectors can be used to distinguish two situations, as is well known in the art. The monologue state assumes that r1 (n) is a subband echo residual signal for subband 1 herein.

The adaptive filter 201 further includes a control device 502 that outputs the coefficient W ₁ (n) at the output 504 of the FIR filter 500. Coefficient W ₁ (n) at output 504 is subband far-end speech signal x ₁ (n) at port 221, near-end speech signal y ₁ (n) at port 210, and port 241. Is generated based on the adaptive speed control signal s1 (n) in.

Referring to Fig. 6, the operation of the adaptive filter 201 will now be described (the description of the adaptive filter 201 applies to each of the other 31 adaptive filters). From the subband far-end signal x ₁ (n) input at 221, the vector X ₁ (n), which is the L maximum current subband far-end signal sample, is block 602, that is, X ₁ (n) = [x1 (n), x1 (n−). 1),... X1 (n-L + 1)] ^T , is stored in a buffer (not shown) in the adaptive filter 201.

The energy estimate E ₁ for the subband far-end signal energy is an adaptive filter (shown in FIG. 5), as indicated in block 604, ie, E ₁ = (1-a) E ₁ + aX ₁ (n) x ₁ (n). Generated at controller 502. Where a is a scalar between 0 and 1, for example a = 0.01.

The adaptive filter 201 sets the adaptation speed in response to the adaptation control signal s ₁ (n) at the input terminal 241. The adaptive filter controller 502 chooses to adapt during the particular sampling instant or use the previous coefficients, as indicated in decision block 606. At the instant of sampling, the coefficient W ₁ (n) updated at [n mod s ₁ (n)] = 1 is equal to input r ₁ (n) at input 260, X ₁ (n) from block 602, and block 604. From the adaptive filter controller 502 as indicated in block 608 based on E ₁ (n) from. In other words,

W ₁ (n) = W ₁ (n-1) + br ₁ (n) X ₁ (n) E ₁ ^-1

to be. Where b is the step size and W _i (n) = [W ¹ ₀ (n) W ¹ ₁ (n) ... W ¹ _L-1 (n)] ^T. In other cases, the coefficient W ₁ (n) does not change when [n mod s ₁ (n)] = 0, as indicated at block 610. In both cases, the coefficient W1 (n) is output at 504 for use by the FIR filter 500.

Thus, for an adaptive filter with normal adaptation rate, s _i (n) = 1 will adapt every sampling instant. For an adaptive filter with a slower adaptation rate, for example a quarter rate, s _i (n) is 4 and will only adapt once every four sample moments.

Accordingly, the adaptive control signals {s _i (n): i = 1, 2, ..., 32} at the output stages 241 to 243 for each of the adaptive filters 201-203 are generated at each sampling instant and respectively. Enable or disable adaptation for &lt; RTI ID = 0.0 &gt; Subbands with high echo residual signal energy will adapt each sampling instant, while subbands with low echo residual energy will adapt only during a portion of the effective sampling instant. By adapting for shorter sampling instants, the complexity of the echo canceller can be greatly reduced without significantly degrading the performance of the subband echo canceller, as reflected by the number of device instructions per second needed to perform echo cancellation. . The computational complexity is reduced while maintaining the original performance which is very desirable for practical applications with limited processing capacity. In addition, using echo residual energy as a measure to determine that the subband receives the maximum adaptive resource has an effect on reducing the complexity of the echo canceller while maintaining user-recognized performance.

Claims (10)

A method of operating an adaptive filter of a subband echo canceller, Determining echo residual energy in each of the plurality of subbands, Assigning each adaptation rate to an adaptive filter associated with each subband as a function of echo residual energy in each subband, and Updating coefficients of an adaptive filter at each respective adaptation rate selected for the subbands Method comprising a. 2. The method of claim 1, wherein specifying an adaptation rate for the M adaptive filters comprises assigning an adaptation rate to the M adaptive filters based on the M energies of the M subband echo residual signals. How to. 3. The method of claim 2, wherein calculating M energies of the M subband echo residual signals comprises low pass filtering the squares of the echo residual signals for each subband. 3. The method of claim 2, wherein specifying an adaptation rate for the M adaptive filters includes dividing the M subband echo residual signals into several groups, wherein the echo residual energy in each group is a group energy set. It is added together to output the method. 3. The method of claim 2, wherein specifying an adaptation rate for the M adaptive filters comprises selecting a maximum value from a group energy set. 3. The method of claim 2, wherein specifying an adaptation rate for the M adaptive filters further comprises outputting a normal adaptive rate control signal for the subband with higher group energy. 3. The method of claim 2, wherein specifying an adaptation rate for the M adaptive filters further comprises outputting a slow adaptation rate control signal for a subband with lower group energy. 7. The method of claim 6, wherein outputting the normal adaptive speed control signal comprises outputting a signal to enable adaptation, wherein the output of the normal adaptive speed control signal corresponds to a corresponding adaptive filter every sample interval. Enabling the coefficient adaptation of a. 8. The method of claim 7, wherein outputting the slow adaptation rate control signal comprises outputting a signal to disable the adaptation during some adaptation intervals, wherein the output of the slow adaptation rate control signal is determined by the corresponding adaptive filter. Wherein the adaptation can be performed once every D sample instants, where D is an integer. In the subband echo canceller, A plurality of adaptive filters associated with the plurality of subbands, and An adaptive speed controller coupled to the plurality of adaptive filters Including but not limited to: The adaptive speed controller measures the echo residual energy in each subband, assigns a normal adaptation rate to subbands with higher echo residual energy, and assigns a slow adaptation rate to subbands with lower echo residual energy. Subband echo canceller, characterized in that specified.