United States Patent 19
`Hoshuyama
`
`54) BEAMFORMER USENG COEFFICIENT
`RESTRANED ADAPTIVE FILTERS FOR
`DETECTING INTERFERENCE SIGNALS
`75 Inventor: Osamu Hoshuyama, Tokyo, Japan
`73) Assignee: NEC Corporation, Tokyo, Japan
`
`21 Appl. No.: 523,059
`22 Filed:
`Sep. 1, 1995
`30
`Foreign Application Priority Data
`Sep. 1, 1994 IJP
`Japan .................................... 6-208635
`(51
`Int. Cl. ... G01S 15/00
`52 U.S. Cl. .......................... 367/121:367/901; 367/119;
`367/905; 381/94
`58. Field of Search .............................. 367/12, 119, 121,
`367/123, 129,901, 905, 103; 128/661.01;
`381/94
`
`56)
`
`References Cited
`U.S. PATENT DOCUMENTS
`3,763,490 10/1973 Hadley et al. .......................... 342/375
`4,956,867 9/1990 Zurek et al. ........................... 38/94.1
`OTHER PUBLICATIONS
`L. Griffiths et al., “An Alternative Approach to Linearly
`Constrained Adaptive Beamforming", IEEE Transactions
`on Antennas and Propagation, vol. AP-30, No. 1, Jan. 1982,
`pp. 27-34.
`S. Nordholm et al., “The Board-Band Wiener Solution for
`Griffiths-Jim Beamformers", IEEE Transactions on Signal
`Processing, vol. 40, No. 2, Feb. 1992, pp. 474-479.
`
`
`
`HIH III
`
`USOO5627799A
`Patent Number:
`11
`45 Date of Patent:
`
`5,627,799
`May 6, 1997
`
`I. Claesson et al., “A Spatial Filtering Approach to Robust
`Adaptive Beaming", IEEE Transactions on Antennas and
`Propagation, vol. 40, No. 9, Sep. 1992, pp. 1093-1096.
`"Processing Signals Carried By Propagating Waves". Mul
`tidimensional Digital Signal Processing, Prentice-Hall,
`Inc., pp. 289-315.
`M.M. Goodwin et al., "Constant Beamwidth Beamforming".
`Proceedings of International Conference on Acoustics,
`Speech and Signal Processing 93, pp. 1-169-1-172.
`Primary Examiner Ian J. Lobo
`Attorney, Agent, or Firm-Sughrue, Mion, Zinn, Macpeak &
`Seas
`ABSTRACT
`57
`In an adaptive array beamformer, a spatial beamforming
`filter is connected to a sensor array for respectively filtering
`and summing array signals to produce a first filter output
`containing a target signal that arrives in a specified direction.
`First adaptive filters provide transversal-filtering the first
`filter output to produce a second filter output not containing
`the target signal, using a first error signal by restraining their
`tap weight coefficients. The array signals are further coupled
`to subtractors. Each subtractor detects a difference between
`the second filter output of the corresponding first adaptive
`filter and the corresponding sensor signal to derive the first
`error signal. Second adaptive filters provide transversal
`filtering the first error signals of the subtractors to produce
`third filter outputs, using a second error signal, by restrain
`ing their tap weight coefficients. The third filter outputs are
`summed and subtracted from the first filter output to produce
`an output of the beamformer, which is supplied as the second
`error signal to the second adaptive filters
