MXPA98010418A

MXPA98010418A - Eco breaker for non-line circuits

Info

Publication number: MXPA98010418A
Application number: MXPA/A/1998/010418A
Authority: MX
Inventors: Douglas Romesburg Eric
Original assignee: Ericsson Inc
Priority date: 1996-07-24
Filing date: 1998-12-09
Publication date: 2000-06-05

Abstract

The present invention relates to a system and method for sprouting the echo, which uses two or more microphones (22, 36 and / or 68) to form echo signal beams (L i), which corresponds to a voice signal (L) that enters, which is the output of a loudspeaker (20). The outputs of the microphones (22, 36 and / or 68) are filtered appropriately and then combined (Figures 3-7 and 10-11), to cancel this echo signal (Lï), even if the voice signal (L) which enters has been distorted non-linearly (for example in a digital (analogue) converter (16), an amplifier (18) and / or the loudspeaker (20) Likewise, the microphones (22, 36 and / or 68) can be selectively placed or directed (eg, as shown in Figures 8-9) to allow linear cancellation of ambient noise (

Description

ECO BREAKER FOR NON-LINEAR CIRCUITS BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to the elimination of echo in communication systems and, more particularly and as an example, to the elimination of echoes in mobile stations equipped with hands-free operation.

Related Technique In communication systems, provided for bidirectional voice transmission (two-way) over a communication link between two users, such as in landline or wireless telephone systems, the acoustic signal from the local user is usually detected by a microphone at the near end of the communication link and then transmitted over this communication link to the far end, where it is reproduced by a loudspeaker at the end for presentation to the remote user. Conversely, the acoustic signal from the remote user is detected by a microphone at the far end of the communication link and then transmitted over this communication link to the near end, where it is reproduced in a speaker at this close end for presentation to the local user At either end of the co-communications link, the original voice signal transmitted from the other end and reproduced by the loudspeaker at this end may be reflected by, or may propagate through, the surrounding environment and may be detected. by the microphone at this end. Therefore, this signal from the loudspeaker will be transmitted back to the user at the other end and will arrive delayed in time, relative to the original voice signal (the amount of delay being equal to the time required for the original voice signal make the "round trip"). The signal delayed in time can produce an annoying "echo", which can be heard by the user at the other end. As is well known in the art, the perceptibility (severity) of the echo signal that is returned, for example, from the near end to the far end of the communication link is a function of two factors: first, the amplitude (volume or intensity) ) of the echo signal, transmitted from the near end to the far end and, second, the amount of delay in the echo signal, received at the far end relative to the original voice signal, transmitted from the far end to the extreme near. In general, an increase in the amplitude or delay of the received echo signal results in an increase in the perceptibility. The amplitude of the echo signal received at the far end, in turn, depends on the sensitivity of the microphone at the near end to the local speaker signal, which forms the echo signal. The delay of the echo signal, on the other hand, depends on the communication medium (for example wire line or wireless, analog or digital, etc.). These two aspects (amplitude and delay) of the echo signal are described below in the following. In conventional handsets for wireline telephones, for example, the microphone is designed to be positioned close to the user's mouth, while the speaker is designed to be essentially covered by the user's ear. In this arrangement, there is no need for significant gain (amplification) in the microphone, in order to pick up the near-end voice signal and, thus, the microphone is not very sensitive to the local signal of the speaker. On the other hand, the continuous reduction in the size of portable telephones and the increased use of loudspeakers has meant that the microphone is further away from the user's mouth and, therefore, must have a relatively high gain in order to preserve the desired level in the near-end voice signal. However, this also means that the microphone is more sensitive to picking up the local signal from the speaker. The situation is even more pronounced for hands-free accessories used with vehicle telephones, where the microphone can be further away from the user's mouth and, therefore, its gain must be even greater, which means that it can be highly sensitive to the pickup of the local speaker signal. In summary, the volume of echo signals, produced by modern phones, is likely to be of significant perception. The delay associated with the echo signal can also be significant in its perception mode. For a given volume of echo, the perceptibility increases in proportion to an increase in the echo delay up to 50 ms. In general, an echo delay of more than 50 ms is considered intolerable in your perception. The echo signals produced by the original analog telephones in the wireline network found are of relatively short delays (ie, less than 50 ms) and thus, not significantly perceptive, or to the extent that they are significant in their mode of perception, they experience only linear distortion and, therefore, can be effectively suppressed within the network. However, recent digital telephones, which include cordless digital phones and cell phones, process voice signals through voice coders, which introduce not only significant delays (for example of the order of 200 ms), but also distortions. non-linear, which prevent the effective cancellation of echo signals in the network. For these modern phones, the echo signal must be suppressed at the source, that is, before transmission.

To avoid the transmission of unwanted echo signals it is necessary to isolate the signal from the near end user, which enters the microphone from the far end user signal, which is generated by the loudspeaker, but which can also be detected by the microphone and to transmit to the far-end user only the signal from the near-end user, so that the far-end user will not hear a delayed version of his or her own voice. This has generally been achieved through an echo cancellation or cancellation process, which is designed to remove the echo signal from the microphone output, thus leaving only the user's signal for transmission (for purposes of this specification, the term "echo cancellation" and "echo cancellation" are used interchangeably to refer to the function of eliminating or reducing echo signals, while the need for echo cancellation is present, to some extent, in all systems telephone systems, including those conventionally used in conventional landline telephone sets, as mentioned above, is particularly acute in hands-free loudspeaker applications and especially severe in car-mounted (ie portable) radiotelephones adapted for hands-free operation The closed vehicle environment is very susceptible to multiple reflections of the alt signal avoz in the high gain microphone, which is used for hands-free operation. However, the task of echo cancellation in this environment is complicated by the movement of the vehicle and by changes in the relative direction and strength of the user and echo signals, when windows are opened or closed or as the user moves his head while driving. Also, modern digital radiotelephones include non-linear components (eg voice encoders) that introduce a distortion or linearity in the echo signal making it more difficult to remove it by simple echo cancellation techniques. Previous attempts at echo cancellation, generally, and in the mobile radiotelephone environment, specifically, can be seen, for example, in U.S. Patent Nos. 4,468,641, 4,584,441, 4,712,235, 5,062,102, 5,084,865, 5,305,309, 5,307,405, 5,131,032, 5,193,112, 5,237,562, 5,263,019, 5,263,020, 5,274,705, 5,280,525, 5,315,585, 5,319,585 and 5,475,731. However, a common approach to the problem of echo cancellation in the digital radiotelephone environment can be seen in the circuit shown in Figure 1. This echo cancellation circuit is connected to a telephone system, such as a telephone digital cellular (not shown). The signal L that enters from the telephone system is received on line 10 of the circuit. The signal L is a modulated pulse key (PCM) or other digital representation of the speech signal that originates from the far-end talkative (not shown). This digital signal is applied to a serial combination of a digital-to-analog converter (DAC) 16, an amplifier 18 and a loudspeaker 20, where it is converted from digital to analog, amplified and converted from an electrical signal to an acoustic signal (audio ), respectively. With continued reference to the echo cancellation circuit of Figure 1, each DAC 16, amplifier 18 and speaker 20, can introduce at least some distortion (non-linear) to the incoming signal L. Consequently, the output of the loudspeaker 20 is not a true replica, but rather a distorted version L1 of the incoming L signal. This audio signal L1 will propagate through the surrounding area, reflecting off one or more surfaces and changing in direction, amplitude, frequency and / or phase, before being detected by a microphone 22, which is actually intended to detect the output signal T from the talkative 4 of the near end. As is well known in the art, the amplitude of multiple routes and dependent on frequency and phase changes and delays, experienced by the signal L1, as it travels through the loudspeaker 20 to the microphone 22, can be described by a simple linear acoustic transfer function, here designated as H] _. Similarly, another acoustic transfer function, here designated as H3, can be defined by the resulting route of the speech signal T from the talker 4 to the microphone 22. Still referring to Figure 1, the microphone 22, which includes or it is connected to an amplifier and analog to digital converter (both of which are not shown in Figure 1 for simplicity purposes), convert the acoustic echo and the conversational audio signals into digital electrical signals. The output of the microphone 22 is a composite signal MQ_ = L * «" H] _ + T »H3 (where the symbol" • "designates the multiplication or convolution of these signals in the frequency or time domain, respectively). M] _ is fed to an input of an adder (or, equivalently, a subtractor) 24. The other input to adder 24 receives the output of an adaptive FIR filter, which is used to model or estimate the transfer function H ^ a through a set of coefficients derived from the filter, as revealed, for example, in the US patent, No. 5,475,731- The input to the filter 14 is the input signal L (before the digital-to-analog conversion in the DAC 16 ) and thus, the output of the filter 14 is L »E, which presumably approaches the echo signal L'H? _ It is assumed that the filter 14 can compensate or correspond to the delay of the signal L through the DAC 16 , the amplifier 18, the speaker 20, the acoustic path H ^ and the microphone 22, so the output the filter 14 is aligned in time with the signal M] _. Thus, subtracting in the adder 24 the output of the filter 14 from the signal Mx the echo signal (acoustic) L '»H? in M? it must effectively be canceled by the (electrical) echo estimate ^ H ^, leaving only the desired signal from the talker »H3 for transmission on line 12 to the telephone system, where it can also be processed and transmitted to the far-end talker. It will be appreciated that the transfer function provided by the filter 14 is an estimated and non-exact replica of H] _ and, likewise, K ^ is a dynamic function that is affected by changes in the vehicle environment (for example, by opening or close the windows). If such changes occur while the incoming L signal is active, the output of the filter 14 may deviate from the estimated true echo and thus, there will be a residual echo signal E; L on line 12. Of course, if the talker T is talking, the Ex signal will also include the voice signal ^ Hß. However, it is assumed that the echo and voice signals are relatively uncorrelated so that the Ex signal can be used as an error feedback signal, to adaptively adapt the coefficients of the filter 14 in order to minimize the signal E ^. The algorithm of the Minimum Average Squares (LMS) is a well-known technique, which can be used for this purpose. Inherent to the echo cancellation circuit of the prior art, according to Figure 1, is the assumption that the echo signal L1, which is produced from the speaker 20, is substantially equal to the incoming signal L, which enters the filter 14. In other words, the cancellation of the echo, according to the circuit of Figure 1, requires that the distortion introduced to the signal L by the DAC 16, the amplifier 18 and the speaker 20 be relatively negligible. If this is true, then while the transfer function H ^ for the signal L 'can be substantially duplicated in the filter 14, the echo signal' * H? can it be effectively removed from the output signal M] _ = L'i + «H3 by subtracting the output L * H from Mx? of the filter 14. Thus, when the near end talkative is silent, the error signal E ^, under these ideal conditions, would be zero. However, the assumption formed regarding the negligible effect of the signal distortion (ie L * «Hi = L« H?) In the circuit of Figure 1 does not hold true in many "real world" applications. In practice, significant distortion may be added to the signal L which enters one or more of the DACs 16, the amplifier 18 and the speaker 20. In many cases, the transfer characteristics of the adaptive filter 14 will be limited to the cancellation of subtraction in adder 24 to only those components of signal L, which have not been distorted, thus leaving a significant residual echo signal? j_ = L1 * Hx - L * H? on line 12, which may be In other words, because the signals L and L 'are not linearly related, the signal Ej_ will contain a non-linear distortion which will be transmitted to the far-end talker. Significant distortion is added to the speaker 20. Typical hands-free accessory speakers, used with cell phones, for example, introduce an amplitude distortion of approximately 10%, due to the resonance in the band The free-to-air transmission can also be as high as 12 dB through the loudspeaker and microphone. This limits the echo suppression with the circuit of Figure 1 to approximately 20 - 12 = 8 dB, which is too small for the minimum industry requirement of 45 dB. Even when a more expensive "linear" (dynamic) loudspeaker with little or no resonance of the passband is used, there will be a distortion of around 1% and the loudspeaker will only limit echo suppression to about 28%. dB. Of course, the other components (DAC and amplifier) will add even more distortion, causing the performance of the echo suppression circuit in Figure 1 to deviate further from the accepted industry standard. In summary, for many practical applications, the idealized circuit of Figure 1 does not meet industry minimum standards or user requirements. Recognizing the limitations of the echo cancellation circuit of Figure 1, the prior art has attempted to overcome these limitations by using smaller distortion components, such as dynamic speakers, or by adding other components to remove or block the residual echo signal on the line 12, such as central suppressors, adaptive attenuators, spectral subtractors, speech detectors, ambiguous language detectors, divergence detectors or noise analysis detectors. Nevertheless, minor distortion components are expensive and, as mentioned before, do not completely eliminate the distortion or significantly improve the overall performance. Also, some of the additional components really add their own distortion, and others can actually interfere with the circuit's own operation. Voice detectors, for example, are often deceived by ambient noise that results in inadvertent removal of voice signals, rather than echo signals. Similar disadvantages are also found with the use of other types of detectors.

