United States Patent [19]
`United States Patent [19]
`Wu
`Wu
`
`111111
`
`1111111111111111111111111111111111111111111111111111111111111
`US006125179A
`US006125179A
`[11] Patent Number:
`[11] Patent Number:
`[45] Date of Patent:
`[45] Date of Patent:
`
`6,125,179
`6,125,179
`*Sep.26,2000
`*Sep. 26, 2000
`
`[54] ECHO CONTROL DEVICE WITH QUICK
`[54] ECHO CONTROL DEVICE WITH QUICK
`RESPONSE TO SUDDEN ECHO-PATH
`RESPONSE TO SUDDEN ECHO-PATH
`CHANGE
`CHANGE
`
`[75] Inventor: Ying Wu, Grass Valley, Calif.
`Inventor: Ying Wu, Grass Valley, Calif.
`[75]
`
`[73] Assignee: 3Com Corporation, Rolling Meadows,
`[73] Assignee: 3Com Corporation, Rolling Meadows,
`Ill.
`Ill.
`
`[ *] Notice:
`[*] Notice:
`
`This patent issued on a continued pros(cid:173)
`This patent issued on a continued pros
`ecution application ?led under 37 CFR
`ecution application filed under 37 CFR
`1.53(d), and is subject to the tWenty year
`1.53( d), and is subject to the twenty year
`patent term provisions of 35 U.S.C.
`patent term provisions of 35 U.S.c.
`154(a)(2).
`154(a)(2).
`
`[21] Appl. No.: 08/571,710
`[21] Appl. No.: 08/571,710
`[22]
`Filed:
`Dec. 13, 1995
`Dec. 13, 1995
`[22] Filed:
`
`Int. CI? .............................. H04M 1/00; H04M 9/00
`[51]
`[51] Int. Cl.7 ............................ .. H04M 1/00; H04M 9/00
`[52] U.S. CI. .......................... 379/388; 379/400; 379/406;
`[52] US. Cl. ........................ .. 379/388; 379/400; 379/406;
`379/410
`379/410
`[58] Field of Search ..................................... 379/406,410,
`[58] Field of Search ................................... .. 379/406, 410,
`379/411, 409, 388, 392, 400; 307/321;
`379/411,409,388,392,400; 307/32.1;
`370/286, 288, 289, 290, 291
`370/286, 288, 289, 290, 291
`
`[56]
`[56]
`
`References Cited
`References Cited
`
`U.S. PATENT DOCUMENTS
`U.S. PATENT DOCUMENTS
`
`4,021,623
`4,021,623
`4,593,161
`4,593,161
`4,697,261
`4,697,261
`4,764,955
`4,764,955
`4,951,269
`4,951,269
`5,117,418
`5,117,418
`5,307,405
`5,307,405
`
`5/1977 Suyderhoud et al. .................. 379/411
`5/1977 Suyderhoud et al. ................ .. 379/411
`6/1986 Desblanche et al. ................... 379/410
`6/1986 Desblanche et al. .
`379/410
`9/1987 Wang et al.
`........................... 370/32.1
`9/1987 Wang et al. .... ..
`.. 370/321
`8/1988 Galand et al. .......................... 379/411
`8/1988 Galand et al.
`379/411
`8/1990 Amano et al. .......................... 367/135
`8/1990 Amano et al.
`367/135
`5/1992 Chaffee et al. ........................ 370/32.1
`5/1992 Chaffee et al.
`.. 370/321
`4/1994 Sih .......................................... 379/410
`4/1994 Sih ........................................ .. 379/410
`
`5,477,535
`12/1995 Lahdemaki ............................. 370/32.1
`5,477,535 12/1995 Lahdemaki ........................... .. 370/321
`10/1996 Genter ... .... ... ... ... ... ... .... ... ... .... 370/32.1
`5,561,668
`5,561,668 10/1996 Genter .... ..
`9/1997 Urbanski ................................. 379/406
`5,668,871
`5,668,871
`9/1997 Urbanski ..
