Fig. 1. Audiogram is taken from www.Earinfo.com website and this
represents a hearing problem to a patient at high frequency.
Design and Implementation of an Area and Power-efficient
Reconfigurable Hearing Aid using Interpolated Sub-band Distribution
Technique
ATRYEE BHUYAN
Department of Electronics and
Communication Engineering
Gauhati University
Guwahati, Assam, INDIA
MANASH PRATIM SARMA
Department of Electronics and
Communication Engineering
Gauhati University
Guwahati, Assam, INDIA,
NIKOS E MASTORAKIS,
Faculty of Engineering,
Technical University of Sofia,
BULGARIA
Abstract: A hearing aid is a compensatory device that helps in overcoming various hearing disabilities. Since different people
possess different hearing problems and the requirement gets changed over time, it is necessary to design a reconfigurable
hearing aid which is generic in nature such that it supports various hearing disabilities without modifying the hardware
components. The objective of the paper is to implement a reconfigurable digital hearing aid which is hardware efficient and it
is auto-adaptable to disabilities ranging from mild to severe intensities. Since multipliers and LUTs are power hungry
elements, we have proposed a design which is multiplier less and LUT-less DA Architecture. The prototype filter is a FIR
filter which is designed using LUT-less Distributed Arithmetic Algorithm that saves 64% of logic elements & memory and
76% of power utilization. As it is a multiplier-less as well as LUT- less architecture, hence this can be claimed to be area and
power efficient design. The delays and matching errors are within the standard limits which are accepted globally. The input
audio spectrum is divided into three regions and for each region there are four different filter banks are proposed using
interpolated sub-bands distribution. Xilinx System Generator is used to implement the proposed design. The proposed design
requires least manual configuration for selection of filter banks for audiogram matching which also minimizes the trial-and-
error method to establish the best match with the person’s audiogram.
Keywords: audiogram, reconfigurable, sub-band distribution
Received: March 21, 2022. Revised: November 11, 2022. Accepted: December 12, 2022. Published: December 31, 2022.
1. Introduction
Hearing aid device benefits lot of people having hearing
loss problems and 90% of users accept it. The hearing aid
performs the function of making low intensity sounds
audible and loud sounds comfortable. These devices match
audiogram gains and dynamic ranges of ear characteristics.
The basic function of hearing aid is amplification of sound,
but designers face challenges of restoring other
characteristics like loudness, speech intelligibility etc. Fixed
signal bands formation plans are used in most of the present-
day hearing-aid systems. Fixed filter banks not have
sufficient flexibility to match audiogram response of
different hearing losses. Reconfigurable filter bank can be a
solution to control parameters for more adaptable hearing
aid device as per user requirements in every situation using
non uniform sub-band distribution. [1]
The most common type of hearing loss is Sensorineural
Hearing Loss (SNHL) where inner sensory nerves are
damaged and this loss can happen at any age so most of the
hearing aid devices are design to compensate SNHL[4]. The
technique of audiogram matching is basically adjustment of
magnitude response of filters in an inverted fashion where
amplification is done to the magnitude response as required .
Unfortunately, only 10% of the hearing aids are able to meet
up the requirement. Thus, a properly adjustable hearing aid
can improve the speech matching of hearing deprived people
[5].
ENT specialist or we can say audiologist perform
audiogram test by the method of pure tone audio frequency
signals. These signals are within the range of 250Hz to
16Khz. The mark ‘O’ is for the left ear and ‘X’ is for right
Ear on Y-axis, and amplitude or loudness of hearing
sensitivity is on X-Axis. The following audiogram in Fig. 1
is for hearing loss at high frequency. There are other
audiograms as well for different range of hearing losses. The
graph is measured from low to high frequencies (low to high
pitches) going from left to right, and the graph is measured
from soft sounds on the top to loud sounds at the bottom
shown in Fig. 2. The above information is important for
achieving reconfigurability. [7]
Finite impulse response (FIR) a r e digital filters
having common DSP functions and are widely used in
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
312
Volume 21, 2022
Fig. 2. Information gathered from audiogram (Source:
www.earinfo.com)
(3)
(2)
(1)
multiple applications like telecommunications,
wireless/satellite communications, video and audio
processing, biomedical signal processing and many others.
