Sie sind auf Seite 1von 4

Review Of Sparse Channel Estimation

Basic focus on the development of the Sparse Channel Estimation using pilot signal &
comparison of different algorithms being employed.

Sanjeev Baghoriya
Electronics & Communication Engineering department
Indraprastha Institute of Information Technology
Delhi, India
sanjeev13161@iiitd.ac.in

AbstractSparsity occurs in various communication applications
which can be infact a boon in estimating the channel to recover the
signal transmitted through it. In addition to this , if the Channel
State Information(CSI) are known at the receiver then the recovery
becomes easier. Therefore , we have taken into account the pilot
signal utilization technique for the channel estimation by
equalisers. This paper reviews the various algorithms & methods
that have been described over the years by many pioneers in their
work. The later portion of the paper compares the algorthms to
discuss the pros ,cons & possible future advancements .
Index Terms Pilot signal, Sparsity
INTRODUCTION
Wireless communication applications basically involves lots
of problems but the main problem is the successful reception &
recovery of the transmitted signal through the channel . So , it is
very much critical to understand the behaviour of the channel in
the sense that how will the signal be affected .Whether it will
degrade or completely lost or have multipath etc .So knowing the
channel characteristics is basically termed as the process of
channel estimation. With the improvements in the channel
estimation techniques it is even possible to transmit & receive
over the various range of frequencies over different fading
phenomenas. We have divided our content in sub-sections where
firstly we will be discussing about the fundamentals of our
problem such as sparsity of a channel , equalisers
Sparsity:
Exploiting the sparsity of channel with different methods has
been a hot topic for the researchers . It not only reduces the work
of the receiver but also complexity & the cost . Sparsity[1] is
basically a notion where the information or the data rate of a
signal is smaller than depicted by its bandwidth or the number of
degrees of freedom on which signal depends is smaller than its
length[2]. Sparse channel is the channel impulse response where
the zero components outnumber the non-zero components. So,it
is not necessary to estimate all the enteries of the sparse signal
rather only the knowledge of the non-zero enteries as well as the
location of those enteries is required for the process of
reconstruction.
The condition can mathematically can be stated as the signal x
is sparse in the basis V. The signal can be represented using only
a small number K<<M of elements from V.

|| Vx||
lo
M .


Fig.1 : Sparse Channel Impulse Response

Equalisers: [3]
Since many practical channels are bandlimited & distort the
transmit signal linearly which causes ISI [3]. Therefore to
mitigate the affects of the channel an intelligent receiver
component is employed which is known as the equaliser. The
complexity of equalisers increases as the channel length
increases ,thus the sparse channels are well suited for their easy
functioning . Let us consider the following block diagram:

Fig.2 Basic Wireless communication block diagram [4]
Channel Estimation Techniques:
Channel estimate allows the receiver to approximate the effects
of the channel on the signal. It is of three types:
1) Pilot Assisted: - It is the most straight forward where the
symbols are known to the receiver.
2)Blind(Without Pilots): - It is based on the channel statistics
employment rather than the pilots.
3)Semi-Blind: - It is the combination of the above where the
initial estimation is pilot based & next on channel tracking.
For example , an ofdm fram contains preambles ,guard bands ,
data symbols,cyclic prefix & a pilot tone which contains the
information of the mentioned parameters so that it becomes easy
for the receiver to recover the signal & process accordingly.
Basically the channel is observed as :
Y =H*X +n
where length of Y is << length of X to fulfill the sparsity
condition.
OVERVIEW OF THE ALGORITHMS
a) Classical Least squares

The classical Least Square channel[5]
estimation ,popular in the 90s ,assisted by the pilots includes two
steps in the frequency domain. Initially , the channel impulse
response for the all the pilots is estimated separately. Then , the
channel impulse response for the non-pilots is obtained through
process of interpolation.
,where J is the cost function,
where X
p
is the pilot symbol , E

p
channel response estimate of
each pilot signal, Y
p
is the received signal.
then the estimator E

p
becomes
E

p
=(X
P
H
X
p
)
-1
X
P
H
Y
p
. For LS , we can say that every tap in this
channel estimate will have a non-zero value which will adversely
affect the equalizer performance. So , to estimate a sparse
channel reponse the non-zero taps are reduced by means of
threshold which is known as thresholdedLS(ThLS). It provides
very slow convergence & is expensive in comparison to other
available methods but will perform efficiently for high SNR.