`.. 379/406
`10/1997 Dent et al. .............................. 379/388
`5,680,450
`5,680,450 10/1997 Dent et al. ............................ .. 379/388
`12/1997 Suizu et al. ............................. 379/390
`5,696,819
`5,696,819 12/1997 SuiZu e161. ........................... .. 379/390
`12/1997 Urbanski ................................. 379/406
`5,696,821
`5,696,821 12/1997 Urbanski
`..
`12/1997 Yoshida et al. ......................... 379/390
`5,699,423
`5,699,423 12/1997 Yoshida e161. ....................... .. 379/390
`
`Primary Examiner—Krista Zele
`Primary Examiner---Krista Zele
`Assistant Examiner-Allan Hoosain
`Assistant Examiner—Allan Hoosain
`Attorney, Agent, or Firm—McDonnell Boehnen Hulbert &
`Attorney, Agent, or Firm-McDonnell Boehnen Hulbert &
`Berghoff
`Berghoff
`[57]
`[57]
`
`ABSTRACT
`ABSTRACT
`
`An Acoustic Echo Control (AEC) device operates to reduce
`An Acoustic Echo Control (AEC) device operates to reduce
`acoustic echo feedback in a speakerphone. An adaptive echo
`acoustic echo feedback in a speakerphone. An adaptive echo
`canceller responds to a far-end speech signal to generate an
`canceller responds to a far-end speech signal to generate an
`estimated echo Which is subtracted from a near-end speech
`estimated echo which is subtracted from a near-end speech
`signal, to generate a compensated near-end speech signal.
`signal, to generate a compensated near-end speech signal.
`An Echo Return Loss (ERL) estimator provides an accurate
`An Echo Return Loss (ERL) estimator provides an accurate
`but gradually adjusted estimate of the ERL of the echo
`but gradually adjusted estimate of the ERL of the echo
`canceller. A non-linear processor responds to the ERL esti(cid:173)
`canceller. A non-linear processor responds to the ERL esti
`mation and provides additional attenuation to maintain the
`mation and provides additional attenuation to maintain the
`overall system ERL. The AEC device also incorporates dual
`overall system ERL. The AEC device also incorporates dual
`talking mode detectors, one for the adaptive filter and one
`talking mode detectors, one for the adaptive ?lter and one
`for the non-linear processor, to detect a plurality of talking
`for the non-linear processor, to detect a plurality of talking
`modes, which are used to control the adaptation of the
`modes, Which are used to control the adaptation of the
`adaptive ?lter and the attenuation provided by the non-linear
`adaptive filter and the attenuation provided by the non-linear
`processor. A convergence indicator responds to the sudden
`processor. A convergence indicator responds to the sudden
`echo path changes, and quickly corrects the ERL estimation;
`echo path changes, and quickly corrects the ERL estimation;
`thereby adjusting the talking mode detector and the non
`thereby adjusting the talking mode detector and the non(cid:173)
`linear processor accordingly to maintain the overall system
`linear processor accordingly to maintain the overall system
`ERL and stability.
`ERL and stability.
`
`22 Claims, 1 Drawing Sheet
`22 Claims, 1 Drawing Sheet
`
`Remote 103
`Station
`
`r·· .. ··········· ...... ········ .. ·········· .... ·······:::::)···· .. · .. · .... ············1·13······· .. · .. ·· .. ···· .. ········ .. ··············· .. · .. ·············· .. ·1
`
`146
`
`100
`
`AlD.
`D/A
`
`alking
`detector
`NLP
`
`116
`
`Talking Mode
`detector for
`adaption control
`
`Adaptive
`flltel'
`
`20
`
`12>
`
`NO.
`D/A
`
`Detector
`
`\
`105
`\
`\
`\
`"-J
`1
`j
`!
`i
`
`i J
`~~
`
`104
`
`1 24 !
`'-- 132
`134
`:
`i. .................................................................................................................................................................. ,
`-11~
`
`101--"""'"
`
`106
`
`RTL607_1023-0001
`
`Realtek 607 Ex. 1023
`
`

`
`\C
`""-l
`~
`....
`Ul
`N
`~
`....