Related to the problems of cancellation of echo in hands-free telephone systems, is the problem of the acoustic noise cancellation, that is, the removal of ambient (background) noise from the signal of the talker that is transmitted. A recent approach to the cancellation of both noise and echo, is revealed in uo and collaborators, "Acoustic Noise Canceling Microphone System and Eco for Desktop Conferences" Procedures of the 6th International Conference on Applications and Process Technology of Signal (ICSPAT), October 24-26, 1995, pages 41-45. As shown in Figure 2, this approach uses two directional microphones, 22 and 26, placed close to each other, but pointing in opposite directions. The first microphone 22 points to the talkative 4 of the near end and is used as the primary microphone. The second microphone 26 points away from the near-end talker 4 and is used as a reference microphone to cancel an N signal from the ambient noise from a noise source 8. An acoustic barrier (not shown) is placed between the microphones 22 and 26 to reduce the escape of the T signal from the talker in the reference microphone 26. With continued reference to the noise and echo cancellation circuit of Figure 2, while this circuit is in idle mode, there will be no signal from the near end or far end talker and the Mp and Mr outputs of the primary and reference microphones, 22 and 26, respectively, will contain only the background noise N of the noise source 8. The noise signal Mp of the primary microphone 22 is supplied to an adder 27. The noise signal Mr from the reference microphone 26 is supplied to the an adaptive filter 28, which has a transfer function A (z). The output of the adaptive filter 28 is subtracted from the noise signal p in the adder 27. During the idle mode, the adaptive filter 28 adjusts its coefficients using the LMS algorithm in order to minimize the error signal Ea at the output of the adder 27 and, thus, cancel the noise signal Mp of the primary microphone 22. The adaptive filter 28 will finally converge to an optimal transfer function A * (z), which minimizes the residual noise Ea. When the circuit of Figure 2 is in the receive mode, and in addition to the background noise N from the noise source 8 there will be an echo signal L1 of the speaker 20. An optimum fixed filter A * (z), whose coefficient is obtained from the previous inactive mode, it is used in place of the adaptive filter 28 to cancel the noise component in the primary mode Mp. The echo components of the signal Mp are canceled in an adder 29, which receives the output of an adaptive filter 30, which has a transfer function B (z). The adaptive filter 30 adjusts its coefficients and cancels the acoustic echo using the LMS algorithm to minimize the signal E ^ of the residual error of the adder 29. The adaptive filter 30 will finally converge to a function B * (z ) of optimal transfer, which minimizes the residual echo E ^. When the circuit of Figure 2 is in the transmission mode, and in addition to the background noise N from the noise source 8, there will be a voice signal T from the near-end talkative 4 (but presumably no echo signal). The fixed filter A * (z) optimal, but not the filter B * (z), is again used to cancel the noise component in the primary signal Mp. Thus, it is assumed that, due to the placement of the microphones, 22 and 26, and their separation with an acoustic barrier, there will be a small escape of the speech signal T in the reference microphone 26. Thus, it is assumed that the T signal from the near end talker will be detected by the primary microphone 22 and transmitted with minimal distortion to the far end talker. When the circuit of Figure 2 is operating in the "double talk" mode (transmitter and receiver), the reference signal M will contain the background noise N from the noise source 8., an echo signal L1 from the speaker 10 and a voice signal T (presumably minimum) from the talker 4. In this mode, the fixed optimal filters A * (z) and B * (z) of the previous modes inactive and of reception, respectively, are used to cancel the noise and echo components, respectively, in the primary signal Mp. The circuit of Figure 2 is designed to allow the use of two low-order filters for noise and echo cancellation. Because the primary and reference microphones, 22 and 26, are placed closely together, their Mp and Mr outputs, respectively, will contain highly correlated non-acoustic and noise signals that can be canceled with two low-order filters. and 30, respectively. However, this design requires the use of robust voice detectors to differentiate between voice and noise in idle, receive, transmit and double talk modes. In a noisy environment, such as in the vehicle telephone environment, such detectors can be fooled by noise, especially during calls from one mobile unit to another mobile unit. Also, the circuit design of Figure 2 requires that the filters 28 and 30 be activated and deactivated during operation in the different modes, resulting in annoying transitions that can be heard by the far-end user. In view of the drawbacks of prior approaches to echo and / or noise cancellation, there is a need for a new echo cancellation circuit, which can effectively remove the echo signals, in the presence of a non-linear distortion , and that does not require the use of expensive components or additional detectors. It is also convenient that this new circuit be able to effectively cancel the echo even in a noisy or changing environment. In addition, it is convenient that this new circuit be capable of effectively canceling the background noise. These objects are met by the present invention.