`
`10 Claims, 11 Drawing Sheets
`
`SPATA
`LOWPASS
`FILTER
`
`SPATAL
`HGHPASS
`FTER
`
`Meta Platforms, Inc. Exhibit 1008
`Page 1 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 1 of 11
`
`5,627,799
`
`FG,
`1
`PRIOR ART
`SPATAL LOWPASS FILTER 2
`
`
`
`TAP-WEIGHT MEMORY
`
`HIGHPASS
`
`LAF = leaky Adaptive Filter
`
`Meta Platforms, Inc. Exhibit 1008
`Page 2 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 2 of 11
`
`5,627,799
`
`FIG. 2 PRIOR ART
`
`SPATAL HIGHPASS FILTER BANK 6
`
`
`
`FILTER
`
`FER
`
`FILTER
`
`Meta Platforms, Inc. Exhibit 1008
`Page 3 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 3 of 11
`
`5,627,799
`
`FG, 3
`PRIOR ART
`32
`GE
`
`FROM DELAY 3
`
`
`
`300
`
`
`
`--------------------- ~~~– – – – – – – – ~- - ---------- – – – – – – – – – – – – – – -1-~~~~- - - - - - - - - - - )
`
`PTIVE FILTER
`EAKY ADA
`
`EAKY ADAPTIVE FILTER
`
`Meta Platforms, Inc. Exhibit 1008
`Page 4 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 4 of 11
`
`5,627,799
`
`
`
`FG, 4
`
`HGHPASS
`
`- ----? ? ? ? ---- » ~~~~• • • • • • • • ?= *
`
`----------------------
`
`Meta Platforms, Inc. Exhibit 1008
`Page 5 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 5 of 11
`
`5,627,799
`
`
`
`FG, 5
`
`NCAF = Norm-Constrained Adaptive Filter
`
`Meta Platforms, Inc. Exhibit 1008
`Page 6 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 6 of 11
`
`5,627,799
`
`FG, 6
`12
`FROMDELAY 3-GE
`
`
`
`
`
`CONSTRANT
`COEFFICENT
`GENERATOR
`
`- A D
`11
`
`FROM
`SUBTR
`ACTORS
`90-9M-1
`
`NORM.CONSTRANED ADAPTIVE FILTER
`13
`
`13M-1
`NORM-CONSTRANED ADAPTIVE FILTER
`
`Meta Platforms, Inc. Exhibit 1008
`Page 7 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 7 of 11
`
`5,627,799
`
`FROM
`UPDATE
`cuts
`
`
`
`
`
`
`
`
`
`
`
`FG, 10
`OUTPUT
`
`74
`
`DIVISION
`CIRCUIT
`
`MNMUM
`SELECTOR
`
`INPUT
`
`Meta Platforms, Inc. Exhibit 1008
`Page 8 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 8 of 11
`
`5,627,799
`
`
`
`FG, 8
`
`- A
`FILTER
`
`CCAF = Coefficient-Constrained Adaptive Filter
`
`Meta Platforms, Inc. Exhibit 1008
`Page 9 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 9 of 11
`
`5,627,799
`
`14
`
`FROMSUBTRACTOR 9,
`
`COEFFICIENT.
`CONSTRAINE)
`ADAPTIVE FILTER
`
`- - - -
`
`A
`
`
`
`(x)
`
`1
`
`FROM - - - -
`FILTER 2
`
`- ----------
`
`TO SUBR
`ACTOR 9
`
`j
`
`Meta Platforms, Inc. Exhibit 1008
`Page 10 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 10 of 11
`
`5,627,799
`
`
`
`FIG 11
`
`SPATAL
`
`CCAF = Coefficient-Constrained Adaptive Filter
`NCAF = Norm-Constrained Adaptive Filter
`
`Meta Platforms, Inc. Exhibit 1008
`Page 11 of 17
`
`

`

`U.S. Patent
`
`May 6, 1997
`
`Sheet 11 of 11
`
`5,627,799
`
`FG, 12
`
`2
`
`
`
`
`
`SPATAL
`LOWPASS
`FILTER
`
`TO DELAY 3
`
`
`
`SPATAL
`LOWPASS
`FILTER
`
`REFERENCE
`SIGNAL
`
`TO ADAPTIVE FILTERS
`80-8M-10R 140-14M
`
`lM-1
`
`/
`
`
`
`
`
`
`
`
`
`TO SUBTRACTORS 90 - 9M-1
`
`Meta Platforms, Inc. Exhibit 1008
`Page 12 of 17
`
`

`

`5,627,799
`
`1.
