Wavelet Packet: A Multirate Adaptive Filter for

De-noising of TDM Signal

V. V. Satyanarayana G. V. Pradeep Kumar Jaya Krishna Sunkara
Tallapragada Assistant Professor, Assistant Professor,
Associate Professor, Department of ECE, Department of ECE,
Department of ECE, C.B.I.T., Gandipet, Hyderabad. Sree Vidyanikethan Engineering
Sree Vidyanikethan Engineering [email protected] College, Tirupati
College, Tirupati [email protected]
[email protected]

Abstract—The transmission of data through a practical using adaptive rather than fixed filters. An adaptive filter is a
channel always introduce some sort of interference. The most self-designing filter that uses a recursive algorithm.
common noise is additive uniform noise. Currently, the de-
noising of various signals is mainly by using traditional adaptive As shown in Fig. 1, an Adaptive Noise Canceller (ANC)
filtering techniques like LMS, NLMS, LSL, DFT and DCT. The has two inputs – primary and reference. The primary input
adaptive filtering techniques are the best choice for the de- receives a signal s from the signal source that is corrupted by
noising application, because the de-noising involves the removal noise n uncorrelated with the signal. The reference input
of random nature, which can be well handled by a iterative receives a noise no uncorrelated with the signal but correlated
procedure. Wavelets, on the other hand, perform the de-noising
by decomposing the signal, applying any thresholding technique in some way with the noise n . The noise no passes through a
and then reconstruct the signal back. In recent times, the filter to produce an output n̂ that is a close estimate of
wavelets are used for de-noising of noisy signals and images as primary input noise. This noise estimate is subtracted from the
well. In this paper, the wavelet packets based on existing
corrupted signal to produce an estimate of the signal at ŝ , the
traditional wavelets as well as new wavelets are utilized for signal
de-noising. First of all, a sharp signal of short duration is
ANC system output. In noise canceling systems a practical
generated, and then it is used as the basis function for wavelet. objective is to produce a system output sˆ = s + n − nˆ that is a
This wavelet is used in wavelet packet for the de-noising. The best fit in the least squares sense to the signal s . Assume that
proposed de-noising by using wavelet packet was compared with s , no , n1 and y are statistically stationary and have zero
that of traditional adaptive filtering techniques, wavelets and
wavelet packets with existing wavelets by using different design means. The signal s is uncorrelated with no and n1 , and n1
metrics. It is found that the de-noising performance of proposed
technique is better than that of remaining. is correlated with no [4].

Keywords— Multirate filter, de-noising, wavelet packet, sˆ = s + n − nˆ

adaptive filter.
sˆ = s + (n − nˆ ) 2 + 2 s (n − nˆ )
2 2

Taking expectation of both sides and realizing that s is

I. INTRODUCTION
uncorrelated with no and n̂ ,
Multirate systems are building blocks commonly used in
digital signal processing (DSP). Their function is to alter the
rate of the discrete-time signals, which is achieved by adding
E[ sˆ2 ] = E[ s 2 ] + E[(n − nˆ ) 2 ] + 2 E[ s (n − nˆ )]
or deleting a portion of the signal samples. Multirate systems
play a central role in many areas of signal processing, such as = E[ s 2 ] + E[(n − nˆ ) 2 ] [4]
filter bank theory and multi-resolution theory. They are
essential in various standard signal processing techniques such The signal power E[ s 2 ] will be unaffected as the filter is
as signal analysis, de-noising, compression and so forth. 2
adjusted to minimize E[ sˆ ] .
During the last decade, they have found large number of
applications in new and emerging areas of signal processing,
as well as in several neighboring disciplines such as digital min E[ sˆ2 ] = E[ s 2 ] + min E[(n − nˆ ) 2 ] [4]
communications. Adaptive filtering techniques are used in a
This is equivalent to produce the output ŝ to be a best least
wide range of applications, including echo cancellation,
adaptive equalization, adaptive noise cancellation, and squares estimate of the signal s . The rest of the paper is
adaptive beam-forming [1-3]. These applications involve organized as follows. Section II reviews traditional adaptive
processing of signals that are generated by systems whose filters. In section III, wavelet and wavelet packet
characteristics are not known a priori. Under this condition, a decomposition and reconstruction processes along with new
significant improvement in performance can be achieved by wavelet packets are presented. The simulation results

978-1-5386-4304-4/18/$31.00 ©2018 IEEE.

including the ANC performance of traditional adaptive filters, the gradient of the mean square error, the weights are updated.
wavelets, existing wavelet packets and new wavelet packets That is, if the MSE-gradient is positive, it implies, the error
are discussed in section IV. Section V concludes the paper. would keep increasing positively, if the same weight is used
for further iterations, which means we need to reduce the
weights. In the same way, if the gradient is negative, we need
to increase the weights. So, the basic weight update equation
is:
Wn +1 = Wn − μΔε [n], [1]
where ε represents the mean-square error. The negative sign
indicates that, we need to change the weights in a direction
opposite to that of the gradient slope. The mean-square error,
as a function of filter weights is a quadratic function which
Fig. 1 Adaptive Noise Canceller [1] means it has only one extreme that minimizes the mean-square
error, which is the optimal weight. The LMS thus, approaches
towards these optimal weights by ascending/descending down
II. TRADITIONAL ADAPTIVE FILTERS the mean-square-error versus filter weight curve [5-7]. The
Least mean squares (LMS) algorithms are a class LMS algorithm for pth order algorithm can be summarized as
follows [7].
of adaptive filter used to mimic a desired filter by finding the
filter coefficients that relate to produce the least mean squares Parameters: p is the filter order,
of the error signal (difference between the desired and the µ is the step size
actual signal). It is a stochastic gradient descent method in that
the filter is only adapted based on the error at the current time. Initialization: hˆ(0) = 0
It was invented in 1960 by Stanford University professor Computation: For n = 0, 1, 2,….
Bernard Widrow and his first Ph.D. student, Ted Hoff [1]. The x(n) = [ x(n), x(n − 1),..., x(n − p + 1)]T
Fig. 2 shows the LMS adaptive filtering.
e( n) = d ( n) − hˆ H ( n) x (n)
hˆ(n + 1) = hˆ( n) + μe * ( n) x ( n) [7]

Normalized Least Mean Squares Filter (NLMS)

The main drawback of the "pure" LMS algorithm is that it is
sensitive to the scaling of its input x (n). This makes it very
hard to choose a learning rate μ that guarantees stability of
the algorithm. The Normalized Least Mean Squares (NLMS)
filter is a variant of the LMS algorithm that solves this
problem by normalizing with the power of the input [8-9]. The
NLMS algorithm can be summarized as follows [8]:
Fig. 2 LMS Adaptive Filtering [1]
Parameters: p - filter order
The realization of the causal Wiener filter looks a lot like µ - step size
the solution to the least squares estimate, except in the signal
processing domain. The least squares solution, for input Initialization: hˆ(0) = 0
matrix X and output vector Y is βˆ = ( X X ) X y . The
T −1 T
Computation: For n = 0, 1, 2, …
FIR Wiener filter is related to the least mean squares filter, but x(n) = [ x(n), x(n − 1),...., x(n − p + 1)]T
minimizing its error criterion does not rely on cross-
correlations or auto-correlations. Its solution converges to the
Wiener filter solution. Most linear adaptive filtering problems
e( n) = d ( n) − hˆ H ( n) x (n)
can be formulated using Fig. 2. That is, an unknown
ˆ ˆ μe* (n) x(n)
system h(n) is to be identified and the adaptive filter attempts h(n + 1) = h(n) + H
to adapt the filter hˆ(n) to make it as close as possible to h(n), x (n) x(n)
while using only observable signals x(n), d(n) and e(n),
but y(n), v(n) and h(n) are not directly observable. Its solution III. WAVELET AS A MULTIRATE ADAPTIVE FILTER
is closely related to the Wiener filter. The basic idea behind Wavelet analysis often speaks about approximations and
LMS filter is to approach the optimum filter weights (R-1P) by details. The approximations are the low-frequency, high-scale
updating the filter weights in a manner to converge to the components of the signal. The details are the high-frequency,
optimum filter weight. The algorithm starts by assuming a low-scale components[10]. The original sequence S, applied to
small weight (zero in most cases), and at each step, by finding
two complementary filters and emerges as two signals as (cA) and detail (cD) components. The wavelet packet analysis
shown in Fig. 3. looks like a complete tree structure. The multistage wavelet
packet analysis looks like as shown in Fig. 8.

Fig. 3 Filtering Process in Wavelet Analysis Fig. 5 Multistage Decomposition

If a digital sequence of say 512 samples is applied to the filter
bank consisting of one low and one high pass filter as
mentioned above, the length of A will be 512 and that of D
will also be 512. Hence, the data to handle was doubled. But
note that in A as well as in D only 256 samples are
irredundant. To remove the redundant samples, downsamplers
are employed as shown in Fig. 4. The outputs are denoted by
cA and cD.
Fig. 6 Reconstruction Stage

Fig. 4 Wavelet processing with downsamplers Fig. 7 Wavelet as a Multirate Adaptive Filter.

This process, i.e., the conversion of S into cA and cD is called

decomposition; the filters at this stage are referred as
decomposition low pass and decomposition high pass filters.
These filters have direct relation to the basis function used in a
specific wavelet. The vectors cA and cD constitutes the DWT
coefficients.
The decomposition process can be repeated, with successive
approximations being decomposed in turn, so that one signal
is broken down into many lower resolution components. This
is called the wavelet decomposition tree shown in Fig. 5. The
maximum number of decomposition stages should be taken so
that the length of the sequence in the last stage is not less than
1. From the wavelet coefficients the original signal need to be
recovered. The process of obtaining the original signal by
using the wavelet coefficients is called reconstruction or
synthesis and is shown in Fig. 6. The downsampling
performed at decomposition stage introduces an aliasing
effect. The reconstruction filters need to be selected so that the
aliasing effect introduced at the decomposition stage should be
cancelled. The overall process of wavelet is depicted in the
Fig. 7. The wavelet packet analysis is an extension of wavelet Fig. 8 Wavelet Packet Analysis
analysis with an inclusion of analysis of both approximation
The wavelet packets use the wavelet filters to decompose and used in designing wavelet packet is given in Fig. 9. The
reconstruct the signals. The wavelet filters correspond to the wavelet filters are shown in the Fig 10. The psi and phi
perfect reconstruction condition as well as to represent the functions of the proposed wavelet packet are shown in the Fig.
data to suite different applications. 11.
IV. RESULTS
Generate short
duration signal In this section the simulation results of Traditional
adaptive filters, Wavelets, Wavelets Packets using existing
Adjust the
Satisfactory Stop
wavelets and Wavelet Packets using new basis functions on
coefficients
de-noising of TDM signal are presented. The flow graph in
Fig. 12 represents the total process that was considered in this
Verify the
Symmetricity
paper.
Calculate Evaluation
Parameters
Verify the area
under the wave

Use the wave as basis

Input Signal function to devise the Denoising
wavelet Packet

Fig. 9 Design process of the proposed wavelet packet

Fig. 12 De-noising of TDM signals based on Wavelet Packet

The two signals that are multiplexed in TDM, multiplexed

signal and noisy signals are shown in the Fig. 13.