On one hand, ASICs have certain restrictio ns
related to high development costs and time-to-market
factors while, on the other hand, programmable DSP
processors are enable to meet desired performance due to
their sequential-execution architecture [9].
The FIR digital filter is presented as:
y[n] =
the FIR filter output, x [n-k] is the input data and ck
represents the filter coefficients. Equation (1) shows the
multiplier-based filter that can be implemented but it may
become highly expensive in terms of area and speed. This
issue has been partially resolved with the use of DA
(Distributed Arithmetic) algorithm.
Initially, sub-bands are distributed uniformly which
provide average audiogram matching but later non-uniform
sub-bands distribution is preferred to get better matching and
less matching error [18]. There are several techniques
defined for sub-bands implementation but the proposed
structure is implemented using multirate signal processing
methods. A 17-band reconfigurable hearing aid is
implemented in this paper using interpolation and
decimation of the prototype filter (HL(z)) which is again
designed using Distributed Arithmetic (DA) Algorithm
which reduces the hardware complexity. This proposed filter
can adapt to optimum bands by itself for various hearing loss
audiograms.
Another factor that is important in a hearing aid is the
noise reduction or noise cancellation technique. It is well
known that noise in background reduces the understanding
of speech and that the greater the level of background noise
the greater the reduction in understanding. There is less
redundancy in the speech signal for a person with hearing
loss since part of the speech is distorted because of the
hearing loss. As a consequence, people with hearing loss
have much greater difficulty than normally hearing people in
understanding speech in noise [23].
In this paper, a reconfigurable hearing aid system is
being proposed which is multiplier-less. The proposed
design uses DA architecture which is LUT-less as well as
multiplier-less design making it area and power efficient.
Then the prototype filter is designed to achieve sub-band
distribution using interpolation to match with the audiogram
of the ailing person. Lastly, the proposed design is
programmed for reconfigurability and automatic adaptability
with the hearing audiogram. This paper is organised as
follows: Section II describes about various theoretical
considerations, Section III describes the implementation of
the proposed design, Section IV constitutes experimental
results, conclusion is drawn in Section V and Section VI
describes the future scope of this proposed filter design.
2. Background
2.1 Distributed Arithmetic Algorithm
Distributed-Arithmetic Algorithm DA algorithm is one
of the well-known multiplier-less methods which involves
use of Logic elements (RAMs, ROMs) or Look-Up Tables
(LUTs) to store pre- computed values of filter coefficients.
It is a powerful technique for replacing the use of multiple
and accumulates operations that make the design hardware-
efficient. This multiplier-less architecture of DA algorithm
is implemented using efficient partition of the function in
partial terms using 2’s complement binary representation
of data. The partial terms can be pre-calculated and stored in
LUTs. The use of memory/LUT capacity increases
exponentially with the order of the filter, due to which DA
implementations require 2K words, K being the number of
taps of the filter. Assuming coefficients ck are known
constants, equation (1) can be rewritten as follows:
y[n] =
The binary decimal form of the variable x[n] can be
represented as follows:
x[n] = ………xb €{0 1}
where xb [n] represents the bth bit of x[n] and B
represents the input width. Finally, the inner product are as
follows:
y[n] xb[k]
= c[0](xB-1[0]2B-1 + xB-2[0]2B-2 + ….+ x0[0]20) +
c[1](xB-1[1]2B-1 + xB-2[1]2B-2 + …+ x0[1]20) + …+ c[N-1]
(xB-1[N-1]2B-1 + xB-2[N-1]2B-2 + … + x0[N-1] 20)
= (c[0]xB-1[0] + c[1]xB-1[1] + ….+ c[N-1]xB-1[N-
1]) 2B-1 + (c[0]xB-2[0] + c [1]xB-2[1] +…+ c[N-1]xB-2[N-1])
2B-2 +….+ (c[0]x0[0] + c[1]x0[1] + …..+ c[N-1]x0[N-1])20
The coefficients in most cases for the multiply
accumulate operation are constants. The partial products are
achieved by multiplying the coefficients c[i] in one bit of
data x[i] at a time using AND operation. These partial
products are then added and the result depends only on the
outputs of the input shift registers. Further, the AND
y[n]
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
313
Volume 21, 2022
Fig. 3 Structure of the proposed reconfigurable structure
functions and adders can be replaced by Look Up Tables
(LUTs) whose outputs are the partial products. Input in
sequence manner is fed to the shift register at the input
sampling rate. The serial output is presented to the RAM
based shift registers at the bit clock rate which is n+1 times
(n is number of bits in a data input sample) the sample rate.
