This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

Design of FIR Half-band Filter with Controllable

Transition-band Steepness
Woon Cho, Daewon Chung, Yunsun Kim, Ingyun Kim, and Joonhyeon Jeon
Department of Electronics and Electrical Engineering, Dongguk University, Seoul, 04620 South Korea
Corresponding author: Joonhyeon Jeon (e-mail: [email protected])
This work was supported by Korea Institute of Energy Technology Evaluation and Planning (no. 20194030202320) and Korea Evaluation Institute Of
Industrial Technology (no. 20012884), which are funded by the Ministry of Trade, Industry & Energy of the Republic of Korea.

ABSTRACT Maximally flat (MAXFLAT) half-band filters usually have wider transition-band than other
filters although the frequency response is maximally flat (i.e., no ripples) in the passband and stopband.
This is due to fact that the maximum possible number of zeros at z = ±1 is imposed in half-band close form
solution, which leaves no degree of freedom, and thus no independent parameters for direct control of the
frequency response. This paper describes a novel method for the design of FIR half-band filters with an
explicit control of the transition bandwidth. The proposed method is based on a generalized Lagrange half-
band polynomial (g-LHBP) with coefficients parameterizing a 0-th coefficient h0 and allows the frequency
response of this filter type to be controllable by adjusting h0. Then, h0 is modeled as a steepness parameter
of the transition-band and this is accomplished through theoretically analyzing a polynomial recurrence
relation of the g-LHBP. This method also provides explicit formulas for direct computation of design
parameters related to choosing a desired filter characteristic (by a reasonable trade-off between the
transition-band sharpness and passband & stopband flatness). The examples are shown to provide a
complete and accurate solution for the design of such filters with relatively sharper transition-band
steepness than other existing half-band filters.

INDEX TERMS Maximally flat FIR filters, FIR digital filters, interpolator, closed-form polynomial,
transition-band steepness

I. INTRODUCTION almost half of their impulse response (IR) coefficients are

Maximally flat (MAXFLAT) filters are one of the most zeros. However, their transition-band is relatively wider than
important types of non-recursive finite impulse response other filters and can be narrowed only by increasing the
(FIR) filters and are applied when high stopband attenuation length of the filter. Most of the popular MAXFLAT FIR half-
or smooth frequency response is desired [1]-[5]. The basic band filters are designed by the Lagrange half-band
idea for the design of MAXFLAT FIR filters is to use a polynomial (LHBP) which has the maximum number of
mathematically proved closed-form solution which satisfies zeros at 𝑧 1 [21], [22]. This class of filters has many
MAXFLAT constraints at the ends of the frequency band and applications such as filter banks, wavelets-based compression,
is mapped to the transfer function for the computation of and multirate techniques [15], [16], [19], [23], [24]. However,
coefficients of filters [6]-[13]. However, classical design similarly to the existing MAXFLAT FIR half-band filters,
involves approximation of the desired frequency response by the LHBP filters do not also have any independent (“free”)
some suitable closed-form polynomial [14]-[18] because parameters. In other words, there is no direct control over the
such a closed-form solution mainly focuses on the flatness of frequency response of LHBP filters in order to achieve a
the filter but not on the exact frequency response [11]-[13]. reasonable trade-off between stopband attenuation and the
Several methods and implementation tricks have been width of the transition-band. This is due to fact that the
proposed for the design of MAXFLAT FIR half-band filters maximum possible number of zeros at 𝑧 1 is imposed on
[15], [16], [19], [20]. These filters have exact cutoff a half-band closed form solution, which leaves no degree of
frequency at the middle of the frequency band ω = π / 2 and freedom, and thus no independent parameters.
allow computationally efficient implementations because

