International Conference on Communication and Signal Processing, July 28 - 30, 2020, India

Design of Multiplierless IFIR based Cosine

Modulated Filter Bank using QPSO
Virat Krishna, Anil Kumar and G.K. Singh

Abstract—This paper presents an improved technique for Later on, closed form methods were presented for devising
designing nearly perfect cosine modulated multiplierless multirate filterbanks [12], [13]. These techniques were also
filterbank, based on interpolated finite impulse response filter suitable for continuous filter coefficients.
(IFIR) and quantum behaved particle swarm optimization
(QPSO). In this technique, the filter coefficients are represented
The evolutionary techniques such as particle swarm
within canonical signed digit (CSD) space to minimize multipliers. optimization (PSO) [14], Artificial bee colony (ABC)
An adjustable window approach is used to build a sample filter, algorithm [15], genetic algorithm [16], hybrid optimization
i.e. IFIR, and then by using quantum behaved particle swarm algorithm [15], and cuckoo search (CS) optimization [17] have
optimization (QPSO), the quantized coefficients are optimized to been used for optimizing the prototype filter coefficients for
achieve magnitude response of 0.707 at cutoff frequency, by FB, by formulating an objective function, based on either filter
changing cut off frequency of the model filter. Hardware
complexity in terms of adders is reduced by CSD, which is a form or filterbank matrices, because of having capability for solving
of sign power of two terms and aids in implementing multipliers multi-dimensional nonlinear and non-differential problem. In
in terms of adders and shifters. Experimental analysis of above discussed methods [14]-[17], the filterbank was
benchmark design examples has been included to determine the designed with continuous coefficients. Since then, fractional
effectiveness of proposed method by evaluating quality derivative was used for designing continuous filter coefficients
measurement parameters such as reconstruction error, aliasing
for FB, using different evolutionary techniques [18]. If the
error, number of adders in the sign power of two terms (SPT)
representation and canonic signed digit representation. filterbank is designed by using the above techniques, the
quantized and realized performances of the filterbanks are
Index Terms—Cosine modulated filter bank, IFIR filter, improved.
QPSO, sign power of two terms, shifters, multipliers, adders, Digital filter requires multipliers and adders, when they are
canonic signed digit. realized in hardware. Thus, the adders and multipliers play the
key role in hardware implementation of a digital filter.
I. INTRODUCTION Though, while realizing both adders and multipliers, there is

T HE filter bank (FB) in digital signal processing (DSP) is a

set of filters with common input and summed output,
which is viable in several signal processing applications, such
always strong motivation to optimize them, as they affect the
space, power, and the cost of a system. Multipliers are more
expensive, and give more complexity than adders [19].
as signal compression, subband coding of audio signals, Therefore, the authors have developed a new technique for
feature extraction, denoising [1]. Two-channel filterbank and representation of filter coefficients using sign power of two
M-channel filterbank are two classes of the filterbank in which, terms (SPT), in which, the multipliers can be enforced by
two-channel filterbank, is also called as quadrature mirror using simple adders and shifters that makes multipliers less
filterbank [2]. The most commonly used filterbank in the circuit, resulting low computational complexity [20].
signal processing is multi-channel cosine modulated Therefore, several techniques, based on SPT were proposed
filterbanks (CMFB) due to simplicity in design, and the design that reduce area, power and complexity of the circuit system
cost is only for designing single prototype filter. [21]. Then, Canonic signed digit representation was further
In the past research, numerous techniques have been used for filter coefficients, which further minimizes the
explored for designing cosine modulated filterbank [3], [4]. number of multipliers i.e. adders and shifters [22]–[25]. The
The first systematic technique, also known as linear search CSD uses signed digit number system, which contains
optimization, has been used for designing of cosine modulated minimum number of SPT terms. Therefore, the number of
filterbank [5]–[11]. In these techniques, either cutoff frequency adders and shifters reduces and efficient circuit
or passband edge frequency was tuned using linear search implementation with low hardware requirement can be done
optimization for designing a prototype filter for CMFB. They [26], [27]. The quantization of continuous filter coefficients in
were efficient for designing continuous filter coefficients. CSD space stagnates the performance of the filter. Therefore,
to overcome this problem, the genetic algorithm (GA) [28],
Virat Krishna and Anil Kumar are with the PDPM Indian Institute of [29], evolutionary optimization technique [30]–[32] have been
Information and Technology, Design and Manufacturing Jabalpur 482005, used for optimal filter design.
India. (email: {vkrishna.viratkrishna; anilkdee} @gmail.com)
G. K. Singh is with Department of Electrical Engineering, Indian Institute
of Technology Roorkee -247667, India. (email: [email protected])