COMPENDIUM OF THE INVENTION The present invention provides the desired level of echo suppression, without costly high-quality audio components, such as linear speakers, which would be required in the circuit of Figure 1, and without the need to use detector circuits. , such as the voice detectors, the limes would be required in the circuit of Figure 2, with its corresponding problems. In general, echo suppression, according to the present invention, is achieved by: (1) using the distorted signal E-j_ in the circuit of a microphone of Figure 1, not as the output of the echo cancellation circuit , but only as an error signal for an adaptive filter, which estimates a desired acoustic transfer function, (2) using at least one other microphone and at least one other adaptive filter, which estimates another desired acoustic transfer function, as in the two-microphone circuit of Figure 2, while also avoiding the use of speech detectors, using the input signal L as the reference signal, to adjust the coefficients of at least one of the adaptive filters, (3) filter the distorted echo signal produced from one or more of the microphones (which by themselves do not add virtually distortions) using the coefficients of one or more of the adaptive filters, and (4) linearly combining the filtered and / or non-filtered outputs the microphones to eliminate the distorted non-linearly echo signal. In this way, the distorted acoustic echo signal, detected by a microphone, can be essentially canceled by the distorted acoustic echo signal, detected by another microphone, instead of being canceled only partially by a non-electrical echo estimated based on the Undistorted echo signal, as is the case, for example, in the prior art circuit of Figure 1. In the various embodiments of the circuit of the present invention, the desired filtering of any microphone output can be performed in a fixed filter, whose coefficients are copied from one or more of the adaptive filters, into an adaptive filter whose coefficients are adjusted with reference to the signal entering L; or in an adaptive filter that uses the output of another adaptive filter as a reference signal. Also, in these modes, the ambient noise can also be automatically canceled, along with the cancellation of the echo signal, by the proper placement or signaling of the microphones of the present invention. Unlike the prior art, this new linear approach to echo and noise cancellation is indifferent to which of the near-end and far-end talkers is active at any time and, likewise, can not be "tricked" by the noise and does not insert any non-linear distortion product by itself. In one aspect, the present invention provides an echo cancellation method in an audio circuit, comprising a microphone and a loudspeaker, the microphone detects a voice signal from a near-end user, the loudspeaker receiving a voice signal from the far end and that generates a corresponding echo signal, which is also detected by the microphone. This method comprises the steps of supplying at least one other microphone in the audio circuit, to detect the echo signal, this at least one other microphone also detects the near-end voice signal; estimating a plurality of acoustic transfer functions in a plurality of adaptive filters, each with a plurality of coefficients, at least one of the adaptive filters uses the far-end voice signal as a reference signal to adapt its coefficients; filter the outputs of one or more of the microphones, using the coefficients of one or more of the adaptive filters; and combining the outputs of microphones, filtered and / or unfiltered, in order to substantially cancel the echo signal, while substantially preserving the near-end voice signal. Several modalities of this method are possible, when the audio circuit includes a first and second microphones, and when the echo signal from the loudspeaker to the first and second microphones is characterized by the acoustic transfer functions, H] _ and H2, respectively. Similarly, several modalities of this method are possible when the audio circuit includes a first, second and third microphones, and when the echo signal from the loudspeaker to the first, second and third microphones is characterized by acoustic transfer functions H] _, H2 and H5 respectively. A number of exemplary embodiments are described here for each of the two microphone circuits and the three microphone circuits. Of course, many other embodiments are considered by the present invention, which include those for circuits of more than three microphones. In a first exemplary embodiment of the method of the invention, for the two-microphone circuit, the method comprises the steps of estimating H ^ on a first adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating H2 in a second adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filtering the output of the first microphone in the first fixed filter, using the coefficients of the second adaptive filter; filtering the output of the second microphone in a second fixed filter, with the use of the coefficients of the first adaptive filter; and subtracting the output of the second fixed filter from the output of the first fixed filter. In a second exemplary embodiment for the two-microphone circuit, the method comprises the steps of estimating H ^ on a first adaptive filter, using the far-end speech signal as a reference signal, to adapt the coefficients of the first adaptive filter to estimate H2 in a second adaptive filter, using the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter to filter the output of the second microphone in a fixed filter, which estimates Hl / H2 'using the coefficients of the first and second adaptable filters; and subtract the output of the fixed filter from the outputs of the first microphone.

In a third exemplary embodiment for the two-microphone circuit, the method comprises the estimation steps 1 / H] _ in a first adaptive filter, using the far-end speech signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating I / H2 in a second adaptive filter, using the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output of the first microphone in the first adaptive filter; filter the output of the second microphone in the second adaptive filter; and subtracting the output of the second adaptive filter from the output of the first adaptive filter. In a fourth exemplary embodiment for the two-microphone circuit, the method comprises the steps of estimating H; j_ in a first adaptive filter, using the far-end voice signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating I / H2 in a second adaptive filter, using the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output of the first microphone in the first adaptive filter; filter the output of the second microphone in the second adaptive filter; and subtracting the output of the second adaptive filter from the output of the first adaptive filter.

In a fifth exemplary embodiment for the two-microphone circuit, the method comprises the steps of estimating H in a first adaptive filter, using the far-end speech signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating Hx / H2 in a second adaptive filter, using the output of the first adaptive filter as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output of the first microphone in the first adaptive filter; filter the output of the second microphone in the second adaptive filter; and subtracting the output of the second adaptive filter from the output of the first adaptive filter. In any of these exemplary embodiments for the two-microphone circuit, the first and second microphones can be placed in relation to the user and the speaker, so that the first microphone receives a substantially greater level of the near-end voice signal. that the second microphone and the second microphone receive a substantially greater level of the echo signal than the first microphone. Alternatively, the first and second microphones may be placed substantially equidistant from the loudspeaker, so as to also suppress ambient noise which is received substantially equally by the first and second microphones.

In a first exemplary embodiment for the three-microphone circuit, the method comprises the steps of estimating Hx in a first adaptive filter, using the far-end speech signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating H2 in a second adaptive filter, using the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; estimating H5 in a third adaptive filter, using the far-end speech signal as a reference signal, to adapt the coefficients of the third adaptive filter; filtering the output of the first microphone in a first pair of fixed filters, using the coefficients of the second and third adaptive filters, respectively; filtering the output of the second microphone in a second pair of fixed filters using the coefficients of the first and third adaptive filters, respectively, filtering the output of the third microphone in a third pair of fixed filters using the coefficients of the first and second adaptive filters, respectively; multiply the output of the first pair of fixed filters by a constant (c) in a first multiplier, where 0 < c < 1; multiply the output of the second pair of fixed filters by a constant (1 -c) in a second multiplier; and subtract the output of the third pair of fixed filters from the outputs of the first and second multipliers. In a second exemplary mode of the three-microphone circuit, the two fixed filters using the coefficients of the third adaptive filter in the first mode are replaced by a single fixed filter through a repositioning of the first and second multipliers at the outputs of the other two fixed filters in the first and second pairs of fixed filters, respectively, and combining the balances of the first and second multipliers before filtering them in this single fixed filter. In any exemplary embodiment for the three-microphone circuit, the value of the constant (s) can be adjusted to follow the direction of the near-end voice signal or to minimize the impact of the noise. These and other aspects and advantages of the present invention will become more readily apparent from the drawings and the accompanying detailed description.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention will be better understood and its numerous objects and advantages will become apparent to those skilled in the art, with reference to the following drawings, in which: Figure 1 is a block diagram of a circuit echo cancellation common of the prior art; Figure 2 is a block diagram of a certain noise and echo cancellation circuit; Figure 3 is a block diagram of one mode of a two-microphone echo cancellation circuit, constructed in accordance with the present invention; Figures 4 to 7 are several alternative embodiments of the echo cancellation circuit, of two microphones, according to Figure 3; Figures 8 and 9 are two different views of an exemplary configuration of the two microphones of any of the circuits of Figures 3 to 7, this configuration being used for cancellation of both noise and echo in a hands-free accessory, for a cell phone mounted on a conventional vehicle; and Figures 10-11 are alternative embodiments of a three-microphone echo cancellation circuit, constructed in accordance with the present invention.