The data is stored in particular address in RAM based shift
registers. The scaling accumulator loads the results from
LUTs from LSB to MSB and the filter output will be
accumulated over the time. For a ‘n’ bit input, n+1 clock
cycles will be required for a symmetrical filter to generate
the output [15].
One disadvantage with original DA architecture is that
its LUT size (2K-words) grows exponentially with
increasing the filter order K. DA offset binary coding (DA-
OBC) can be used to overcome exponentially increasing
memory burden of DA. For a K-tap FIR filter, DA-OBC
requires a 2K−1-word LUT. The modified DA-OBC can
reduce the LUT size from 2K−2 to as low as 2 by exploiting
the observation that if the single term inside the LUT can be
relocated outside the LUT, then the mirrored version in the
lower half of the LUT is the upper half of the LUT where
the signs are reversed. The LUT size decreased by half at
every iteration and at last the LUT-less DA architecture can
be achieved. Thus, the LUT-less DA architecture enables
more high-order FIR filter implementation on a given FPGA
platform.
TABLE 1 Filter coefficients as they are defined in LUT for traditional DA
Architecture
B3B2B1B0
LUT data
0000
0
0001
h[0]
0010
h[1]
0011
h[1] + h[0]
0100
h[2]
0101
h[2] + h[0]
0110
h[2] + h[1]
0111
h[2] + h[1] + h[0]
1000
h[3]
1001
h[3] + h[0]
1010
h[3] + h[1]
1011
h[3] + h[1] +h[0]
1100
h[3] + h[2]
1101
h[3] + h[2] + h[0]
1110
h[3] + h[2] + h[1]
1111
h[3] + h[2] + h[1] + h[0]
2.2 Subband Distribution
Considering both the complexity and the performance,
non-uniformly spaced filter banks are mainly preferred in
digital hearing aids. For sub-band distribution, two factors
are needed to be taken into account. One fact is the hearing
resolution is more at lower frequencies than in higher
frequencies. The other fact is associated with the
transmission process of sound signal in ears (cochlea).
Sound waves create vibration which travels from base to
apex when waves enter cochlea. In this process, all the high
frequency components of the signal pass through the base
but only few of the low frequency components of the signal
reach the apex, so the screen near the base suffers more
damage than the screen near the apex. Based on these two
facts, better adaptability can be obtained if more sub-bands
in low frequency range as well as in high frequency range
are created. The common problem of existing non- uniform
filter banks is that is to achieve both low complexity and
low delay at the same time.
To maintain low complexity, the number of sub-bands is
usually kept as small as possible, which will minimize the
matching performance. Additionally, it is observed that low
complexity is usually achieved with the disadvantage of
long delays. Large interpolation factor results in long
processing delays in the filter bank. This is not a good thing
as long delays can cause mismatch in speech and lip-
reading.