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

To address such a narrow transition band issue in starting with a closed-form half-band polynomial. Design
designing FIR filters so far, there have been a lot of efforts in regularity is to impose zeros at 𝑧 1 on the half-band
the direction of frequency response control by utilizing polynomial, and then, the number of zeros must be less than
various design technologies [14]-[19], [25]-[29]. Rodrigues the maximum possible. First, we present a generalized
and Pai introduced a sharp transition FIR filter design using Lagrange half-band polynomial (g-LHBP) whose all IR
sinusoidal functions of frequency to evaluate the impulse coefficients are represented with a 0-th IR coefficient h0 for a
response coefficients in closed form [25]. This method given order of filter. Then, through analyzing a linear
allows closed-form parameters for simple and direct recursive relation of the g-LHBP, h0 is parameterized to
computation, but there occur non-negligible amplitude directly control the transition-band steepness (or width) of
distortions in the stopband and passband. San-José-Revuelta this filter type. Using the new approach, we develop a design
et al. have reported an intelligence metaheuristic-based procedure that is computationally more efficient and accurate
iterative method using multi-fitness function combined with than the previous methods. Also, this new technique provides
a weighted error function [18]. However, this filter design explicit formulas for the performance evaluation of a
has the disadvantage of requiring complicated and enormous resulting filter and consequently, allows unusual flexibility in
computation processing for adjusting the ripples of the bands choosing a best filter with a desired magnitude response
and the width of the transition-band. In contrast to previous characteristic (namely, with trade-off between the transition-
studies [18], [25], Frequency Response Masking (FRM)- band sharpness and passband & stopband flatness). Design of
based filters outperform other designed filters in terms of the FIR half-band filters through this new approach gives an
transition bandwidth and maximum passband ripple for a additional insight into the physical significance of some
given order of filter within an acceptable limit [17], [19], [28]. independent parameters for an explicit control of the
Recently, Roy and Chandra reported enhanced FRM method frequency response.
using an interpolated band-pass filter, resulting in novel This paper is organized as follows. In Section II we
FRM-based filters with more excellent frequency derive a generalized Lagrange half-band polynomial with h0.
characteristics [28]. However, this filter bank structure-based In Section III, an objective control function is derived from
design requires high design complexity to obtain an a recursive relation of the g-LHBP and analyzed to
interpolated prototype filter and two masking filters which parameterize h0 as a steepness control factor of the
configure the FRM filter bank. Moreover, there occur serious transition-band. Additionally, various formulas are
problems such as high group delay and the generation of proposed to design this filter type efficiently and accurately
aliasing band, and the magnitude response of this filter type with a narrow transition band. In Section IV, design
never passes through the half-band cutoff frequency ω = π / 2. examples that demonstrate the power of the new technique
Design methods mentioned above focus on realizing narrow are shown. In addition, to show the effectiveness of the
transition band filters, but not half-band filters with narrow proposed method, g-LHBP filters are compared to existing
transition bandwidth. FIR half-band filter design should state-of-art filters. Conclusions are drawn in Section V.
allow frequency control factors for narrow transition-band by
considering limited frequency characteristics such that half of II. Generalized Lagrange Half-band Polynomial
their coefficients are zero as well as being symmetrical. Khan When Let H(z) be a general symmetric FIR half-band filter
[30], [31] reported FIR half-band filters with narrow of type II (odd number of odd-symmetric coefficients) with
transition-band that have their points of flatness at the middle the real impulse response hn of order 4K – 2, which can be
point between the passband and stopband, but such a filter written as
does not have any independent parameters to control the
𝐻 𝑧 𝑧 𝑄 𝑧 (1)
frequency response and the sharpness of the transition-band
can be achieved with a comparatively higher value of filter by using the transfer function
length. Ma et al. have proposed a cascaded half-band filter
𝑄 𝑧 0.5 ∑ ℎ 𝑧 𝑧 (2)
design by controlling stopband attenuation for a fixed
transition bandwidth [16]. This multistage algorithm allows Where QK(z) represents a zero-phase half-band lowpass
computation reduction of more than 4% per input sample as filter. Design of MAXFLAT FIR half-band filters for the
compared to conventional filters, but there exists the filter type of (2) can be easily realized by using some suitable
limitation of half-band filter design because a heuristic closed-form polynomial [10]-[13], [22] which is then
threshold value is required to meet a specified narrow mapped to the filter function by certain transformations.
transition band. Hence, the half-band design is needed for Regularity is imposed in the design of QK(z) by focusing
highly accurate filters with controllable frequency QK(z) to have zeros at 𝑧 1, i.e., terms of the form (1 + z-1).
characteristics – i.e., with a reasonable trade-off between the One of most popular methods for designing MAXFLAT FIR
transition-band sharpness and passband & stopband ripples. half-band filters of order 4K – 2 is to use a Lagrange half-
In this paper, our aim is to design FIR half-band filters band polynomial (LHBP) [21], [22] as below
with controllable frequency responses. This is achieved by

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

ℓ Computation of IR coefficients hn 's, using (10), was reported

𝑄 𝑧 𝑧 ∑ℓ 𝑑 ,ℓ (3)
in [13], [24]. In a similar way, mapping (9) into (2), we can
get
where dK,ℓ is
2 𝐾 1
ℎ ∑ℓ 𝑔ℓ
𝐾 ℓ 1 ℓ ! 𝑛 ℓ
𝑑 ,ℓ (4) 2 𝐾 1
ℓ ! ℓ!
∑ℓ 𝑔 ℓ , 𝑛 1, 2, 3, . . . , 2𝐾 2 (11a)
𝑛 ℓ 𝐾
The LHBP has a maximum number of zeros at 𝑧 1, and
thus, it has a maximally flat response at ω = π, i.e., or equivalently,

ℎ 0 𝑎𝑛𝑑
0, 𝑘 0,1,2, . . . ,2𝐾 1 (5)
2 𝐾 1
ℎ ∑ℓ 𝑔ℓ (11b)
2𝑛 ℓ
It is shown that the LHBP filter does not have any 2 𝐾 1
∑ℓ 𝑔 ℓ , 𝑛 1, 2, 3, … , 𝐾 1
independent (“free”) parameters as described in (3) and there 2𝑛 ℓ 𝐾
is no direct control over the frequency response of the filter
𝐴
obtained by the LHBP. where 0, if A < B or B < 0.
𝐵
Note that when substituting (10) into (11a), the odd number
Let us define that QK(z) shown in (2) has 2(K – 1) zeros at indexed coefficients are 0, as shown in (11b) – i.e., h2n-1 ’s =
𝑧 1: i.e., 0. From (7) the frequency response of this class of filters can
be also expressed in terms of K and h0 as
0, 𝑘 0, 1, 2, . . . , 2𝐾 3 (6)
ℓ
𝑄 𝜔 𝑐𝑜𝑠 ∑ℓ 𝑑 ,ℓ 𝑠𝑖𝑛
The condition of (6) are imposed on (2), and using Lagrange
interpolation at coincident points [30], [31] so that QK(z) has 1 2 ℎ 𝑠𝑖𝑛 𝑠𝑖𝑛 (12)
a recursive relationship similar to (3), a closed-form half-
band polynomial, called a generalized LHBP (g-LHBP), can It is shown from (12) that the frequency response is ever
be obtained in terms of h0 as (see (A.12) in Appendix A) controllable by introducing h0 as a parameter and thus,
various g-LHBP filters with trade-off between transition
ℓ
𝑄 𝑧 𝑧 ∑ℓ 𝑑 bandwidth and magnitude flatness can be obtained. In
,ℓ
addition, (11), using (10), also allows direct computation of
1 2 ℎ (7) the coefficients if h0 is chosen. However, for a given order of
filter 4K – 2, there are an infinite number of FIR half-band
filters due to the large dynamic range of h0. Thus, h0 must be
It is seen that (7) is identical to LHBP of (3) if h0 is given as
modeled as a controllably independent parameter which can
,
be optimized and determined to obtain such a filter with the
ℎ 1 (8) desired magnitude characteristic. This will be explained in
the next section.
where (8) is obtained by additionally imposing a zero at 𝑧
1 on (7). For a general closed-form expression, mapping (7) III. Objective Control Function of g-LHBP: Transition-
to Band Steepness Parameter h0
Now, through the analysis of (12), we consider how to
derive h0 as a steepness parameter for the direct control of the
𝑄 𝑧 𝑧 𝑔 ∑ℓ 𝑔 ℓ 𝑧ℓ 𝑧 ℓ
transition-band edges. Design regularity is here to determine
(9) h0 so that g-LHBP filters have tolerant magnitude distortion
(ripples) but narrow transition band.
we can obtain the interpolation coefficients gℓ 's (ℓ = 0, 1, 2, From the recursive relation of QK(z) (see (A.2) in
…, K) in terms of h0 and K as (see (A.13) in Appendix A) Appendix A) it is shown that AK(z) plays a role as an
objective function for the extension of QK−1(z)LHBP to QK(z)
ℓ ∑ℓ , 2 𝐾 𝑗 (where QK−1(z)LHBP and AK(z) are also indicated in (A.3) and
𝑔ℓ 1
ℓ 𝑗 (A.12) of Appendix A, respectively). Remarkably, QK(ω)
described in (12) can be also rewritten as a recursive relation
ℓ 2 𝐾 1 2𝐾
1 2 ℎ (10)
ℓ 1 ℓ 𝑄 𝜔 𝑄 𝜔 𝐴 𝜔 (13)