978-1-7281-4988-2/20/$31.00 ©2020 IEEE 0715

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.
 Therefore, from the above context, in this paper, a
multiplierless cosine modulated filter bank using IFIR filter is H(z)
presented. In order to make the FB multiplier less, the
coefficients of IFIR filter are quantized. The quantized
coefficients degrade the performance of filters. In order to
improve the performance of multiplier less filterbank, QPSO x(n) G(zL) I(z) y(n)
optimization technique is used. The rest of the paper is
structured as follows. A brief summary of CMFB is given in
Section II. In Section III, the description of filter design, CSD
Fig. 1. IFIR Filter
representation of filter coefficients and QPSO optimization are
bank, the magnitude response of filter is given by [33]–[38]:
given respectively. The proposed methodology is shown in 2
Section IV. The outcomes of the proposed method are 2 S S
A(w)  A(w  1, for 0  w  , (6)
illustrated in Section V. In the last, the conclusion is devoted M M
to Section VI. S
and A( w) 0 , for w t . (7)
M
II. DESCRIPTION OF COSINE MODULATED FILTER BANK In cosine modulated filter bank, if Eq. (6) is satisfied in
Cosine modulated filterbank is mostly used in M channel analysis and synthesis bank, amplitude distortion is eliminated,
maximally decimated filterbank [2]-[4]. In CMFB, only one while Eq. (7) leads to proper sampling, so aliasing error is also
prototype filter coefficients are built and remaining of the eliminated. Linear phase filter is chosen to avoid the phase
analysis and synthesis banks is generated by cosine modulation distortion. The impulse response of CMFB in terms of analysis
of the prototype filter coefficients. The advantage of using and synthesis filters i.e. ai (n) and si (n) respectively is given
CMFB is that high stopband attenuation can be achieved. as [35]–[38]:
The reconstructed signal O( z ) relates to input I ( z ) by the i ­°Re( zi ) ˜ cos >S ˜ (2i  1) ˜ (2n  1) / 4 ˜ M @ ½°
ai (n) ® ¾ ˜ a (n) (8)
¯° Im( zi )sin >S ˜ (2i  1) ˜ (2n  1) / 4 ˜ M @¿°
relationship given by [33]–[38]:
M 1  j 2S l
O( z ) D0 ( z ) ˜ I i ( z )  ¦ Dl ( z ) ˜ I i ( z ˜ e M
). (1) ­°Re( zi ) ˜ cos >S ˜ (2i  1) ˜ (2n  1) / 4 ˜ M @ ½°
and si (n) ® ¾ ˜ a ( n) (9)
¯° Im( zi )sin >S ˜ (2i  1) ˜ (2n  1) / 4 ˜ M @¿°
l 1