DETAILED DESCRIPTION Referring first to Figure 3, the conventional echo cancellation circuit of Figure 1 has been modified, according to one embodiment of the present invention. In this embodiment and in addition to the microphone 22, a second microphone 36 is used to detect the echo signal L 'from the loudspeaker 20 and the signal T from the talker from the near end talker 4. As will be appreciated after a review of the operation of Figure 3 and noted below, the microphone 36 is used in a different manner from that of the microphone 26 of Figure 2. It will further be appreciated that while the distortion introduced by the components 16 to 20 prevent the estimated output of the echo from the filter 14 from completely canceling the acoustic echo in the adder 2 in Figure 1, the filter 14, however, can be used in Figure 3 to estimate the acoustic transfer function Hx, since that their coefficients will converge on the same values, even in the presence of such distortion. as shown in Figure 3, the audio path between the loudspeaker 20 and the microphone 36 is characterized by an acoustic transfer function labeled H2 • Similarly, the audio path between the near end talker 4 and the microphone 36 is characterized by an acoustic transfer function, labeled H4. The output M2 = L '«H2 + T» H4 of the microphone 36 is applied to an input of an adder 38. The other input of the adder 38 receives the output of an adaptable FIR filter, 40, which models the H2 function • The input The adaptive filter 40 is the signal that enters L on line 10 from the telephone system. Therefore, the output of the adaptive filter 40 is the signal L »H2, which is subtracted in the adder 38 of the signal M2. The adaptive filter 40 uses the LMS algorithm to minimize the error signal E2 at the output of the adder 38. With continued reference to Figure 3, the output Mx = L '* H1 + »H3 of the first microphone 22 is supplied to a fixed FIR filter 42, whose coefficients are copied from the leads for the adapter filter 40, as shown by the dashed line 44, between the filters 40 and 42 Therefore, the output of the fixed filter 42 is an echo composite signal and the talker, represented by the combination L '«HX« H2 + »H3» H2. Similarly, the output M2 = L! »H2 +» H of the second microphone 36 is applied to a fixed FIR filter 32, whose coefficients are copied from the leads for the adaptive filter 14, as shown by the dashed line 34, between the filters 14 and 32. Therefore, the output of the fixed filter 32 is a signal composed of the echo and the talker, represented by the combination L '»H2 # H1 +« H4 * Hi. As shown in Figure 3, the output of the fixed filter 32 is subtracted from the output of the fixed filter 42 in an adder 46.

Since, by definition, L '»H * H-2 = L' * H2 * HX, the echo components in the outputs of the fixed filters 32 and 42, will cancel linearly with each other in the adder 46, leaving a composite signal of the talker , echo-free, T »H-3« H2-T »H4» H1 = T (H3 * H2-H4 »H) at the output F of adder 46 to transmit on line 12 to the telephone system. It will be appreciated that the circuit of Figure 3 achieves the cancellation of the linear echo in the adder 46, independent of the relative configuration or placement of the two microphones 22 and 36 and the speaker 20. However, it is ideally desirable that the frequency response of the circuit, as reflected in the output F of the adder 46, corresponds, as closely as possible, to the signal T of the talker, detected in the microphone 22. Mathematically speaking, it is convenient that the following equation remains true: F = T (H3 «H2-H4« H1) = T «H3 As will be easily recognized, this equation holds true if H = ly H »HX = 0. The first condition can be fulfilled if the second microphone 36 is placed directly in front of and close to the speaker 20, so that the_ energy ratio signal L1 direct to the reflected, detected by the microphone 36 is quite high, and H2 will be very close to being a pulse function (ie, H2 = l) • The second condition is met if the second microphone 36 is placed a lot closer (for example 10 times closer) to the speaker 20 than the first microphone 22, so that H < < H2, and its first microphone 22 points towards the talker 4 while the second microphone 36 points away from the talker 4 and towards the speaker 20, so that H4 << H3 «Therefore, H4 * HX« H3 »H2 and H4« HX can be treated as being close to 0. Under these conditions, the combination (H3 »H2_H4 * Hx) will be effectively reduced to H3, and the F signal of output will resemble the desired input signal T «H3. Referring now to Figure 4, an alternative embodiment of the echo cancellation circuit of Figure 3 is shown. In this embodiment, the Mx output of the microphone 22 is provided directly to the adder 46 without passing through any filter, such as the fixed filter 42 shown in Figure 3. This arrangement eliminates any delay in processing the T signal from the near end talkative through the filter 42, in the event that such a delay is noticeable by the far-end talkative. Likewise, any delay in the signal T through the Hx route can be minimized if the microphone 22 is placed near - from the close-end talker 4. To supply the cancellation of the echo in the adder 46, the output M2 of the microphone 36 is passed through a filter 48 of infinite impulse response (IIR), fixed, which performs a rational acoustic transfer function Hx / H2 (where the symbol "/" designates division or deconvolution in the frequency or time domain, respectively.The numerator Hx of this function is provided by copying the coefficients of the adaptive filter 14, as shown by the line of dashes 34. The denominator H2 is provided by copying the coefficients of the adaptive filter 40, as shown by the dashed line 44. With continued reference to Figure 4, the output M of the microphone 22 is a signal composed of echo and of the talker, represented by the combination L '»HX + T« H3. The output of the filter 48 is an echo composite signal and the talker, represented by the combination of L '»HX +» H4 »Hx / H2. The output of the filter 48 is subtracted from the signal Ml in the adder 46. Assuming that the delay for the signal L 'through the route H2 is less than or equal to the delay through the route H (which would be the case, for example, if the microphone 36 is much closer to the speaker 20 than the microphone 22, or its microphone 36 and loudspeaker 20 are contained within the same set), the causality will be maintained for the filter 48. The echo component L "HX in the signal Mx will be effectively canceled by the equivalent component from the output of the fixed filter 48. Thus, the signal F on the output line 12 to the telephone system will be free of echo. Referring now to Figure 5, another alternative embodiment of the echo cancellation circuit of Figure 3 is shown. This modality can be considered more efficient than the modality shown in Figure 4, since it avoids the process requirements and memory associated with the copying of filter coefficients, such as filters 14 and 40 in Figure 4. Also, this mode avoids the risk associated with the use of an IIR filter, which may become unstable at some frequency , as may occur if H2 in the denominator of filter 48 in Figure 4 goes to zero at any frequency. In Figure 5, the outputs of the microphones 22 and 36 are supplied to the adaptable filters FIR 50 and 52, respectively, which estimate the rational transfer functions 1 / HX and I / H2, respectively, and which use the LMS algorithm to minimize error signals E-1 and E2 at the output of adders 24 and 38, respectively. -tea. As will be appreciated by persons skilled in the art, since the FIR filters are unconditionally stable, the adaptive filters, 50 and 52, which represent approximations of the -FIR of the IIR functions, will be stable at all frequencies.

Also shown in Figure 5 are the delay elements 54 and 56, which are used to ensure the causality of the filters 50 and 52, respectively, and for the time alignment of the input signals to each of the adders. and 38, respectively. In other words, without the delay elements 54 and 56, the filters 50 and 52 will to estimate a negative delay in order to compensate the positive delay of the signal L through the part of the circuit consisting of the components 16- 20, the acoustic path Hx or H and the microphone 22 or 36, as applicable. The inclusion of the delay elements 54 and 56 allows the filters 50 and 52, respectively, to converge to a positive delay equal to the difference between the amount of delay in the elements 54 and 56, respectively, and the remaining part of the delay. circuit. As will be appreciated by ordinary persons skilled in the art, the delay elements 54 and 56 can operate linearly so as not to introduce any distortion in the L signal. It will also be appreciated that the amount of delay through the elements 54 and 56 should be the same in order to ensure the time alignment of the outputs of the filters 50 and 52, which form the inputs to the adder 46. Thus, it is possible to replace the delay elements 54 and 56 by a single delay element, through the which input signal L passes, before branching to adders 24 and 38.