`0\
`
`C
`C
`C
`N
`~~
`N
`~ '?
`'JJ.
`
`~ = .....
`~ .....
`~
`•
`rJl
`d •
`
`106
`
`Fig. 1
`
`.
`124.
`
`--
`
`, ••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••• 1
`
`101---"
`
`.
`i 134.) ~'-132
`
`-11~
`
`104
`
`126
`
`: J
`
`\
`\
`\
`
`\
`105
`
`~~
`~
`J
`
`D/A
`AID,
`
`•
`
`Adaptive
`
`120 r.t;>..,
`filter
`
`Detector
`onvergencel~
`
`adaption control
`
`detector for
`Talking Mode
`
`116
`
`detector for
`alking Mod
`
`NLP
`
`ntrolle
`echo
`line
`
`D/A
`AID,
`
`r··-1~·;;5"··~~············~:··-·········Z·113· .... ·····-· ............... _ ... _ ............... -: ... !
`
`100
`
`102
`.. ,.. ....••••
`.....
`.'
`r·t PSTN)
`............... ,
`
`RTL607_1023-0002
`
`

`
`1
`1
`ECHO CONTROL DEVICE WITH QUICK
`ECHO CONTROL DEVICE WITH QUICK
`RESPONSE TO SUDDEN ECHO-PATH
`RESPONSE TO SUDDEN ECHO-PATH
`CHANGE
`CHANGE
`
`6,125,179
`6,125,179
`
`2
`2
`By the principle of orthogonality, the cross-correlation
`By the principle of orthogonality, the cross-correlation
`should be nearly Zero When the ?lter is converged. HoWever,
`should be nearly zero when the filter is converged. However,
`for a practical environment, and especially in the presence of
`for a practical environment, and especially in the presence of
`double-talk (double-talking operation), false divergence
`double-talk (double-talking operation), false divergence
`detection is frequent. This is because speech from indepen
`5 detection is frequent. This is because speech from indepen(cid:173)
`dent sources (near end, far end) has similar spectral and
`dent sources (near end, far end) has similar spectral and
`temporal characteristics, and detectors Which employ short
`temporal characteristics, and detectors which employ short-
`term estimation tend to predict a non-zero cross correlation.
`term estimation tend to predict a non-Zero cross correlation.
`10 Only if averaged for a substantial time can the cross(cid:173)
`Only if averaged for a substantial time can the cross
`10
`correlation be guaranteed to approach zero. Thus, a conver(cid:173)
`correlation be guaranteed to approach Zero. Thus, a conver
`gence detector based solely on cross-correlation is neces(cid:173)
`gence detector based solely on cross-correlation is neces
`sarily a compromise between accuracy and response time.
`sarily a compromise betWeen accuracy and response time.
`
`BACKGROUND OF THE INVENTION
`BACKGROUND OF THE INVENTION
`The present invention relates generally to voice commu(cid:173)
`The present invention relates generally to voice commu
`nication systems and, more particularly, to an electronic
`nication systems and, more particularly, to an electronic
`system for reducing acoustic echo. Speakerphones, Which
`system for reducing acoustic echo. Speakerphones, which
`employ one or more microphones together with one or more
`employ one or more microphones together With one or more
`speakers to enable “hands-free” telephone communication,
`speakers to enable "hands-free" telephone communication,
`can allow "hands-free" communications as well as partici(cid:173)
`can alloW “hands-free” communications as Well as partici
`pation in a conversation by a number of persons. Modern
`pation in a conversation by a number of persons. Modern
`speakerphones are capable of operating in a variety of
`speakerphones are capable of operating in a variety of
`modes, Which include, single-talking mode in Which the
`modes, which include, single-talking mode in which the 15
`15
`transmission of voice information is in a single direction,
`transmission of voice information is in a single direction,
`and double-talking mode, in Which the voice information is
`and double-talking mode, in which the voice information is
`transmitted by both sides, Which increases the interactivity
`transmitted by both sides, which increases the interactivity
`of the conversation, but also causes a phenomena known as
`of the conversation, but also causes a phenomena knoWn as
`"acoustic feedback echo", in which acoustic energy trans(cid:173)
`“acoustic feedback echo”, in Which acoustic energy trans
`mitted by the speaker of the speakerphone is picked up by
`mitted by the speaker of the speakerphone is picked up by
`the microphone of the same speakerphone. Typically, speak
`the microphone of the same speakerphone. Typically, speak(cid:173)
`erphones utilize an Acoustic Echo Control (AEC) device to
`erphones utiliZe an Acoustic Echo Control (AEC) device to