`BEAMFORMER USING COEFFICIENT
`RESTRANED ADAPTIVE FILTERS FOR
`DETECTING INTERFERENCE SIGNALS
`
`2
`canceling capability. The difference between the assumed
`direction and the actual arrival direction of the target signal,
`or a look-direction error, is of another concern because it
`degrades the target signal, or a look-direction error, is of
`another concern because it degrades the target signal. In
`order to compensate for this shortcoming, the spatial high
`pass filter bank of the prior art needs as many spatial
`highpass filters as is necessary to provide a wide range of
`angles to reject the target signal to prevent its leakage into
`the interference path of the beamformer.
`SUMMARY OF THE INVENTION
`It is therefore an object of the present invention to provide
`an adaptive array beamformer with a reduced number of
`sensors while allowing a look-direction error.
`According to the present invention, there is provided an
`adaptive array beamformer comprising an array of spatially
`distributed sensors, and a spatial beamforming filter con
`nected to the sensors for respectively filtering output signals
`of the sensors and summing the filtered output signals to
`produce a first filter output containing a target signal arriving
`at the array in a specified direction. A plurality of first
`adaptive filters are provided, each having a tapped-delay line
`connected to receive the first filter output, a coefficient
`update circuit for producing tap weight coefficients indicat
`ing correlations between tap signals from the tapped-delay
`line and a first error signal applied thereto, a plurality of
`multipliers for weighting the tap signals with the
`coefficients, respectively, and means for summing the
`weighted tap signals to produce a second filter output not
`containing the target signal. The coefficient update means
`includes restraining means for preventing the coefficients
`from increasing indefinitely. A plurality of first subtractors
`are provided, each detecting a difference between a corre
`sponding sensor signal and the second filter output of the
`corresponding first adaptive filter and supplying the differ
`ence to the coefficient update circuit of the corresponding
`first adaptive filter as the first error signal. A plurality of
`second adaptive filters are provided, each having a tapped
`delay line connected to receive the error signal from a
`corresponding one of the first subtractors, a coefficient
`update circuit for producing tap weight coefficients indicat
`ing correlations between tap signals from the tapped-delay
`line and a second error signal applied thereto, a multiply
`and-sum circuit for weighting the tap signals with the
`coefficients respectively and Summing the weighted tap
`signals to produce a third filter output. The coefficient update
`circuit includes restraining means for preventing the coef
`ficients from increasing indefinitely. An adder is provided
`for summing the third filter outputs from the second adaptive
`filters. A second subtractor detects a difference between the
`first filter output and the output of the adder and supplying
`the difference to the coefficient update circuit of the second
`adaptive filters as the second error signal.
`In a preferred embodiment, a second spatial beamforming
`filter is connected to the sensors for respectively filtering
`output signals of the sensors and summing the filtered output
`signals to produce a secondfilter output containing the target
`signal, the second spatial beamforming filter having a
`greater beam width than a beam width of the first spatial
`beamforming filter. The first adaptive filters are connected to
`the output of the second spatial beamforming filter, instead
`of to the output of the first-named spatial beamforming filter.
`BRIEF DESCRIPTION OF THE DRAWINGS
`The present invention will be described in further detail
`with reference to the accompanying drawings, in which:
`
`BACKGROUND OF THE INVENTION
`1. Field of the Invention
`The present invention relates generally to interference
`cancelers, and more particularly to a generalized sidelobe
`canceler, or adaptive beamformer for an array of sensors
`such as microphones and the like.
`2. Description of the Related Art
`It is known that wideband signals propagating across an
`array of sensors in directions that are different than the beam
`steering direction of the array suffer a distortion that is
`similar to lowpass filtering.
`According to a prior art microphone array, signals
`detected by an array of microphones are lowpass filtered and
`Summed together to detect a target signal that arrives in a
`particular direction. The adaptive microphone array beam
`former is one form of the generalized sidelobe canceler as
`described in an article "An alternative Approach to Linearly
`Constrained Adaptive Beamforming", Lloyd J. Griffiths and
`Charles W. Jim, the IEEE Transactions on Antenna and
`Propagation, Vol. AP-30, No. 1, January 1982, pages 27–34.