With continued reference to Figure 5, the output of the adaptive filter 50 is an echo and talkative composite signal, represented by the combination of L '+ T »H3 / HX. Similarly, the output of the adaptive filter 52 is an echo and talkative composite signal, represented by the combination of L '+ T «H4 / H2. The output of the filter 52 is subtracted from the output of the filter 50 in the adder 46, whereby cancels the echo component L1, and leaves the talk signal T (H3 / HX-H4 / H2). Consequently, the signal F on the output line 12 to the telephone system will be echo-free. It should be noted that, although the microphones 22 and 36 are placed somewhat equidistant from the loudspeaker 20 in Figure 5, such an arrangement is not required for effective echo cancellation in the adder 46, and is for purposes of illustration only. The only requirement for the proper operation of the circuit in Figure 5 is that H3 / HX »H2 / H2 so the T signal from the talker will be preserved substantially after the cancellation of the echo in the adder 46. This requirement can be fulfilled ,, for example, if the microphone 22 is directed towards the near-end talker 4 and away from the speaker 20, while the microphone 36 is directed away from the near-end talker 4 and towards the speaker 20, so that H3 »H4 and Hx «H2. Referring next to Figure 6, another alternative modality of the echo cancellation circuit of Figure 3 is shown. This modality can be considered as something of a hybrid of the circuits shown in Figures 3 and 4, since use some components of each of those two circuits. The circuit of Figure 6 avoids the use of potentially unstable IIR filters, such as filter 48 in Figure 4, but does not require any copy of coefficients, although a stable FIR filter 32 was used in Figure 3. The circuit in the Figure 6 uses the delay element 56 for the same purpose as in Figure 5 (ie, the time alignment of the inputs to the summer 38), but does not require the use of the delay element 54 of Figure 5, since the adaptive filter 14 in Figure 6 can appropriately count for the delay of the signal L in that part of the circuit (ie for the purpose of the time alignment of the inputs to the adder 24). The circuit of Figure 6, however, may require the use of another linear delay element 58 to ensure that the Mx output of the microphone 22 reaches the adder 46 in time alignment with the output of the FIR filter 52, so that the components of the echo in these two outputs, will effectively cancel each other in the adder 46. It will be appreciated that, for this purpose the delay through the element -58 must correspond to the delay through the element 56 which, in turn, is equivalent to the delay through the acoustic path H2 and filter 52. Since the filter 32 incorporates an estimate of the delay through the acoustic path H, the delayed signal of the microphone 36 to the adder 46 will find some delay as the delayed signal of the microphone 22 to the adder 46 if the delay element 56 and 58 have matching delays. As can be seen from Figure 6, the M output of the microphone 22 is an echo and talkative composite signal, represented by the combination L '»HX + T« H3. The output of filter 32 is an echo and talkative composite signal, represented by the combination of LI »HX + T» H4 »HX / H2. The output of the filter 32 is challenged in the adder 46 from the Mx (delayed) output of the microphone 22, thus canceling the echo component L '* HX and leaving the talkative signal T (H3 ~ H4 »HX / H2). Consequently, the signal F on the output line 12 of the telephone system will be echo-free. As with Figure 5, although the microphones 22 and 26 are placed somewhat equidistant from the loudspeaker 20 in Figure 6, such an arrangement is not required for effective echo cancellation in the adder 46 and is for illustrative purposes only. The only requirement for the proper operation of the circuit in Figure 6 is that H3 »H4» HX / H2 so that the T signal of the talker will be substantially preserved after the cancellation of the echo in the adder 46. This requirement is met, example, if the microphones 22 and 35 are placed or directed in some way that the microphone 22 receives substantially greater energy for the T signal from the conversationalist than the microphone 36, which means that H3 »H4. Referring now to Figure 7, a more alternative modality of the echo cancellation circuit of Figure 3 is shown. This modality combines the advantages while avoiding any potential disadvantages, associated with the modalities discussed previously. In particular, this mode prevents the copying of the filter coefficients or the use of the IIR filters or the delay elements while providing the desired echo cancellation in the adder 46. In Figure 7, the Mx output of the microphone 22 it is fed directly to the adder 46. This means that, if the microphone 22 is placed close to the near end talker 4, there will be a minimum delay of the signal T, which is detected by the microphone 22. An adaptive FIR 60 filter estimates the function of rational transfer H / H2 based on the output of the adaptive filter 14 and the output M2 of the microphone 36. As mentioned above, in spite of performing a rational function, the filter 60 will be stable, because it is a filter FIR. With continued reference to Figure 7, the M output of the microphone 22 is an echo and talkative composite signal, represented by the combination L '* HX + T * H3. The output of the filter 60 is an echo and talkative composite signal, represented by the combination L, »HX + T« H4. The output of the filter 60 is subtracted from the signal Mx in the adder 46, thus canceling the echo component L '* HX and leaving the signal of the talker (H3-H4). Consequently, the signal F on the output line 12 to the telephone system will be echo-free. Also, as long as the microphones 22 and 36 are placed in relation to the talker 4 so that H3 >; > H4, the F signal at the circuit output will be very close to the desired conversational signal »H3 that appears at the circuit input. In summary, the circuit of Figure 7 achieves the cancellation of the echo with minimum delay or distortion of the signal T of the conversationalist. For illustrative purposes, Figure 7 shows the microphone 36 placed directly on the front of the speaker 20, which means that H2 must be very close to 1 and the H1 / H2 function will closely approximate H1. Under these circumstances, the adaptive filter 60 will essentially model the echo path Hl from the speaker 20 to the microphone 22. However, it should be noted that the circuit of Figure 7 can achieve the desired echo cancellation even if the microphones 22 and 36 is placed, for example, substantially equidistant from the speaker 2. While the talker 4 is closer to the microphone 22 than to the microphone 36, H3 will be larger than H4 and the T signal from the talker will be preserved at the circuit output. However, if the microphones 22 and 36 are positioned relative to the speaker 20 so that the delay through the acoustic path H2 is greater than the delay through the acoustic path H1, it may be necessary to apply a delay to the signal Mx (for example, as shown by the delay element 58 in Figure 6), in order to ensure the causality of the filter 60 in Figure 7 (otherwise the filter 60 would have to estimate a negative delay). Although the effect of the background noise N from a noise source 8, as shown in Figure 2, has not been specifically discussed in connection with Figures 3-7 it will be readily appreciated that the ambient noise will be canceled to the adder 46 at a degree that is dependent on the relative amplitude, frequency and phase of the noise signals detected by the microphones 22 and 36, respectively, and on the nature of the filter transfer functions applied to these noise signals, before arriving to the adder 46. However, if the gains for the echo acoustic paths, Hl and H2, to the microphones 22 and 36, respectively, can be made substantially identical to each other and to the gains for the corresponding acoustic noise paths, the Noise components received in the adder 46 can be canceled in the same way and using the same filters as for the echo components (at least for those noise components in the low frequencies). These are, for example, the range of 300 to 800 Hz, which must be detected in phase and at an equal amplitude in each of the microphones 22 and 36. This double cancellation of noise and echo can be achieved by placing the microphones 22 and 36 equidistant from speaker 20 so that Hl = H2. To ensure that the T signal from the near end talkative is not also canceled in this adder 46, the microphone 22 can be placed much closer to the talker 4 than the microphone 36, so that H3 »H4. An example of such an arrangement is shown in Figures 8 and 9. Referring now to Figures 8 and 9, two generalized views of an exemplary configuration of the microphones 22 and 36 and the loudspeaker 20 of Figures 3 to 7 are shown. in a hands-free cell phone application. The cell phone (not shown) is mounted on a conventional vehicle 62, which has an instrument panel 64 and a windscreen 66. The microphones 22 and 36 can be provided as part of an accessory hands-free kit for use with the telephone cell phone. The loudspeaker 20 may also be provided as part of the accessory equipment or may be part of an original radio equipment in the vehicle 62. In the example shown in Figures 8 and 9, the loudspeaker 20 is mounted centrally below the instrument panel 64. , and the microphones 22 and 36 are placed in the opposite upper corners of the windshield 66. The microphone 22 is seen to be closer to the near end talker 4, who, in this case, is the driver of the vehicle 62. The microphone 36, on the other hand, it is closer to the other near end talker 6, in this case, it is the passenger of the front seat in the vehicle 62. This arrangement allows both speakers of 1 near end, 4 and 6, to participate in a conversation with hands-free Also illustrated in Figures 8 and 9, are the acoustic transfer functions H1-H4, which correspond to those shown in Figures 3 to 7 for the echo signal L 'from the speaker 20, and the signal from the talker 4 of the near end, as applicable. For simplicity, the acoustic transfer functions for the voice signal from the near end talker 6 are not shown or discussed here, although it will be understood that the noise and echo cancellation analysis not for the near end talker 6 will be a mirror of the analysis for the near-end talker 4, as generally presented in connection with Figures 3 to 9. In the example shown in Figures 8 and 9, the audio paths of the speaker 20 to the microphones 22 and 36, respectively , they are illustrated as being of equal length and, therefore, the gains for the functions Hl and H2 must be substantially equal (ie Hl = H2). On the other hand, the audio path from the talker 4 to the microphone 22 is substantially shorter than the audio path from the talker 4 to the microphone 36 and, therefore, the gain for the function H3 must be substantially greater than for the function H4 (ie H3 »H4). Since