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

where QK−1(z)LHBP and AK(ω) are given, respectively, as indicated that using (20) to get h0 in (12) yields QK(ω) that
exactly passes through QK(ωp+) and QK(ωp−) at ω = ωp+ and
ℓ ωp−. Thus, using ωp+ and ωp− as two transition-band edge
𝑄 𝜔 𝑐𝑜𝑠 ∑ℓ 𝑑 ,ℓ 𝑠𝑖𝑛 (14)
frequencies of this filter (i.e., upper and lower edge
frequencies), we can define a transition-band slope of this
𝐴 𝜔 1 2 ℎ 𝑐𝑜𝑠 𝜔 𝑠𝑖𝑛 𝜔 (15) class of g-LHBP filters as below

(Here AK(ω) has been simplified with 𝑠𝑖𝑛 𝜔 2 𝑠𝑖𝑛 𝑐𝑜𝑠 ).

𝑠𝑙𝑜𝑝𝑒 (21)
Hence, we can be sure from (13) that for a given K, since
QK−1(ω)LHBP is a known MAXFLAT lowpass function, AK(z)
characterizes QK(ω) in terms of h0. In other words, we can It is seen, based on (21), that h0 shown in (20) can be used as
describe the effect of h0 on QK(ω) through the analysis of a steepness parameter to directly control the transition-band
AK(ω). From (15) AK(ω) passing through zero value at ω = 0, slop represented with QK(ωp+) and QK(ωp−) for a given K.
ω = π / 2, and ω = π exhibits a “anti-symmetric sinusoid-like Consequently, substituting the upper edge QK(ωp+) =  (then,
shape” with respect to π / 2 as the center point in the range of the lower edge becomes QK(ωp−) = 1 −  from (19)) into
0≤ω≤π. Thus, there are two anti-symmetric peak values that respectively (20) and (21), we can rewrite h0 and slopeK in
are available to assist in identifying influential observations terms of  as
of AK(ω) in terms of h0 on QK(ω). If ωp+ and ωp− are the two
peak frequencies of AK(ω), we can obtain in terms of K, by ℎ , (22)
solving ∂AK(ω) / ∂ω = 0 at ω = ωp±, as √

𝜔 𝑎𝑟𝑐 𝑐𝑜𝑠 ,0 𝜔 𝜋/2 𝜔 𝜋 (16) 𝑠𝑙𝑜𝑝𝑒 , (23)

√

and then, the two peak values AK(ωp+) and AK(ωp−) are where ωp−−ωp+ has been obtained by using (16) on product-
obtained, by substituting (16) into (15), as to-sum transformation (cosωp−) (cosωp+) = ½{cos(ωp− − ωp+)
cos(ωp−+ωp+)}. Then, it can be seen from (23) that the
𝐴 𝜔 1 2 ℎ 1 (17) steepness of the transition-band is controllable by changing
√
the upper-edge parameter  for a given K. To determine h0,
where ωp± (double signs in same order) denotes ωp+ and ωp−, according to (22), so that a g-LHBP filter has a relatively
and is related to ωp+ + ωp− = π from cos(ωp+ + ωp−) = −1. narrower transition band than a MAXFLAT g-LHBP filter, 
Especially, since both QK−1(z)LHBP and AK(ω) is anti- has to be chosen within the limits of QK(ωp+)MAXFLAT <  ≤ 1
symmetric with respect to ω = π / 2 as the center point in 0≤ where QK(ωp+)MAXFLAT can be obtained according to (8), (12),
ω ≤ π, QK−1(ωp±)LHBP and AK(ωp±) satisfy the following and (16) (respectively, h0, QK(ω)MAXFLAT, and ωp+). Then, note
properties: that for a given K, h0, computed by substituting  =
QK(ωp+)MAXFLAT into (22) is equal to that h0 by (8), and QK(ω)
𝑄 𝜔 𝑄 𝜔 1 (18a) using h0 with  > 1 has sharper transition-band but larger
distortion response in the passband and the stopband. Based
on the results so far, Fig. 1 shows a design procedure to
𝐴 𝜔 𝐴 𝜔 0 (18b)
permit direct and simple computation of coefficients of g-
LHBP filters.
From the relationship of (13), these anti-symmetric properties
at ω = ωp+ and ωp− result in
h0 h0,
𝑄 𝜔 𝑄 𝜔 1 (19)
K
 p QK 1 ( p ) LHBP g
Substituting (17) into (13) with ω = ωp± and simplifying with
respect to h0, we can obtain
QK ( p )  
h2n
  QK ( p ) MAXFLAT
ℎ (20) QK ( p ) MAXFLAT    1
√

where two h0’s (double signs in same order) have an FIGURE 1. Design procedure to obtain g-LHBP filters with narrow
transition band.
equivalent value since QK(ωp+) − QK−1(ωp+)LHBP = −QK(ωp−) +
QK−1(ωp−)LHBP from the properties of (18a) and (19). It is