`As described in an article "The Broad-Band Wiener Solution
`for Griffiths-Jim Beamformers', S. Nordholm, I. Claesson
`and P. Eriksson, the IEEE Transactions on signal Processing,
`Vol. 40, No. 2, February 1992, pages 474-478 (hereinafter
`referred to as Document 1), the generalized sidelobe can
`celer comprises, a spatial lowpass filter connected to an
`array of microphones for filtering signals from the array and
`Summing the filtered signals so that only the desired signal
`is contained in the summed signal. A plurality of spatial
`highpass filters are provided to form a spatial highpass filter
`bank. Each spatial highpass filer is connected to a selected
`pair of microphones for filtering and Summing the sensor
`signals to detect the interference signals. A plurality of
`adaptive filters are provided for using the interference sig
`nals as reference signals to detect those components having
`high correlation with the interference signals contained in
`the detected target signal.
`Since the spatial highpass filters of Document 1 are of
`nonadaptive type and each uses two microphone outputs, the
`range of signals which must be rejected is very narrow. As
`a result, a slight departure from the intended direction causes
`a leakage of the desired signal into the interference path of
`the beamformer.
`To overcome the prior art shortcoming, a proposal has
`been made to implement a spatial highpass filter for receiv
`ing more than two microphone outputs as described in an
`article "A Spatial Filtering Approach to Robust Adaptive
`Beaming”, I. Claesson et al., the IEEE Transactions on
`Antennas and Propagation, Vol. 40, No. 9, September 1992,
`pages 1093 to 1096 (hereinafter referred to as Document 2).
`According to Document 2, each of the highpass filters that
`comprise the spatial highpass filter broadens the range of
`arrival angles by receiving multiple spatial samples from a
`selected set of microphone outputs using a plurality of leaky
`adaptive filters.
`However, a large number of microphones (the Qvalue)
`are required to implement a beamformer having a wide
`range of rejection angles, for each group of spatial highpass
`filters in the filter bank. If a sufficient number of micro
`phones is not provided, the degree of design freedom must
`be sacrificed, resulting in a beamformer having a low noise
`
`15
`
`25
`
`30
`
`35
`
`45
`
`50
`
`55
`
`65
`
`Meta Platforms, Inc. Exhibit 1008
`Page 13 of 17
`
`

`

`5,627,799
`
`5
`
`O
`
`15
`
`20
`
`30
`
`3
`FIG. 1 is a block diagram of a prior art adaptive array
`beamformer;
`FIG. 2 is a block diagram of the spatial highpass filter of
`the FIG. 1 prior art;
`FIG. 3 is a block diagram of the leaky adaptive filters of
`the FIG. 1 prior art;
`FIG. 4 is a block diagram of an adaptive array beam
`former according to a first embodiment of the present
`invention;
`FIG. 5 is a block diagram of an adaptive array beam
`former according to a second embodiment of the present
`invention;
`FIG. 6 is a block diagram of the norm constraint adaptive
`filters of the second embodiment;
`FIG. 7 is a block diagram of the constraint coefficient
`generator used in FIG. 6;
`FIG. 8 is a block diagram of an adaptive array beam
`former according to a third embodiment of the present
`invention;
`FIG. 9 is a block diagram of the coefficient-constrained
`adaptive filters of the third embodiment;
`FIG. 10 is a graphic representation of the input/output
`characteristic of the limiters of FIG. 9;
`25
`FIG. 11 is a block diagram of an adaptive array beam
`former according to a fourth embodiment of the present
`invention; and
`FIG. 12 is a block diagram of a modification of the present
`invention.
`
`35
`
`45
`
`DETALED DESCRIPTION
`Before proceeding with the detailed description of the
`present invention, it may provide helpful to provide an
`explanation of the prior art with reference to FIGS. 1 to 3.