`reduce this echo by generating an estimate of the eXpected
`reduce this echo by generating an estimate of the expected
`feedback (“acoustic echo”) betWeen the speaker and the
`feedback ("acoustic echo") between the speaker and the
`microphone, and subtracting the eXpected echo from the
`microphone, and subtracting the expected echo from the
`signal produced by the microphone (the “near-end signal”)
`signal produced by the microphone (the "near-end signal")
`before transmission of the signal to a remote communica(cid:173)
`before transmission of the signal to a remote communica
`tions station. Generally, the AEC device is adaptive in the
`tions station. Generally, the AEC device is adaptive in the 30
`sense that changes in the acoustic echo path are accounted
`sense that changes in the acoustic echo path are accounted
`for in generating the estimated echo.
`for in generating the estimated echo.
`Typical AEC devices consist of an echo canceller filter
`Typical AEC devices consist of an echo canceller ?lter
`cascaded with a non-linear processor. The echo canceller 35
`cascaded With a non-linear processor. The echo canceller
`35
`?lter generates a linearly corrected near-end signal and the
`filter generates a linearly corrected near-end signal and the
`non-linear processor, in conjunction with a talking mode
`non-linear processor, in conjunction With a talking mode
`detector, Which detects various talking modes (single
`detector, which detects various talking modes (single(cid:173)
`talking, double-talking, etc.), provides additional echo
`talking, double-talking, etc.), provides additional echo
`attenuation for certain talking modes. The additional attenu(cid:173)
`attenuation for certain talking modes. The additional attenu
`ation provided by the nonlinear processor increases the echo
`ation provided by the nonlinear processor increases the echo
`cancellation performance, also known as the echo return loss
`cancellation performance, also knoWn as the echo return loss
`enhancement, but reduces the degree of the double-talking
`enhancement, but reduces the degree of the double-talking
`operation, therefore, reducing the interactivity of the con
`operation, therefore, reducing the interactivity of the con- 45
`45
`versation. Thus a typical echo control device strikes a
`versation. Thus a typical echo control device strikes a
`compromise betWeen the interactivity and the echo return
`compromise between the inter activity and the echo return
`loss performance.
`loss performance.
`Under steady state conditions, the echo canceller con(cid:173)
`Under steady state conditions, the echo canceller con
`verges to very nearly cancel the echo, tracking only gradual
`verges to very nearly cancel the echo, tracking only gradual 50
`changes to avoid instability. Sudden changes occurring in
`changes to avoid instability. Sudden changes occurring in
`the echo path, such as a relative repositioning of the speaker
`the echo path, such as a relative repositioning of the speaker
`and microphone disturb the system. Typical echo-cancellers
`and microphone disturb the system. Typical echo-cancellers
`and nonlinear processors are slow to respond to a sudden 55
`and nonlinear processors are sloW to respond to a sudden
`55
`change in the echo path. Thus, When a change happens, the
`change in the echo path. Thus, when a change happens, the
`system performance, such as echo return loss enhancement
`system performance, such as echo return loss enhancement
`and stability, is signi?cantly degraded until the system
`and stability, is significantly degraded until the system
`eventually re-converges.
`eventually re-converges.
`To respond to sudden changes in the echo path, some AEC 60
`To respond to sudden changes in the echo path, some AEC
`devices utilize a convergence detector to monitor the con(cid:173)
`devices utiliZe a convergence detector to monitor the con
`vergence of the echo canceller filter. Such detectors rely on
`vergence of the echo canceller ?lter. Such detectors rely on
`the principle that the adaptive ?lter Will diverge When
`the principle that the adaptive filter will diverge when
`sudden echo path changes occur. The degree of convergence
`sudden echo path changes occur. The degree of convergence 65
`65
`of the filter can be detected by examining the cross(cid:173)
`of the ?lter can be detected by examining the cross
`correlation between the estimated echo and estimation error.