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

Performance evaluation: Choosing h0, with  within

0.2
= 1.0

A3()
QK(ωp+)MAXFLAT <  ≤ 1.0 may cause overshoot and 1.2
max
= 0.0605 MAXFLAT= 0.8667
undershoot in the passband and the stopband, respectively. 0.0
1.0 = 1.0
Hence, it is necessary to verify whether such distortions are 0.2952

tolerant or not. To calculate the passband and stopband peak 0.8 

 d p
 
p 
d 
errors due to using (22), let ωd+ and ωd− be the peak 0.6
-0.2
0.0 0.5 1.0

Q3()
overshoot and undershoot frequencies. Then, QK(ωd+) and 0.4 MAXFLAT= 0.8667
QK(ωd−) become peak values in the passband and the
0.2
stopband, respectively. Solving ∂QK(ω) / ∂ω = 0 at ω = ωd±, max

0.2952
ωd± can be obtained in terms of K and h0, as 0.0 
d p
 
p  
d
-0.2
𝜔 0.0 0.2 0.4 0.6 0.8 1.0

2 𝐾 2
𝑎𝑟𝑐 𝑐𝑜𝑠 1 1 FIGURE 2. The frequency responses of two g-LHBP filters for a given
ℎ , 𝐾 2 K=3; MAXFLAT filter (black solid line) and narrow transition band filter
(24) (red-bold line).

where ωd±(double signs in same order) denotes ωd+ and ωd−, IV. Design Examples
satisfying such that ωd+ + ωd− = π from cos(ωd+ + ωd−) = −1. In this section, through the design examples of g-LHBP
In the similar way of deriving (19), QK−1(ωd±)LHBP and AK(ωd±) lowpass filters, we demonstrate the usefulness of the
have the following properties, respectively proposed method, and verify that that the performance
parameters derived above are accurate.
𝑄 𝜔 𝑄 𝜔 1 𝑎𝑛𝑑 For example, in the case of K = 4 (i.e., the order of filter is
such that 4K – 2 = 14), a general form for the g-LHBP of
𝐴 𝜔 𝐴 𝜔 0 (25) order 14 can be given from (9) as

to yield 𝑄 𝑧 𝑧 𝑔 ∑ℓ 𝑔 𝑧ℓ 𝑧 ℓ
(28)
ℓ

𝑄 𝜔 𝑄 𝜔 1 (26)
Then, gℓ 's (ℓ = 0, 1, 2, .., 4) are obtained from (10) as
From (26) it can be seen that the maximum overshoot ripple
𝛿 , due to QK(ωd+) is equal to the magnitude of the peak 𝑔 2 ℎ , 𝑔 6⋅2 ℎ , 𝑔 3/8 16 ⋅ 2 ℎ ,
𝑔 9/4 26 ⋅ 2 ℎ , 𝑔 19/4 30 ⋅ 2 ℎ (29)
undershoot QK(ωd−) as follows:

𝛿 , 𝑄 𝜔 1 𝑄 𝜔 (27)
to yield the transfer function of the form shown in (2), which
is expressed as
where 𝛿 , is zero (i.e., 𝛿 , = 0) if h0, is chosen according
to (8) for the design of g-LHBP filters with MAXFLAT 3 25
response. From (16) and (24) the inequality relation between 𝑄 𝑧 ℎ 𝑧 5ℎ 𝑧 9ℎ 𝑧
2 2
ωp± and ωd± is 0 < ωd+ < ωp+ < π/2 < ωp− < ωd− < π and this 75 75
results in QK(ωd+) > QK(ωp+) > QK(ωp−) > QK(ωd−) due to 5ℎ 𝑧 0.5 5ℎ 𝑧
2 2
AK(ωp+) > AK(ωd+) > AK(ωd−) > AKωp−). Fig. 2, in the case of 25 3
K=3, shows these parameters on the frequency responses of 9ℎ 𝑧 5ℎ 𝑧 ℎ , 𝑧
2 2
two g-LHBP filters where two h0's have been chosen (30)
according to (22) with  = 1.0 and (8) (equal to (22) with
MAXFLAT = 0.8667), respectively. It is shown that using (22) where h2n (n = 1, 2, 3) is obtained by substituting (29) into
allows g-LHBP filters with narrow transition band but (11b) with K = 4. Note that the odd number indexed
distortion such as overshoot and undershoot in the passband coefficients of the half-band filter given by (2) are zero – i.e.,
and the stopband, Hence, there still remains whether or not h2n-1 (n = 1, 2, 3). The frequency response of Q4(z) is such
this undesired distortion due to the use of h0, is within the that from (13) Q4(ω) = Q3(ω)LHBP + A4(ω) where Q3(ω)LHBP
limit acceptable to the design of such filters with tolerant and A4(ω) are given, respectively from (14) and (15), as
magnitude distortion but narrow transition band. Such
performance evaluations are verified through design
examples discussed in the next section. 𝑄 𝜔 𝑐𝑜𝑠 1 3 𝑠𝑖𝑛 6 𝑠𝑖𝑛 (31)

𝐴 𝜔 ℎ 2 𝑐𝑜𝑠 𝜔 𝑠𝑖𝑛 𝜔 (32)

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

Then, h0 are obtained, by substituting K = 4 into (22), as 214 that is equal to substituting MAXFLAT = 0.8592 into (33).
Fig. 3 and Fig. 4 shows A4(ω) and Q4(ω) due to four h0, 's
ℎ
√
𝛾 𝑄 𝜔 (33) and the related parameters are also indicated in Table I. It is
, ⋅ shown that the g-LHBP filters can have, by controlling h0, in
(30), tolerant distortions but relatively narrow transition
where choosing K = 4 yields ωp+ = 0.3766π and Q3(ωp+)LHBP bands – i.e., 0 < 𝛿 , ≤ 0.0610 in 0.8592 <  ≤ 1.0 but
= 0.8220 from (16) and (31), respectively. To show the
effectiveness of the proposed method, the performance 2.9117 < |Slope4,| ≤ 4.0527. Especially, taking  = 0.9 yields
evaluation is carried out with four g-LHBP filters using four the g-LHBP filter that has an approximately flat magnitude
response similar to the MAXFLAT FIR filter (substituting
h0, 's with  = MAXFLAT, 0.9, 0.95, and 1.0 where  =
MAXFLAT) but relatively narrower transition band.
MAXFLAT means the use of (8) for h0, – i.e., h0, = h0 = −d4,3 /

TABLE I
RELATED PARAMETERS OF FIG. 3 AND FIG. 4

Characteristics of four g-LHBP filters due to various 𝛾 𝑄 𝜔

Parameters Related Equation
𝛾 0.8592 𝛾 0.9 𝛾 0.95 𝛾 1.0

order of filter 14 𝐾 4 4𝐾 2

𝜔 /𝜋 0.3766
(16)
𝜔 /𝜋 0.6234

𝑄 𝜔 0.8220
(31)
𝑄 𝜔 0.1780

ℎ , -0.00122070 -0.00255885 -0.004201355 -0.00584116 (8) and (33)

𝐴 𝜔 0.0372 0.0780 0.1280 0.1780

(32)
𝐴 𝜔 -0.0372 -0.0780 -0.1280 -0.1780

𝑄 𝜔 0.8592 0.9000 0.9500 1

(31) + (32) by (13)
𝑄 𝜔 0.1408 0.1000 0.0500 0

𝜔 /𝜋 0 0.2335 0.2847 0.3079

(24)
𝜔 /𝜋 1 0.7665 0.7153 0.6921

𝑄 𝜔 1 0.9826 0.9516 0.9288

(31)
𝑄 𝜔 0 0.0174 0.0484 0.0712

𝐴 𝜔 0 0.0219 0.0757 0.1322

(32)
𝐴 𝜔 0 -0.0219 -0.0757 -0.1322

𝑄 𝜔 1 1.0045 1.0273 1.0610

(31) + (32) by (13)
𝑄 𝜔 0 -0.0045 -0.0273 -0.0610

𝑠𝑙𝑜𝑝𝑒 , -2.9117 -3.2421 -3.6474 -4.0527 (23)

𝛿 , 0 0.0045 0.0273 0.0610 (27)

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

Table II indicates gℓ 's and hn 's of two filters using h0, 's with considered for comparison purposes. It can be seen that the
MAXFLAT and  = 1.0 where gℓ 's and hn 's have been chosen g-LHBP filters outperform the other designed filters in terms
according to (10) and (11), respectively. of transition bandwidth and peak-to-peak passband/stopband
For additional examples, Fig. 5 and Fig. 6 show g-LHBP ripples. Particularly, it also appears the FRM-based filter
filters with various K, and the related parameters of the filters suggested by Roy and Chandra [28] has relatively narrow
are also indicated in Table III. It can be found that as  transition bandwidth as compared to the g-LHBP filter, but
increases for a given K, the steepness of the transition-band non-negligible ripples take place in the stopband apart from
slope increases rapidly but the amplitude distortion (𝛿 , ) complicated problems due to FRM bank structure-based filter
increases very slightly. The more K increases, the greater form. Consequently, these examples demonstrate that the
effect it can have, as shown in Fig. 7 and Fig. 8 that proposed method derives flexible FIR half-band filters with
demonstrates the performance of g-LHBP filters for a given controllable frequency characteristics – i.e., with a reasonable
K and . In addition, Table IV shows the comparison of the trade-off between the transition-band sharpness and passband
proposed method with the previous state-of-the-art works, & stopband ripples.
where examples given in [16]-[19], [25], [28] have been

1.2
0.2  =1.0
1.0  =0.95
 =1.0  =0.9
MAXFLAT
0.1  =0.95 0.8 Q3 ( ) LHBP
 =0.9
MAXFLAT 0.6

Q4 ()
A4()

0.0
0.4

-0.1 0.2

p
 
p
0.0 
p p

-0.2
-0.2
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
 

FIGURE 3. The frequency responses due to  for a given K = 4: A4(ω) FIGURE 4. The frequency responses due to  for a given K = 4: Q4(ω)

TABLE II
COEFFICIENTS OF THE TWO G-LHBP FILTERS FOR K = 4

ℎ by (8) (: 𝛾 0.8592) ℎ , . by (37) with 𝛾 1.0

g0 -0.078124998 -0.373834144

g1 0.468749990 2.243004864

g2 -0.874999974 -5.606346304

g3 -0.218750042 7.469687744

g4 2.406250048 -6.465024320

h0 -0.001220703 -0.005841156

h2 0.011962891 0.035065168

h4 -0.059814453 -0.101398552

h6 0.299072266 0.322174543

h7 0.5 0.5

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

0.3
0.15

AK ()
K=4
K=4 K=3
1.2

AK ()
K=3 K=2
1.2 K=2
K=6
K=6 K=6
K=5 0.00 K=6 K=5 0.0 K=5
K=5
K=4 K=4
0.8
0.8

QK ()
 
QK ()

-0.15 -0.3
0.0 0.5 1.0 0.0 0.5 1.0
0.4
0.4

K=3
K=2 K=3
K=2

0.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
 

FIGURE 5. The frequency responses of g-LHBP filters due to K:  = 0.9 FIGURE 6. The frequency responses of g-LHBP filters due to K:  = 1.0

TABLE III
RELATED PARAMETERS OF FIG. 5 AND FIG. 6 FOR MAXFLAT,  = 0.9, AND  =1.0
𝐾 (4K - 2) a
Parameters
2(6) 3(10) 5(18) 6(22)

ℎ , -0.03125000 0.00585938 -0.00026703 -0.00006008

MAXFLAT
𝑠𝑙𝑜𝑝𝑒 , -1.9646 -2.4848 -3.2832 -3.6166
by (8)
𝛿 , 0.0000 0.0000 0.0000 0.0000

ℎ , . -0.03615381 0.00949352 -0.00068793 -0.00018373

𝛾 0.9 𝑠𝑙𝑜𝑝𝑒 , . -2.0417 -2.7103 -3.6978 -4.1029

𝛿 , . 0.0018 0.0040 0.0048 0.0049

ℎ , . -0.06862976 0.02041182 0.00162649 -0.00044455

𝛾 1.0 𝑠𝑙𝑜𝑝𝑒 , . -2.5521 -3.3879 -4.6222 -5.1287

𝛿 , . 0.0581 0.0605 0.0613 0.0614

a
The (4K - 2) denotes the order of filter and the related parameters of the filters, in the case of K = 4, have been indicated in Table I.

5.5 0.075
 =1.0
5.0  =1.0
4.5 0.050
 =0.9
|slopeK, r|

4.0
MAXFLAT
K.max

3.5 0.025
3.0
 =0.9
2.5 0.000 MAXFLAT
2.0
1.5 -0.025
2 3 4 5 6 2 3 4 5 6
K K
FIGURE 7. The performance comparison (from Table III) for a FIGURE 8. The performance comparison (from Table III) for a
given 𝑲 and 𝜸 : |𝒔𝒍𝒐𝒑𝒆𝑲,𝜸 | given 𝑲 and 𝜸 : 𝜹𝒎𝒂𝒙
𝑲,𝜸

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

TABLE IV
COMPARISON OF PROPOSED G-LHBP FILTER AND PREVIOUS NARROW TRANSITION-BAND FILTERS

Peak-to-peak ripple in the Peak-to-peak ripple in the

Design Method Filter Order (N) Transition-bandwidth (rad/ π)
passband stopband

Ref.[18] 70 0.077 0.17234 6.4 x10-6

g-LHBP 70 0.051 3.81x10-5

Ref.[16] 86 0.081 0.00138 0.00231

g-LHBP 86 0.044 4.07x10-5

Ref.[17] 100 0.024 0.00277 0.15882

-5
g-LHBP 98 0.039 4.28x10

Ref.[28] 100 0.015 0.00104 0.01413

g-LHBP 102 0.017 0.00039

Ref.[25] 270 0.15 0.03442 0.02377

g-LHBP 270 0.022 4.91x10-5

Ref.[19] 306 0.034 0.00024 0.00028

-5
g-LHBP 306 0.015 5.14x10
※A peak-peak ripple (overshoot) value in the passband of a g-LHBP filter is equal to that (undershoot) in the stopband, as shown in figures and Table III.
In addition,  of g-LHBP filters is calculated by applying the transition bandwidth and filter order of the corresponding compared filter into (23), and it is
determined within the limit of 𝛾 𝛾 1.

V. CONCLUSIONS similar to (3) and consequently, can be expressed by using an

The Problems with wide transition-band always arise in objective function AK(z) for the extension of QK-1(z)LHBP to
MAXFLAT FIR half-band filter design which leaves no QK(z) as below
degree of freedom (i.e., independent parameters) to control
the frequency response by some closed-form polynomial. 𝑄 𝑧 𝑄 𝑧 𝐴 𝑧 (A.2)
In this paper, we have proposed a new method to design
FIR half-band filters with an explicit control of the transition- where QK-1(z)LHBP is a LHBP filter of order 4K − 6, which is
band steepness. For this purpose, we have developed a obtained from (3) as
generalized Lagrange half-band polynomial parameterizing
0-th coefficient h0 and have provided a solution to use h0 as a 𝑄 𝑧
ℓ
transition-band steepness parameter of this filter type. In 𝑧 ∑ℓ 𝑑 (A.3)
,ℓ
addition, new formulas have been given for direct and simple
computation of parameters in closed form. The examples
and the objective function AK(z) is defined by using two
were also shown to verify the performance of this class of
unknown coefficients cK−1 and cK as
filters with tolerant ripple but relatively narrower transition
band. Hence, a solution to the problem encountered in the
previous methods is found. 𝐴 𝑧 𝑧

APPENDIX A: Derivation of the g-LHBP 𝑐 𝑐 (A.4)

From (2) the transfer function of order 4K – 2 (=2N) can
be rewritten by From (A.2) QK(z) can be rewritten as
𝑄 𝑧 ℎ 𝑧 ℎ 𝑧 ... ℎ 𝑧 ℓ
ℎ 𝑧 0.5 ℎ 𝑧 ℎ 𝑧 𝑄 𝑧 𝑧 ∑ℓ 𝑑 ,ℓ
... ℎ 𝑧 ℎ 𝑧 (A.1)
𝑐 𝑐 (A.5)
The flatness condition of (6) is imposed on (A.1)-i.e., 2(K − 1)
zeros at z = −1, and using Lagrange interpolation at For mapping this polynomial into a general form of (11),
coincident points [24], [30], QK(z) has a recursive relation using the transformation

