1518

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

Acoustic Noise and Echo Canceling

with Microphone Array
Mattias Dahl, Associate Member, IEEE, and Ingvar Claesson, Member, IEEE

Abstract A novel method of performing acoustic echo cancelling using microphone arrays is presented. The method employs a digital self-calibrating microphone system. The calibration process is a simple indirect on-site calibration that adapts
to the particulars of the acoustic environment and the electronic
equipment in use. Primarily intended for handsfree telephones in
automobiles, the method simultaneously suppresses the handsfree
loudspeaker and car noise. The system also continuously takes
into account disturbances such as fan noise. Examples from an
extensive evaluation in a car are also included. Typical performance results demonstrate 20-dB echo cancellation and 10-dB
noise reduction simultaneously.
Index Terms Adaptive array, array signal processing, echo
suppression, microphone array, speech enhancement.

I. INTRODUCTION

CHO-RELATED problems are extremely common in

telephone systems. Speech originating from the far-end
talker echoes back with a time delay, thereby causing perception problems (see Fig. 1). Perception is further impaired
when the near-end talker is situated in a noisy car in the
handsfree mode. Increased use of mobile telephones in cars
has created a greater demand for handsfree in-car installations. The advantages of handsfree telephones are safety and
convenience. The disadvantages are poor sound quality and
acoustic feedback of the far-end speech signal introduced by
the handsfree loudspeaker.
A conventional method for decreasing the acoustic feedback of the far-end talker during handsfree communication is
adaptive echo cancellation (EC) [1], [2], and a considerable
amount of effort has been put into this field. Most work has
been devoted to single-signal solutions [3][5], although some
papers have proposed systems with two microphones [6], [7].
The multimicrophone acoustic echo canceller introduced in
this paper is not based on array theory and spatial a priori
information [8], [9]; instead it relies on an indirect calibration
[10]. The method is capable of simultaneously suppressing
the handsfree loudspeaker and car noise. Alternative methods
for increasing the signal-to-noise ratio in handsfree mobile
telephones are spectral subtraction [11][13] and active noise
control (ANC) [14][16].
The proposed system is intended for use with mobile
handsfree communication equipment, and it thus also takes
into account the near field in a small enclosure. A near field
Manuscript received December 13, 1996; revised March 3, 1998.
The authors are with the Department of Signal Processing, University of
Karlskrona/Ronneby, S-372 25 Ronneby, Sweden.
Publisher Item Identifier S 0018-9545(99)05747-3.

and enclosed situation is difficult to describe in an a priori

model, and for this reason, reference signals gathered from the
real target and jammer positions have been used. These signals
contain useful information about the acoustic environment, as
well as electronic equipment, such as microphones, amplifiers,
A/D converters, anti-aliasing filters, etc. Microphone element
geometry and other spatial and spectral information is also
inherent in the target and jammer signals gathered. The main
problem with all adaptive filtering is to obtain the desired
signal for the adaptive filters. Since it is impossible to isolate
the near-end talkers speech from the car noise and jammer
noise in the real situation in the car, we are forced to take
second best. We adapt the filters in a given situation using
perfect prerecorded signals matching the real signals as
closely as possible and adding actual real car noise each time
the near-end talker is silent.
II. WORKING SCHEME FOR THE ECHO-CANCELING
ADAPTIVE MICROPHONE ARRAY
The working scheme for the microphone array can be
divided into two phases: phase 1, which is the gathering phase,
and phase 2, which is the continuous filtering and adaptation
phase. During phase 2, the system utilizes the calibration
signals gathered in phase 1.
A. Phase 1The Gathering Process
The gathering phase takes place on-site in the actual environment by emitting representative sequences from each
jammer and target position while the car is parked. In this
way, a fair signal-to-noise ratio is obtained during data acquisition.
Array signals for each sequence and channel are then
stored as digital samples in the memory for later use as
training signals. For the target signal, this can be performed
by allowing a loudspeaker to emit colored noise, or by letting
the near-end talker read a representative sequence from a
desired position in the car (see Fig. 2). The procedure is
repeated for the jammer signal from the handsfree loudspeaker.
The multichannel calibration signals will later be used
to form the input and reference for the echo canceller in
Phase 2, the operating phase. The signals contain information
on the acoustical environment, variations in the electrical
equipment, and spatial and frequency responses. The main idea
of this straightforward solution is that the calibration signals
themselves will be the best tutor. Instead of calculating a

00189545/99$10.00 1999 IEEE

DAHL AND CLAESSON: ACOUSTIC NOISE AND ECHO CANCELLING WITH MICROPHONE ARRAY

1519

Fig. 1. Two-way handsfree communication conversation between the far- and near-end speaker.