`correlation betWeen the estimated echo and estimation error.
`
`SUMMARY OF THE INVENTION
`SUMMARY OF THE INVENTION
`
`In a principle aspect, the present invention takes the form
`In a principle aspect, the present invention takes the form
`of an acoustic echo device, exhibiting a high degree of
`of an acoustic echo device, exhibiting a high degree of
`stability and provides quick and accurate response to sudden
`stability and provides quick and accurate response to sudden
`changes in the echo path. The acoustic echo device includes
`20 changes in the echo path. The acoustic echo device includes
`an adaptive echo canceller which adaptively modifies a
`an adaptive echo canceller Which adaptively modi?es a
`near-end speech signal to cancel an acoustic feedback echo
`near-end speech signal to cancel an acoustic feedback echo
`component in the signal to generate a modified near-end
`component in the signal to generate a modi?ed near-end
`signal; the acoustic feedback being generated by a far-end
`signal; the acoustic feedback being generated by a far-end
`25
`speech signal received by the device. An echo return loss
`25 speech signal received by the device. An echo return loss
`estimator provides an estimate of the echo return loss of the
`estimator provides an estimate of the echo return loss of the
`echo canceller. A means, responsive to the estimated echo
`echo canceller. A means, responsive to the estimated echo
`return loss proportionally attenuates the modi?ed near-end
`return loss proportionally attenuates the modified near-end
`speech signal to substantially cancel acoustic feedback con
`speech signal to substantially cancel acoustic feedback con(cid:173)
`tained in the modi?ed near-end signal.
`tained in the modified near-end signal.
`Thus, an object of the present invention is to provide a
`Thus, an object of the present invention is to provide a
`rescue device to detect sudden echo path changes and to
`rescue device to detect sudden echo path changes and to
`adjust the non-linear processor accordingly to compensate
`adjust the non-linear processor accordingly to compensate
`for such changes. A further objection is to provide a reliable
`for such changes. A further objection is to provide a reliable
`?lter convergence detector for the various talking modes by
`filter convergence detector for the various talking modes by
`introducing an energy normalization factor. It is a further
`introducing an energy normaliZation factor. It is a further
`object to provide a unified mechanism to react to a sudden
`object to provide a uni?ed mechanism to react to a sudden
`40 echo-path change. It is a further, more specific object to be
`echo-path change. It is a further, more speci?c object to be
`able to both respond quickly to the true detection, and to
`able to both respond quickly to the true detection, and to
`recover quickly from the false detection.
`recover quickly from the false detection.
`In additional aspects, the acoustic echo control device
`In additional aspects, the acoustic echo control device
`employs a convergence detector which is responsive to
`employs a convergence detector Which is responsive to
`sudden echo-path changes and Which corrects the estimated
`sudden echo-path changes and which corrects the estimated
`echo return loss of the adaptive echo canceller to maintain
`echo return loss of the adaptive echo canceller to maintain
`the overall echo return loss and system stability.
`the overall echo return loss and system stability.
`These and other objects, features, and advantages of the
`These and other objects, features, and advantages of the
`present invention are discussed or apparent in the following
`present invention are discussed or apparent in the folloWing
`detailed description.
`detailed description.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`BRIEF DESCRIPTION OF THE DRAWINGS
`
`The preferred embodiment of the present invention is
`The preferred embodiment of the present invention is
`described herein with reference to the drawings wherein:
`described herein With reference to the draWings Wherein:
`FIG. 1 is a block diagram of an electronic system utilizing
`FIG. 1 is a block diagram of an electronic system utiliZing
`the present invention.
`the present invention.