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

ℓ REFERENCES
2ℓ
[1] J.F. Kaiser, “Design subroutine (MAXFLAT) for
ℓ
ℓ symmetric FIR lowpass digital filters with maximally-
2ℓ flat pass and stop bands,” in Program for Digital Signal
2 ∑ℓ 1 (A.6)
ℓ 𝑗 Processing. New York, NY, USA: IEEE Press, 1979,
ch.5, sec.3, pp. 1-6.
[2] P.P. Vaidyanathan, “Efficient and multiplierless design
on (A.5) and simplifying in a similar way of [10], we can of FIR filters with very sharp cutoff via maximally flat
obtain gℓ 's which are expressed in terms of cK−1 and cK as building blocks,” IEEE Trans. Circuits. Syst., vol. 32,
no. 3, pp. 236–244, Mar. 1985.
2 𝐾 𝑗 [3] P.P. Vaidyanathan, “Optimal design of linear phase FIR
ℓ ∑ℓ ,
𝑔ℓ 1 digital filters with very flat passbands and equiripple
ℓ 𝑗 stopbands,” IEEE Trans. Circuits. Syst., vol. 32, no. 9,
ℓ 2 𝐾 1 2𝐾 pp. 904–917, Sep. 1985.
1 (A.7)
ℓ 1 ℓ [4] C.C. Tseng and S.L. Lee, “Design of linear phase FIR
filters using fractional derivative constraints,” Signal
𝐴 Process., vol. 21, no. 5, pp. 1317–1327, May. 2010.
where 0, if 𝐴 𝐵 or 𝐵 0. [5] M.D. Kušljević, M.D, J.J. Tomić and P.D Poljak,
𝐵 "Maximally Flat-Frequency-Response Multiple-
From (A.7) and (13) g0, g1, h0, and h1 are given respectively Resonator-Based Harmonic Analysis," IEEE Trans.
as Instrum. Meas., vol. 66, no. 12, pp. 3387–3398, Sep.
2017.
2𝐾 [6] T. Yoshida and N. Aikawa, "Low-Delay Band-Pass
𝑔 1 and 𝑔 1 Maximally Flat FIR Digital Differentiators," Circuits
1 Syst. Signal Process., vol. 37, no. 8, pp. 3576–3588,
(A.8) Dec. 2018.
[7] O. Herrmann, “On the approximation problem in
2 𝐾 1 nonrecursive digital filter design,” IEEE Trans. Circuit
ℎ and ℎ 𝑔 𝑔 0 Theory., vol. 18, no. 3, pp. 411–413, May. 1971.
1 [8] J.A. Miller, “Maximally flat nonrecursive digital filters,”
(A.9) Electron. Lett., vol. 8, no. 6, pp. 157–158, Mar. 1972.
[9] L. Rajagpoal and S.C. Dutta Roy, “Design of
where h1 = 0 results from h2n−1 's = 0 given in (A.1). maximally-flat FIR filters using the Bernstein
Substituting (A.8) into (A.9) yields cK−1 and cK which are polynomial,” IEEE Trans. Circuits. Syst., vol. 34, no. 12,
pp. 1587–1590, Dec. 1987.
expressed in terms of h0 as [10] M.U.A. Bromba and H. Ziegler, “Explicit formula for
filter function of maximally flat nonrecursive digital
𝑐 1 2 ℎ (A.10) filters,” Electron. Lett., vol. 16, no. 24, pp. 905–906,
Nov. 1980.
[11] P. Thajchayapong, M. Puangpool and S. Banjongjit,
𝑐 1 2 ℎ (A.11) “Maximally flat FIR filter with prescribed cutoff
frequency,” Electron. Lett., vol. 16, no. 23, pp. 514–515,
Jun. 1980.
By substituting these results into (A.4), (A.5), and (A.7), we [12] B.C. Jinaga and S.C. Dutta Roy, “Coefficients of
can have respectively maximally flat low and high pass non-recursive digital
filters with specified cut-off frequency,” Signal Process.,
vol. 9, no. 2, pp. 121–124, Sep. 1985.
1 𝑧 [13] J. Jeon and D. Kim, “Design of nonrecursive FIR filters
𝐴 𝑧 1 2 ℎ 𝑧
2 with simultaneously MAXFLAT magnitude and
prescribed cutoff frequency,” Digital Signal Process.,
(A.12) vol. 22, no. 6, pp. 1085–1094, Dec. 2012.
[14] C.C. Tseng and S.L. Lee, “Closed-form designs of
digital fractional order Butterworth filters using discrete
transforms,” Signal Process., vol. 137, pp. 80–97, Aug.
𝑄 𝑧 𝑧 2017.
ℓ
[15] X. Huang, B. Zhang, H. Qin and W. An, “Closed-form
1 2 design of variable fractional-delay FIR filters with low
∑ℓ 𝑑 ,ℓ ℎ
or middle cutoff frequencies,” IEEE Trans. Circuits.
Syst. I: Regular Papers., vol. 65, no. 2, pp. 628–637,
(A.13) Aug. 2017.
[16] R. Ma, Y. Wang, W. Hu, W. Zhu and K. Zhang,
“Optimum design of multistage half-band FIR filter for
ℓ ∑ℓ 2 𝐾 𝑗 , audio conversion using a simulated annealing
𝑔ℓ 1 algorithm,” 2018 13th IEEE Conf. Indust. Electron.
ℓ 𝑗
Appli., pp. 74–78, May. 2018.
ℓ 2 𝐾 1 2𝐾
1 2 ℎ [17] N. Haridas and E. Elias "Reconfigurable Farrow
ℓ 1 ℓ Structure Based FRM Filters for Wireless
ℓ ∑ℓ , 2 𝐾 𝑗 Communication Systems," Circuits Syst. Signal
1
ℓ 𝑗 Process., vol. 36, no. 1, pp. 315–338, Jan. 2017.
ℓ 2 𝐾 1 2𝐾 [18] L.M. San-José-Revuelta and J.I. Arribas, “A new
1 2 ℎ (A.14) approach for the design of digital frequency selective
ℓ 1 ℓ