3. Implementation
In all the reconfigurable hearing aid devices that were
mentioned in the literature are highly complex and the
coefficient multiplier which is the most resource consuming
in the hardware is used by the other filter bank methods.
Thus, the Distributed Arithmetic Algorithm is implemented
using modified technique where multipliers as well as LUTs
are not used which reduce the design complexity and make
it area and power efficient. The diagram of the proposed
reconfigurable hearing aid system is shown in Fig 3. Each
filter bank shown in the figure below is a combination of
interpolated FIR filters and each filter is designed using DA
algorithm, it is explained in the upcoming sub-sections.
3.1 FIR filter design using multiplier-less DA
algorithm
Basically, selection of filter coefficients with respect to
the input signal is the main idea taken behind
implementation of the LUT-less DA architecture. The filter
coefficients as shown in Table 1 is for a 4-bit filter order
where, B3 B2 B1 B0 are the input bits and h[0] h[1] h[2]
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
314
Volume 21, 2022
Fig. 4 A 4-Tap FIR Filter using Distributed Arithmetic
Algorithm
h[3] are the filter coefficients obtained from FDA tool in
MATLAB. This LUT coefficient data is important factor to
implement a LUT-less DA architecture. Instead of LUT,
Multiplexer and slice blocks replace the LUTs. As it is seen
in Table 1 with different values of B3 B2 B1 B0 filter
coefficients are selected and added to get the desired results
of the Filter. Therefore, the following diagram is shown in
Fig 4 which is implemented in this paper in place of
traditional DA architecture.
The structure shown in Fig 4 is considered to be basic
design for FIR filter, HL(z). HL(z) is a low pass filter which
will be used further to design reconfigurable hearing aid
system. One of the advantages of FIR filter is the
symmetricity of its coefficients due to which FIR filter HL(z)
is used for interpolation and decimation operations.
3.2 Band formation in FIR filter
Frequency Response Masking is considered in this
proposed design. Multirate signal processing is used in these
systems where up-sampling and down-sampling is
performed to get variation in the width of the passbands. An
interpolation by M will increase the sampling rate in time
domain and Decimation by N will decrease the sampling
rate in the time domain. HL(z) is the basic prototype filter
with a bandwidth of π/3 using Distributed Arithmetic
Algorithm. Further, a high pass filter HH(z) and mid
frequency range filter HM(z) is designed from HL(z) as given
in the equations (4) and (5).
HH(z) = z-N/2 – HL(z) ............... ...(4)
HM(z) = 1 – HL(z) – HH(z) ..........(5)
where N is the length of the filter. The equations used for
generating sub-filters from the prototype filters are given in
Table 2.
TABLE 2 Equations for generating interpolated filters
HH(z2) = z−N/2 − HL(z2)
HM(z2) = 1 − HL(z2) − HH(z2)
HH(z4) = z−N/2 − HL(z4)
HM(z4) = 1 − HL(z4) − HH(z4)
HH(z8) = z−N/2 − HL(z8)
HM(z8) = 1 − HL(z8) − HH(z8)
HC(z2∕3) = z−N/2 − HL(z2∕3)
HC (z4∕3) = z−N/2 − HL(z4∕3)
Using the above equations, a 16-band FIR filter can be
constructed as shown in the Fig. 5. The above diagram can
be also be modified to implement 3-band/8-band/9-band/17-
band FIR Filter system. The bandwidths of different sub-
bands are π/3, π/6, π/12, π/24 and these bandwidths are
selected differently as per design requirements in Table 3.
In the proposed design a comparison is done by
implementing different sub-band system in order to check
the minimum matching error which should be within the
standard limit ± 3 dB for different Audiograms. Eight types
of audiograms are taken into consideration in this paper. A
comparison table is shown in the Table 4 from which it can
be considered which filter is optimum for different
Audiograms depending on their matching errors and delays.