`In FIG. 1, a linear array of microphones 1-1 of identical
`operating characteristics are located at sufficient distances
`from signal sources of interest so that the wavefront of each
`signal at the microphones is considered to be linear. The
`microphones are connected to FIR transversal filters
`40
`20-20 of a spatial lowpass filter 2, the outputs of the
`filters 20 being summed by an adder 26 to produce an output
`containing the target signal from a particular (assumed)
`direction and signals from other directions which are uncor
`related with the target signal. The outputs of filters 20 are
`applied through a timing adjustment delay circuit 3 to a
`Subtractor 32 of a canceler 4.
`The outputs of the M microphones are further connected
`to a spatial highpass filter bank 6 to produce (M-Q-1)
`output signals. The filter bank 6 operates so that the signals
`including the target signal as well as signals in the neigh
`borhood of the assumed direction are rejected. The outputs
`of the filter bank 6 thus contain the undesired signals as
`dominant components. The outputs of filter bank 6 are fed
`through leads Fo-F,
`to leaky adaptive filters 300-30
`of the canceler 4. Leaky adaptive filters 30 of the canceler
`detect undesired signals contained in the output signal of the
`beamformer atterminal 5 having a high correlation with the
`undesired signals detected by the spatial highpass filter 6 by
`adaptively updating their tap weight values using the output
`of the beamformer as a signal indicating the amount of
`correction error. The high correlation signals detected by the
`leaky adaptive filters 30 are combined by an adder 31 and
`fed to the subtractor 32 where it is subtracted from the
`time-coincident signal from spatiallowpass filter 2, whereby
`the undesired signals are canceled at the outputterminal 5 of
`the beamformer.
`
`50
`
`55
`
`65
`
`4
`Each of the filters 20 has a tapped delay line formed by
`delay elements 22-22 forming (G-1) delay-line taps
`which are connected to corresponding tap weight multipliers
`23 for respectively weighting the tap signals with particular
`tap weight coefficients supplied from a tap weight memory
`24 (where G is equal to or greater than 2), the weighted tap
`signals being summed by an adder 25 and fed to the adder
`26. The tap weight memory 24 of each filter 20 stores a set
`of tap weight coefficients whose values are determined so
`that filters 20 exhibit particular characteristics which result
`in an output containing the target signal. If the assumed
`direction is normal to the length of the microphone array, the
`integer G-2 is used and the tap weight coefficient of the
`multiplier 23 is set equal to "1". Other design approaches
`are described in "Multidimensional Digital Signal
`Processing", Prentice-Hall, Inc, pages 289-315, 1984 and
`IEEE, Proceedings of International Conference on
`Acoustics, Speech and Signal Processing 93, pages
`169-172.
`Spatial highpass filter bank 6 of the type described in
`Document 2 is shown in FIG. 2. Filter bank 6 is made up
`of(M-Q-1) groups 40 of Q highpass filters 41 each, and an
`adder 42, which each group forming a spatial highpass filter,
`where Q is equal to or greater than "3". Each spatial filter 40
`receives a selected set of the microphone outputs such that
`the signals from the microphones positioned closer to the
`center of the array are coupled to an increasing number of
`filters 41. Thus, the signals incident on the center area of the
`microphone array fare filtered through a greater number of
`filters 41 than the signals incident on the edges of the array
`are. Highpass filters 41 are basically of the same transversal
`filter configuration as the filters 20, but with different delay
`line lengths (G) and different filter characteristics.
`The characteristics of the highpass filters 41 offilter bank
`6 are those of a rejection filter wherein a group of signals
`propagating in the assumed direction are rejected at the
`output of adder 42 of each spatial highpass filter 40. A basic
`design method for this type of spatial filter is described in
`Document 2. One important consideration is the degree of
`design freedom which is determined by the number of
`microphones used. For an M-microphone array, it is repre
`sented by M-Q+1. With the use of a large number of
`microphones a beamformer having a wide rejection angle
`with high attenuation can be implemented. Advantageously,
`the target signal can be rejected in the interference path of
`such beamformers even though the assumed direction differs
`from its actual arrival direction.