`
`DETAILED DESCRIPTION OF THE
`DETAILED DESCRIPTION OF THE
`PREFERRED EMBODIMENT
`PREFERRED EMBODIMENT
`
`FIG. 1 of the drawings shows a block diagram of a
`FIG. 1 of the draWings shoWs a block diagram of a
`preferred AEC device 100, for use in a speakerphone, seen
`preferred AEC device 100, for use in a speakerphone, seen
`generally at 101, to reduce acoustic feedback echo, seen
`generally at 101, to reduce acoustic feedback echo, seen
`schematically at 105, from speaker 108 to microphone 106.
`schematically at 105, from speaker 108 to microphone 106.
`As seen in FIG. 1, the speakerphone 101 may be coupled via
`As seen in FIG. 1, the speakerphone 101 may be coupled via
`
`RTL607_1023-0003
`
`

`
`6,125,179
`6,125,179
`
`3
`3
`the Public Switched Telephone Network (PSTN) 102 to a
`the Public Switched Telephone Network (PSTN) 102 to a
`remote station 103 which may take the form of a conven(cid:173)
`remote station 103 Which may take the form of a conven
`tional telephone or a speakerphone. The speakerphone 101
`tional telephone or a speakerphone. The speakerphone 101
`includes a microphone 104 for detecting acoustical energy to
`includes a microphone 104 for detecting acoustical energy to
`produce a near-end speech signal 106 and a speaker 108 for
`produce a near-end speech signal 106 and a speaker 108 for 5
`generating acoustical energy in accordance With far-end
`generating acoustical energy in accordance with far-end
`speech signal 110. The far-end speech signal 110 is gener
`speech signal 110. The far-end speech signal 110 is gener(cid:173)
`ated by module 112 Which digitiZes the signal received from
`ated by module 112 which digitizes the signal received from
`the remote station 103 and cancels line echo if such echo
`the remote station 103 and cancels line echo if such echo
`eXists. Module 114 digitiZes the incoming signal from
`exists. Module 114 digitizes the incoming signal from
`microphone 104 to generate the near-end speech signal 106.
`microphone 104 to generate the near-end speech signal 106.
`Far-end speech signal 110 is adjusted at 146, in accor
`Far-end speech signal 110 is adjusted at 146, in accor(cid:173)
`dance With output generated by speaker volume control
`dance with output generated by speaker volume control
`module 144 to compensate for dynamic characteristics of
`module 144 to compensate for dynamic characteristics of 15
`15
`speaker 108. The resulting volume compensated signal 113
`speaker 108. The resulting volume compensated signal 113
`is used to drive speaker 108 and as an input to far-end speech
`is used to drive speaker 108 and as an input to far-end speech
`energy estimator 136 to detect energy in the far-end speech
`energy estimator 136 to detect energy in the far-end speech
`signal 110. The far-end speech energy estimate provided by
`signal 110. The far-end speech energy estimate provided by
`estimator 136 is used by talking mode detectors 116 and 118,
`estimator 136 is used by talking mode detectors 116 and 118, 20
`and by far-end noise ?oor estimator 140. A near-end speech
`and by far-end noise floor estimator 140. A near-end speech
`energy estimator 138 responds to compensated signal 126 to
`energy estimator 138 responds to compensated signal 126 to
`provide a near-end speech energy estimate for use by
`provide a near-end speech energy estimate for use by
`near-end noise ?oor estimator 148, talking mode detector
`near-end noise floor estimator 148, talking mode detector 25
`25
`118 and echo return loss estimator 142 and talking mode
`118 and echo return loss estimator 142 and talking mode
`detector 116. Estimators 136 and 138 preferably operate by
`detector 116. Estimators 136 and 138 preferably operate by
`taking the mean square root, or magnitude of the respective
`taking the mean square root, or magnitude of the respective
`input signal, and utiliZing a loW-pass ?lter, Which produces
`input signal, and utilizing a low-pass filter, which produces
`a value indicative of the energy contained in the respective 30
`a value indicative of the energy contained in the respective
`input signal.
`input signal.