The maximum delay should be limited to 20ms otherwise it
will cause a mismatch in synchronizing the visual lip
reading and the audio being processed. If the audiogram of
any human is within the range of 20dB then the person is not
suffering from any hearing losses. But different person can
have different audiograms for different hearing losses. Filter
banks having lesser sub-bands shows more matching errors
as seen in Table 4.
3.3 Implementation of the complete proposed
system design
In order to achieve automatic reconfigurability study of
the audiogram is necessary. The values achieved in different
Audiograms are considered for auto- reconfigurabilty. The
Audiogram can be divided into 3 parts based on frequency
ranges 0-2.5 kHz, 2.5-5.5 kHz and 5.5 – 8 kHz. So, changes
in graph in these regions can be taken care of by the allotted
filter banks. When the graph is almost flat then filter banks
with lower sub-bands are selected and when the graph
shows sharp variation then filter banks with higher sub-
bands are selected. An audiogram describes the mildest
sound that can be heard at test frequencies of 250 Hz,
500Hz, 1kHz, 2kHz, 4kHz and 8kHz by the hearing
impaired. The audiograms have six distinct hearing threshold
values which are represented as ‘ti’ at different octaves. The
gradients are calculated as ‘gi = ti+1 – ti’ and the maximum
gradient in any frequency range is considered for optimum
filter bank selection. The maximum value of gradients in kth
region can be termed as slope, ‘Sk’. There will be 5 gradients
which can be further divided (considering the normalised
frequency ranges) as-
For Region 1, frequency range - 0 – 2.5 kHz
Slope Value = S1 = max(|g1|,|g2|,|g3|)
For Region 2, frequency range - 2.5 – 5.5 kHz
Slope Value = S2 = max(|g4|,|g5/3|)
For Region 3, frequency range - 5.5 – 8 kHz
Slope Value = S3 = max(|2g5/3|)
Four type of filter banks are chosen based on Table 4
looking at their audiogram matching performances, while 1st
type of filter bank is a combination of HL(z), HH(z) and HM(z)
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
315
Volume 21, 2022
Fig 5. 16-band FIR architecture
which gives 3 sub-bands. The optimum filter bank selection
is done from the slope values Sk of maximum compounded
gradients. If Sk is within the upper threshold of 5dB, filter
bank 1 is suggested in the region ‘k’. Similarly, when Sk is
within the upper threshold of 10dB, filter bank 2 is suggested
in the corresponding region. Again, an upper threshold of
15dB suggests filter bank 3 and that of 30dB suggests filter
bank 4 in the respective regions. The following are the design
specifications –
TABLE 3 Design Specifications
Design Specifications
Value
Sampling Frequency
16 KHz
Passband ripples
0.05 dB
Stopband Attenuation
50 dB
Transition Bandwidth
0.175
Filter Order of the prototype
35
Bandwidth of subbands
π/3, π/6, π/12, π/24
Passband edges
0.2459
Stopband edges
0.4209
When the variations in the graph is sharp then most of the
time higher sub-band system is considered while for the
graphs with minimum variation lower sub-band systems are
preferred. So, depending on the slope values, different filter
bank will be selected for the respective regions. The whole
concept mentioned above is programmed in the MATLAB.
In MATLAB data is described in the floating-point form
while described in the fixed- point form in this FPGA
system. After quantizing the filter coefficients using 12-bit-
width signed binary[23], we can obtain the final coefficients.
.
TABLE 4. Comparison table of matching errors for different sub- band
system
Hearing loss
Type
6 band non
uniform
8 band non
uniform
9
band non
uniform
16 band non
uniform
17 band non
uniform
MME
MME
MME
MME
MME
Type 1
3.85
1.71
6.14
7
2.7
Type 2
4.14
1
2
2.52
2.71
Type 3
1.28
1
2.53
2.9
2.28
Type 4
3.52
1
1.28
1.42
1.28
Type 5
1.56
0.42
0.71
1.28
1.71
Type 6
5.52
0.71
1.57
1
1.28
Type 7
2.42
2.5
2
3.28
0.71
Type 8
1.14
1
1.42
2.57
1.42
Delay
(ms)
15.37
7.67
12.88
6
11.78
In Table 4, TYPE 1 mild hearing loss in the high
frequencies, TYPE 2 mild to moderate hearing loss in low
frequencies, TYPE 3 mild hearing loss in all frequencies.