the noise signal N and the echo signal L 'must be highly correlated in each microphone 22 and 36 in Figures 8 and 9, they will be similarly canceled in the adder 46 in any of the circuits shown in Figures 3 to 7. Further, since the microphone 22 receives a signal from the talker much larger than the microphone 36, the signal from the talker will be preserved substantially after the cancellation of the noise and echo not in the adder 46. However, in some facilities it can it is necessary to use an automatic gain control (AGC) in the output signal F, in order to ensure an appropriate signal level (signal-to-noise ratio) on line 12 to the telephone system. The AGC will also supply the far end converter with a more consistent volume level received because the near end talker adjusts its voice in correspondence with the conditions of the changing noise. Also, AGC may be necessary to reduce the effect of quantization-type noise, which results from the use of codes in the digital telephone system. While Figures 3 to 9 have illustrated the present invention in circuits using only two microwaves, it will be appreciated that the approach of the present invention to linearly combine different microphone signals to achieve echo suppression can be used with three or more microphones. Figures 10 and 11 provide two examples of echo cancellation circuits using three microphones in accordance with the present invention. These two examples extend the basic circuit in Figure 3 from two microphones to three microphones. It should be clearly understood, however, that the circuits in Figures 4 to 9 can be similarly extended and, likewise, will be readily apparent to persons of ordinary skill in the art, that it is possible, in accordance with the present invention, to build many different circuits that use more than three microphones. Referring now to Figure 10, the adaptive filters 14 and 40 model the acoustic transfer functions Hl and H2 for the echo paths from the loudspeaker 20 to the microphones 22 and 36, respectively and use the LMS algorithm to minimize the error signals El and E2 at the output of the adders 24 and 38, respectively, in the same manner as discussed in relation to the circuit shown in Figure 3. The circuit of Figure 10, however, uses a third microphone 68, an adder 70 and an adaptive filter 72, which models the acoustic transfer function H5 for the echo path from the loudspeaker 20 to the microphone 68 and which also uses the LMS algorithm to minimize the error signal E3 at the output of the adder 70. As before, the objective is to manipulate the echo components in the outputs of the microphones 22, 36 and 68, so that they can be combined linearly and canceled in 1 adder 46. For this purpose, the circuit of the Figure 10 uses two fixed filters 32, whose coefficients are copied along the line 34 from the adaptive filter 14, two fixed filters 42 whose coefficients are copied along the line 44 from the adaptive filter 40 and two fixed filters 74, whose coefficients are copied along the line 76 of the adaptive filter 72. In other words, the fixed filters 32, 42 and 74 perform the transfer functions Hl, H2 and H5, respectively, as shown in Figure 10. With continuous reference Figure 10, the output Mx of the microphone 22 is passed through two successive stages of filters 42 and 74, respectively, and then multiplied by a constant (c) in a multiplier 78 (where 0 <; c < 1) . The echo component at the output of the multiplier 8 will be the signal c »L *» Hl »H2 * H5, which is fed to the adder 46. The output M2 of the microphone 36 is passed through two successive stages of filters 32 and 74 , respectively, and then multiplied by a constant (1 - c) in a multiplier 80. The echo component in the output of multiplier 80 will be the signal (1 - c) »L '» H2 »Hl * H5, which is also fed to the adder 46. The output M3 of the microphone 68 is passed through the two successive stages of the filters 32 and 42, respectively, and the resulting echo component L '»H5 * H1« H2 is then fed directly to the adder 46 for subtraction from the other echo signals received by the adder 46. In the adder 46, the received echo components will cancel each other and the output F will be free of echo, as shown by the following formula: F (that) = c «L, * Hl * H2» H5 + (l-c) L «» H2 * Hl »H5 -L '« H5 «H1« H2 = 0 Figure 11 is a variant of Figure 10, aimed at reducing the number of fixed filters 32, 422 and 74 and the associated amount of coefficient copying. As can be seen from a comparison of Figures 10 and 11, the circuit of Figure 11 replaces the two fixed filters 74 in the circuit of Figure 10, with a fixed filter 74 and an adder 82, combining the outputs of the multipliers 78 and 80, before passing the combination through the single fixed filter 74 and over the adder 46. The output of the fixed filter 74 in Figure 11, therefore, is equivalent to the sum of the outputs of the multipliers 78 and 80 in Figure 10, and echo cancellation will occur in adder 46 of Figure 11, in a manner equivalent to Figure 10. Figures 10 and 11 also illustrate that the echo signal will be subtracted in adder 46 and can be supplied to a negative (inverted) conductor of the adder 46, in the manner as shown in Figure 10 or, alternatively, multiplied by the constant (-1) in a multiplier 84 before being supplied to the adder 46, as shown in Figure 11. An examination of Figures 10-11 discloses the basic teachings of the present invention, which can be extended to circuits with an arbitrary number of microphones (ie greater than three). To achieve the cancellation of the echo in the adder 46, a similar approach to "minimum common denominator" is used in that each of the outputs of the microphones is filtered through a plurality of filtering stages, which impart in these outputs a common set of transfer functions. As necessary, one or more microphone outputs can be multiplied by the appropriate constants, so that when they are added (or subtracted) in the adder 46, do not leave any echo signal. Returning briefly to Figure 10, for example, the Mx output of the microphone 22, the echo path which has the transfer function Hl, is passed through the filters 42 and 74, which perform the transfer functions H2 and H5 , which correspond to the echo paths to the two other microphones 36 and 68, respectively. Similarly, the M2 output of the microphone 36, the echo path to which it has the transfer function H2, is passed through the filters 32 and 74, which perform the transfer functions Hl and H5, which correspond to the echo paths to the other two microphones 22 and 68, respectively. A similar process is followed for the M3 output of the microphone 68. Selective multiplication by these constants (c) and (1-c) in Figure 10 ensures that the various echo signals are properly evaluated and can be completely canceled from each other in the adder 46. It will further be appreciated that the value of the constant (c) in Figures 10 and 11 may vary between 0 and 1 in order to "steer" the microphones 22 and 36 towards the speech signal T from the near-end talker 4 moving away from the noise signal N from the noise source 8, without affecting the cancellation of the echo in the adder 46. For example, the microphones 22 and 36 can be placed on the windscreen 66, as generally shown in FIGS. and 9, while the microphone 68 can be placed close to and directed towards the loudspeaker 20 inside the vehicle 62. If c = 0, the output of the multiplier 78 in Figures 10-11 will be zero, and only the output M2 of the microphone 36 will be zero. will be passed to the adder 46 Conversely, if s = 1, the output of the multiplier 80 in Figures 10-11 will be zero, and only the output Mx of the microphone 22 will be passed to the adder 46. Between these two extremes, the value of (c) can to be adjusted to increase the sensitivity of one of the microphones 22 and 36 to the voice signal T, in order to automatically guide the movements of the head 4 of the near end. Similarly, the value of (c) can be adjusted to decrease the sensitivity of one of the microphones 22 and 36 to the noise signal N, so as to minimize noise where, for example, this microphone is closest to source 8 of the noise. The most appropriate value of (c) at any time can be determined, for example, through techniques that form beams and microphone arrays, as is well known in the art. It should now be apparent from Figures 3 to 11 and the accompanying discussion that each of the circuits of the present invention achieves the desired cancellation of the echo signal detected through a first microphone using at least one other microphone, to establish another route for the echo signal to an appropriate node in the circuit, where the echo signal detected by the first microphone can be canceled by the echo signal detected by one or more of the other microphones. Indeed, a beam pattern is formed for the echo signal through the use of a plurality of microphones, such that there is an override for this signal at a given node in the echo cancellation circuit. In this node, the distortion components, both linear and non-linear, of the echo signal cancel out as well as a significant portion of the ambient noise. It can also be seen that the present invention achieves the suppression of echo and noise with two or more microphones in a wide variety of configurations, and that when three or more microphones are used, the sensitivity of at least some of these microphones can be controlled by the appropriate evaluation of the microphone outputs in order to maximize the output level of the voice signal or to minimize the level of the noise signal output, without affecting the echo cancellation carried out by the circuit . It will be noted that, for purposes of illustration, the adaptable filters in the circuits of Figures 3 to 11, use the Minimum Average Squares (LMS) algorithm to estimate the desired functions. However, it will be appreciated by those skilled in the art, that many other estimation algorithms can also be used. In particular, two categories of algorithms are suitable for this purpose. The first category of algorithms, which is known as the Descending Gradient algorithms, include the LMS, normalized LMS (NLMS) and block LMS (BLMS). The second category of algorithms, known as the Least Squares Estimation (LSE) algorithms, include Kalman Filtration, Minimum Recursive Squares (LRS) and Fast Transverse Filter (FTF). It will also be noted that while only one speaker has been shown in the circuits of Figures 3 to 11, the teachings of the present invention can be easily extended to echo cancellation circuits that incorporate a plurality of speakers. In general, those skilled in the art will readily recognize that many modifications and variants can be made in the embodiments of the present invention disclosed herein, without departing substantially from the spirit and scope of the present invention. Therefore, the form of the invention disclosed herein is exemplary and does not attempt to limit the scope of the invention, as defined in the following claims.