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

FIR filters using an FPA-based algorithm,” Expert Syst. Seoul, South Korea. From 2016 to the present, he has
Appli., vol. 106, pp. 92–106, Sep. 2018.
[19] T.S. Bindiya and E. Elias, “Design of totally multiplier-
lectured on computers and networks. His current research
less sharp transition width tree structured filter banks interests include digital design filter, medical image
for non-uniform discrete multitone system,” AEU Int. J. processing, edge detection, deep learning.
Electron. Commun., vol. 69, no. 3, pp. 655–665, Mar.
2015.
[20] S.C. Pei and P.H. Wang, “Closed-form design and Daewon Chung received the
efficient implementation of generalized maximally flat bachelor’s degree from the Department
half-band FIR filters,” IEEE Signal Process. Lett., vol. of Information and Communication
7, no. 6, pp. 149–151, Jun. 2000.
[21] M.M.J. Yekta, “Equivalence of the Lagrange Engineering, Dongguk University, in
interpolator for uniformly sampled signals and the 2015. He is currently pursuing the
scaled binomially windowed shifted sinc function,” integrated master's and doctor's degree
Digital Signal Process., vol. 19, no. 5, pp. 838–842, Sep.
2009. in Electronics and Electrical
[22] R. Ansari, C. Guillemot and J.F. Kaiser, “Wavelet Engineering at Dongguk University,
construction using Lagrange halfband filters,” IEEE Seoul, South Korea.
Trans. Circuits. Syst., vol. 38, no. 9, pp. 1116–1118,
Sep. 1991.
From 2016 to the present, he has lectured on computers
[23] G. Anbarjafari and H. Demirel, “Image super resolution application and visual programming. His current research
based on interpolation of wavelet domain high interests include digital filter design, signal & image
frequency subbands and the spatial domain input image,” processing (enhancement, restoration), edge detection, deep
ETRI Journal., vol. 32, no. 3, pp. 390–394, jun. 2010.
[24] B.-R. Horng and A.N. Willson, “Lagrange multiplier Learning with computer vision.
approaches to the design of two-channel perfect-
reconstruction linear-phase FIR filter banks,” IEEE Yunsun Kim recreceived the
Trans. Signal Process., vol. 40, no. 2, pp. 364–374, Feb.
1992. bachelor’s degree from the Department
[25] J. Rodrigues and K.R. Pai, “New approach to the of Electronics Engineering, Chosun
synthesis of sharp transition FIR digital filter,” University, in 2019. He is currently
Proceedings of the IEEE International Symposium on
Industrial Electronics., vol. 3, pp. 1171–1173, Jun.
pursuing the combined master's and
2005. doctor's degree in Electronics and
[26] X. Huang, S. Jing, Z. Wang, Y. Xu, and Y. Zheng, Electrical Engineering at Dongguk
“Closed-form FIR filter design based on convolution
University, Seoul, South Korea. His
window spectrum interpolation,” IEEE Trans. Signal
Process., vol. 64, no. 5, pp. 1173–1186, Oct. 2015. current research interests include computer vision, super
[27] X. Huang, Y. Wang, Z. Yan, H. Xian and M. Liu, resolution, digital image processing based on AI and deep
“Closed-form FIR filter design with accurately learning.
controllable cut-off frequency,” Circuits, Syst. Signal
Process., vol. 36, no. 2, pp. 721–741, Feb. 2017.
[28] S. Roy and A. Chandra, “Design of narrow transition Ingyun Kim received the bachelor’s
band digital filter: An analytical approach,” Integration., degree from the Department of
vol. 68, pp. 38–49, Sep. 2019.
[29] S. Roy and A. Chandra, “A Survey of FIR Filter Design Electronics and Electrical Engineering,
Techniques: Low-complexity, Narrow Transition-band Dongguk University, in 2020. He is
and Variable Bandwidth,” Integration., vol. 44, pp. currently pursuing the master's degree in
193–204, Mar. 2021.
[30] I.R. Khan and R. Ohba, “Efficient design of halfband
Division of Electronics and Electrical
low/high pass FIR filters using explicit formulas for tap Engineering at Dongguk University,
coefficients,” IEICE Trans. Fundament., vol. 83, no. 11, Seoul, South Korea. His current research
pp. 2370–2373, Nov. 2000. interests include time series forecasting, object detection,
[31] I.R. Khan, “Flat magnitude response FIR halfband
low/high pass digital filters with narrow transition digital image processing based on AI and deep learning.
bands,” Digital Signal Process., vol. 20, no. 2, pp. 328–
336, Mar. 2010. Joonhyeon Joon received the B.S.
degree in electronic engineering from
Dongguk University, Korea, in 1984 and
the M.S. and Ph.D. degrees in electric
Woon Cho received the bachelor’s and electronic engineering from Korea.
degree from the Department of Advanced Institute of Science
Information and Communication Technology, Korea, in 1986 and 1991,
Engineering, Dongguk University, in respectively. From 1991 to 1999, he was
2014. He is currently pursuing the a Project Manager in the Image Processing Department of
combined master's and doctor's degree Korea Telecom Research Center, where he was responsible
in Electronics and Electrical for the development of network-oriented video service
Engineering at Dongguk University, systems. He is currently an Associate Professor with the

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3069985, IEEE
Access

Department of Electronics and Electrical Engineering,

Dongguk University, Seoul, South Korea. His current
research interests include digital image processing and
transmission, theory and application of wavelet transforms,
and digital filter design.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/