TYPE 4 Most common type of hearing loss caused due to
ageing specially in the consonant areas. TYPE 5 Common
type of hearing loss seen in older workers working in noisy
industries. TYPE 6 Moderate type of hearing loss. Patients
lost much of loudness in speech. TYPE 7 High hearing loss
in low frequencies, severe hearing loss in middle frequency
and total hearing loss in high frequencies. TYPE 8 Profound
or severe hearing loss
4. Results and Observations
In all the existing hearing aid automatic adoption to
optimum bank selection was lacking, most of the hearing
devices depend on manual interventions to achieve the best
hearing bank. The proposed design can automatically
reconfigure to the best-suited bank. The frequency responses
and matching error results of some audiograms are shown
in Fig. 6 (a) – Fig. 6 (l). The best filter bank is assigned for a
particular audiogram for different regions of the signal
without manual interference. Even the audiograms with
sharp variations in different regions can also be matched
successfully. The programmable block in Fig. 3 is a
MATLAB function block which is an interface between
MATLAB coding script and Xilinx environment. The logic
used in this programmable block is Implementation section.
The proposed reconfigurable filter has very less
complexity as the most power and area component i.e.,
multiplier is not used in the implementation. The proposed
FIR filter with filter order ranging to 1024 (35 in this case)
can be implemented using zero multiplier. The hardware
complexity, delay and maximum matching error of different
filter bank implementation method is listed in Table 6.
TABLE 5. Comparison results for all the Audiogram of the proposed
system
Type
MME
(in dB)
Filter bank
selection
Delay
(in ms)
R1
R2
R3
Audiogram 1
2.90
1
2
1
4.00
Audiogram 2
1.51
1
2
2
3.89
Audiogram 3
2.00
1
4
3
12.89
Audiogram 4
1.53
1
3
2
5.62
Audiogram 5
2.28
2
2
4
8.46
Audiogram 6
2.70
2
2
3
8.51
Audiogram 7
1.76
2
3
4
18.64
Audiogram 8
1.29
3
4
2
17.77
The Audiogram matching of the proposed system for
different hearing losses and their matching error are shown in
the previous figures, also in Table 5 the detailed results are
shown where R1, R2, R3 denotes Region 1, Region 2,
Region 3 of the input signal and the corresponding filter
bank selected for that particular region. MME denotes
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
316
Volume 21, 2022
Fig 6. (a) – (d) TYPE 1 Audiogram, Matching curve & Matching error, (b) – (e) TYPE 3 Audiogram, Matching curve & Matching error, (c) – (f)
TYPE 4 Audiogram, Matching curve & Matching error, (g) – (j) TYPE 5 Audiogram, Matching curve & Matching error, (h) – (k) TYPE 7
Audiogram, Matching curve & Matching error, (i) – (l) TYPE 8 Audiogram, Matching curve & Matching error.