Claims

CLAIMS 1. In an audio circuit, comprising a microphone and a loudspeaker, this microphone detects a voice signal from a near-end user, the loudspeaker receives a far-end voice signal and generates a corresponding echo signal, which is also detected by the microphone, a method for canceling the echo signal, comprising the steps of: supplying at least one other microphone in the audio circuit, to detect the echo signal, this at least one other microphone also detects the signal of Near end voice; estimating a plurality of acoustic transfer functions in a plurality of adaptive filters, each having a plurality of coefficients, at least one of the adaptive filters uses the far-end voice signal as a reference signal, to adjust its coefficients; filter the outputs of one or more of the microphones, which use the coefficients of one or more of the adaptive filters; and combining the outputs of microphones, filtered and / or unfiltered, in order to substantially cancel out the echo signal, while substantially preserving the near-end voice signal.
2. The method of claim 1, wherein the filtering step comprises filtering at least one microphone output in the fixed filter, whose coefficients are copied from one or more of the adaptive filters.
3. The method of claim 1, wherein the filtering step comprises filtering at least one microphone output in an adaptive filter, whose coefficients are adapted using the far-end voice signal, as a reference signal.
4. The method of claim 1, wherein the filtering step comprises filtering at least one microphone output in an adaptive filter, which uses the output of another adaptive filter as a reference signal.
5. The method of claim 1, wherein the audio circuit includes a first and second microphones, the echo signal from the loudspeaker and the first and second microphones have acoustic transfer functions, Hx and H2, respectively, and in which the method it comprises the steps of: estimating Hx in a first adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating H2 in a second adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filtering the output of the first microphone in a first fixed filter, using the coefficients of the second adaptive filter; filtering the output of the second microphone in a second fixed filter, which uses the coefficients of the first adaptive filter; and subtracting the output of the second fixed filter, from the output of the first fixed filter.
6. The method of claim 5, further comprising the step of placing the first and second microphones in relation to the user of the near end and the speaker, so that the first microphone receives a substantially greater level of the end voice signal. close, compared to the second microphone, and this second microphone receives a substantially higher level of the echo signal than the first microphone.
7. The method of claim 5, further comprising the step of placing the first and second microphones substantially equidistant from the loudspeaker, so as to also suppress the ambient noise, which is received, substantially equally, by the first and second microphones .
8. The method of claim 1, wherein the audio circuit includes a first and second microphones, the echo signal from the loudspeaker to the first and second microphones has the acoustic transfer functions, Hx and H2, respectively, and wherein the method comprises the steps of: estimating Hx in a first adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the first adaptive filter; estimating H2 in a second adaptive filter, which uses the far-end voice signal, as a reference signal to adapt the coefficients of the second adaptive filter; filter the output of the second microphone in a fixed filter, which estimates Hx / H2, using the coefficients of the first and second adaptive filters; and subtract the output of the fixed filter from the output of the first microphone.
9. The method of claim 8, further comprising the step of placing the first and second microphones in relation to the user of the near end and the speaker so that the first microphone receives substantially a higher level of the near-end voice signal than the user. second microphone, and this second microphone receives a substantially greater level of the echo signal than the first microphone.
10. The method of claim 8, further comprising the step of placing the first and second microphones substantially equidistant from the loudspeaker, so that it also suppresses ambient noise, which is received substantially equally by the first and second microphones.
11. The method of claim 1, wherein the audio circuit includes a first and second microphones, the speaker echo signal to the first and second microphones, which has the acoustic transfer functions, Hx and H2, respectively, and where the method it comprises the steps of: estimating 1 / HX in a first adaptive filter, which uses the far-end voice signal as a reference signal to adapt the coefficients of the first adaptive filter; estimating I / H2 in a second adaptive filter, which uses the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output of the first microphone in the first adaptive filter; filter the output of the second microphone in the second adaptive filter; and subtracting the output of the second adaptive filter from the output of the first adaptive filter.
12. The method of claim 11, further comprising the step of placing the first and second microphones in relation to the near end user and the loudspeaker so that the first microphone receives a substantially greater level of the near end voice signal than the second microphone, and this second microphone receives a substantially greater level of the echo signal than the first microphone.
13. The method of claim 11, further comprising the step of placing the first and second microphones substantially equidistant from the loudspeaker, so as to also suppress ambient noise, which is received in substantially equal form by the first and second microphones.
14. The method of claim 1, wherein the audio circuit includes a first and second microphones, the echo signal of the loudspeaker to the first and second microphones, have acoustic transfer functions, Hx and H2, respectively, and wherein the method comprises the steps of: estimating Hx in a first adaptive filter, with the use of the far-end voice signal as a reference signal to adapt the coefficients of the first adaptive filter; estimate 1 / H2 in a second adaptive filter, with the use of the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output of the second microphone in the second adaptive filter; filter the output of the second adaptive filter, with the use of the coefficients of the first adaptive filter; and subtract the output of the fixed filter from the output of the first microphone.
15. The method of claim 14, further comprising the step of placing the first and second microphones in relation to the user of the near end and the speaker, so that the first microphone receives a substantially greater level of the near-end voice signal than the second microphone, and the second microphone receives a substantially greater level of the echo signal than the first microphone.
16. The method of claim 14, further comprising the step of placing the first and second microphones substantially equidistant from the loudspeaker, so as to also suppress ambient noise, which is substantially uniformly received by the first and second microphones.
17. The method of claim 1, wherein the audio circuit includes a first and second microphones, the echo signal from the loudspeaker to the first and second microphones has the acoustic transfer functions, Hx and H2, respectively, and wherein the method comprises the stages of: estimating H in a first adaptive filter, with the use of the far-end voice signal as a reference signal to adapt the coefficients of the first adaptive filter; estimate Hx / H2 in a second adaptive filter, with the use of the output of the first adaptive filter, as a reference signal, to adapt the coefficients of the second adaptive filter; filter the output "of the second microphone in the second adaptive filter; subtract the output of the second adaptive filter from the output of the first microphone.
18. The method of claim 17, further comprising the step of placing the first and second microphones in relation to the user and the speaker, so that the first microphone receives a substantially greater level of the near-end voice signal than the second microphone , and this second microphone receives a substantially greater level of the echo signal than the first microphone.
19. The method of claim 17, further comprising the step of placing the first and second microphones substantially equidistant from the loudspeaker, so as to also suppress ambient noise, which is received substantially uniformly by the first and second microphones.
20. The method of claim 1, wherein the audio circuit includes a first, second and third microphones, the echo signal of the loudspeaker to the first, second and third microphones has the acoustic transfer functions, Hx H2 and H5, respectively, and wherein the method comprises the steps of: estimating H in a first adaptive filter, with the use of the far-end voice signal as a reference signal to adapt the coefficients of the first adaptive filter; estimating H in a second adaptive filter, with the use of the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; estimating H5 in a third adaptive filter, with the use of the far-end voice signal, as a reference signal, to adapt the coefficients of the third adaptive filter; filtering the output of the first microphone in a first pair of fixed filters, with the use of the coefficients of the second and third adaptable filters, respectively; filtering the output of the second microphone in a second pair of fixed filters, with the use of the coefficients of the first and third adaptable filters, respectively; filtering the output of the third microphone in a third pair of fixed filters, with the use of the coefficients of the first and second adaptive filters, respectively; multiply the output of the first pair of fixed filters by a constant (c), in a first multiplier, where 0 < c < 1; multiply the output of the second pair of fixed filters by a constant (1 - c) in a second multiplier; and subtract the output of the third pair of fixed filters from the outputs of the first and second multipliers.
21. The method of claim 20, wherein the value of (c) is adjusted so as to minimize the noise detected by the microphones.
22. The method of claim 20, wherein the value of (c) is adjusted so as to vary the sensitivity of the microphones to the near-end speech signal.
23. The method of claim 1, wherein the audio circuit includes a first, second and third microphones, the echo signal from the loudspeaker to the first, second and third microphones has the acoustic transfer functions, Hx, H2 and H5, respectively , and wherein the method comprises the steps of: estimating Hx in a first adaptive filter, with the use of the far-end voice signal as a reference signal to adapt the coefficients of the first adaptive filter; estimating H in a second adaptive filter, with the use of the far-end voice signal as a reference signal, to adapt the coefficients of the second adaptive filter; estimating H5 in a third adaptive filter, with the use of the far-end voice signal, as a reference signal, to adapt the coefficients of the third adaptive filter; filtering the output of the first microphone in a first fixed filter of the second adaptive filter; filter the output of the second microphone in a second pair of fixed filters, with the use of the coefficients of the first adaptive filter; multiply the output of the first fixed filter by a constant (c), in a first multiplier, where 0 <; c < i; multiply the output of the second fixed filter by a constant (1 - c) in a second multiplier; add the outputs of the first and second fixed filters in an adder; filter the output of the adder in a third fixed filter, with the use of coefficients of the third adaptive filter; filtering the output of the third microphone in a pair of fixed-filters, with the use of the coefficients of the first and second adaptable filters, respectively; and subtract the output of the pair of fixed filters from the output of the third fixed filter.