Fig. 10 Wavelet Filters of proposed wavelet packet

Fig. 13 Input, TDM and Noisy signals

The Peak Signal to Noise Ratio (PSNR), Mean Square Error
(MSE), Maximum Squared Error (MAX ERR), Ratio of
Squared Norms (L2RAT) are calculated and tabulated for
Fig. 11 psi and phi functions of new wavelet different types of techniques. Table I present the performance
of adaptive filters. The performance of existing wavelets is
In this paper, a new set of wavelet filters are proposed to suite presented in the Table II. The basic denoising technique of the
the de-noising of TDM signals. The wavelet filters are wavelet is used [11-16]. The performance of the wavelet
designed based on the principle of adaptation of coefficients to packets with existing wavelets is presented in the Table III.
suit the application by feedback connection. The algorithm The performance of Wavelet packets with new wavelets is
presented in the Table IV. Because the uniform noise that was TABLE III: Performance of Wavelet packets with existing
added with the TDM signal varies with each run, the algorithm wavelets on ANC of TDM Signal
was run for five times. Each time different signal (noise) will PSNR MSE
be added to the TDM and de-noising will be performed. To FS SS FS SS
analyze the algorithm, the results are taken and tabulated for 5
runs. Haar 54.02 54.11 Haar 0.25 0.25
db10 54.50 53.75 db10 0.23 0.27
TABLE I: Performance of Traditional adaptive filters on ANC
db45 54.63 54.14 db45 0.22 0.25
of TDM Signal
sym6 54.41 54.08 sym6 0.23 0.25
PSNR MSE
coif4 54.88 53.55 coif4 0.21 0.28
FS SS FS SS
bior2.4 54.28 54.06 bior2.4 0.24 0.25
NLMS 52.30 52.32 NLMS 0.38 0.38
dmey 54.01 54.10 dmey 0.25 0.25
RLS 53.37 52.39 RLS 0.29 0.37
rbio1.3 54.39 54.05 rbio1.3 0.23 0.25
DFT 53.30 53.04 DFT 0.30 0.32
Average 54.39 53.98 Average 0.23 0.26
DCT 53.88 51.62 DCT 0.26 0.44
MAX ERR L2RAT
LSL 54.02 53.87 LSL 0.25 0.26
FS SS FS SS
Average 53.37 52.65 Average 0.30 0.35
Haar 1.77 1.87 Haar 1.08 1.12
MAX ERR L2RAT
db10 1.33 1.86 db10 1.06 1.05
FS SS FS SS
db45 1.39 1.53 db45 1.03 1.05
NLMS 2.03 3.16 NLMS 0.95 0.93
sym6 1.25 1.54 sym6 1.03 1.02
RLS 1.93 2.73 RLS 0.98 1.02
coif4 1.56 1.66 coif4 1.08 1.05
DFT 1.80 2.14 DFT 1.01 0.93
bior2.4 1.51 1.78 bior2.4 1.02 1.07
DCT 2.07 4.15 DCT 0.91 0.88
Dmey 1.61 1.76 dmey 1.01 1.09
LSL 1.56 2.12 LSL 0.91 0.95
rbio1.3 1.70 1.78 rbio1.3 1.03 1.06
Average 1.88 2.86 Average 0.95 0.94
Average 1.51 1.72 Average 1.04 1.06
TABLE II: Performance of Traditional Wavelets on ANC of
TDM Signal TABLE IV: Performance of New Wavelet Packets on ANC of
PSNR MSE TDM Signal
PSNR MSE
FS SS FS SS
FS SS FS SS
Haar 53.43 53.10 Haar 0.29 0.31
WPN-1 54.85 55.11 WPN-1 0.21 0.20
db10 53.58 53.83 db10 0.28 0.26
WPN-2 54.81 54.55 WPN-2 0.21 0.22
db45 52.09 52.15 db45 0.40 0.39
WPN-3 55.48 55.60 WPN-3 0.18 0.17
sym6 52.98 52.90 sym6 0.32 0.33
WPN-4 55.80 54.96 WPN-4 0.17 0.20
coif4 53.56 53.16 coif4 0.28 0.31
WPN-5 55.09 55.25 WPN-5 0.20 0.19
bior2.4 51.93 52.41 bior2.4 0.41 0.37
Average 55.21 55.09 Average 0.19 0.20
Dmey 53.30 53.41 dmey 0.30 0.29
MAX ERR L2RAT
rbio1.3 54.16 53.61 rbio1.3 0.24 0.28
FS SS FS SS
Average 53.13 53.07 Average 0.32 0.32
WPN-1 1.56 1.50 WPN-1 0.93 1.50
MAX ERR L2RAT
WPN-2 1.60 2.19 WPN-2 0.93 0.93
FS FS FS
WPN-3 1.30 1.56 WPN-3 0.93 0.94
FS SS FS SS
WPN-4 1.28 1.25 WPN-4 0.97 1.01
Haar 1.69 1.87 Haar 0.88 0.87
WPN-5 1.24 1.54 WPN-5 1.02 0.94
db10 1.84 1.66 db10 0.89 0.93
Average 1.40 1.61 Average 0.96 1.07
db45 1.78 1.85 db45 0.81 0.89
sym6 2.18 1.83 sym6 0.83 0.90
The performance of all these techniques is compared by
coif4 1.63 1.65 coif4 0.85 0.94 considering signal wise. The Fig. 14 and 15 shows the PSNR
bior2.4 1.77 1.63 bior2.4 0.83 0.88 values of first signal with different techniques.
Dmey 1.54 1.82 dmey 0.91 0.87
rbio1.3 1.33 1.63 rbio1.3 0.93 0.94
Average 1.72 1.74 Average 0.87 0.90
 average PSNR obtained is 53.01dB, with existing wavelets it
is 53.105dB and with the existing wavelet packets 54.19dB.
The average Maximum error with the above techniques is
2.375, 1.737 and 1.623 respectively. Wavelet packet with new
wavelet function has produced better results than that of the
above mentioned techniques. The PSNR and maximum error
are 55.15dB and 0.508. The wavelet packets with new basis
functions has shown better performance mainly because the
basis function has irregular and sharp coefficient values which
resists and cancels the effect of noise by itself. The remaining
noise is dealt by the threshold schemes. Hence the wavelet
packets with the proposed basis functions outperforms the
existing techniques.

References
Fig. 14 PSNR of First Signal with different techniques
[1] B. Widrow and M. E. Hoff, “Adaptive switching circuits,” WESCOM
Conv. Rec., pt. 4, pp.96-140, 1960.
[2] B.Widrow, J. M. McCool, M. G. Larimore, and C. R. Johnson, Jr.,
“Stationary and nonstationary learning characteristics of the LMS
adaptive filters,” Proceedings of the IEEE, vol. 64, pp.1151-1162, Aug.
1976.
[3] G. Ungerboeck, “Theory on the speed of convergence in adaptive
equalizers for digital communication,” IBM Journal on Research and
Development, vol. 16, pp. 546-555, Nov. 1972.
[4] B.Widrow and S. D. Stearns, Adaptive Signal Processing, Prentice Hall,
Englewood Cliffs, NJ, 1985.
[5] S. Haykin, Adaptive Filter Theory, Prentice Hall, Englewood Cliffs, NJ,
4th edition, 2002.
[6] Qi Zhang, Yuancheng Yao, Mingwei Qin, “An Uncorrelated Variable
Step-size LMS Adaptive Algorithm”, Journal of Emerging Trends in
Computing and Information Sciences, VOL. 3, NO.11 Nov, 2012.
[7] J. E. Mazo, “On the independence theory of equalizer convergence,” The
Bell System Technical Journal, vol. 58, pp. 963-993, May 1979.
[8] M. G. Bellanger, Adaptive Digital Filters and Signal Analysis, Marcel
Fig. 15 PSNR of Second Signal with different techniques Dekker, Inc., NewYork, NY, 2nd edition, 2001.
[9] Paulo S.R. Diniz, Adaptive Filtering Algorithms and Practical
Implementation, Springer, 2008.
[10] Jaya Krishna Sunkara, Purnima Kuruma, Ravi Sankaraiah Y, “Image
From the above figures it is clear that the PSNR is
Compression Using Hand Designed and Lifting Based Wavelet
well ahead with new wavelet packets when compared with Transforms”, International Journal of Electronics Communications and
traditional adaptive filters, wavelets or existing wavelet Computer Technology (e-ISSN: 2249-7838, IF: 1.2456), vol. 2 (4),
packets. The MSE is directly linked with PSNR and the values 2012.
are given in the Tables I to IV. Sometimes, though the overall [11] DeVore, R.A.; B. Jawerth, B.J. Lucier (1992), "Image compression
MSE is less, the maximum difference or maximum error at a through wavelet transform coding," IEEE Trans. on Inf. Theory, vol. 38,
No 2, pp. 719–746.
particular point in time makes it difficult for the destination [12] Donoho, D.L. (1993), "Progress in wavelet analysis and WVD: a ten
party to interpret the overall signal. This is why the MAX minute tour," in Progress in wavelet analysis and applications, Y. Meyer,
ERR is considered. S. Roques, pp. 109–128. Frontières Ed.
[13] Donoho, D.L.; I.M. Johnstone (1994), "Ideal spatial adaptation by
wavelet shrinkage," Biometrika, vol. 81, pp. 425–455.
V. CONCLUSION [14] Donoho, D.L.; I.M. Johnstone, G. Kerkyacharian, D. Picard (1995),
Wavelets, with its powerful strength of adaptive nature, are "Wavelet shrinkage: asymptopia," Jour. Roy. Stat. Soc., series B, vol. 57
no. 2, pp. 301–369.
applicable to many applications where adaptive filters are in [15] Donoho, D.L.; I.M. Johnstone, "Ideal de-noising in an orthonormal basis
use. In this paper, the de-noising of TDM signal is considered. chosen from a library of bases," C.R.A.S. Paris, t. 319, Ser. I, pp. 1317–
Two signals sampled at different rates are time multiplexed. 1322.
The resulting TDM signal has undergone a noisy channel. The [16] Donoho, D.L. (1995), "De-noising by soft-thresholding," IEEE Trans.
channel is assumed to have uniform noise. The signal at the on Inf. Theory, 41, 3, pp. 613–627.
output of the channel is given as input to the de-noising unit.
The signal from the de-noising unit is given to the de-
multiplexing section of TDM; two signals are separated and
compared with the original input signals. Traditional adaptive
filters, existing wavelets, wavelet packets with existing
wavelets and wavelet packet with new wavelet function are
utilized for de-noising. With the traditional adaptive filters, the