Maximum Matching Error in Table 5. The design and
analysis of the proposed structure are done using the Xilinx
System Generator in MATLAB R2017b where software
simulation is done. The device utilization and power
dissipation factors are evaluated using Xilinx Vivado
2018.3 software and it is shown in Fig 7 and Fig 8.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
317
Volume 21, 2022
Table 6 Comparison of the proposed system with existing methods
Fig 7. Device Resource Utilization of the proposed system
Fig 8. Power Utilization of the proposed system
5. Conclusion
A multi-filter reconfigurable hearing aid system is
implemented using frequency masking response based on
interpolation and decimation techniques which further
divide the audio signal into several sub-bands. The
prototype filter is implemented using Distributed Arithmetic
Algorithm which is a multiplier-less technique making the
system area- efficient and power-efficient. This
reconfigurable filter is also automatic where the audio
spectrum is divided into three regions in order to study the
variations in the audiogram more properly and the best-
suited filter bank is assigned to the hearing profile with
maximum matching error 2.71 dB which is within the
standard limit (± 3 dB). The delay analysis of the proposed
filter is done and it is seen that the maximum delay is
18.6ms. A general comparison with existing filter bank
designs is shown in Table 6 which depicts the efficient
working of the proposed filter compared to the previously
reported filter designs. Finally a multiplier-less filter bank
for hearing aid is successfully implemented with sub-bands
ranging from 3 to 16. This proposed design saves the device
resource utilization upto 64% as seen in Fig 7 and power
consumption upto 24% as presented in Fig 8. Thus, a
simplified automatic reconfigurable filter is successfully
implemented which avoids the tiresome manual matching of
bands with hearing profiles. This method not only reduces
hardware replacement with change in hearing profile but
also proves to be efficient in terms of hardware utilization.
6. Future Scope
Further, in this system noise cancellation feature
can also be introduced by studying accurate estimate of the
noise spectrum as it varies over time. Various filtering
methods like spatial filtering, adaptive noise cancellation,
weiner filtering can be used to avoid the effect of noise on
the hearing aid. One such method that can be effective in
this proposed system is adaptive noise cancellation
technique. Adaptive noise cancellation requires at least two
directional microphones and, under ideal conditions, at least
one microphone must be placed at the noise source.
However, both microphones are placed near the head with
one microphone picking up more speech than noise and the
Method
Type of Filter
Number
of Bands
MME
Max.
Delay
(ms)
No of
Multipliers
Variable
Bandwidth
Filters [8]
Reconfigurable
3 to 8
2.47
dB
1.1
216
Interpolation
[9]
Reconfigurable
17
5.62
dB
21.6
76
Frequency
Response
Masking [4]
Reconfigurable
16
2.26
dB
6
30
Quasi-ANSI
Filter [18]
Fixed
18
3.26
dB
10
226
Sub-band
Distribution
[1]
Reconfigurable
3 to 8
4.56
dB
5
120
3-level octave
interpolation
[2]
Reconfigurable
3 to 17
2.39
dB
17.8
18
Proposed
System
Reconfigurable
3 to 16
2.71
dB
18.6
-Nil-
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
318
Volume 21, 2022
other microphone picking up more noise than speech. The
output of the adaptive filter which is the noise signal will be
subtracted from the combined signal where speech signal
and noise signal are added. This will reduce the effect of
noise to some extent and an improved speech-to-noise ratio
will result within the upper range of 20dB, with improved
intelligibility [23].
References
[1] Manik Sonawane & Sangita R.Chougule ,“Study and Analysis of
Audiogram Matching for Reconfigurable FIR Filter Bank Use in
Hearing Aids”, International Journal of Applied Engineering
Research , Volume 15, 2020
[2] Tomson Devis, Manju Manuel, “Hardware-efficient auto-
reconfigurable hearing aid using 3-level octave interpolated filters for
auditory compensation applications”, Physical and Engineering
Sciences in Medicine, 2021
[3] Tong Ma, Changpeng Shen, Ying Wei, “Adjustable Filter Bank
Design for Hearing aids System”, IEEE, 2019
[4] Tomson Devis, Manju Manuel, “A 17-Band Non-uniform
Interpolated FIR Filter Bank for Digital Hearing Aid”, International
Conference on Communication and Signal Processing, Volume 19,
2018
[5] Harry Levyitt, “Digital hearing aids: A tutorial review”, Journal of
Rehabilitation Research and Development, Volume 24, 1987.