Fig. 2. On-site calibration of the array system by using the existing handsfree loudspeaker (the jammer). The figure shows the procedure for the jammer.
The corresponding operation is performed for the target.

large number of statistical features from the prerecorded data

gathered, we merely inject the reference signals during the
operating/adapting phase.
The microphone elements and placement can be chosen
arbitrarily, but they should not be altered or moved unless
a new calibration is made.

1) Calibration Signals: The calibration signals gathered arrive from the desired and unwanted talker and handsfree loudspeaker positions, respectively, and should have approximately
the same spectral content as the true signals. There are different
methods to facilitate this. A very simple approach is to collect
and superpose human utterances from the target position and

1520

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

TABLE I
OPERATION MODES

III. CONTROL ALGORITHM

The beamformer consists of
finite-duration impulseresponse (FIR) filters, one for each microphone (see Fig. 3).
has taps, and all FIR filters are collected
Each FIR filter
where
denotes transpose.
in
traveling to the far-end side from the
The system output
upper beamformer is given by
(1)

noise from the handsfree loudspeaker. These signals are stored

in the target and jammer memory, respectively.
A more cumbersome procedure is also to place a
loudspeaker in the desired position and let the loudspeakers
emitone at a timeflat, colored noise or gathered speech.
However, this method is well suited for repeating the
experiments during an evaluation.

, and
is a
where
most recent samples of
in
vector containing the
reverse order.
The task for the adaptive lower beamformer filters during
the I and Rx modes is to make the lower beamformer output
resemble a linear combination
of the memory target
signals. This is achieved by minimizing the composite error
between the desired signal
and the output
The lower beamformer
from the lower beamformer
input
(2)

B. Phase 2The Operating Mode

A telephone conversation is normally divided into four
modes: idle (I), receive (Rx), transmit (Tx), and double talk
(DT) (see Table I).
During the I mode, only car noise exists, i.e., there is no
speech from the near- or far-end talker. This is the time to
obtain a good car noise estimate and to adapt the array by
means of the real incoming noise and the signals stored from
phase 1 (see Fig. 3).
Compartment noise impinges on each of the microphone
elements in the microphone array, and is added with prerecorded virtual near- and far-end speech signals, i.e., the stored
signals. Note that the noise signals have passed through the
same electronic equipment as the stored signals. The target
speech signals gathered represent a person talking in the actual
environment and desired position, while the jammer memory
signals represent the handsfree loudspeaker. Observe also that
the two signal constellations were gathered one at a time under
conditions with almost no background noise.
In the Tx mode, only near-end talker speech and car noise
are present. When near-end speech is detected, the adaptation
is turned off. Incoming microphone signals from the microphone array are now processed by the fixed upper beamformer
using the latest filter coefficients adapted to the latest actual
situation. In this way, the filter coefficients suit the actual
disturbance situation.
In the Rx mode, only far-end (handsfree loudspeaker) signal
and car noise are present. The algorithm behavior is similar
to the I mode described above. The main difference is that
the microphone signals in this mode consist of real noise and
speech from the handsfree loudspeaker.
In the DT mode, near- and far-end speech, as well as car
noise, are present on the near-end side. In the DT mode,
no adaptation is made; the beamformer coefficients are fixed
and the output from the upper beamformer is transmitted.
The output consists of enhanced near-end talker speech and
suppressed far-end speech signal.

is given by a weighted sum of the target and jammer calibraand

tion signals stored
respectively, and actual
as defined in Fig. 3. The significance of the
car noise
and is discussed at the end of this section.
weights
from the adapting lower beamformer is, thus,
The output
given by
(3)
The desired
where
for the adaptive filters is formed by a suitable
signal
combination of the stored target signals only. An effective
combination is to use the sum of all the target signals as
a desired signal. The control algorithm now has access to
all information needed to indirectly calibrate the microphone
array system by means of the real incoming noise and the
stored signals from the on-site calibration. The beamformer
so obtained are used continuously in the
filter weights
upper beamformer to reduce the real jammer and car noise
and enhance the real near-end talker. It should be noted that
from the lower beamformer is an output
the output
signal which is based on prerecorded calibration and actual
microphone signals and is useful for adaptation purposes only.
The optimal least-squares (LS) solution for the array system
parallel FIR filters is given by minimizing
using
(4)
is given
where the composite error
, and
denotes the total data matrix
by
[2]. This is done by solving the matrix form of the normal
equations for the linear LS filter
(5)
, and
where
the optimal LS solution