`Dual talking mode detectors 116 and 118 detect an
`Dual talking mode detectors 116 and 118 detect an
`operational mode from a group of operational modes Which
`operational mode from a group of operational modes which
`include: (1) quiescent mode, in Which neither speakerphone
`include: (1) quiescent mode, in which neither speakerphone 35
`35
`101 nor remote station 103 is transmitting; (2) transmission
`101 nor remote station 103 is transmitting; (2) transmission
`mode, in Which speakerphone 101 is transmitting and
`mode, in which speakerphone 101 is transmitting and
`remote station 103 is quiescent; (3) receive mode, in Which
`remote station 103 is quiescent; (3) receive mode, in which
`remote station 103 is transmitting and speakerphone 101 is
`remote station 103 is transmitting and speakerphone 101 is
`quiescent; and (4) double-talking mode, in Which both
`quiescent; and (4) double-talking mode, in which both 40
`speakerphone 101 and remote station 103 are transmitting.
`speakerphone 101 and remote station 103 are transmitting.
`Talking mode detectors 116 and 118 both preferably operate
`Talking mode detectors 116 and 118 both preferably operate
`in a manner described further below.
`in a manner described further beloW.
`Separate talking mode detectors for the non-linear pro
`Separate talking mode detectors for the non-linear pro(cid:173)
`cessor 128 and the adaptive filter 120 are advantageously
`cessor 128 and the adaptive ?lter 120 are advantageously
`employed in order to provide for the different requirements
`employed in order to provide for the different requirements
`of the non-linear processor and the adaptive filter. It is
`of the non-linear processor and the adaptive ?lter. It is
`desirable to stop the adaptation of the filter 120 while the
`desirable to stop the adaptation of the ?lter 120 While the
`AEC 100 is operated in either quiescent or double-talking 50
`AEC 100 is operated in either quiescent or double-talking
`mode when the ratio of the near-end signal and far-end
`mode When the ratio of the near-end signal and far-end
`signal is greater than a predetermined value. In order to
`signal is greater than a predetermined value. In order to
`accurately do so, the mode detection for the adaptation filter
`accurately do so, the mode detection for the adaptation ?lter
`120 requires different characteristics than that for the non-
`120 requires different characteristics than that for the non
`
`4
`4
`linear processor 128. Ideally, the nonlinear processor
`linear processor 128. Ideally, the nonlinear processor
`requires quick response to reduce the cut-off of the conver(cid:173)
`requires quick response to reduce the cut-off of the conver
`sation occurring betWeen device 101 and remote station 103.
`sation occurring between device 101 and remote station 103.
`It is also less tolerant to the false detection of double-talking
`It is also less tolerant to the false detection of double-talking
`mode. A talking mode detector for the adaptive ?lter, on the
`mode. A talking mode detector for the adaptive filter, on the
`other hand, requires a third mode for the double-talking
`other hand, requires a third mode for the double-talking
`mode With loW near-end speech energy. In this mode, the
`mode with low near-end speech energy. In this mode, the
`adaptive filter is still able to converge with a smaller update
`adaptive ?lter is still able to converge With a smaller update
`gain. Thus, it is dif?cult to optimiZe a single talking mode
`10 gain. Thus, it is difficult to optimize a single talking mode
`detector for both functions.
`detector for both functions.
`Talking mode detector 116 generates an output indicative
`Talking mode detector 116 generates an output indicative
`of one of the four aforementioned talking modes to an
`of one of the four aforementioned talking modes to an
`adaptive ?lter 120 Which generates an estimated echo signal
`adaptive filter 120 which generates an estimated echo signal
`122 Which has characteristics approximately that of actual
`122 which has characteristics approximately that of actual
`echo signal seen at 105. The estimated echo signal 122 is
`echo signal seen at 105. The estimated echo signal 122 is
`subtracted at 124 from the near-end signal 106 to generate
`subtracted at 124 from the near-end signal 106 to generate
`a ?rst compensated signal 126.
`a first compensated signal 126.