Input signal attenuation given by aliased function is given by

Dl ( z ) For 0 d i d M where a(n) is the impulse response prototype
M 1  j 2S l
filter of N-length, the phase factor is given by [35]-[38]:
1
Dl ( z )
M
¦ S ( z) ˜ A ( z ˜ e
k 0
k k
M
), (2)
zi (n)
2
i  j , for i even (10)
N 1 N 1
2
where, Ak ( z ) ¦ a (n) ˜z
n 0
k
n
and Sk ( z ) ¦ s (n) ˜z
n 0
k
n
. 2
i  j , for i odd
2
Now D0 ( z) is the overall distortion transfer function for
unaliased input. For perfect reconstruction and to cancel III. FILTER DESIGN
aliasing D0 ( z) c ˜ z  nd , c z 0 where nd is positive integer
and Dl ( z) 0 Where, l is the positive integer with A. Prototype design using IFIR
range l 1,2,3,..., M  1 . Two FIR filters are connected in a cascade arrangement for
In CMFB, generally there are three errors, which interfere narrow band FIR filter design, also known as Interpolated FIR
with the output. The distortion in the amplitude and the phase filter, which is shown in Fig. 1 [37]. The linear phase FIR filter
is known as amplitude and phase deviation. The slow sampling xlp(n) with windowing technique of order N is designed by
of the input signal generates aliasing error, within the output of xlp(n) = w(n)
hilp(n), where hilp(n) ideal low pass filter impulse
the filter bank. The peak to peak reconstruction error is given response and w(n) is the window function, hilp(n) is given by
by [33]–[38]: [33]–[38]:
sin(Zc (n  0.5 N ))
E pp max w ( M ˜ D0 (e jw ))  min w ( M ˜ D0 (e jw )) . (3) hilp (n) (11)
S (n  0.5 N )
The peak aliasing error is given by [36]–[38], Zc is the cut off frequency of ideal low pass filter. For a given
Ea max w E (e jw )) , (4) stop-band attenuation As , order of filter is given by [35]–[38]:
As  7.95
N (12)
1
14.36('Z )
ª M 1 2
º 2
where, E (e jw ) « ¦ D1 (e ) » . where, Zs is the stop-band frequency of pass filter, and Z p is
jw
(5)
«¬ l 1 »¼
the pass-band frequency of low pass filter [35]-[38]:
Eq. (3) and Eq. (4) are the approximation to reconstruction Zs  Z p
and aliasing error. For perfect reconstruction M-band CMFB 'Z (13)
2

0716

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.
 FIR filter at higher order is computationally complex and area. In hardware implementation of a multiplierless
requires more number of adders and multipliers so IFIR filterbank, the continuous coefficients are quantized and
technique is used, because it is computationally efficient. The represented in SPT form. The CSD is a technique, which
main idea is to design IFIR as a cascade of linear phase FIR utilizes the minimum number of SPT terms. Therefore, the
filter, i.e. G(zL) with periodic amplitude response and I(z) that CSD representation reduces the number of adders and shifters.
suppresses the undesired pass band of the periodic filter. This CSD is a three digit system -1, 1 or 0. CSD also reduces the
can be achieved by single pass band filter with sharper non-zero digits by n/2 where, n is the number of non-zero
transition band and a narrower pass band than the parent FIR. digits in binary expression of 2’s compliment [24].
The width of transition band and pass band of overall cascade
are 1/Lth of those of G(z). Basically, transfer function h(z) = TABLE I
THREE BIT CANONICAL SIGNED DIGIT NUMBER
G(zL)
I(z). G(zL) is the shaping filter, which is realized from
model filter G(z), by adding L-1 no of zeros in the samples of Number 2’s comp. CSD
G(z). The value of stretch factor or upsampling factor ‘L’ 3 011 101’
should be chosen, such that it reduces the computational 2 010 010
complexity, provided by the IFIR filter depending on the 1 001 001
applications for which it is designed. Lopt. for minimum number 0 000 000
of multipliers in hardware implementation is given by Mehrnia -1 111 001’
and Wilson [39]: -2 110 01’0
2S -3 101 1’01
Lopt . (14)
ws  w p  2S ( ws  w p ) N

FIR filters provides the linear phase, and therefore was used Coefficient ‘z’ in CSD can be defined as z s ¦ bz
i M
i
i
where
in most of the applications. The increased order of FIR filter
makes the system computational complex. Therefore, IFIR ‘M’ and ‘N’ are positive numbers, which represents the integer
filter was used at higher order of FIR, which reduces the and fractional part of z respectively [35]–[38]:
n n
number of multipliers and also the computational cost [35]-
[38].
D ¦i 0
bi ˜2i ¦d ˜ 2
i 0
i
i
(17)

(multiFIR  multiIFIR ) CSD encoding scheme is same as Booty’s encoding method,

%Computationalreduction (15)
multiFIR where comparative analysis of the adjacent pairs is done from
lower significant bit to higher significant bit plus one. The
B. CSD representation of IFIR filter coefficient digit di is corresponding CSD digit of 2’s compliment no bi
CSD is the sign digit number system, which increases the and bi+1. The method of conversion of 2’s compliment to
computation speed [25], [26].CSD has advantages such as low CSD is given in Fig. 2. The detailed analysis of CSD can be
power consumption, high speed and efficient handling of chip found in [18], [40]. Table I shows the 3-bit CSD number.
C. Optimization using QPSO
START
QPSO is covariant of PSO which has fever control
parameters and outperforms PSO in searching ability [41],
Get B = bn-1, bn-2 ,…, b1, b0 [42]. QPSO has better global convergence. The problem is
formulated to achieve specified strength with multiple
i=0 objectives. The number of layers, its stacking sequence and
Initial carry ci = 0 thickness of each layer are primary optimization variables.
The QPSO is special type of swam based algorithm, which
is the expanded version of particle swarm optimization as
ci+1 = carry(bi+bi+1+ci)
position and velocity of a particle cannot be defined at the
same time. The theory of wave function of quantum mechanics
di = bi + ci - 2ci+1 put forward for consideration, which enhance the inspection
and optimization is designated as quantum particle search
i = i+1
optimization. The search space is given by [42]–[44].
X ik vik r jik ˜ log §¨ 1 ·¸ , (18)
Yes © \u ¹
i<n where, vi shows how fast the individual solution is moving
towards the best solution. ji is the mean difference of
No individual solution to space computed by [42]–[44],
STOP \ 1Pbestik  \ 2Gbestik
vik , (19)
\1 \ 2
Fig. 2. Flow chart for 2,s compliment to CSD conversion

0717

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.
 START TABLE II
PERFORMANCE EVALUATION OF M CHANNEL COSINE MODULATED FILTER
BANK
Initialize iteration, population, Nm Ni M L RC Aliasing %CR SPT CSD
tolerance. error error ’s
×10-3
Set the range of wc, MR=0.707 84 26 4 2 2.1 1.12×10-4 43.00 157 122
92 26 4 2 1.9 2.9×10-4 43.54 164 119
84 26 8 2 6.3 1.59×10-5 43.00 148 102
Design prototype filter 92 26 8 2 2.1 3.84×10-5 43.54 150 114
100 26 8 2 4.2 1.99×10-4 44.00 156 119
108 42 16 2 6.3 1.7×10-4 41.63 176 136
Quantization and CSD conversion of filter 124 42 16 2 8.4 3.7×10-4 42.56 212 151
coefficients 124 74 16 2 14 9.9×10-5 38.31 218 153
108 84 32 4 12 4×10-4 62.71 275 203
172 84 32 4 55 5×10-4 66.79 375 302
Calculate magnitude response at cut off frequency.
Error = MR – Measure MR
deteriorates the performance. Using QPSO, the performance of
the system is upgraded by designing a better prototype filter.
Yes
Error < tolerance Store the Pbest The cut-off frequency approach differs in order to minimize
error in magnitude response at π/M relative to the ideal
No magnitude response (0.707), when the error is minimum, the
Iterate for new population program is terminated and the frequency is stated. Once a
prototype filter is constructed, the cosine modulated filter bank
No iteration ≤
is produced by cosine modulating the filter coefficients. To
maxiterations reduce hardware complexity, the experimental filter
coefficients are implemented in CSD form, so that number of
Yes
Evaluate the values of Gbest from the
multipliers can be reduced. The reconstruction error, aliasing
obtained Pbest values error and number of adders are calculated. The technique
suggested is shown in Fig. 3 and the observational results are
Optimum cutoff frequency = Gbest shown in the Table II.
Error=minimum
0 -6.4603

Design prototype filter at Gbest

-20 -38.0442
Magnitude (dB)

Stop

Phase (radians)
-40 -69.628

Fig. 3. Flowchart for proposed method

jik P Pmeanik  J ik , (20) -60 -101.2118

W
1
and Pmeank
W
¦ Pbest
i 1
i,c , for all c. (21) -80 -132.7957

where, \ 1 , \ 2 and \ 3 are random variables with uniform -100 -164.3795

distribution in the range (0,1).

-120 -195.9634
0 0.1 0.2
0.3 0.4 0.5 0.6 0.7 0.8 0.9
D. Parameter setting Normalized Frequency (uS rad/sample)
The design parameter of prototype include mutation Fig. 4. Plot of magnitude and phase response of experimented filter
coefficients in CSD space
probability (0.25), cut off frequency and order of both model
and interpolation filter. The population size is 100 and the
V. RESULT AND DISCUSSION
maximum and minimum set values for frequency are 1 and 0
respectively. The update equation for beta is given by [42]– A MATLAB program based on the proposed method,
[44]: described above, was implemented. Several examples showing
E 1  ((0.5 u (i  1)) / max iter ); (22) the effectiveness of proposed method in the CSD space are
mentioned. The performance evaluating parameters from Eq.
IV. PROPOSED METHODOLOGY (3), Eq. (4) and Eq. (15) have been calculated and mentioned
in Table II. In the Fig. 4, the respective plot of magnitude and
In this paper, an IFIR based cosine modulated filter bank is
phase response of a filter after conversion into the CSD space
designed using quantum behaved particle swarm optimization.
is given. As depicted in Fig. 4, the phase response is almost
Initially, model filter, up sampled filter and the interpolation
linear in nature and so the coefficient represents the
filter are built according to the specification provided. The up
symmetricity. Thus, the calculations are done on the half of its
sampled and the interpolation filter is convolved to form a
coefficients. The Table III shows a comparison between the
prototype filter. The coefficients of filter are quantized which

0718

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.
 20
50 Subband 1 0.08
0 Subband 2
0.06
Magnitude in db [H (Z) ]

Magnitude in db [H(Z) ]
Subband 3

Magnitude in db [H(Z) ]
-20 0 Subband 4 0.04
-40 Subband 5 0.02
Subband 6
-60 -50 0
Subband 7
-80 Subband 8 -0.02

-100 -0.04
-100
-0.06
-120
-150 -0.08
-140
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) Normalized Frequency ( Z) (b) Normalized Frequency ( Z) (c) Normalised frequency ( Z)
Fig. 5. (a) Plot of frequency response of experimented filter (b) Plot of frequency response of filter bank for all the 8 channels (c) Plot of reconstruction error
TABLE III banks based on Hopfield neural network,” IOP Conf. Ser. Mater. Sci.
PERFORMANCE COMPARISON OF METHOD MOOTED WITH EXISTING METHODS Eng., vol. 569, no. 3, 2019, doi: 10.1088/1757-899X/569/3/032034.
[5] C. D. Creusere and S. K. Mitra, “A Simple Method for Designing
As Aliasin High-Quality Prototype Filters for M-Band Pseudo QMF Banks,” IEEE
Ref. N Nm Ni R.E.
(db) g error Trans. Signal Process., vol. 43, no. 4, pp. 1005–1007, 1995, doi:
1.18×10 10.1109/78.376856.
Kalathil et al.[35] 62 - - - - -5 [6] F. Cruz-Roldán, P. Amo-López, S. Maldonado-Bascón, and S. S.
Dohare and Datar [36] - 101 - - 0.0612 0.0603 Lawson, “An efficient and simple method for designing prototype
Chakrabarti et al. [21] - - 30 28 0.0054 0.0016 filters for cosine-modulated pseudo-QMF banks,” IEEE Signal
Rayavarapu and Process. Lett., vol. 9, no. 1, pp. 29–31, 2002, doi: 10.1109/97.988722.
100 - - - - 4×10-3 [7] F. Cruz-Roldán, P. Martín-Martín, J. Sáez-Landete, M. Blanco-
Prakash [38]
1.59× 2.8×10- Velasco, and T. Saramäki, “A fast windowing-based technique
Proposed method - 193 84 26 exploiting spline functions for designing modulated filter banks,” IEEE
10-3 5
Trans. Circuits Syst. I Regul. Pap., vol. 56, no. 1, pp. 168–178, 2009,
doi: 10.1109/TCSI.2008.925350.
mooted method and existing one. In Fig. 5, the magnitude [8] A. Datar, A. Jain, and P. C. Sharma, “Design and performance analysis
of adjustable window functions based cosine modulated filter banks,”
response of prototype filter and cosine modulated filter bank
Digit. Signal Process. A Rev. J., vol. 23, no. 1, pp. 412–417, 2013, doi:
are depicted along with the reconstruction error. 10.1016/j.dsp.2012.07.007.
Example: An eight band CMFB is designed using given [9] A. Kumar, G. K. Singh, and B. Kuldeep, “An Improved and Simplified
specifications, which are stated as stopband frequency Approach for Designing Cosine Modulated Filter Bank Using Window
Technique,” J. Math. Model. Algorithms, vol. 10, no. 3, pp. 213–226,
Zs=1/band, passband frequency Zp = 1/(4×band) and cutoff 2011, doi: 10.1007/s10852-011-9150-8.
frequency Zc = (Zp+Zs)/2. Order of model and interpolation [10] A. Kumar, G. K. Singh, and S. Anurag, “Design of nearly perfect
filter Nm = 84 and Ni =26 respectively. The obtained values of reconstructed non-uniform filter bank by constrained equiripple FIR
technique,” Appl. Soft Comput. J., vol. 13, no. 1, pp. 353–360, 2013,
filter bank are aliasing error 2.8×10 -5 and reconstruction error doi: 10.1016/j.asoc.2012.08.024.
6.3×10-3 respectively with 43% computational reduction. [11] K. Shaeen and E. Elias, “Prototype Filter Design Approaches for Near
Perfect Reconstruction Cosine Modulated Filter Banks - A Review,” J.
Signal Process. Syst., vol. 81, no. 2, pp. 183–195, 2015, doi:
VI. CONCLUSION 10.1007/s11265-014-0929-5.
In the paper, multiplierless CMFB have been formed, with [12] A. Kumar, R. Pooja, and G. K. Singh, “An efficient closed-form design
method for nearly perfect reconstruction of non-uniform filter bank,”
the fair reduction in the hardware complexity by limiting the ISA Trans., vol. 61, pp. 167–178, 2016, doi:
number of adders and shifters for multipliers. The QPSO is 10.1016/j.isatra.2016.01.013.
used to improve the IFIR filter performance made by [13] A. Kumar, R. Pooja, and G. K. Singh, “Design and performance of
closed form method for cosine modulated filter bank using different
quantizing the coefficients. As a result, IFIR based design of
windows functions,” Int. J. Speech Technol., vol. 17, no. 4, pp. 427–
CMFB minimizes the hardware requirement. The obtained 441, 2014, doi: 10.1007/s10772-014-9242-8.
negligible values of aliasing and reconstruction error are [14] B. Kuldeep, A. Kumar, and G. K. Singh, “PSO based optimized
accepted in most of the signal processing applications, and fractional derivative constraints for designing M-Channel filter bank,”
Proc. 2015 Int. Conf. Signal Process. Comput. Control. ISPCC 2015,
thus the M -band NPR CMFB is designed analyzed and tested. pp. 140–144, 2016, doi: 10.1109/ISPCC.2015.7375013.
[15] I. Sharma, A. Kumar, G. K. Singh, and H. N. Lee, “Design of
multiplierless cosine modulated filterbank using hybrid technique in
sub-expression space,” Int. Conf. Digit. Signal Process. DSP, vol. 0,
REFERENCES pp. 360–364, 2016, doi: 10.1109/ICDSP.2016.7868579.
[1] H. N. L. Kumar, A., B. Kuldeep, I. Sharma, G. K. Singh, “Advances in [16] S. C. Chan, W. Liu, and K. L. Ho, “Multiplierless perfect
multirate systems,” Adv. Multirate Syst., pp. 1–221, 2017, doi: reconstruction modulated filter banks with sum-of-powers-of-two
10.1007/978-3-319-59274-9. coefficients,” IEEE Signal Process. Lett., vol. 8, no. 6, pp. 163–165,
[2] Vaidhyanathan P.P., Vaidyanathan_1993_Multirate systems and filter 2001, doi: 10.1109/97.923040.
banks.pdf. Prentice hall P T R, 1993. [17] K. Baderia, A. Kumar, and G. K. Singh, “An improved method for
[3] R. D. Koilpillai and P. P. Vaidyanath, “New Results on Cosine- designing cosine modulated filter bank using polyphase components,”
Modulated FIR Filter Banks Satisfying Perfect Reconstruction,” IEEE, 3rd Int. Conf. Signal Process. Integr. Networks, SPIN 2016, no. 8, pp.
vol. 2, no. 3, pp. 1793–1796, 1991. 9–13, 2016, doi: 10.1109/SPIN.2016.7566653.
[4] W. Xu, X. Gao, and J. Zhao, “Design of sparse cosine modulated filter [18] I. Sharma, A. Kumar, D. Kumar, and G. K. Singh, “Performance of

0719

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.
 CSE Techniques for Designing Multiplier-Less FIR Filter Using [32] S. Chiardonna, F. Di Giandomenico, and Nadir Murru, “On enhancing
Evolutionary Algorithms,” Circuits, Syst. Signal Process., vol. 37, no. efficiency and accuracy of particle swarm optimization algorithms,”
6, pp. 2574–2590, 2018, doi: 10.1007/s00034-017-0679-2. Int. J. Innov. Comput. Infromation Control, vol. 11, no. 4, pp. 1165–
[19] Z. Luo and R. B. Lee, “Cost-effective multiplication with enhanced 1189, 2015.
adders for multimedia applications,” Proc. - IEEE Int. Symp. Circuits [33] I. Sharma, A. Kumar, and G. K. Singh, “Adjustable window based
Syst., vol. 1, pp. 4–7, 2000, doi: 10.1109/iscas.2000.857179. design of multiplier-less cosine modulated filter bank using swarm
[20] J. B. E. Yong Ching Lim, Bede Liu, “VLSI circuits for decomposing optimization algorithms,” AEU - Int. J. Electron. Commun., vol. 70,
binary integers into signed power-of-two terms,” IEEE, vol. 2868, no. no. 1, pp. 85–94, 2016, doi: 10.1016/j.aeue.2015.10.008.
m, 1990. [34] I. Sharma, A. Kumar, and G. K. Singh, “An Efficient Method for
[21] Debajyoti Mishra, S. Dhabal, Roshni Chakrabati, and P. Venkteswaran, Designing Multiplier-Less Non-uniform Filter Bank Based on Hybrid
“Canonical Signed Digit Representation of Quadrature Mirror Filter Method Using CSE Technique,” Circuits, Syst. Signal Process., vol.
using Genetic Algorithm,” in IEEE, 2012, pp. 65–68. 36, no. 3, pp. 1169–1191, 2017, doi: 10.1007/s00034-016-0351-2.
[22] R. Bolton, P. C. Craig, and L. C. Westphal, “Computer-Aided Design [35] S. Kalathil and E. Elias, “Design of multiplier-less sharp non-uniform
of Recursive Digital Filters with Coefficients Having Restricted cosine modulated filter banks for efficient channelizers in software
Minimal Representation,” IEEE Trans. Acoust., vol. 29, no. 6, pp. defined radio,” Eng. Sci. Technol. an Int. J., vol. 19, no. 1, pp. 147–
1205–1208, 1981, doi: 10.1109/TASSP.1981.1163698. 160, 2016, doi: 10.1016/j.jestch.2015.06.003.
[23] A. G. Dempster and M. D. Macleod, “Use of Minimum-Adder [36] R. S. Dohare and P. S. Datar, “Quantized Coefficient F.I.R. Filter for
Multiplier Blocks in FIR Digital Filters,” IEEE Trans. Circuits Syst. II the Design of Filter Bank,” Int. J. Eng. Trends Technol., vol. 4, no.
Analog Digit. Signal Process., vol. 42, no. 9, pp. 569–577, 1995, doi: August, pp. 3271–3276, 2013.
10.1109/82.466647. [37] Roshni Chakrabati, S. Dhabal, Debajyoti Mishra, and P. Venkteswaran,
[24] R. T. Hartley, “Subexpression sharing in filters using canonic signed “An IFIR Approach for Designing M-band NPR Cosine Modulated
digit multipliers,” IEEE Trans. Circuits Syst. II Analog Digit. Signal Filter Bank with CSD,” IEEE, pp. 77–80, 2012, doi:
Process., vol. 43, no. 10, pp. 677–688, 1996, doi: 10.1109/82.539000. 10.1109/CODIS.2012.6422140.
[25] R. M. Hewlitt and E. S. Swartzlander, “Canonical signed digit [38] N. Rayavarapu and N. R. Prakash, “An efficient IFIR filter based
representation for FIR digital filters,” IEEE Work. Signal Process. Syst. prototype filter design for cosine modulated transmultiplexers,” 2010
SiPS Des. Implement., pp. 416–426, 2000, doi: Int. Conf. Signal Acquis. Process. ICSAP 2010, no. 2, pp. 315–318,
10.1109/sips.2000.886740. 2010, doi: 10.1109/ICSAP.2010.76.
[26] H. Uppalapati, H. Rastgar, M. Ahmadi, and M. A. Sid-Ahmed, “Design [39] A. Mehrnia and A. N. Willson, “On optimal IFIR filter design,” Proc. -
of quadrature mirror filter banks with canonical signed digit IEEE Int. Symp. Circuits Syst., vol. 3, 2004, doi:
coefficients using genetic algorithm,” 2005 Int. Conf. Commun. 10.1109/iscas.2004.1328701.
Circuits Syst. - Proc., vol. 2, pp. 682–686, 2005, doi: [40] R. Guo and L. S. DeBrunner, “A novel fast canonical-signed-digit
10.1109/ICCCAS.2005.1495205. conversion technique for multiplication,” ICASSP, IEEE Int. Conf.
[27] M. Aktan, A. Yurdakul, and G. Dündar, “An algorithm for the design Acoust. Speech Signal Process. - Proc., vol. 32310, pp. 1637–1640,
of low-power hardware-efficient FIR filters,” IEEE Trans. Circuits 2011, doi: 10.1109/ICASSP.2011.5946812.
Syst. I Regul. Pap., vol. 55, no. 6, pp. 1536–1545, 2008, doi: [41] N. Agrawal, A. Kumar, V. Bajaj, and H. N. Lee, “Controlled ripple
10.1109/TCSI.2008.917997. based design of digital IIR filter,” Int. Conf. Digit. Signal Process.
[28] S. Y. Park and N. I. Cho, “Design of multiplierless lattice QMF: DSP, vol. 0, pp. 627–631, 2016, doi: 10.1109/ICDSP.2016.7868634.
Structure and algorithm development,” IEEE Trans. Circuits Syst. II [42] S. N. Omkar, R. Khandelwal, T. V. S. Ananth, G. Narayana Naik, and
Express Briefs, vol. 55, no. 2, pp. 173–177, 2008, doi: S. Gopalakrishnan, “Quantum behaved Particle Swarm Optimization
10.1109/TCSII.2007.910969. (QPSO) for multi-objective design optimization of composite
[29] V. J. Manoj and E. Elias, “Design of multiplier-less nonuniform filter structures,” Expert Syst. Appl., vol. 36, no. 8, pp. 11312–11322, 2009,
bank transmultiplexer using genetic algorithm,” Signal Processing, doi: 10.1016/j.eswa.2009.03.006.
vol. 89, no. 11, pp. 2274–2285, 2009, doi: [43] S. Dhabal and S. Sengupta, “Efficient design of high pass FIR filter
10.1016/j.sigpro.2009.05.006. using quantum-behaved particle swarm optimization with weighted
[30] X. S. Yang and S. Deb, “Multiobjective cuckoo search for design mean best position,” Proc. 2015 3rd Int. Conf. Comput. Commun.
optimization,” Comput. Oper. Res., vol. 40, no. 6, pp. 1616–1624, Control Inf. Technol. C3IT 2015, no. 1972, pp. 0–5, 2015, doi:
2013, doi: 10.1016/j.cor.2011.09.026. 10.1109/C3IT.2015.7060145.
[31] P. Civicioglu and E. Besdok, A conceptual comparison of the Cuckoo- [44] N. Agrawal, A. Kumar, and V. Bajaj, “A New Method for Designing of
search, particle swarm optimization, differential evolution and artificial Stable Digital IIR Filter Using Hybrid Method,” Circuits, Syst. Signal
bee colony algorithms, vol. 39, no. 4. 2013. Process., vol. 38, no. 5, pp. 2187–2226, 2019, doi: 10.1007/s00034-
018-0959-5.

0720

Authorized licensed use limited to: PES University Bengaluru. Downloaded on September 03,2022 at 10:31:45 UTC from IEEE Xplore. Restrictions apply.