[2]. This yields

given

DAHL AND CLAESSON: ACOUSTIC NOISE AND ECHO CANCELLING WITH MICROPHONE ARRAY

1521

Fig. 3. Balance between inputs during phase 2, the operating phase.

a specific training set for the multiple-input/single-output

system. The optimal LS calculations can be solved by using the
recursive least-squares algorithm (RLS); in a nontrivial array
system this is not, however, suitable for real-time processing
and should be used for off-line calculations only. In the
evaluation part of this paper, the FIR filters are updated by
the normalized least-mean-squares (NLMS) algorithm [1]. The
objective in the LMS case is to minimize
(6)
denotes the expectation operation. The adaptive
Here,
is updated in the direction of the negative
weight vector
gradient of the cost function. The recursive relation of the
adaptive weights in the steepest descent algorithm can be
, where
is
written as
the convergence factor and step-size parameter which controls
the stability and convergence of the algorithm. By omitting
the expectation from the objective function (6) in deriving the
gradient
(7)
we obtain the well-known LMS algorithm gradient estimate
for each adap[1]. By replacing the convergence factor

tive filter by an individual normalized coefficient

, the stability properties influenced by the length
of the filter and power
in the input signal will be
released. However, to obtain a stable algorithm, the normalized
must satisfy the criterion
convergence factor
The power estimate
is normally computed for each
as an exponential average.
individual microphone channel
In this evaluation, we have used one and the same averaged
for all the adaptive filters given by
power estimate
(8)
and is related to the integration time
in the samples.
The NLMS algorithm used is finally summarized as follows:

where

(9)
The performance of the adaptive algorithm can be controlled
and where
by
controls the near-end memorized speech
1) factor
amplification/attenuation;
signal

1522

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

Fig. 4. Linear microphone geometry [one-dimensional (1-D)] with six microphones. The distance between elements is 50 mm.

2) factor

controls the far-end memorized speech signal

amplification/attenuation;
controls the incoming environment noise
3) factor
from the
amplification/attenuation for signals
microphone array.
The mix between the components will control the adaptive
filter suppression or amplification of the near-end talker, the
acoustic echo from the handsfree loudspeaker, and the car
noise. Since the calibration signals from the desired and
unwanted positions, respectively, have approximately the same
spectral content as the true signals, the array suppression or
amplification is achieved both in the spatial and the temporal
and
versus
will cause the
domain. A high value of
adaptive filter to emphasize cancellation of both the far-end
speech and/or car noise. However, too large a suppression will
produce degradation of the near-end speech signal.

Fig. 5. Two-dimensional (2-D) microphone geometry, with six microphones.

The distance between the elements is 50 mm.

IV. EVALUATION CONDITIONS

A. Car Environment
The performance evaluation of the acoustic echo-cancelling
beamformer was carried out in a Volvo 940 station wagon.
Data was gathered on a multichannel digital audio tape (DAT)
recorder with a sample rate of 12 kHz, and with 5-kHz
bandwidth. In order to facilitate simultaneous driving and
recording, the near-end talker was simulated with a loudspeaker mounted in the passenger seat.
B. Microphone Configurations
Two microphone brands were evaluated: a conventional
handsfree mobile telephone microphone (Ericsson RLC 509
11/03 RA) and a high-quality microphone from Sennheiser.
Since no major differences in performance were obtained,
we have restricted the number of figures in this paper by
including the results for the cheaper alternative only. The
Ericsson microphones were mounted flat on the visor, while
the Sennheiser microphones were mounted 20 mm below
the visor using a fixture. These are not ideal positions for
a commercial implementation, since the positions are not
stationary.
The distances between the normal near-end talker position
(i.e., the loudspeaker mounted in the passenger seat) and microphones were 330 mm (Ericsson) and 350 mm (Sennheiser),
respectively. Two different microphone geometries were evaluated (see Figs. 4 and 5). For a typical car compartment
installation using a linear microphone geometry, see Fig. 6.
V. IMPLEMENTATION