`In each of the leaky adaptive filters 30 (FIG. 3), a
`corresponding output signal from the filter bank 5 is suc
`cessively shifted through delay-line taps formed by delay
`elements 50-502 and the tap signals are weighted respec
`tively by (L-1) multipliers 51 with tap weight coefficients
`Supplied from update circuits 53-53, and then summed
`by adder 52 for coupling to the adder 31. Each update circuit
`53 operates in accordance with the least mean square (LMS)
`algorithm. The output of beamformer from subtractor 32,
`representing a correction error, is weighted by a stepsize pu
`in a multiplier 54 and applied to a multiplier 55 of each
`update circuit 53 for detecting a correlation between the
`weighted error and a corresponding tap signal. Each update
`circuit 53 includes a leaky integrator formed by an adder 56,
`a multiplier 57 and a delay element 58. The correlation
`output of multiplier 55 is summed with a feedback signal
`from multiplier 57 and delayed by a symbol interval by
`delay element 58. The delayed symbol is applied to the
`corresponding tap weight multiplier 51 as an updated tap
`weight coefficient as well as to the multiplier 57 where it is
`
`Meta Platforms, Inc. Exhibit 1008
`Page 14 of 17
`
`

`

`S
`scaled down by a factor ol, (equal to less than unity) and fed
`back to the adder 56. Because of this scale-down feedback,
`the integrator operates as a leaky integrator which differs
`from normal integrators where the scale factor is unity. The
`leaky integration prevents the tap weight coefficient from
`growing indefinitely when the target signal, when there is a
`leakage of the target signal to the interference path (i.e., the
`outputs of filter bank 6) of the beamformer due to the
`inherent variability of microphone characteristics and posi
`tional errors of the microphones. Otherwise, the interference
`signals produced by the adaptive filters would become
`identical to the components of the signal in the main path of
`the beam former, and the resulting cancellation would sub
`stantially remove the target signal.
`However, in order to implement a beamformer having a
`wide range of rejection angles, a large number of micro
`phones (the Qvalue) are required for each group of spatial
`highpass filters in the filter bank. If a sufficient number of
`microphones is not provided, the degree of design freedom
`must be sacrificed, resulting in a beamformer having a low
`noise cancelling capability.
`Referring now to FIG. 4, there is shown an adaptive array
`beamformer according to a first embodiment of the present
`invention in which parts corresponding to those of FIG.1 are
`marked by the same numerals as those used in FIG. 1, the
`description thereof being omitted for simplicity. The adap
`tive array beamformer of this embodiment comprises a
`spatial highpass filter 16 and a canceler 17. Spatial highpass
`filters 16 includes M delay circuits 7-7
`connected
`respectively to the microphones 1-1, M leaky adaptive
`filters 8-8
`and M subtractors 9-9
`connected
`respectively to the outputs of the M delay circuits 7.
`The spatial lowpass filter 2, connected to the microphone
`array, provides spatial lowpass filtering of the individual
`microphone signals and summing the lowpass-filtered sig
`nals in the same manner as in the prior art beamformer to
`detect the target signal. The output of the spatial lowpass
`filter 2 is applied to all the leaky adaptive filters 8 as a
`reference signal as well as to the delay 3. The outputs the
`microphone array are passed through corresponding delay
`circuits 7 to subtractors 9 to which the outputs of leaky
`adaptive filters 8 are also supplied to be subtracted from the
`corresponding microphone outputs. The output of each
`subtractor 9 is coupled to the corresponding leaky adaptive
`filter 8 as an error signal to update their tap weight values.
`The M delay circuits 7 provide a delay to the microphone
`outputs so that they are time coincident at the inputs of
`corresponding subtractors 9 with the output signals of leaky
`adaptive filters 8.