24. The method of claim 23, wherein the value of (c) is adjusted to minimize the noise detected by the microphones.
25. The method of claim 23, wherein the value of (c) is adjusted to vary the sensitivity of the microphones to the near-end speech signal.
26. The method of claim 1, wherein the audio circuit is part of an accessory, hands-free, for a cordless telephone
27. The method of claim 1, wherein the adaptive filters use a Gradient Descent or Minimum Squares Estimation (LSE) algorithm to estimate the acoustic transfer functions.
28. The method of claim 1, wherein the far-end voice signal is distorted in the loudspeaker and / or in an amplifier or digital-to-analog converter (DAC), connected to the speaker.
29. The method of claim 1, wherein the adaptive filters are finite impulse response (FIR) filters.
30. A circuit for canceling an echo signal from an audio source, this circuit comprises: a first microphone for detecting the echo signal along a first path, defined by first acoustic transfer functions, Hx; a second microphone, for detecting the echo signal along a second path, defined by second acoustic transfer functions, H2; a first adaptive filter, to estimate Hx; a second adaptive filter, to estimate H2; a first fixed filter, to filter the output of the first microphone, which uses the estimation of H; a second fixed filter, to filter the output of the second microphone, which uses the Hx estimation; and an element for subtracting the output of the second fixed filter from the output of the first fixed filter, in order to cancel the echo signal.
31. The circuit of claim 30, wherein the coefficients of the first fixed filter are copied from the coefficients of the second adaptive filter, and wherein the coefficients of the second fixed filter are copied from the coefficients of the first adaptive filter.
32. The circuit of claim 30, wherein the first and second microphones are placed substantially equidistantly from the audio source, so as to suppress ambient noise, which is substantially uniformly received by the first and second microphones.
33. The circuit of claim 30, wherein the echo signal is generated from an incoming signal, received by the audio source, and where the first and second adaptive filters use the incoming signal as a reference signal to estimate Hx and H, respectively.
34. The circuit of claim 33, wherein the echo signal is a distorted version of the incoming signal.
35. A circuit for canceling an echo signal from an audio source, this circuit comprises: a first microphone for detecting the echo signal along a first path, defined by first acoustic transfer functions, Hx; a second microphone, for detecting the echo signal along a second path, defined by second acoustic transfer functions, H2; a first adaptive filter, to estimate Hx; a second adaptive filter, to estimate H; a first fixed filter, to filter the output of the first microphone, which uses an acoustic transfer function, Hx / H2, based on the estimates of Hx and H2; an element to subtract the output of the fixed filter from the output of the first microphone, in order to cancel the echo signal.
36. The circuit of claim 35, wherein the coefficients of the first fixed filter are copied for use by the fixed filter.
37. The circuit of claim 35, wherein the first and second microphones are placed substantially equidistantly from the audio source, so as to suppress ambient noise, which is substantially uniformly received by the first and second microphones.
38. The circuit of claim 35, wherein the signal is generated from an incoming signal, received by the audio source, and wherein the first and second adaptive filters use the incoming signal as a reference signal to estimate Hx and H2, respectively.
39. The circuit of claim 38, wherein the echo signal is a distorted version of the incoming signal.
40. A circuit for canceling an echo signal from an audio source, this circuit comprises: a first microphone for detecting the echo signal along a first path, defined by a first acoustic transfer function, Hx; a second microphone, for detecting the echo signal along a second path, defined by a second acoustic transfer function, H; a first adaptive filter, to estimate 1 / H and to filter the output of the first microphone; a second adaptive filter, to estimate 1 / H2 and to filter the output of the second microphone; and an element for subtracting the output of the second fixed filter from the output of the first fixed filter, in order to cancel the echo signal.
41. The circuit of claim 40, wherein the first and second microphones are placed in substantially equidistant form from the audio source, so as to also suppress ambient noise, which is received substantially uniformly by the first and second microphones.
42. The circuit of claim 40, wherein the echo signal is generated from an incoming signal, received by the audio source, and wherein the first and second adaptive filters use the incoming signal as a reference signal, to estimate 1 / HX and 1 / H2, respectively.
43. The circuit of claim 42, wherein the echo signal is a distorted version of the incoming signal.
44. A circuit for canceling an echo signal from an audio source, this circuit comprises: a first microphone for detecting the echo signal along a first path, defined by first acoustic transfer functions, Hx; a second microphone, for detecting the echo signal along a second path, defined by second acoustic transfer functions, H2; a first adaptive filter, to estimate Hx; a second adaptive filter, for estimating H2 and for filtering the output of the second microphone; a fixed filter, to filter the output of the second adaptive filter, which uses the Hx estimation; and an element to subtract the output of the fixed filter from the output of the first microphone, in order to cancel the echo signal.
45. The circuit of claim 44, wherein the coefficients of the fixed filter are copied from the coefficients of the first adaptive filter.
46. The circuit of claim 44, wherein the first and second microphones are placed in substantially equidistant form from the audio source, so as to also suppress ambient noise, which is received in substantially equal form by the first and second microphones.
47. The circuit of claim 44, wherein the echo signal is generated from an incoming signal, received by the audio source, and where the first and second adaptive filters use the incoming signal as a reference signal, to estimate Hx and H2, respectively.
48. The circuit of claim 47, wherein the echo signal is a distorted version of the incoming signal.
49. A circuit for canceling an echo signal from an audio source, this circuit comprises: a first microphone for detecting the echo signal along a first path, defined by first acoustic transfer functions, Hx; a second microphone, for detecting the echo signal along a second path, defined by second acoustic transfer functions, H2; a first adaptive filter, for estimating Hx a second adaptive filter, for estimating Hx / H and for filtering the output of the second microphone; and an element for subtracting the output of the second fixed filter from the output of the first microphone, in order to cancel the echo signal.
50. The circuit of claim 49, wherein the first and second microphones are placed substantially equidistantly from the audio source, so as to also suppress ambient noise, which is received in substantially equal manner by the first and second microphones .
51. The circuit of claim 49, wherein the echo signal is generated from an incoming signal, received by the audio source, and where the first and second adaptive filters use the incoming signal as a reference signal, to estimate Hx and the second adaptive filter uses the output of the first adaptive filter to estimate H? / H2.
52. The circuit of claim 51, wherein the echo signal is a distorted version of the incoming signal.
53. In a voice communication system, which includes a loudspeaker, which generates an echo signal of an incoming signal, an echo cancellation circuit, which comprises: a plurality of microphones, to detect the echo signal from the speaker; an element for estimating a plurality of acoustic transfer functions, which use the signal that enters as a reference signal; an element for filtering one or more of the outputs of the microphones, which uses one or more of the estimated acoustic transfer functions; and an element to combine the filtered outputs of the microphones to cancel the echo signal.
54. The circuit of claim 53, wherein: the microphones comprise a first, second and third microphones, for detecting the echo signal along the first, second and third routes, respectively, defined by the first, second and third functions of acoustic transfer, Hx, H and H5, respectively; this estimating element comprises a first, second and third adaptive filters for estimating Hx, H and H5, respectively; this filtering element comprises: a first pair of fixed filters, for filtering the output of the first microphone, which uses the estimates of H2 and H5, respectively, a second pair of fixed filters, to filter the output of the second microphone, which uses the estimates of Hx and H5, respectively; and a third pair of fixed filters, to filter the output of the third microphone, which uses the estimates of Hx and H2, respectively; and the combining element comprises: a first multiplier, for multiplying the output of the first pair of fixed filters by a constant (c), where 0 < c < 1; a second multiplier, to multiply the output of the second pair of fixed filters by a constant (1-c); and a subtractor, to subtract the output of the third pair of fixed filters from the outputs of the first and second multipliers.
55. The circuit of claim 54, wherein the value of (s) is adjusted to minimize the noise detected by the microphones.
56. The circuit of claim 54, wherein the value of (c) is adjusted to vary the sensitivity of the microphones to a user signal.
57. The circuit of claim 53, wherein: the microphone comprises a first, second and third microphones, for detecting the echo signal along the first, second and third routes, respectively, defined by the first, second and third functions of acoustic transfer, Hx, H2 and H5, respectively; the estimation element comprises a first, second and third adaptive filters, to estimate Hx, H and H3, respectively; and the filtering element and the combining element comprise: a first fixed filter, for filtering the output of the first microphone, using the estimation of H2, a second fixed filter, to filter the output of the second microphone, with the use of Hx estimation; a first multiplier, to multiply the output of the first fixed filter by a constant (c), where 0 < c 1; a second multiplier, to multiply the output of the second fixed filter by a constant (1 - c); an adder, to add the outputs of the first and second multipliers; a third fixed filter, to filter the output of the adder, which uses the estimate of H5; a pair of fixed filters, to filter the output of the third microphone, which uses the estimates of Hx and H2, respectively; and a subtractor, to subtract the output of the pair of fixed filters from the output of the third fixed filter.
58. The circuit of claim 57, wherein the value of (c) is adjusted to minimize the noise detected by the microphones.
59. The circuit of claim 57, wherein the value of (c) is adjusted to vary the sensitivity of the microphones to a user's signal.
60. The circuit of claim 53, wherein the system comprises a hands-free accessory, for a cordless telephone.
61. The circuit of claim 53, wherein the incoming signal is distorted non-linearly in the loudspeaker or in an amplifier or digital-to-analog converter (DAC), connected to the loudspeaker.
62. The circuit of claim 53, wherein the estimating element comprises a plurality of adaptive filters.
63. The circuit of claim 62, wherein the adaptive filters use a Gradient Descent or Minimum Squares Estimation (LSE) algorithm to estimate the acoustic transfer functions.
64. The circuit of claim 62, wherein the filtering element comprises a plurality of fixed filters, each having coefficients that are copied from one or more of the adaptive filters.
65. The circuit of claim 64, wherein each adaptive filter and each fixed filter are finite impulse response (FIR) filters.
66. The circuit of claim 53, wherein the combining resource comprises: