IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002
1583
Adaptive Null-Forming Scheme in
Digital Hearing Aids
Fa-Long Luo, Senior Member, IEEE, Jun Yang, Senior Member, IEEE, Chaslav Pavlovic, and
Arye Nehorai, Fellow, IEEE
Abstract—We propose an effective adaptive null-forming
scheme for two nearby microphones in endfire orientation that
are used in digital hearing aids and in many other hearing devices.
This adaptive null-forming scheme is mainly based on an adaptive
combination of two fixed polar patterns that act to make the null
of the combined polar pattern of the system output always be
toward the direction of the noise. The adaptive combination of
these two fixed polar patterns is accomplished by simply updating
an adaptive gain following the output of the first polar pattern
unit. The value of this gain is updated by minimizing the power
of the system output, and related adaptive algorithms to update
this gain are also given in this paper. We have implemented this
proposed system on the basis of a programmable DSP chip and
performed various tests. Theoretical analyses and testing results
demonstrated the effectiveness of the proposed system and the
accuracy of its implementation.
Fig. 1. Common direction processing system with two omnidirectional
microphones in endfire orientation.
Index Terms—Adaptive signal processing, array signal processing, beamforming, hearing aids, microphones, noise reduction,
speech enhancement.
I. INTRODUCTION
H
EARING devices with directionality make use of the direction difference between the target signal and the noise.
There are two types of directional devices: one with fixed directionality (or, say, with the fixed polar pattern) [1]–[4] and
the other with adaptive directionality that can track the varying
or moving noise sources [5]–[8]. Many techniques have been
used to achieve directionality in both fixed mode and adaptive
mode [3], [7], [9], [10]. However, most of these techniques can
be neither immediately employed nor implemented in instruments such as hearing aids because of the limit of hardware
size, the number and distance of microphones, computational
speed, mismatch of microphones, power supply, and other practical factors relating to these hearing instruments [7], [11], [12].
Manuscript received September 18, 2000; revised March 13, 2002. The work
of A. Nehorai was supported by the Air Force Office of Scientific Research
under Grants F49620-99-1-0067 and F49620-00-1-0083, the National Science
Foundation under Grant CCR-0105334, and the Office of Naval Research under
Grant N00014-01-1-0681. The associate editor coordinating the review of this
paper and approving it for publication was Dr. Brian Sadler.
F.-L. Luo was with the R&D Department, GN ReSound Corporation, Redwood City, CA 94063 USA. He is now with Quicksilver Technology, Inc., San
Jose, CA 95119 USA.
J. Yang was with the R&D Department, GN ReSound Corporation, Redwood
City, CA 94063 USA. He is now with ForteMedia, Inc., Cupertino, CA 95014
USA.
C. Pavlovic was with the R&D Department, GN ReSound Corporation, Redwood City, CA 94063 USA. He is now with Sound ID, Palo Alto, CA 94303
USA
A. Nehorai is with the Electrical and Computer Engineering Department, University of Illinois at Chicago, Chicago, IL 60607 USA.
Publisher Item Identifier S 1053-587X(02)05634-9.
Fig. 2. Three typical polar patterns obtained by the system in Fig. 1 with
using different delay values. From left to right, the patterns are bidirectional,
hypercardioid, and cardioid.
For example, in common hearing aids such as behind-the-ear
aids, there can be only two microphones, and the distance between these two microphones is only about 10 mm [13].
The most common technique in use in hearing aids is a directional microphone or a dual-omnimicrophone system with some
fixed polar patterns, as shown in Fig. 1. The directional system
in Fig. 1 can provide different polar patterns by selecting different values of delay . By way of example, Fig. 2 shows three
,
polar patterns with the value of delay being 0,
, respectively, where is the distance between the two
and
microphones, and is the sound speed. The direction directly
in front of the hearing-aid wearer is represented as 0 , whereas
180 represents the direction directly behind the wearer. The
thick line stands for the gain as a function of direction of the
sound arrival where the gain from any given direction is represented by the distance from the center of the circle. These
three polar patterns are called bidirectional pattern (with null
at 90 and 270 ), hypercardioid pattern (with null at 110 and
250 ), and cardioid pattern (with null at 180 ), respectively. Obviously, the cardioid system attenuates sound the most from directly behind the wearer, whereas the bidirectional system attenuates sound the most from directly to the left and to the right of
the wearer. In different listening environments, users select one
of these three polar patterns using control buttons to achieve the
best noise reduction performance, given the specific listening
1053-587X/02$17.00 © 2002 IEEE
1584
environment. However, for time-varying and moving-noise environments, this fixed directional system delivers degraded performance and therefore, the system with adaptive directionality
is highly desirable.
For a system with two nearby microphones in endfire orientation, the direct way to achieve adaptive directionality is to adaptively change the delay of the system in Fig. 1 so that its value
is equal to the transmission delay value of the noise between the
two microphones. From a performance and complexity point of
view, the key problem in this method is in effectively estimating
the delay value of the noise when noise and the target speech are
both present. Another problem related to this method is how to
implement this delay unit in real time since this delay is usually
a fractional sample delay. For example, if the sampling rate is
16 000 Hz, one sample interval is 62.5 s, but the largest transmission delay value of the noise between the two microphones
and about 29.1 s with
mm. In the scheme of
is
Fig. 1 with fixed mode, this fractional-sample delay is usually
implemented during the A/D stage by processing that is similar
to an up-sampling-shift method [13]. In the adaptive mode, the
up-sampling-shift method is no longer suitable, and the adaptive implementation of a fractional-sample delay is instead accomplished by adaptively updating the coefficients of a specific
filter (ideally, an all-pass filter with linear phase) [14]. These
problems prevent this idea from being of practical value in a
hardware implementation.
On the basis of these problems, this paper proposes a more
practical and effective adaptive directionality system for hearing
devices with two nearby endfire orientation microphones. This
adaptive directionality system is based mainly on an adaptive
combination of two fixed polar patterns that are arranged to
make the null of the combined polar pattern of the system output
always be toward the direction of the noise. The null of one of
these two fixed polar patterns is at 0 (straight ahead of the subject) and the other’s null is at 180 . Both polar patterns are cardioid. The first fixed polar pattern can be implemented by deand
laying the front microphone signal with the value being
subtracting it from the rear microphone signal. Likewise, the
second polar pattern is implemented by delaying the rear miand subtracting it from
crophone signal with the value being
the front microphone signal. The adaptive combination of these
two fixed polar patterns is accomplished by adding an adaptive
gain following the output of the first polar pattern. Different gain
values will provide the combined polar pattern with nulls at different degrees. The value of this gain is updated by minimizing
the power of the system output. Related adaptive algorithms to
update this gain are given in this paper. The hardware implementation of this proposed system has been completed and tested.
The test results demonstrated the accuracy and effectiveness of
the proposed scheme.
It should be noted that for systems with two nearby microphones, there have been also some very effective algorithms
such as the scheme proposed by Vanden Berghe and Wouters
[11] and the scheme proposed by Elko and Pong [15]. In comparison with the two-stage scheme in [11], our proposed system
and algorithm are very simple because ours requires only an
adaptive gain (one-tap adaptive filter).
As a matter of fact, the structure of Fig. 3 in this paper is the
same as that of Fig. 2 in [15], and this paper presents some sim-
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002
Fig. 3.
Schematic of the proposed adaptive null-forming system.
ilar discussions concerning the algorithms, implementations,
and tests, as in [15], although we did our work independently.
In comparison with [15], the major contributions of this paper
could be summarized as follows.
1) The unique mapping relationship between the time delay
and the adaptive gain for forming a null is established in
our paper.
2) We prove that this unique relationship between the time
delay and the adaptive gain is independent of frequency
for the related configurations.
3) The entire shape of the polar pattern resulting from
changing the adaptive gain has been proved to be exactly
the same as the entire shape of the polar pattern resulting
from changing the time delay.
4) The optimum gain value and related adaptive algorithms
are derived in the case where the target signal and noise
are both present and the corresponding system has been
proved to provide the maximal signal-to-noise ratio
(SNR).
The rest of this paper is organized as follows. Section II
presents the proposed scheme and makes related theoretical
analyses. Hardware implementation of this proposed scheme
is dealt with in Section III. Section IV will give various test
results demonstrating the effectiveness and accuracy of the
proposed scheme. In Section V, we will give some conclusions.
II. PROPOSED ADAPTIVE NULL-FORMING SYSTEM
The proposed scheme is shown in Fig. 3, where the received
signals at the front microphone and the rear microphone are
and
, respectively.
delay unit in two channels;
output of the system;
adaptive gain;
output of the adaptive gain processing unit.
From the scheme of Fig. 1, it can be easily seen that the polar
is cardioid with the null at 180 ; likewise, the
pattern of
is cardioid but the null is at 0 . The polar
polar pattern of
is a combination of
pattern of the whole system output
and
and determined by the gain
. The relationwith the gain
ship of the null of the system output
is as follows:
(1)
is the angle of
where is the frequency of the signal, and
the null along the line between the two microphones. To prove
LUO et al.: ADAPTIVE NULL-FORMING SCHEME IN DIGITAL HEARING AIDS
1585
TABLE I
RELATIONSHIP BETWEEN THE NULL AND THE GAIN WITH DIFFERENT FREQUENCIES
this relationship, we use the front microphone as a reference
,
, and
as
channel and have
The null at the polar pattern means that the output power
at this direction, that is, we have
(2)
where is the angle of the source along the line between the
two microphones. It should be noted that all right-side terms
, which is
of (2) should be multiplied by a variable
the signal received in the front microphone. However,
could be considered as 1 for the sake of simplicity using the front
microphone as a reference channel. From (2), we can obtain the
output power of the proposed system as
(3)
Furthermore, using the related trigonometric identities yields
Rearranging the above equation yields (1) and concludes the
proof of (1).
, we will consider only the
Because
in the following, realizing that all related
interval
. It can been
conclusions apply also to the interval
, then
seen from (1) that for all frequencies, if
; if
, then
. This result
can also be obtained directly from Fig. 3 because
means
, the polar pattern of
is the same as that
, and because
means that
and
of
will be identical and will make
be zero when the signal
comes from 90 . Except for these two special cases, (1) shows
that the relationship between the null and the gain depends on
the frequency of the signal. However, with the approximation
for the frequency range of interests as used for obtaining the polar pattern of the scheme in Fig. 1, (1) can be approximated as
(4)
This result means that the relationship between the null and the
gain will be independent of the frequency. In effect, the difference between (1) and (4) is very small, especially when
is in the range from 90 to 180 . The lower the frequency, the
smaller this difference will be. To illustrate this, Table I shows a
mm and at frequencies 500 Hz, 1000
set of results with
1586
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002
patterns, respectively, as shown in Fig. 2. In effect, the corresponding null can also be obtained from (4) or Table I.
These results suggest that the polar-pattern null of the system
output
for all frequencies can be moved by simply updating
and keeping the shape of the polar pattern
the gain value
the same as that of the scheme in Fig. 1 with the corresponding
delay value determined by (8).
Next, we show how to adaptively update the gain value so that
the corresponding null can always be toward the noise instead
of toward the target signal in the case where the target signal
and noise are both present. In this case, we have
(9)
(10)
(11)
Fig. 4. Comparison of the result (solid line) of (4) with the result (dashed line)
of (1) at 6000 Hz frequency.
Hz, 2000 Hz, 4000 Hz, 6000 Hz, and 8000 Hz, respectively. For
further illustrations, Fig. 4 compares the results of (1) and (4) at
6 kHz frequency.
From (4), we also have
for
(5)
This shows that the relationship between the null and the gain
, we
is a monotonic function. In other words, given a gain
can get the unique null for all frequency ranges of interest.
Previously, we have discussed the relationship of the polarwith the gain
. Now,
pattern null of the system output
we will investigate the shape of the whole polar pattern. Before
this, we first present the relationship of the output power
of the fixed system in Fig. 1 against the angle using the approxas follows [1].
imation
(12)
(13)
(14)
and
are the desired signal part and the noise
where
part in the front microphone, respectively, and is the delay
of the noise transmission from the front microphone to the rear
microphone. In the above, we also assume that the desired signal
comes straight ahead, that is, its angle is at 0 . By substituting
(9) and (10) into (12), we get
(15)
contains only the noise part. This property shows
that is,
is equivalent to minimizing
that minimizing the power of
because
the power of the noise part contained in the output
the target signal and the noise are not correlated. The optimal
that minimizes the power of
:
gain
(16)
(6)
Similarly, the output power
approximated from (3) as
of the proposed system can be
can be obtained by
(17)
(7)
A comparison of (6) with (7) shows that these two systems will
provide exactly the same polar pattern with mapping relationship
(8)
, 0.4903, and 0, reFor example, if we choose
,
spectively, then the corresponding delay equals 0,
, respectively, according to (8), and these two systems
and
will provide the bidirectional, hypercardioid, and cardioid polar
,
, and
are the power of
, the cross-corwhere
and
, and the power of
, respectively
relation of
[16]. Moreover, the objective function (16) is a quadratic func. On the basis of all
tion with unique minimization point
the above properties, we can conclude that the null of the system
, which is determined by minimizing the system output
power (16) or determined by (17), will always be toward the
direction of the noise when the target signal and the noise are
both present.
Now, the problem becomes how to adaptively update the opti, given the samples of
and
[as obmization gain
tained by (11) and (12), respectively] rather than the cross-corand the power
. We discuss this problem in
relation
Section III together with other issues of the hardware implementation of the system.
LUO et al.: ADAPTIVE NULL-FORMING SCHEME IN DIGITAL HEARING AIDS
1587
III. IMPLEMENTATION OF THE PROPOSED ADAPTIVE
NULL-FORMING SYSTEM
Having the sample of
and
, we can obtain the optimized gain using any available adaptive algorithms, such as the
LMS, NLMS, LS, or RLS algorithm, because (16) is a typical
quadratic optimization problem with only one coefficient. The
fact that only one coefficient need be calculated makes related
adaptive algorithms very simple and makes the real-time hardware implementation of the proposed system possible.
The LMS version for getting the adaptive gain can be written
as
(18)
where is a step parameter that is positive constant less than
, and is the power of the input
. For better performance and faster convergence speed, can also be time varying
(as used by the normalized LMS algorithm [16]). Based on this,
an algorithm with varying-step can be obtained:
(19)
is the estiwhere is a positive constant less than 2, and
.
mated power of
Equations (18) and (19) are suitable for the sample-by-sample
adaptive mode. However, in our hardware implementation of
this proposed scheme, we use the frame-by-frame adaptive
mode because of other processing in hearing aids is also based
on the frame-by-frame mode. With this mode, the following
steps are used to calculate the adaptive gain [16]. First, we
and
and the
estimate the cross-correlation between
at the th frame by
power of
(20)
(21)
is the number of all samples in a frame
respectively, where
and
of
and equals 56 in our implementation. Second,
and
, and then, the
(17) are replaced with the estimated
estimated adaptive gain is obtained by (17).
We have implemented this proposed scheme in one programmable (assembly code) DSP chip together with other
necessary processing such as A/D and D/A. Other practical
factors considered in our hardware implementation include the
following.
1) In this proposed scheme, the two microphones are assumed to have the same frequency response. However,
this is not the case since in real-world devices, these two
microphones are always mismatched. In order to overcome this problem, we first measure the frequency responses of the two microphones, and we then design and
add a matching filter at the rear channel on the basis
of the measured mismatch. From our measurements, a
first-order IIR filter can compensate well for the misand
have the desired carmatch and can make
dioid polar pattern with null at 180 and 0 , respectively.
Fig. 5 presents an example related to matching filter de-
Fig. 5. Upper curve is the measured mismatch, the bottom curve is the
frequency response of the designed matching filter, and the middle curve is the
frequency response difference of the two channels after matching.
sign where the upper curve is the measured mismatch between the front and rear microphones, the bottom curve
is the frequency response of the designed matching filter,
and the middle curve is the frequency response difference
of the two channels after matching.
2) In order to get a better estimate and make the frame-byframe processing smoother, we estimate the cross-correand
and the power of
by
lation between
(22)
(23)
and are two adjustable parameters such that
,
, and
. Obviously, if
and
, (22) and (23) becomes (20) and (21),
respectively.
would be at de3) The null of the total system output
gree less than 90 (that is, in the front hemisphere) when
is larger than 1.
the absolute value of the gain
This has the potential of canceling the target signal in the
front hemisphere. To avoid this problem, we limit the dynamic range of the gain to the range 1 to 1. This limit is
also convenient for the hardware implementation because
in hardware implementation, all digital values are in the
range from 1 to 1.
4) In the hardware implementation of this proposed system,
the total program memory usage is 129 words, the processing time is 0.32 ms, the propagation delay is 0.04 ms,
and the current is 38 A, which shows that the power cost
of this scheme is very small.
where
IV. TESTING RESULTS
With the DSP code to implement this algorithm being completed and being pulled to GN ReSound behind-the-ear (BTE)
devices, we made extensive tests to verify the algorithm and its
implementation. This kind of device used GN ReSound’s own
DSP chip and related platform (operation systems and assembly
languages). In these tests, a Knowles EM 4346 microphone and
a Knowles EM 3356 microphone were put in the endfire config-
1588
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002
Fig. 6. (Left) Measured polar pattern of x(n) and (right) measured polar pattern of y (n).
Fig. 7.
(Left) Measured polar patterns of the system output z (n) with the gain value being
uration to the upper front and upper rear part of a BTE device,
respectively. The distance of these two microphones was about
1.2 cm. This BTE device was connected with a 2-cc coupler. The
sampling rate in this processing was 16 000 Hz. The distance
between the device and the speaker was about 1.0 m. The input
of the speaker was a frequency-sweep signal with frequencies
from 200 Hz to 8000 Hz and with 70 dB SPL. The room noise
was about 50 dB SPL. A Brüel & Kjær Type 2012 Time-Frequency Analyzer was used to measure and analyze related input
and output signals. In order to get the polar patterns, we changed
the angle between the device and speaker at each 10 by turning
the device, which is easier and more accurate than by moving
the speaker. With this setup, 36 sets of data with all frequencies
from 200 Hz to 8000 Hz, corresponding to 36 positions (angles)
of the device, can be obtained. These 36 sets of data were used
to obtain the polar patterns in combination with our polar-pattern generation software. The main testing can be summarized
as follows.
and
were measured to
1) The polar patterns of
verify that they were the desired fixed polar patterns, that
was cardioid with the
is, that the polar pattern of
was carnull at 180 and that the polar pattern of
dioid with the null at 0 . Fig. 6 shows the measured polar
and
. It should be noted that the polar
pattern of
was measured by reversing the position of
pattern of
00.2174 and (right) 00.4903, respectively.
the two microphones, and hence, its measured polar pattern should be the same as the measured one of
.
2) With a given specific value of the adaptive gain, the polar
was measured to see
pattern of the system output
if the null of the polar pattern and the given value of the
gain met the unique relationship given in (4) and Table I.
Two examples are shown in Fig. 7 with values of the gain
being 0.2174 and 0.4903, respectively.
3) A source (the target signal) was set straight ahead (0 ),
and another source (the noise) was set at specific degree
(toward which the null of the proposed adaptive system
should be) to see if the gain value obtained by the adaptive algorithm [(17), (22), and (23)] met the relationship
(4) and Table I. As a matter of fact, when we measured
each polar pattern, we moved the second source from 0
to 360 successively, as mentioned above. With this measurement method, the measured polar pattern obtained by
the proposed adaptive null-forming system is shown in
Fig. 8. This pattern results because the null of the system
should always be toward the degree of this moving source.
Note that we limit the gain value to the range from 1 to
zero.
4) A speaker with pure-tone sound (the target signal) was
set at 0 and a wide band noise with adjustable frequency
bandwidth and sound pressure level was set at different
LUO et al.: ADAPTIVE NULL-FORMING SCHEME IN DIGITAL HEARING AIDS
1589
a gain calculation is involved in the adaptive algorithm, the
computational complexity is low, which makes the proposed
scheme realizable in hardware implementation. Theoretical
analyses and various tests demonstrated the effectiveness of
the proposed system. As a result, this proposed scheme can
serve as a new and practical tool for noise reduction and signal
enhancement in many hearing devices.
ACKNOWLEDGMENT
Fig. 8. Measured polar pattern of the system output corresponding to testing 3.
angels. The angles of the noise source were changed,
(omni-directionality),
and then, the outputs of
(fixed cardioid directionality), and
(adaptive directionality) were measured to further see the performance
improvement of the proposed adaptive system over the
fixed system of Fig. 1. Moreover, in order to see how this
scheme function under real-world conditions, we have
also devised the hearing in noise test (HINT). The HINT
was utilized to assess speech understanding of sentences
at threshold level under varying noise conditions. The
HINT was an adaptive procedure whereby the masker
(uninterrupted speech-shaped noise) remained at 65 dBA
and the speech signal (sentences) level varied. The sentences were delivered at 0 azimuth, and the noise was
delivered at
1) 90 and 270 ;
2) 180 ;
3) 90, 180, and 270 ;
simultaneously. Subjects were tested in three different
conditions:
1) omni-directional system;
2) fixed directionality system in Fig. 1;
3) proposed adaptive directionality system.
All these tests showed that there was more mean benefit
over the omni-directional system for the proposed adaptive directionality system than for the fixed directional
system in Fig. 1.
V. CONCLUSIONS
In this paper, we propose an effective adaptive null-forming
scheme for two nearby microphones in endfire orientation
and deal with its implementation and tests. The null can be
adaptively moved by simply changing a gain value instead of
changing a fractional-sample delay value. We presented and
proved the unique mapping relationship between the null and
the gain value, and we discussed related adaptive algorithms
for updating the gain. The null corresponding to the gain
obtained by the given adaptive algorithm can be guaranteed
to be always toward the noise source in the case where the
target signal and the noise are both present. Because only
The authors are grateful to the anonymous reviewers and the
associate editor for their very useful suggestions and valuable
comments. They also thank N. Michael, S. Petrovic, C. Struck,
and other co-workers at the R&D Department, GN ReSound
Corporation, for their contributions to the implementation and
testing of this proposed system. A U.S. patent application based
on this proposed system is pending.
REFERENCES
[1] M. Valente, “Use of microphone technology to improve user performance in noise,” Trends Amplificat., vol. 4, no. 3, pp. 112–135, 1999.
[2] W. Soede, A. J. Berkhout, and F. A. Bilsen, “Development of a directional hearing instrument based on array technology,” J. Acoust. Soc.
Amer., vol. 94, no. 2, pp. 785–798, 1993.
[3] J. M. Kates and M. R. Weiss, “A comparison of hearing-aid array-processing techniques,” J. Acoust. Soc. Amer., vol. 99, no. 5, pp. 3138–3148,
1996.
[4] J. G. Desloge, W. M. Rabinowitz, and P. M. Zurek, “Microphone-array
hearing aids with binaural output-part I: Fixed-processing systems,”
IEEE Trans. Speech Audio Processing, vol. 5, pp. 529–542, Nov. 1997.
[5] D. P. Welker, J. E. Greenberg, J. G. Desloge, and P. M. Zurek, “Microphone-array hearing aids with binaural output-part II: A two-microphone adaptive system,” IEEE Trans. Speech Audio Processing, vol. 5,
pp. 543–551, Nov. 1997.
[6] J. E. Greenberg, “Modified LMS algorithm for speech processing with
an adaptive noise canceller,” IEEE Trans. Speech Audio Processing, vol.
6, pp. 338–351, July 1998.
[7] J. E. Greenberg and P. M. Zurek, “Evaluation of an adaptive beamforming method for hearing aids,” J. Acoust. Soc. Amer., vol. 91, no.
3, pp. 1662–1676, 1992.
[8] L. J. Griffiths and C. W. Jim, “An alternative approach to linearly constrained adaptive beamforming,” IEEE Trans. Antennas Propagat., vol.
AP-30, pp. 27–34, Jan. 1982.
[9] M. W. Hoffman, T. D. Trine, K. M. Buckley, and D. J. Van Tasell, “Robust adaptive microphone array processing for hearing aids: Realistic
speech enhancement,” J. Acoust. Soc. Amer., vol. 96, no. 2, pp. 759–770,
1994.
[10] H. Saruwatari, S. Kajita, K. Takeda, and F. Itakura, “Speech enhancement using nonlinear microphone array with complementary
beamforming,” in Proc. Int. Conf. Acoust., Speech, Signal Process.,
1999, pp. 69–72.
[11] J. Vanden Berghe and J. Wouters, “An adaptive noise canceller for
hearing aids using two nearby microphones,” J. Acoust. Soc. Amer.,
vol. 103, no. 6, pp. 3621–3626, 1998.
[12] F.-L. Luo, J. Yang, C. Pavlovic, and A. Nehorai, “An FFT Based
Algorithm for Adaptive Directionality of Dual Microphones,” Dept.
Elect. Eng. Comput. Sci., Univ. Illinois at Chicago, Chicago, IL,
UIC-EECS-00-8, 2000.
[13] B. W. Edwards, Z. Hou, C. J. Struck, and P. Dharan, “Signal processing
algorithms for a new, software based, digital hearing device,” Hearing
J., vol. 51, no. 9, pp. 44–52, 1998.
[14] T. I. Laakso, V. Vaelimaeki, M. Karjalainen, and U. K. Laine, “Splitting
tools for fractional delay filter design,” IEEE Signal Processing Mag.,
vol. 13, pp. 30–60, Jan. 1996.
[15] G. W. Elko and A.-T. Nguyen Pong, “A simple adaptive first-order differential microphone,” in Proc. IEEE Workshop Appl. Signal Process.
Audio Acoust., 1995.
[16] S. Haykin, Adaptive Filter Theory. Englewood Cliffs, NJ: PrenticeHall, 1996.
1590
Fa-Long Luo (SM’95) received the B.S., M.S.
and Ph.D. degrees, all with honors, in electronics
engineering from Xidian University, Xi’an, China,
in 1983, 1989, and 1992, respectively.
From 1983 to 1986, he was an engineer with
Changfeng Electronic Systems Corporation. He
was with Tsinghua University, Beijing, China, as a
Research Fellow from December 1991 to December
1993. From May 1994 to October 1998, he was with
University of Erlangen-Nuremberg, Nuremberg,
Germany, as a Principal Research Scientist, where
he was first supported by the Alexander von Humboldt Foundation of Germany
and then by the German Research Foundation. From October 1998 and
February 1999, he was a Senior Project Leader of Cybernetics InfoTech,
Inc. From March 1999 to May 2001, he was with GN ReSound Corporation,
Redwood City, CA, as a Senior Research Scientist and Project Manager. In
June 2001, joined Quicksilver Technology as a Senior Member of Technical
Staff. He has seven U.S. patents pending. He has authored two books: Applied
Neural Networks for Signal Processing (Cambridge, UK: Cambridge Univ.
Press, 1997, 1998, and 1999) and Neural Networks and Signal Processing
(Beijing, China: National Electronic Industrial Publishing House of China,
1993). He has also written more than 80 articles in journals and conferences.
As a principal investigator, he has proposed and conducted more than 20
research and industry projects in signal and data processing with various
applications and their implementation. His inventions on the spectral contrast
enhancement and adaptive microphone array system have been successfully
used and implemented in hearing aid products, and significant performance
improvements have been achieved.
Dr. Luo received the National Young Investigator Award of China, which is
biennially granted to up to 100 individuals who have made internationally or nationally recognized extraordinary contributions in science and technology and
is nominated from young scientists, engineers, educators, and technical executives over all China, in 1994. He received the National Outstanding Science
and Technology Book Award of China in 1995. He has been a technical committee member (Neural Networks Technical Committee) of IEEE Neural Networks Council since 1998. From June 1997 to December 2000, he was a technical committee member (Neural Networks for Signal Processing) of the IEEE
Signal Processing Society. As a Guest Editor, he edited a special issue of Signal
Processing on neural networks (vol. 64, no. 3, 1998). He is the Executive Guest
Editor of a special issue of Speech Communication on speech processing for
hearing aids. He was an Associate Editor of an IEEE Control Systems Society
Conference. As a reviewer, session chair, and technical committee member, he
has served on many international journals and conferences. He is an editorial
board member of International Journal of Information Fusion and IEEE Communication Surveys and Tutorials.
Jun Yang (SM’99) received the B.S. degree from
Huazhong (Central China) University of Science
and Technology in 1985, the M.S. degree from
Northwestern Polytechnic University in 1988, and
the Ph.D. degree from Xidian University, Xi’an,
China, in 1991, all in electronic engineering.
From November 1991 to October 1993, she was
with Tsinghua University, Beijing, China, where
she was promoted to Associate Professor in 1993.
She was with Oldenburg University of Germany and
Viennatone Corporation of Austria as a Research
Fellow and Principal Engineer, respectively, from October 1993 to November
1998. In December 1998, she joined GN ReSound Corporation, Redwood
City, CA, as a Senior Research Engineer and then became the Chief Scientist
of Spatializer Audio Laboratories, Inc. Since March 2001, she has been a
Senior Audio Engineer with VM Labs, Inc. and a Principal DSP Engineer
with ForteMedia, Inc., respectively. She has five U.S. patents (pending,
co-inventor) and has authored 40 journal articles and conference papers on
signal processing with its various applications such as audio, communications,
and hearing aids. The hearing devices with some of her inventions are well
sold in the world market. As a principal investigator, she has also conducted
several research projects on communications, speech coding and processing,
and an audio interface for blind computer users. Most of her projects, because
of the importance and excellence, were granted by government agencies or
foundations of Germany and China, respectively. Her works and contributions
have received wide attention and were reported by Science and Technology
Daily (the national science and technology newspaper of China) and published
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002
by the State Science and Technology Committee of China. She has been a reviewer for Speech Communication
Dr. Yang has also received several university-level awards for her excellent
work in science and technology. She has been a reviewer for the IEEE
TRANSACTIONS ON SPEECH AND AUDIO PROCESSING. She is also a Member of
Audio Engineering Society and was a Senior Member of the Chinese Institute
of Electronics.
