Wireless Pers Commun

DOI 10.1007/s11277-017-5013-6

Block and Fast Block Sparse Adaptive Filtering

for Outdoor Wireless Channel Estimation
and Equalization

Harish Kumar Sahoo1 • Basabadatta Mohanty2 • Bijayananda Patnaik2

 Springer Science+Business Media, LLC 2017

Abstract Rayleigh’s distribution is mainly used when fading wireless medium does not
have proper line of sight (LOS) path and is dominated by a large number of non-line of
sight (NLOS) paths due to reflections of the received signal. Also because of relative
motion of the base station and mobile station, a random frequency shift is generally
introduced in the carrier, which can be realized in terms of Doppler spread. In case of
Rayleigh’s fading channels, there are two critical problems for receiver design that is
accurate estimation of channel coefficients followed by mitigation of channel impairments
like inter symbol interference and fading in presence of user mobility. The accuracy of
estimated channel state information is really crucial to design robust equalizer for
reconstruction of bit sequence and the equalizer performance is affected by the fading rate
and Doppler spread. The main research contributions of the paper is based on the
exploitation of underlying sparseness of block adaptive filters through l0 -norm penalty for
accurate estimation with stable convergence which helps to design computationally effi-
cient adaptive models for estimation. The accuracy of the proposed sparse block and fast
block models is tested using 16 QAM modulation format with Rayleigh’s fading wireless
channel for outdoor environments. With the help of MATLAB simulations, the perfor-
mance of the proposed sparse BLMS and FBLMS adaptive filtering are verified and the
detail comparison results are presented.

Keywords ISI  Doppler spread  Sparse block adaptive filter  LOS  CSI  Rayleigh
fading

& Harish Kumar Sahoo

harish_etc@vssut.ac.in
1
Department of Electronics and Telecommunication Engineering, Veer Surendra Sai University of
Technology (VSSUT), Burla, Sambalpur, India
2
Department of Electronics and Telecommunication Engineering, International Institute of
Information Technology (IIIT), Bhubaneswar, Bhubaneswar, India

123
 H. K. Sahoo et al.

1 Introduction

Mobile radio propagation environments are normally affected by fading effects that
describes the characteristics of wireless communication channel. These are also called
Rayleigh’s fading if the multiple reflections from obstacles are large in number with the
absence of LOS transmission and the received signal envelope statistics is described by
Rayleigh distribution. Also the cellular wireless network is affected by user mobility which
is more prominent in outdoor channel as compared to a static or quasi-static indoor
channel. Thus design of practical model with time varying channel gain is very important
to address the issues of outdoor channel. The relative motion of a user in an outdoor
environment causes a shift in the frequency of the transmitted signal in terms of Doppler
shift along each individual multi path [1]. If these effects of fading and Doppler shift are
collectively incorporated in the channel impulse response of outdoor wireless channel
model, then the channel estimation becomes quite difficult in a practical wireless envi-
ronment. The accuracy of estimated channel state information (CSI) in presence of fading
and Doppler spread helps to design efficient equalizer model which is a crucial block of
wireless receiver [2, 3]. Based upon the estimated CSI, the receiver will be able to
reconstruct the transmitted data sequence. But many times the practical use of adaptive
channel estimator and equalizer is limited due to large computational complexity and slow
error convergence as the inherent sparseness of a communication channel is not taken into
consideration [4, 5]. Also development of adaptive channel estimator and equalizer is
dependent on robust beam forming structure and algorithm [6, 7] due to orientation of
antenna elements. Exploiting that sparseness of adaptive filters by modifying the cost
function through norm penalty makes the model quite efficient in terms of computational
complexity and estimation accuracy [8, 9]. Analysis of stability and convergence of Block-
sparse LMS (BS-LMS) adaptive algorithm [10] shows that the converge rate is faster for a
given set of data samples as compared to that of l0 -norm based LMS algorithms [11]. The
proposed adaptive filtering model is developed through l0 -norm penalty introduced to the
cost function of Block LMS in frequency domain as the frequency domain processing has
advantages of faster convergence. Rayleigh fading wireless channel and 16-QAM modu-
lation format are used to generate data samples for the receiver block. The proposed model
utilizes the benefits of block processing and sparseness in frequency domain to achieve
substantial reduction in bit error rate (BER) and mean square error (MSE) to design
efficient equalizer for time varying outdoor communication environment.
The paper is organized as follows: Sect. 2 explains mathematical modelling of outdoor
wireless channel. Section 3 explains the sparse adaptive models for estimation of the
channel coefficients and equalization of received symbols. Section 4 briefly explains norm
penalty based sparse adaptive filtering models. Section 5 presents mathematical formula-
tion of sparse-BLMS and sparse-FBLMS algorithms. Thorough comparison between
algorithms is presented in Sect. 6 in terms of simulation results considering the practical
scenarios of outdoor environments. Concluding remarks are given in Sect. 7 followed by
appendix.