[6] Ryota Shimokura, “Hearing Aids”, Intechopen, 2018
[7] Merlin Livingston L.M., Agnel Livingston L.G.X, Nagercoil Jenila
Livingston L.M., “Design of Hearing Aid Using Variable Bandwidth
Filter”, International Journal of Scientific & Engineering Research,
Volume 7, Issue 10, October-2016.
[8] James T. George & Elizabeth Elias, “A 16-Band Reconfigurable
Hearing Aid using Variable Bandwidth Filters”, Global Journals Inc.
(USA), Volume 14, Issue 1, 2014.
[9] Dhanya N, D Raja Godwin, “Reconfigurable filter bank structures for
hearing aid applications”, International Journal of Pure and Applied
Mathematics, Volume 118, 2018.
[10] Nayan Modhave, Yepuganti Karuna, Sourabh Tonde,“Design of
Matrix Wiener Filter for Noise Reduction and Speech Enhancement
in Hearing Aids”, IEEE, Volume 7, 2016
[11] Shankarayya G. Kambalimath, Prem C. Pandey, Pandurangarao N.
Kulkarni, Shivaling S. Mahant-Shetti, and Sangamesh G.
Hiremath,“FPGA-Based Design of a Hearing Aid with Frequency
Response Selection through Audio Input”, IEEE, 2016
[12] Anurag Aggarwal, Astha Satija, Tushar Nagpal, “FIR Filter
Designing using Xilinx System Generator”, International Journal of
Computer Applications, Volume 68, Issue no 11, 2016.
[13] Igor Dzukleski and Cesar Ortega-Sanchez,“Reconfigurable FIR
FILTER in FPGA”, Curtin University of Technology, 2005
[14] Nisha Haridas, Elizabeth Elias, “Efficient variable bandwidth filters
for digital hearing aid using Farrow structure”, Journal of Advanced
Research, Volume 7, 2016
[15] Hsi-Shou Wu, Zhengya Zhang, Marios C. Papaefthymiou, “A
13.8μW Binaural DualMicrophone Digital ANSI S1.11 Filter Bank
for Hearing Aids with Zero-Short- CircuitCurrent Logic in 65nm
CMOS”, ISSCC, Issue 114, 2017.
[16] Lama Shaer, Ihab Nahlus, Jawad Merhi, Ayman Kayssi & Ali
Chehab, “Low-Power Digital Signal Processor Design for a Hearing
Aid”, IEEE, 2018
[17] Shin-Chi Lai, Chih-Hao Liu, Ling-Yi Wang, Shin-Hao Chen, and Ke-
Horng Chen, “11.25-ms-Group-Delay and Low- Complexity
Algorithm Design of 18-Band Quasi-ANSI S1.11 1/3 Octave Digital
Filterbank for Hearing Aids”, IEEE, 2015.
[18] Rainy Chaplot, Anurag Paliwal, “Low Power Reconfigurable FIR
Filter Based on Window Techniques for On Chip Network”, IEEE,
2013
[19] Lena L. N. Wong, Yuan Chen, Qianran Wang, and Volker Kuehnel,
“Efficacy of a Hearing Aid Noise Reduction Function”, National
Library of Medicine, 2018.
[20] Mr. Mayur B. Kachare, Prof. D. U. Adokar, “A Distributed
Arithmetic (DA) Based Digital FIR Filter Realization”, International
Conference on Global Trends in Engineering, Technology and
Management, 2016
[21] Priyanka Nain, Dr. G.S. Virdi, “Look-Up Table based Scaling
Accumulator in MAC using VHDL”, IJEE, Vol 9, Dec 2017
[22] Yajun Zhou, Pingzheng Shi, “Distributed Arithmetic for FIR Filter
implementation on FPGA”, IEEE, 2019
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the Creative
Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en_US
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2022.21.34
Atryee Bhuyan, Manash Pratim Sarma, Nikos E Mastorakis
E-ISSN: 2224-2678
319
Volume 21, 2022