AND

EVALUATION

The data gathered from the multichannel DAT recorder

was converted into Matlab format. The evaluation section of

Fig. 6. Car compartment installation of the handsfree system using linear

microphone array geometry.

this paper is based on this data. We have also performed an

exhaustive real-time DSP evaluation, with the main difference
being the fix-point implementation. The fix-point restriction
necessitates extra care being taken, but the results comply with
the off-line results based on recorded data. The DSP system is
equipped with eight TMS 320C25 DSPs; in this implementation, these are configured to serve six A/D converters and four
D/A converters simultaneously. The system is controlled from
an ordinary PC, which acts as a host computer and operator
interface.
Performance plots from the Volvo are illustrated in Figs.
713, showing the results from an adapted (4 s) and frozen
upper beamformer. The sound files corresponding to all figures
are available in wave format via the Internet.1 Observe that
during the filtering phase, the nominal values, i.e., the true signal levels in the car, were used with approximately SNR
dB and SIR
dB, where the memory-signal-to-interference
ratio (MSIR) is defined by

MSIR

(10)

and
denote the stored calibration signals for
where
target and jammer, respectively. In the handsfree mode, this
1 http://www.its.hk-r.se.

DAHL AND CLAESSON: ACOUSTIC NOISE AND ECHO CANCELLING WITH MICROPHONE ARRAY

1523

Fig. 7. Output power versus number of filter taps with two microphones, flat noise training signals. Number of taps
the bottom of the figure.

= 16; 32; 1 1 1 ; 2048 noted at

Fig. 8. Output power versus number of filter taps with six microphones, flat noise training signals. Number of taps
at the bottom of the figure.

= 16; 32; 1 1 1 ; 2048 is noted

corresponds to the speaker, usually placed in the front seat,

and the handsfree loudspeaker, directed toward the driver.
In addition, the memory-signal-to-noise ratio (MSNR) for
the calibration phase is defined as

MSNR

(11)

and
denote the stored calibration signals, i.e.,
where
the target and the actual car noise. In the car environment,
corresponds to environmental car noise, which, in this
evaluation, includes radio music, fan noise, or noise from a
side window wound down. Common information for all plots
is as follows.
Near-end speech, coming from the target position is
denoted Speaker.
Far-end speech signal, i.e., handsfree loudspeaker, is
denoted Echo.

The recorded calibration signals are flat noise. Speechcolored noise or human speech overlayed give similar
results.
All sequences are of 7 s, and are subsequently merged
together, i.e., a new sequence starts at 0, 7, 14
s.
1) The near-end talker is active from 1.5 to 3.5 s.
2) The far-end talker is active from 4.5 to 6.5 s.
3) In the remaining time, only background noise is
present, unless otherwise declared.
All figures begin with a 7-s sequence with an unadapted
single microphone signal, i.e., a plain unfiltered singlechannel microphone signal.
The results are presented as short-time (20 ms) power
estimates in decibels.
All signals are limited to telephone bandwidth
(3003400 Hz).
The number of filter taps needed is a crucial parameter;
we found that 128256 filter taps are sufficient (see Figs. 7
and 8). Since the evaluation was performed at the sample
rate 12 000 Hz, the number of taps could be reduced by

1524

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

Fig. 9. Output power versus MSIR for two- and six-microphone array, left- and right-hand plots, respectively. MSIR in decibels (+15; +5; 0;
is noted at the bottom of the figure.

Fig. 10.

05 015)
;

Different microphone placement results, 1-D array. Used microphones in the linear geometry, see Fig. 4, are noted at the bottom of the figure.