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

2 Jacks Model Based Outdoor Channel Modelling

A fading channel characterized by varying channel gain with Rayleigh’s distribution and
for a given Doppler spectrum [12–14] can be realized with the help of Jacks model. The
real part and imaginary part of the complex channel hI ðtÞ and hQ ðtÞ, for N number of plane
waves for a Jack’s model can be represented as:
X
N0 pffiffiffi
hI ðtÞ ¼ 2 ðcos /n cos wn tÞ þ 2 cos /N cos wo t ð1Þ
n¼1

X
N0 pffiffiffi
hQ ðtÞ ¼ 2 ðsin /n cos wn tÞ þ 2 sin /N cos wo t ð2Þ
n¼1

where /n and /N are the initial phases of the nth sinusoid with Doppler-shifted and the
maximum Doppler frequency fo , respectively and N0 ¼ ðN=2  1Þ=2, N=2 is an odd
number.
The initial phase can be set to
/N ¼ 0 and /n ¼ pn=ðN0 þ 1Þ for::n ¼ 1; 2; . . .; N0

The complex channel impulse response for Jack’s model is expressed as:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
hðtÞ ¼ E0 = 2N0 þ 1  hI ðtÞ þ jhQ ðtÞ ð3Þ
where E0 is the average of the fading channel.
The Doppler-shifted frequency fwn gNn¼1
0
may be given as (Fig. 1):
wn ¼ wo cos hn ¼ 2pfo cosð2pn=NÞ ð4Þ

2 cos φN0 cos( wN0 t )

2 cos φN0 −1 cos( wN0 −1t )

hI (t )
+

2 cos φ1 cos( w1t )

h(t ) = hI (t ) + jhQ (t )
2 cos φN cos( wd t ) +

2 sin φN cos( wd t )
π
+
2sin φ1 cos( w1t ) 2
hQ (t )
f m cos(2π N 0 / N ) f m cos(2π 1/ N )

2sin φN0 cos( wN0 t )

Fig. 1 Jacks model for outdoor enviroments

123
 H. K. Sahoo et al.

3 Sparse Adaptive Estimation and Equalization Models

Adaptive channel estimation helps to provide CSI to design robust equalizer that helps in
reduction of BER and accurately reconstruct the bit sequence at the receiving end [15].
Accurate CSI information obtained from the transmitter side plays an important role to
improve the BER performance of the receiver. The sparse nature of the channel impulse
response due to the near zero weight coefficients reduces the computational burden and the
cost function can be modified by using l0 -norm penalty as discussed in Sect. 4.

3.1 Channel Estimation Model

Adaptive channel estimation model as shown in Fig. 2 mainly focuses on the proper
tracking of the time variation of channel coefficients in a wireless medium to get reduced
BER at the receiver [16, 17]. The transmitted sequence is passed through a multi path
fading channel [18] and the channel output is corrupted by additive Gaussian noise w(k).
The signal x(k) is transmitted over a multi path fading channel having impulse response hL .
The difficulty lies in the detection of transmitted bit sequence from the received symbols as
the channel coefficient vector is unknown and time varying [19]. The received signal Y1 ðkÞ
is expressed as:
Y1 ðkÞ ¼ xðkÞ  hL þ wðkÞ ð5Þ

where hL is given as: hL ¼ ½h0 ; h1 ; . . .hL T . wðkÞ is the additive white Gaussian noise. Then
error signal is generated by the difference of measured and estimated output vector as
shown in Eq. (6).
^
erðkÞ ¼ Y1 ðkÞ  YðkÞ ð6Þ

3.2 Channel Equalization Model

The adaptive equalizer model is designed by using a combination of three blocks i.e. a
single layer artificial neural network [20, 21] with fifth order functional expansion, FIR
Filter and a feedback filter. The DFE model proposed in this paper, explores the advantages
of each block to get improvement in the performance of the equalizer model. The error
signal is generated considering the overall response of all the sections.

w( k )
+
Y1(k)
x(k ) Y( k )
hL +
+ er (k)
+
-
ˆ
<N
W

Sparse Adaptive Filtering

Fig. 2 Channel estimation model

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

erðkÞ ¼ dr ðkÞ  uo ðkÞ ð7Þ

The desired response dr ðkÞ is the delayed form of the 16-QAM modulated data input.
uo ðkÞ, the combination of output for neural network and FIR filter can mathematically be
represented as:
uo ðkÞ ¼ zf ðkÞ þ zfn ðkÞ ð8Þ

Output of functional block zfn ðkÞ can be expressed as:

zfn ðkÞ ¼ bðuðkÞÞ ð9Þ

X
N X
q
uðkÞ ¼ Wi ðkÞUi ðXR ðkÞÞ þ Bi ðkÞvi ðkÞ ð10Þ
i¼1 i¼1

The output of FIR block is given as

zf ðkÞ ¼ W2 ðkÞXR ðkÞT ð11Þ
T
where W2 ðkÞ ¼ ½wNþ1 ðkÞ; wNþ2 ðkÞ; . . .wNþm ðkÞ is the coefficient vector of FIR block.
The received data corrupted by white Gaussian noise can be expressed as:
XR ðkÞ ¼ ½xðkÞ; xðk  1Þ; . . .xðk  m þ 1ÞT ð12Þ
XR ðkÞ is the simultaneous input for neural network as well as FIR filter whose overall
output combination is fed to the feedback section via signð:Þ function. The feedback filter
output directly reaches to the input layer of the functional expansion block.

VðkÞ ¼ ½v1 ðkÞ; v2 ðkÞ; . . .vq ðkÞT ¼ ½signðuo ðk  1ÞÞ; signðuo ðk  2ÞÞ; . . .signðuo ðk  qÞÞT
ð13Þ
Equation (13) represents the feedback signal vector. The sign(.) function is a threshold,
used for reconstruction of the discrete data sequence. The single layer functional expansion
network structure [22, 23] performs trigonometric expansion of the input sequence as in the
Eq. (21):
XR ¼ ½ðxR ; cosðpxR Þ; sinðpxR Þ; cosð2pxR Þ; sinð2pxR Þ; . . .Þ ð14Þ

The expansion comprises of ‘k’ orthogonal basis functions i.e.

UðkÞ ¼ ½U1 ðkÞ; U2 ðkÞ; . . .UK ðkÞT ð15Þ

To test the adaptive model, l0 -norm based sparseness is introduced to BLMS and
FBLMS which are used to train the model and adjust the equalizer coefficients as shown in
Fig. 3.

4 Background of Sparse Adaptive Filtering

There are various norm penalty functions which can be introduced to the cost function of
adaptive filters to favour sparsity. These includes l0 -norm, l1 -norm,l2 -norm in general lp -
norm (0 \ p \ 1). More smaller the value of p for the lp -norm more it resembles for the l0 -

123
 H. K. Sahoo et al.

Decission Device

B1 ( k )
Z −1
d r (k )
i

Z −1
Bq ( k ) uO (k )

X R (k )
Φ1 ( k ) ∑ β (u (k ))
z fn (k )
∑ ∑
w1 ( k ) u (k )
Φ 2 (k )
Z −1 FN w2 ( k )

Φ n (k )
wN ( k )
wN +1 ( k )
z f (k )
er (k )
wN + 2 ( k ) ∑

wN + m ( k )
Sparse Adaptive
Algorithms

Fig. 3 Adaptive equalizer model

norm [24]. For a class of LMS adaptive filters l0 -norm penalty can be used to make the
model sparse by generating a zero attractor [25, 26]. The general weight update equation
for a traditional LMS adaptive filter is given as:
wi ðk þ 1Þ ¼ wi ðkÞ þ lerðkÞxðk  iÞ for 0  i\l ð16Þ

erðkÞ ¼ dr ðkÞ  xT ðkÞwðkÞ ð17Þ

where l is the step-size parameter. The l0 -norm penalty can be incorporated to the
quadratic cost function and can be represented as

EðkÞ ¼ jerðkÞj2 þckwðkÞk0 ð18Þ

where k:k0 denotes the l0 -norm that counts the number of non-zero entries in wðkÞ and
c [ 0 balances the new penalty and estimation error. A general approximation for l0 -norm
is given as:
L1 
X 
kwðkÞk0  1  ebjwi ðkÞj ð19Þ
i¼0

when b approaches infinity, the two sides are strictly equal so the Eq. (19) can be written as
L1 
X 
EðkÞ ¼ jerðkÞj2 þc 1  ebjwi ðkÞj ð20Þ
i¼0

By minimising the cost function of Eq. (20), the weight update equation is given as:

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

wi ðk þ 1Þ ¼ wi ðkÞ þ lerðkÞxðk  iÞ  Kbsgnðwi ðkÞÞebjwi ðkÞj ð21Þ

where K ¼ lc and sgnð:Þ is a component wise sign function defined as:

x=jxj; x 6¼ 0
sgnðxÞ ¼ ð22Þ
0; elsewhere

The computational complexity caused by last exponential term of the Eq. (21) can be
reduced by the first order tailor series expansion of the exponential function as:

bjxj 1  bj xj; jx  1=bj
e ¼ ð23Þ
0; elsewhere

As the above exponential function is larger than zero so the approximation of Eq. (21) is
bounded to be positive and again the weight update equation can be remodified as:
wi ðk þ 1Þ ¼ wi ðkÞ þ lerðkÞxðk  iÞ þ Kfb ðwi ðkÞÞ ð24Þ
where
8 2
< b x þ b; 1=b  x\0
fb ðxÞ ¼ b2 x  b; 0\x  1=b ð25Þ
:
0; elsewhere

5 Sparse Model Based Block and Fast Block Adaptive Filters

l0 -norm based sparsity [27, 28] as shown in Eq. (24) can be combined with block and fast
block processing of data samples as described below. The proposed models are developed
as sparse BLMS and sparse block FBLMS for efficient estimation of coefficients. The
convergence of the algorithms is achieved by minimising the cost function as given in
Eq. (20).