`Talking mode detector 118 generates an output indicative
`Talking mode detector 118 generates an output indicative
`of one of the four aforementioned talking modes to Non(cid:173)
`of one of the four aforementioned talking modes to Non
`Linear Processor (NLP) 128 Which provides additional echo
`Linear Processor (NLP) 128 which provides additional echo
`attenuation to ?rst compensated signal 126. The output 130
`attenuation to first compensated signal 126. The output 130
`of the NLP 128 is multiplied at multiplier 132 With the ?rst
`of the NLP 128 is multiplied at multiplier 132 with the first
`compensated signal 126 to generate transmittable speech
`compensated signal 126 to generate transmittable speech
`signal 134 Which is converted into analog form by a digital
`signal 134 which is converted into analog form by a digital(cid:173)
`to-analog (D/A) converter in module 112 and transmitted
`to-analog (D/A) converter in module 112 and transmitted
`over the PSTN 102 to remote station 103.
`over the PSTN 102 to remote station 103.
`Signal 130 is advantageously generated by NLP 128 to
`Signal 130 is advantageously generated by NLP 128 to
`proportionally attenuate signal 126 based on the estimated
`proportionally attenuate signal 126 based on the estimated
`echo return loss (ERL) 149, as provided by module 142, of
`echo return loss (ERL) 149, as provided by module 142, of
`the adaptive ?lter 120. Thus, When the adaptive ?lter 120
`the adaptive filter 120. Thus, when the adaptive filter 120
`does not provide sufficient attenuation, for example during
`does not provide sufficient attenuation, for example during
`an initial condition or sudden echo-path change, then the
`an initial condition or sudden echo-path change, then the
`output of the non-linear processor changes to increase the
`output of the non-linear processor changes to increase the
`attenuation of signal 126. The enhancement of the overall
`attenuation of signal 126. The enhancement of the overall
`system stability and echo return loss are thus maintained at
`system stability and echo return loss are thus maintained at
`the cost of lower interactivity. When the filter 120
`the cost of loWer interactivity. When the ?lter 120
`converges, the output of the nonlinear processor changes to
`converges, the output of the nonlinear processor changes to
`decrease the attenuation of signal 126 in order to increase the
`decrease the attenuation of signal 126 in order to increase the
`interactivity of the conversation. To achieve an optimal
`interactivity of the conversation. To achieve an optimal
`trade-off betWeen ERL and interactivity, the additional
`45 trade-off between ERL and inter activity, the additional
`45
`attenuation provided by the nonlinear processor is advanta
`attenuation provided by the nonlinear processor is advanta(cid:173)
`geously determined by the estimated ERL 149 of the adap
`geously determined by the estimated ERL 149 of the adap(cid:173)
`tive ?lter 120 and the talking mode detected by the detector
`tive filter 120 and the talking mode detected by the detector
`118. The echo return loss estimator 142 determines the
`118. The echo return loss estimator 142 determines the
`estimated echo return loss by detecting the valley, using a
`estimated echo return loss by detecting the valley, using a
`valley detector, of the ratio betWeen the signal 126 and
`valley detector, of the ratio between the signal 126 and
`reference signal, seen at 113. The following table shows the
`reference signal, seen at 113. The folloWing table shoWs the
`operation of the talking mode detector, NLP and adaptive
`operation of the talking mode detector, NLP and adaptive
`?lter for the various talking modes of the speakerphone:
`filter for the various talking modes of the speakerphone:
`
`Talking Mode
`Talking Mode
`
`Quali?cation
`Qualification
`
`Quiescent Mode
`Quiescent Mode
`
`Ens < Lnf + TIn and
`Ens < Lnf + Tln and
`Efs < Fnf + Tfn
`Efs < Fnf + Tfn
`
`Transmitting Mode
`Transmitting Mode
`
`Ens> Lnf + TIn and
`Ens > Lnf + Tln and
`Efs < Fnf + Tfn and
`Efs < Fnf + Tfn and
`(Ens - Lnf)/(Efs - Fnf) > Ler! +
`(Ens — Lnf)/(Efs — Fnf) > Lerl +
`Ter!
`Terl
`
`Nonlinear Processor
`Nonlinear Processor
`Attenuation (Gnlp)
`Attenuation (Gnl