Chaslav Pavlovic received the M.S. and B.S. degrees
in electrical engineering and the Ph.D. degree in audiology.
He is currently the Executive Vice President for
Research and Development at Sound ID. Previously,
for eight years, he was with GN ReSound (formerly ReSound Corporation), a hearing healthcare
company, where he was the Senior Vice President
for Research and Development. During his tenure,
ReSound (subsequently GN ReSound) established
itself as a technology leader in hearing products and
grew to become the second largest hearing instrument company in the world.
He was a Full Professor of speech technology at Aix en Provence, France, and
an Associate Professor of Audiology at the University of Iowa, Iowa City, He
has held other academic appointments. He has more than 100 publications on
hearing aids, speech intelligibility, and speech quality.
Dr. Pavlovic was the Chair of the 1997 ASA work group that developed ANSI
S3.5: the Speech Intelligibility Index Standard. He has also been the Coordinator
of the European Audiological Tests and Station (EURAUD) project; Chair of the
American National Standards Institute S3-79 Writing Group (Calculation of the
Articulation Index); USA representative to the International Standards Organization ISO/TC 43/SC1; Coordinator of the Overall Quality Assessment Subgroup European Consortium for Speech Assessment Methods (SAM, Project
Esprit); Coordinator of participating French laboratories on projects TIDE and
OSCAR (pattern extraction hearing aids); Member of the American National
Standards Institute S12-8 Writing Group (rating noise with respect to speech
interference); Member of the Editorial Board of Acoustics; Staff Editor of the
Journal D’Acoustique; Board of Directors of the Journal D’Acoustique; and
Member of the Technical Committee on Speech Communication of the Acoustical Society of America.
Arye Nehorai (S’80–M’83–SM’90–F’94) received
the B.Sc. and M.Sc. degrees in electrical engineering
from the Technion—Israel Institute of Technology,
Haifa, in 1976 and 1979, respectively, and the Ph.D.
degree in electrical engineering from Stanford University, Stanford, CA, in 1983.
After graduation, he worked as a Research Engineer for Systems Control Technology, Inc., Palo Alto,
CA. From 1985 to 1995, he was with the Department of Electrical Engineering, Yale University, New
Haven, CT, where he became an Associate Professor
in 1989. In 1995, he joined the Department of Electrical Engineering and Computer Science, The University of Illinois at Chicago (UIC), as a Full Professor.
From 2000 to 2001, he was Chair of the Department’s Electrical and Computer
Engineering (ECE) Division, which is now a new department. He holds a joint
professorship with the ECE and Bioengineering Departments at UIC. His research interests are in signal processing, communications, and biomedicine. He
is on the Editorial Board of Signal Processing and was an Associate Editor for
Circuits, Systems, and Signal Processing.
Dr. Nehorai is Editor-in-Chief of the IEEE TRANSACTIONS ON SIGNAL
PROCESSING. He is also a Member of the Publications Board of the IEEE Signal
Processing Society. He has previously been an Associate Editor of the IEEE
TRANSACTIONS ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, the IEEE
SIGNAL PROCESSING LETTERS, the IEEE TRANSACTIONS ON ANTENNAS AND
PROPAGATION, and the IEEE JOURNAL OF OCEANIC ENGINEERING. He served
as Chairman of the Connecticut IEEE Signal Processing Chapter from 1986
to 1995 and is currently the Chair and a Founding Member of the IEEE Signal
Processing Society’s Technical Committee on Sensor Array and Multichannel
(SAM) Processing. He was the co-General Chair of the First IEEE SAM Signal
Processing Workshop, held in 2000, and will again serve in this position in
2002. He was corecipient, with P. Stoica, of the 1989 IEEE Signal Processing
Society’s Senior Award for Best Paper. He received the Faculty Research
Award from the UIC College of Engineering in 1999. In 2001, he was named
University Scholar of the University of Illinois. He has been a Fellow of the
Royal Statistical Society since 1996.