30%, as the ordinary telephone sample rate of 8000 Hz was

used. The acoustic echo suppression is shown for two and six
microphones. The suppression of the handsfree loudspeaker
with two microphones is 18 dB (Sennheiser 17 dB). When
using six microphones, the improvement is 24 dB (Sennheiser
19 dB).
Different choices of calibration signals were also evaluated.
For the Ericsson microphones, the best choice of calibration
signals appeared to be flat noise, while the Sennheiser microphones performed best when using speech-colored noise. The
most practical approach is, however, to use human speech.
The mix of calibration components will cause the adaptive
filter to amplify or suppress the sources differently in both the
frequency and spatial domains. In this evaluation, it is necessary to examine subjectively results such as speech quality
by means of listening. We have, however, also investigated
the speech quality of the adaptive microphone array system in
cars using speech recognition. The results are most promising
dB will give a notable degradation
[10], [17]. An MSIR
the degradation
of the near-end speech. When MSIR
of the near-end speech is low, and the suppression of the

handsfree loudspeaker is considerable. We have chosen this

as a subjective optimum (see Fig. 9).
Alternative subsets of the two array geometries were also
tested. The evaluation indicates that the aperture between
the microphones used is just as important as the number of
microphones (see Figs. 1011). Microphones 1 and 6 yield
significantly better results than two adjacent microphones. We
observe that three microphones seems to be sufficient. This
is, however, only true when judging by short-time power estimates. The speech quality and distortion are further improved
when the number of microphones is increased. The performance of the linear mount (Fig. 10) gives better performance
than the nonlinear mount (Fig. 11), and this is due to the
increased aperture in the linear geometry.
Four different disturbance situations were investigated: 1)
car noise only; 2) car noise and fan noise; 3) car noise
(90 km/h) and music; and 4) car noise (90 km/h) and side
window wound down. The array maintained its acoustic echocancelling ability, even in the presence of environmental
disturbance. In addition, the array suppresses the environmental disturbances by 710 dB (see Figs. 12 and 13).

DAHL AND CLAESSON: ACOUSTIC NOISE AND ECHO CANCELLING WITH MICROPHONE ARRAY

Fig. 11.

1525

Different microphone placement results, 2-D array. Microphones used in the nonlinear geometry, see Fig. 5, are noted at the bottom of the figure.

Fig. 12. Car noise (90 km/h) and car noise

are noted at the bottom of the figure.

+fan noise results using two and six microphones. Microphones used in the linear geometry, see Fig. 4,

Fig. 13. Car noise (90 km/h)

music noise and car noise
geometry, see Fig. 4, are noted at the bottom of the figure.

VI. SUMMARY

AND

side window noise using two and six microphones. Microphones used in the linear

CONCLUSIONS

A cumbersome part of microphone array implementation

is the calibration phase. The on-site calibrated acoustic echo

canceller in this paper employs a self-calibrating process,

which does not rely on a priori modeling. The acoustic
echo canceller gives substantial suppression of the hands-

1526

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

free loudspeaker in a car compartment, while simultaneously

suppressing environmental disturbance. The placement of the
microphones seems to be an important factor, whereas the
microphone quality is less important.
In order to allow full duplex conversation, approximately
40-dB suppression of the far-end signal (handsfree loudspeaker) is needed. We believe that approximately 20 dB can
be reached with array processing, and the remaining 20 dB
can be achieved with careful placement of loudspeakers and
microphones.
The issue of speech detection for controlling adaptation
is not crucial, since we do not switch the speech; only the
adaptation is switched on and off. If one is uncertain, one
can always assume a cautious strategy; we only adapt when
no near-end speech is present. This does not hamper the
performance seriously; the beamformer has exhibited robust
behavior with respect to changes in driving conditions. We
can, for example, train the array at 90 km/h and use these
coefficients at all speeds with only negligible loss, even when
a near-end talker speaks continuously for a long time. We
conclude with the following remarks.
Good suppression, 19 dB, of the handsfree loudspeaker
was reached with only two microphones and 256 filter
taps.
Target distortion decreased when the number of microphones was increased.
The acoustic echo-cancelling method also yielded good
suppression of the ambient noise in the car.
The calibration signals were either flat noise, speechcolored noise, or human speech. The choice of calibration
signals had little influence on results.
The placement of the microphones was even more important than we had expected.