`Each of the leaky adaptive filters 8 is identical in structure
`to that shown in FIG. 3. Correlations between the reference
`signal and each of the error signals are detected by the leaky
`adaptive filters 8. As described previously in connection
`with the prior art, the strength of a leaky adaptive filter for
`restraining the growth of tap weight is proportional to the
`magnitude of the tap weight value itself. As a result, if the
`optimum value for the tap weight coefficient (which mini
`mizes the error input of the leaky adaptive filter) is relatively
`large, the tap weight value cannot converge to the optimum
`value, resulting in a substantial amount of error from the
`optimum value. This implies that depending on the tap
`weight value the correlation capability of the leaky adaptive
`filters 8 differs significantly. Therefore, those signal
`components, which require a greater tap weight value for
`enabling their correlation to be detected, cannot sufficiently
`be removed, while those signals requiring allower tap weight
`value can be removed sufficiently.
`
`40
`
`45
`
`50
`
`55
`
`65
`
`5,627,799
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`6
`With respect to the signal arriving in the assumed direc
`tion as well as to those arriving in near-assumed directions,
`the output of spatial lowpass filter 2 contains the same
`amount of such signal components as those detected by the
`microphone array, and the maximum tap weight value
`necessary for removing them from the interference path of
`the beamformer is as small as “1”. The leaky adaptive filters
`8 are therefore designed with a low maximum tap weight
`value so that the target signal components are completely
`removed at the outputs of subtractors 9.
`With respect to the interference signals, on the other hand,
`the output of the spatial lowpass filter 2 contains a smaller
`amount of interference signals than those detected by the
`microphone array. Therefore, the tap weight value necessary
`for the leaky adaptive filters 8 to remove the interference
`signals is much higher than “1”. Thus, the amount of
`removal at the outputs of subtractors 9 is much less in the
`case of the interference signals than in the case of the target
`signal components. If normal adaptive filters are used
`instead of the leaky adaptive filters 9, their tap weight values
`would be allowed to grow indefinitely, and as a result, not
`only the interference signals but the target signal compo
`ments are removed.
`Canceler 17 includes M leaky adaptive filters 10-10
`connected respectively to the outputs of corresponding sub
`tractors 9 to receive the interference signals detected in a
`manner just described. Each of the leaky adaptive filters 10
`is identical in characteristic to the prior art leaky adaptive
`filters. Although most of the target signal components are
`removed, there is still a small amount of their leakage at the
`outputs of subtractors 9. Due to the adaptive leaky integra
`tion offilters 10, the growth of their tap weight values due
`to the presence of such small amount of leakage of the target
`signal are restrained. The outputs of leaky adaptive filters 10
`are summed by adder 11 and supplied to subtractor 12 for
`canceling the interference signals contained in the main path
`of the beamformer.
`Since the output of each subtractor 9 contains only a small
`amount of the target signal, the latter is not canceled in the
`subtractor 12 even though there is a look-direction error.
`the leaky adaptive filters 8 of the spatial highpass filter 16
`operate in effect as variable spatial highpass filters. The
`degree of design freedom of the present embodiment is not
`less than that of Document 2 and a large look-direction error
`is allowed using a smaller number of microphones than in
`the case of Document 2.
`A second embodiment of the present invention is shown
`in FIG. 5 in which parts corresponding to those in FIG. 4 are
`marked with the same numeral as those used in FIG. 4. The
`beamformer of FIG. S differs from the first embodiment in
`that the leaky adaptive filters 10 of FIG. 4 are replaced with
`norm-constrained adaptive filters 13-13.
`As shown in detail in FIG. 6, each norm-constrained
`adaptive filter 13 comprises a tapped-delay line formed by
`delay elements 600-60
`tap weight multipliers 61-61
`connected to the delay-line taps, and adder 62 for summing
`the weighted tap signals. Update circuits 630-63,
`are
`provided which are connected to a constraint coefficient
`generator 69. Each update circuit 63 receives an error signal
`from the output of the beamformer from subtractor 12 via
`multiplier 64 where it weighted by the stepsize u. Correla
`tion between the corresponding tap signal and the weighted
`error signalis taken by a multiplier 65 and summed by adder
`66 with a tap weight value of a previous sample supplied
`from multiplier 67. The output of delay element 68 is scaled
`down by multiplier 67 with a constraint control parameter B
`
`Meta Platforms, Inc. Exhibit 1008
`Page 15 of 17
`
`

`

`5,627,799
`
`7
`from the constraint coefficient generator 69. To the con
`straint coefficient generator 69 is connected the output of
`multiplier 65 of each update circuit 63. The output of the
`adder 66 is supplied to the constraint coefficient generator 69
`as the output of the update circuit 63.