5.1 BLMS and Sparse BLMS

Instead of processing the input data on sample basis, BLMS algorithm [29, 30] process the
input data samples through L-point blocks by doing serial to parallel conversion and the
blocks are applied to FIR filter of length M,one block at a time. The weights of BLMS
algorithm are updated using input blocks with the implementation steps as given below:
The weight vector W is given with the expression
WðkÞ ¼ ½w0 ðkÞ; w1 ðkÞ; w2 ðkÞ; . . .wM1 ðkÞT ð26Þ

The output of the filter for the input vector xðkL þ iÞ can be given as:
Y1ðkL þ iÞ ¼ wT ðkÞxðkL þ iÞ ð27Þ

In the analysis, k is the block index and L is the length of each individual block I case of
BLMS.

123
 H. K. Sahoo et al.

If dr ðkL þ iÞ denotes the desired response then the signal, then the error signal erðkL þ
iÞ can be given as:
erðkL þ iÞ ¼ dr ðkL þ iÞ  Y1ðkL þ iÞ ð28Þ

The expression for weights update equation for BLMS algorithm is represented as
follows:
XL1
wi ðk þ 1Þ ¼ wðkÞ þ l i
xðkL þ iÞerðkL þ iÞ ð29Þ

The proposed sparseness in the block processing approach can be introduced simply by
introducing a modification in the update Eq. (29) as explained in section-4 to develop
sparse BLMS algorithm and the modified equation is given as:
XL1
wi ðk þ 1Þ ¼ wðkÞ þ l i
xðkL þ iÞerðkL þ iÞ þ Kfb ðwi ðkÞÞ ð30Þ

The factor Kfb ðwi ðkÞÞ in Eq. (30) forces some of the channel coefficients values to zero
during weight update process.

5.2 FBLMS and Sparse FBLMS

In contrary to traditional FBLMS [31, 32], the sparse model with frequency domain block
processing enhances the error convergence rate with a less computational burden. In case
of sparse-FBLMS some of the weight coefficients which contribute less towards gradient
recursion will be zero in frequency domain. The implementation steps of FBLMS [33–35]
and sparse-FBLMS are similar except the modification present in Eq. (37) as explained
below:
1. Diagonalization of blocks are performed by Fourier transform, to generate a matrix of
dimension 2M  2M as shown in Eq. (31)
MðkÞ ¼ diagonalðFFT½mððk  1ÞMÞ. . .mðkM  1Þ; mðkMÞ; . . .mððk þ 1ÞN  1ÞT Þ
ð31Þ
only last ‘N’ elements of dimension M  1 are retained as given in Eq. (32)
ur ðkÞ ¼ last N elements of IFFT½MðkÞwðkÞ
^ ð32Þ

2. Desired signal vector of required dimension is generated from the same block
DðkÞ ¼ ½dr ðkMÞ; dr ðkM þ 1Þ. . .dr ððk þ 1ÞN  1ÞT ð33Þ

3. Error vector is generated by taking the difference between the desired and estimated
output signal vector, which is transformed to frequency domain by using Fourier
transform.
erðkÞ ¼ dr ðkÞ  ur ðkÞ ð34Þ

0
Ef ðnÞ ¼ FFT ð35Þ
erðkÞ

4. Average power estimation and coefficient update will be carried out using Eq. (36) and
(37).

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

pðkÞ ¼ kpðk  1Þ þ ð1  kÞM H ðkÞMðkÞ ð36Þ

" #
wðkÞ
wi ðk þ 1Þ ¼ wi ðkÞ þ aFFT þ Kfb ðwi ðkÞÞ ð37Þ
0

wðkÞ ¼ first N elements of IFFT½PðkÞM H ðkÞEf ðkÞ ð38Þ

where


PðkÞ ¼ p1 ðkÞ ¼ diag p1 1 1
0 ðkÞ p1 ðkÞ. . .p2N1 ðkÞ ð39Þ

The sparseness is introduced to FBLMS in Eq. (37) which is obtained from modified
cost function by incorporating norm penalty.

6 Simulation Results and Discussions

All the simulations are carried out using MATLAB 2015 for Rayleigh’s fading channel
models and the practical parameter values used in outdoor channel modelling are given in
Table 1.

6.1 Channel Estimation Results

The estimation results of the channel coefficients by BLMS and FBLMS with and without
sparsity are presented through sub figures as Fig. 4a–d. These subfigures reflect the con-
vergence behaviour of both non sparse and sparse algorithms showing how the estimated
coefficients acquire values nearer to the actual time varying values. The transfer function
of Jacks model used to represent the wireless channel for simulation purpose can be given
as:
HðzÞ ¼ ð0:3430þ2:7510i )Z1 þð0:3430þ2:7510i)Z2 þð0:3429þ2:7509i)Z3
ð39Þ
þð0:3428þ2:7508i)Z4 þð0:3427þ2:7506i)Z5

Table 1 Practical Parameter

Delay parameter 2
Values for Simulation
Doppler shift 900 Hz
Mobility 100 km/h
Noise level 30 dB
Modulation 16 QAM
Number of multi path 8
Carrier frequency 500 MHz
Sampling frequency 15,000 MHz
Forward FIR filter order 4
Feedback FIR filter order 2
Order of functional expansions of NN 5

123
 H. K. Sahoo et al.

(a) (b)
1.5
0.3

Weights value
1
Weights value

0.2
0.5
0.1
0
0
-0.5
-0.1
-1
200 400 600 800 1000 1200
200 400 600 800 1000 1200
Number of Iterations
Number of Iterations

(c) (d)
1
Weights value

0.4

Weights value
0.5

0 0.2

-0.5 0
-1
-0.2

200 400 600 800 1000 1200

200 400 600 800 1000 1200
Number of Iterations Number of Iterations

Fig. 4 Estimation results for weight convergence. a Estimated weight convergence for BLMS algorithm,
b Estimated weight convergence for FBLMS algorithm, c estimated weight convergence for sparse-BLMS
algorithm, d estimated weight convergence for sparse-FBLMS algorithm

The coefficients estimation results for all four algorithms to approach the above transfer
function reveal that sparse-FBLMS is more robust against the noise effects among all the
algorithms.
To perform the block processing, block length is divided by the length of the input
sample which reduces the number of iterations as compared to LMS. Fast block approach
has the added advantage of frequency domain processing rather than time domain. Due to
sparse modeling, some of the weight coefficients tend to zero during the process of iter-
ations which leads to substantial reduction in computational complexity.
All the four algorithms are compared as shown in Fig. 5 in terms of MSE, calculated
from the estimated coefficients of the channel which indicates that the mean square error
performance of sparse-BLMS is BLMS are nearly equal but computational complexity is
less. Sparse-FBLMS based model has substantial reduction in estimation error than the
other three algorithms and faster error convergence behaviour which is clearly visible in
Fig. 5. Mathematically MSE can be calculated as:
X
N
MSE ¼ 1=N kerðkÞk2 ð40Þ
k¼1

where, k is number of iterations

6.2 Channel Equalization Results

The channel state information estimated in the previous section are used to design an
adaptive equalizer model as shown in Fig. 3. The BER and MSE values are evaluated for

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

Fig. 5 MSE comparison results 0.25

FBLMS
BLMS

0.2 SPARSE BLMS

SPARSE FBLMS

0.15

MSE
0.1

0.05

0
0 2000 4000 6000 8000

Number of iteration

(a) (b)
0 0
10 10
BLMS sparse-BLMS
FBLMS
sparse-FBLMS
BER
BER

-1 -1
10 10

-2 -2
10 10
0 5 10 15 20 0 5 10 15 20
SNR in dB SNR in dB

Fig. 6 BER comparison results

(a) (b)
0.2
0.1
BLMS sparse-FBLMS

FBLMS sparse-BLMS
MSE

0.1
MSE

0.05

0
0 0.5 1 1.5 2 2.5 3 3.5 4
4 0
Number of Iterations 10 0 0.5 1 1.5 2 2.5 3 3.5 4
4
Number of Iteration 10

Fig. 7 MSE comparison results

equalized sequence with 16-QAM format and the corresponding plots are shown in Figs. 6
and 7. The BER values are low and reduced to lower value with increase in SNR in case of
sparse FBLMS based equalizer as shown in Fig. 6b. Similar performance can also be seen
in terms of MSE as shown in Fig. 7b.

123
 H. K. Sahoo et al.

(a) (b)
4 4
= Transmitted Data = Transmitted Data
= Received Data
Imaginary part

= Received Data

Imaginary part
2 2

0 0

-2 -2

-4 -4
-4 -2 0 2 4 -4 -2 0 2 4
Real part Real part

(c) (d)
4
= Transmitted Data 4
= Received Data = Transmitted Data
2
Imaginary part

= Received Data

Imaginary Part
2

0
0

-2
-2

-4 -4
-4 -2 0 2 4 -4 -2 0 2 4
Real part
Real Part

Fig. 8 Equalized constellation diagram. a Constellation diagram of BLMS, b constellation diagram of

sparse-BLMS, c constellation diagram of FBLMS, d constellation diagram of sparse-FBLMS

The symbol constellation plots are generated from equalized sequence and compared
with original constellation. For sparse-FBLMS the Euclidean distances between the
transmitted data points and received data points have been reduced to a great extent as
compared to other three algorithms considered for comparison as shown in Fig. 8a–d. This
clearly indicates low BER and the ability of adaptive equalizer to reconstruct the bit
sequence with minimum error using sparse FBLMS algorithm.

7 Conclusion

The proposed sparse estimation and equalization models can be used effectively for
practical outdoor wireless environments with user mobility. From BER and MSE analysis,
it is quite apparent that block processing approach in frequency domain along with sparse
modelling improves the overall symbol detection capability of the digital receiver with a
reduction in computational complexity which is a primary requirement to design modern
wireless receiver.

Acknowledgements Authors acknowledge the support of Veer Surendra Sai University of Technology,
Burla and IIIT Bhubaneswar for providing rich resources in terms of e-journals and reference books.

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

Appendix

Base Band Channel Model for QAM

QAM modulated transmitted signal passing through the channel can be expressed as:
X
xbp ðtÞ ¼ sc ðkÞpðt  kTÞ cos xc t
k
X
 ss ðkÞpðt  kTÞ sin xc t
k
X
¼ Re ðsc ðkÞ þ jss ðkÞÞpðt  kTÞejxc t ðA:1Þ
k

where sc ðkÞ,ss ðkÞ and p(t) are real. p(t) is the base band signal and sc ðkÞ þ jss ðkÞ represents
the complex QAM symbol. The complex base band equivalent signal is
X
xlp ðtÞ ¼ ðsc ðkÞ þ jss ðkÞÞpðt  kTÞ ðA:2Þ
k

Assuming xlp ðtÞ to be transmitted over a band pass channel hbp ðtÞ with low pass
equivalent hlp ðtÞ:
hbp ðtÞ ¼ Reðhlp ðtÞejxc t Þ ðA:3Þ

So the channel out is:

" ! #
X
ybp ¼ ðhbp  xbp ÞðtÞ ¼ Reððxlp  hlp ÞðtÞejxc tÞ ¼ Re ðsc ðkÞ þ jss ðkÞÞpðt  kTÞ  hlp ðtÞ ejxc t
k
" ! #
X

¼ Re ðsc ðkÞ þ jss ðkÞÞhlp;e ðt  kTÞ ejxc t ¼ Re ylp ðtÞejxc t
k

ðA:4Þ
where hlp ; e ¼ ðhlp  pÞðtÞ and ylp ðtÞ denotes the complex signal
" !#
X
ylp ðtÞ ¼ ðsc ðkÞ þ jss ðkÞÞhlp;e ðt  kTÞ
k ðA:5Þ
¼ yr ðtÞ þ jyi ðtÞ

So


ybp ðtÞ ¼ Re ylp ðtÞejxc t ¼ yr ðtÞ cos xc t þ jyi ðtÞ sin xc t ðA:6Þ

The QAM demodulator separates yr ðtÞ and yi ðtÞ. But as hlp;e ðtÞ is complex,yr ðtÞ is a linear
combination of sc ðkÞ and the interference component ss ðkÞ. Similarly yi ðtÞ is the linear
combination of ss ðkÞ and sc ðkÞ. As yr ðtÞ and yi ðtÞ are sampled at the output of QAM
demodulator,ylp ðnTÞ is the discrete version of ylp ðtÞ. The equivalent complex discrete-time
channel is given as hd ðnÞ ¼ hlp;e ðnTÞ and the corresponding sampled signal is
yd ðnÞ ¼ ylp ðnTÞ. Thus the expression of output signal is

123
 H. K. Sahoo et al.

X
yd ðnÞ ¼ ðsc ðkÞ þ jss ðkÞÞhd ðn  kÞ þ qðnÞ ðA:7Þ
k

where qðnÞ represents the channel noise.

Sparsity Introduced Through Norm Penalty

The size of a vector ~ w is called the norm of ~w, which can be represented as L1 -norm, L2 -
norm, Lp -norm, … L1 -norm etc.
~w ¼ ðw1 ; w2 ; w3 . . .wn Þ, (wi 2 C for i ¼ 1; 2; . . .n)
In general the L1 -norm is defined as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
wk1 ¼ 1
k~ w1
i i
ðA:8Þ

Let wj is the highest entry in the vector wi , then by the property of the infinity,
w1 1
j ¼ wi ; 8i 6¼ j
X
which implies w1 1
i ¼ wj ðA:9Þ
i
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P p ffiffiffiffiffiffiffi
wk1 ¼ 1 i w1
and k~ i ¼
1
w1 j ¼ jwi j
So the L1 -norm is defined as:
wk1 ¼ maxðjwi jÞ
k~ ðA:10Þ
In that context the Lp -norm or lp -norm for p 2 N; 1  p\1 can be defined as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X pffi
wkp ¼ p
k~ w
i i
ðA:11Þ

The cost function for LMS family using lp -norm penalty can be represented as:

Elp ðkÞ ¼ 1=2jerðkÞj2 þcp kwðkÞklp ðA:12Þ

Using gradient descent method the p-norm based update equation can be derived as:
 1p 
 
wðk þ 1Þ ¼ wðkÞ þ lerðkÞxðkÞ  qp wðkÞlp  sgnðwðkÞÞ jwðkÞj1p ðA:13Þ

where qp ¼ lcp
when the entry of wðkÞ approaches to zero then the above equation is modified as:
 1p 
 
wðk þ 1Þ ¼ wðkÞ þ lerðkÞxðkÞ  qp wðkÞlp  sgnðwðkÞÞ 2p þjwðkÞj1p

ðA:14Þ
here 2p indicates the bounds.

References
1. Nakhai, M. R. (2008). Multicarrier transmission. IET Signal Processing, 2(1), 1–14.

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

2. Zhuang, W. (1999). Adaptive channel equalization for wireless personal communications. IEEE
Transactions on Vehicular Technology, 48(1), 126–136.
3. Gunther, J., & Moon, T. (2009). Minimum bays risk adaptive linear equalizers. IEEE Transactions on
Signal Processing, 57(12), 4788–4799.
4. Chen, R., Jr., & Wu, W.-R. (2004). Adaptive asymptotic bayesian equalization using a signal space
partitioning technique. IEEE Transactions on Signal Processing, 52(5), 1376–1386.
5. Laot, C., Glavieux, A., & Labat, J. (2001). Turbo equalization: Adaptive equalization and channel
decoding jointly optimized. IEEE Journal on Selected Areas in Communications, 19(9), 1744–1752.
6. Zhou, X., Liu, F., Yin, Y., Li, Q., & Qin, J. (2017). Robust beam forming for simultaneous wireless
information and power transfer in MISO interference channels. Wireless Personal Communications, 92,
1545–1557.
7. Jiang, S. Peng, J., Lu, Z., Jiao, J., & Jiang, S., (2014). 802.11ad key performance analysis and its
application in home wireless entertainment. In 2014 IEEE 17th international conference on compu-
tational science and engineering (CSE).
8. Naylor, P. A., Cui, J., & Brookes, M. (2006). Adaptive algorithms for sparse echo cancellation. Signal
Processing, 86(6), 1182–1192.
9. Martin, R. K., Sethares, W. A., et al. (2002). Exploiting sparsity in adaptive filters. IEEE Transactions
Signal Processing, 50, 1883–1894.
10. Jiang, S., & Gu, Y. (2015). Block-sparsity-induced adaptive filter for multi-clustering system identifi-
cation. IEEE Transactions on Signal Processing, 63(20), 5318–5330.
11. Gu, Y., Jin, J., & Mei, S. (2009). Norm constraint LMS algorithm for sparse system identification. IEEE
Signal Processing Letters, 16(9), 774–777.
12. Poggioni, M., Rugini, L., & Banelli, P. (2008). A novel simulation model for coded OFDM in doppler
scenarios. IEEE Transactions on Vehicular Technology, 57(5), 2969–2980.
13. Ko, Y.-C., & Jeong, G. (2002). Doppler spread estimation in mobile communication systems. In IEEE
Vehicular Technology Conference (pp. 1941–1945).
14. Banelli, P., Cannizzaro, C. R., & Rugini, L. (2007). Data-aided Kalman Tracking for channel estimation
in Doppler-affected OFDM systems. In Proceedings of IEEE ICASSP (Vol. 3, pp. 133–136).
15. Wu, W. R., Lee, C. F., & Chen, Y. F. (2009). Time domain equalization for discrete multitone
transceivers. IET Communication, 3(7), 1186–1200.
16. Datta, A., Hari, K. V. S., & Hanzo, L. (2014). Channel estimation relying on the minimum bit error ratio
criterion for BPSK and QPSK signals. IET Communications, 8, 69–76.
17. Di Lorenzo, P., Barbarossa, S., Banelli, P., & Sardellitti, S. (2016). Adaptive least mean squares
estimation of graph signals. IEEE Transactions on Signal and Information Processing, 2(4), 555–568.
18. Chang, M.-X., & Su, Y. T. (2002). Model-based channel estimation for OFDM signals in rayleigh
fading. IEEE Transactions on Communications, 50(4), 1963–1968.
19. Li, L., He, X., & Zhang, H. (2016). Bandwidth efficient design and channel estimation over time-
varying fading environments. In 9th IEEE international congress on image and signal processing,
biomedical engineering and informatics (CISP-BMEI) (pp. 1091–1095).
20. Patraj, C., & Pal, N. R. N. (1995). A functional link artificial neural network for adaptive channel
equalization. Digital Signal Processing, 43(2), 81–195.
21. Hussein, A., Soraghan, J., & Durrani, T. S. (1997). A new adaptive functional-link neural-network-
based DFE for overcoming co-channel interference. IEEE Transactions on Communications, 45(11),
1358–1362.
22. Patra, J. C., Pal, R. N., Chatterji, B. N., & Panda, G. (2009). Identification of nonlinear dynamic systems
using functional link artificial neural network. IEEE Transactions on Systems, Man and Cybernetics,
29(2), 254–262.
23. Sicuranza, G. L., & Carini, A. (2011). A generalized FLANN filter for nonlinear active noise control.
IEEE Transactions on Audio, Speech and Language Processing, 19(8), 2412–2417.
24. Taheri, O., & Vorobyov, S. A. (2011). Sparse channel estimation with lp-norm and reweighed l1-norm
penalized least mean squares. In IEEE ICASSP (pp. 2864–2867).
25. Angelosante, D., Bazerque, J. A., & Giannakis, G. B. (2010). Online adaptive estimation of sparse
signals: Where RLS meets the l1-norm. IEEE Transactions on Signal Processing, 58, 3436–3447.
26. Chen, Y., Gu, Y., Hero, A. O. (2009). Sparse LMS for system identification. In Proceedings IEEE
ICASSP, Taipei (pp. 3125–3128).
27. Martin, R. K., Sethares, W. A., et al. (2002). Exploiting sparsity in adaptive filters. IEEE Transactions
on Signal Processing, 50, 1883–1894.
28. Etter, D. M. (1985). Identification of sparse impulse response systems using an adaptive delay filter. In
ICASSP (Vol. 85, pp. 1167–1172).

123
 H. K. Sahoo et al.

29. Rugini, L., Banelli, P., & Berioli, M. (2007). Block equalization for single-carrier satellite communi-
cations with high mobility receivers. In Proceedings of IEEE Globecom (pp. 5021–5025).
30. Gross, J. H., Etter, D. M., Margo, V. A., & Carlson, N. C. (1992, August) A block selection adaptive
delay filter algorithm for echo cancellation. In Proceedings of the 35th Midwest Symposium on Circuits
and Systems (pp. 895–898).
31. Rugini, L., & Banelli, P. (2008). Frequency-domain extended models for equalization of doubly-
selective channels. In Proceedings of the IEEE SPAWC (pp. 520–524).
32. Dentino, M., McCool, J., & Widrow, B. (1978). Adaptive filtering in the frequency domain. Pro-
ceedings of the IEEE, 66, 1658–1659.
33. Bershad, N. J., & Feintuch, P. L. (1979). Analysis of the frequency domain adaptive filter. Proceedings
of the IEEE, 67, 1658–1659.
34. Narayan, S. S., & Peterson, A. M. (1981). Frequency domain LMS algorithm. Proceedings of the ZEEE,
69, 124–126.
35. Sahoo, H. K., & Mohanty, B. (2016). Adaptive decision feedback equalizer for SISO communication
channel using combined FIR-neural network and fast block LMS algorithm. In IEEE INDICON.

Harish Kumar Sahoo, Ph.D. is currently working as Associate Pro-

fessor in the Department of Electronics and Tele-Communication
Engineering, Veer Surendra Sai University of Technology, Burla,
Sambalpur, India. Prior to this assignment, he worked as Asst. Pro-
fessor at IIIT Bhubaneswar, India. He is a senior member of IEEE. He
completed his M.Tech. from National Institute of Technology, Rour-
kela and Ph.D. from Sambalpur University. He has more than 16 years
of teaching and research experience. He has more than 12 research
publications in different journals and conferences from Elsevier and
IEEE. He is an active reviewer of IEEE Transactions on Instrumen-
tation and Measurement, IEEE Transactions on Power Delivery,
Elsevier ISA Transactions, Springer, Taylor and Francis etc. His
research interest includes adaptive filtering and soft computing appli-
cations in wireless communication, system identification and power
system.

Basabadatta Mohanty received her M.Tech. Degree from BPUT,

Rourkela, India. She is currently pursuing her Ph.D. at IIIT Bhuba-
neswar India. Her research interest includes adaptive filtering and soft
computing applications in wireless communication.

123
Block and Fast Block Sparse Adaptive Filtering for Outdoor…

Bijayananda Patnaik, Ph.D. is currently working as Asst. Professor at

IIIT Bhubaneswar,India. He received Ph.D. degree from IIT Bhuba-
neswar India. He published his research works in Elsevier and IET
journals. His research interest includes wireless communication and
Free space optics.

123