[8] I. Claesson and S. Nordholm, A spatial filtering approach to robust

adaptive beamforming, IEEE Trans. Antennas Propagat., vol. 40, no.
9, pp. 10931096, Sept. 1992.
[9] S. Nordebo, I. Claesson, and S. Nordholm, Adaptive beamforming:
Spatial filter designed blocking matrix, IEEE J. Oceanic Eng., vol. 19,
pp. 583590, Oct. 1994.
[10] S. Nordebo, I. Claesson, and S. Nordholm, An adaptive microphone
array employing calibration signals recorded on-site, in Proc. Int. Conf.
Signal Processing Applications and Technology (ICSPAT 94), Dallas,
TX, Oct. 1994, pp. 2529.
[11] J. R. Deller, Jr., J. G. Proakis, and J. H. L. Hansen, Discrete-Time
Processing of Speech Signals. New York: Macmillan, 1993.
[12] S. F. Boll, Suppression of acoustic noise in speech using spectral subtraction, IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-27,
pp. 113120, Apr. 1979.
[13] J. Yang, Frequency domain noise suppression approaches in mobile telephone systems, in Proc. ICASSP-93, Apr. 1993, vol. II, pp.
363366.
[14] S. M. Kuo and D. R. Morgan, Active Noise Control Systems. New
York: Wiley, 1996.
[15] T. J. Sutton, S. J. Elliott, A. M. McDonald, and T. J. Saunders, Active
control of road noise inside vehicles, Noise Control Eng. J., vol. 42,
no. 4, pp. 137147, 1994.
[16] S. J. Elliott, P. A. Nelson, I. M. Stothers, and C. C. Boucher, In-flight
experiments on the active control of propeller-induced cabin noise, J.
Sound Vibration, vol. 140, no. 2, pp. 219238, 1990.
[17] S. Nordebo, S. Nordholm, B. Bengtsson, and I. Claesson, Noise
reduction using an adaptive microphone array in a carA speech
recognition evaluation, in Proc. 1993 IEEE Workshop on Applications
of Signal Processing to Audio and Acoustics, New Paltz, NY, Oct. 1993.

REFERENCES

Mattias Dahl (S94A95) was born in Uddevalla, Sweden. He received the B.S. degree
in electrical engineering from the Chalmers
Institute of Technology, Gothenburg, Sweden,
the M.S. degree in telecommunication and signal
processing from Lulea University of Technology,
Lulea, Sweden, and the Licentiate degree in
signal processing from Lund University of
Technology, Lund, Sweden, in 1988, 1993, and
1997, respectively.
Since 1993, he has been with the Department
of Signal Processing, University of Karlskrona/Ronneby, Ronneby, Sweden,
where he is involved in adaptive beamforming, speech enhancement, and
active noise control research projects.

[1] B. Widrow and S. D. Stearns, Adaptive Signal Processing. Englewood

Cliffs, NJ: Prentice-Hall, 1985.
[2] S. Haykin, Adaptive Filter Theory. Upper Saddle River, NJ: PrenticeHall, 1996.
[3] M. Sondhi and D. A. Berkley, Silencing echoes in the telephone
network, Proc. IEEE, vol. 68, pp. 948963, Aug. 1980.
[4] D. G. Messerschmidt, Echo cancellation in speech and data transmission, IEEE J. Select. Areas Commun., vol. SAC-2, pp. 283297,
1984.
[5] M. Sondhi and W. Kellermann, Adaptive echo cancellation for speech
signal, in Advances in Speech Signal Processing, S. Furui and M.
Sondhi, Eds. New York: Marcel Dekker, 1992, ch. 11.
[6] S. M. Kuo and Z. Pan, Adaptive acoustic echo cancellation microphone, J. Acoust. Soc. Amer., vol. 93, no. 3, pp. 16291636, Mar.
1993.
[7] M. Rainer and P. Vary, Combined acoustic echo cancellation dereverberation and noise reduction: A two microphone approach, Ann.
Telecommun., vol. 49, nos. 78, pp. 429438, 1994.

Ingvar Claesson (M91) was born in Broby, Sweden, in 1957. He received the Dipl.Eng. and Ph.D.
degrees from Lund University, Lund, Sweden, in
1980 and 1986, respectively.
He was appointed Senior Lecturer in Telecommunication Theory at Lund University in 1986 and
became an Associate Professor in 1992. Since May
1998, he has held the Chair of Signal Processing
at the University of Karlskrona/Ronneby, Ronneby,
Sweden. He is also currently the Head of Research
and Principal Supervisor in Signal Processing. In
1990, he was one of the founders of the Department of Signal Processing. His
current research interests are in adaptive signal processing, blind equalization,
adaptive beamforming, speech enhancement, active noise control, filter design,
and antenna arrays.