`Constraint coefficient generator 69 controls the constraint
`control parameter B such that the p-th power of norm L.
`(where p is an integer equal to or greater than unity) of the
`tap weight coefficients does not exceed a positive integer O
`using the following Equation:
`
`O
`
`(1)
`
`8
`A second spatial lowpass filter 15 may be provided as
`shown in FIG. 12. This second filter is connected to the
`microphone array to produce a signal which can be used as
`a reference signal, instead of using the output of the first
`spatiallowpass filter 2, for the leaky adaptive filters 8 as well
`as for the coefficient-constrained adaptive filters 14 of the
`previous embodiments. In this embodiment, the first spatial
`lowpass filter 2 is designed to form a mainlobe of greater
`widthin the assumed direction in comparison with the width
`of the mainlobe formed by spatiallowpass filter 15. With the
`wider mainlobe of the first spatial lowpass filter 2, the
`overall characteristic of the beamformer is fit to the char
`acteristic of this filter. This arrangement is particularly
`useful when there is a large look-direction error.
`What is claimed is:
`1. An adaptive array beamformer comprising:
`an array of spatially distributed sensors;
`a spatial beamforming filter connected to said sensors for
`respectively filtering output signals of the sensors and
`summing the filtered output signals to produce a first
`filter output containing a target signal arriving at said
`array in a specified direction;
`a plurality of first adaptive filters, each having a tapped
`delay line connected to receive said first filter output,
`coefficient update means for producing tap weight
`coefficients indicating correlations between tap signals
`from the tapped-delay line and a first error signal
`applied thereto, a multiply-and-sum circuit for weight
`ing said tap signals with said coefficients respectively
`and Summing the weighted tap signals to produce a
`second filter output not containing said target signal,
`said coefficient update means including restraining
`means for preventing said coefficients from increasing
`indefinitely;
`a plurality of first subtractors, each detecting a difference
`between an output signal of a corresponding one of said
`sensors and the second filter output of a corresponding
`one of said first adaptive filters and supplying the
`difference to the coefficient update means of the cor
`responding first adaptive filter as said first error signal;
`a plurality of second adaptive filters, each having a
`tapped-delay line connected to receive said first error
`signal from a corresponding one of said first
`Subtractors, coefficient update means for producing tap
`weight coefficients indicating correlations between tap
`signals from the tapped-delay line and a second error
`signal applied thereto, a multiply-and-sum circuit for
`weighting said tap signals with said coefficients respec
`tively and summing the weighted tap signals to produce
`a third filter output, said coefficient update means
`including restraining means for preventing said coef
`ficients from increasing indefinitely;
`an adder for summing the third filter outputs from the
`second adaptive filters; and
`a second subtractor for detecting a difference between the
`first filter output and a summed signal from said adder
`and supplying the difference to the coefficient update
`means of said second adaptive filters as said second
`error signal.
`2. An adaptive array beamformer comprising:
`an array of spatially distributed sensors;
`a first spatial beamforning filter connected to said sensors
`for respectively filtering output signals of the sensors
`and Summing the filtered output signals to produce a
`first filter output containing a target signal arriving at
`said array in a specified direction;
`
`where w is the tap weight coefficient at the i-th delay-line
`tap. By constraining the Lp value below the O-value, the
`growth of tap weights is restrained.
`As shown in detail in FIG. 7, the constraint coefficient
`generator 69 includes a calculator 70 for calculating the p-th
`power of norms. This calculator is formed by a plurality of
`circuits 71-71 for raising the corresponding outputs of
`the update cir