b)Zero Tap Detection using Approximate Maximum Likelihood

This[6] known as AMLE (Maximum likelihood
estimation) is generally used in the detection problems &
outperforms LS at low to moderate SNR but they have a
disadvantage of having high computational costs than classical
LS. Let us assume irrespective of the channel structure the first
estimate as per LS is :
h=[U
T
U]
1
U
T
x . But here the assumption is based only on the
fact that the priori is available only for the LS estimator after
that the whole process depends on the channel statistical
measurements , which is not the area of our interest but to
explain the development & approaches over the years it is
critical to undertstand the comparison.

from [6,section 3.1] we have ,
,where z is the optimising parameter
&p=q=1/2 is assumed for the approximate detection.

b
`
,hence this problem can be solved using the Viterbi algorithm .
They have a problem of unstable output at the high SNR wjere
the LS has upperhand. Below is the BER comparison of both for
a single user bound.

Fig.3 LS vs AMLE BER comparison.. [6]
The advancements in AMLE will be discussed in the inferences
section of all the algorithms.

b) Matching Pursuit Algorithm [3]

Let the stationary channel impulse impulse response be denoted
as C(m), m=0,.,M-1 is sent through the channel & the
received entries are r(m), m=0,,M-1& for sparse channel
C(m) 0 for very few values of m. The response is given as the .
As per [3],

where b denotes the residual matrix & A the column matrix
which when has maximum correlation with b provides
orthogonal projection & the iterations go on till the solution is
reached.
where K represents the
tap value position & correspondingly the tap coefficients. It has
the faster convergence rate advantage over the classical least
squares . But it has the problem of basis re-selection as it
sequentially selects the column matrix from the dictionary which
makes it of less use for equaliser in various applications.
c) Orthogonal Matching pursuit

More accurate estimation of the channel is derived in [8] which
completely eliminates the problem of the basis-reselection. The
work in [8] explains that orthogonal matching pursuit at the
equaliser employs an intelligent method that selects the base
vector such that it has maximum correlation with the residual
vector. This makes the job of the decision feedback equaliser a
lot easier in finding the approximate number of tap.

d) Block Orthogonal Matching Pursuit[9]


Fig.4 Cluster Sparse channel response

In various channel environments there exist large obstacles like
mountains , hills , buildings which tends to give sparse cluster-
structure in the channels which was not considered in the
application of the MP as well as the OMP algorithms. They
loose their importance when large multipath interference arises
due to the long duration pulse shapes caused by these cluster
sparse channels.
The channel vector h is , ||h||
Lo
=K<<L where it has the K-
sparsity. Lo norm counts the non-zero taps in a vector. Also ,
Y=HX+n where :

problem :

which depicts that the cluster is counted only if it is exceeding
the noise floor otherwise the calculated & estimated values are
considered as a part of the noise matrix ,n. All the cluster
components are assigned same tape values as they exhibit same
characteristics thereby reducing the complexity & calculations
required by the equaliser. BOMP works in the same manner as
the OMP except that it consider the correlation of the block of
sparse clusters channel matrix.

e) OMP using Sensing Measurement matrix[10],[11]

Linear Inverse problems are encountered in the channel
estimation applications especially for the sparse channels ,where
the accurate estimation by using the shorter designed
training/pilot sequence is a major problem to solve. It should be
desgned to satisfy the restricted isometric property [12].
Conventionally it has been a problem due to coherence
interference of columns in training signal which is defined as :

where p is the mutual incoherence .
Input: Observation signal vector
Output: Sparse signal Vector [
SMM

Step1: Initialise the residual r0= & set the selected variable
X=0/ . Iteration=1 count start.
Step2: Find variable X
i
that solves maximisation problem
where W is SMM
& uses .
Add the variable X
i
to the seet of selected variables & update X.
Step 3: Update r
i

Step 4: If the stopping condition is attained stop algo else move
to step 2.


where signal length is N=48 & 1000 monte-carlo runs are used.

f) Compress Sensing Technique[12][13][14]

Considering the conventional linear model for the measurement
: y=x= V 0 where V 0 denoted the effective matrix for
the estimation of the k sparse vector 0 .

Fig 6. Compressed sensing mathematical representation.

Methods:
RIP(Restricted
Isometric
Property
It provides the basis to not to deteriorate the
compressed samples but it doesnt signy
how to recover the sparse vector.
Lo norm It estimates the sparse solution accurately
but it is not feasible.
L
1
norm Gives correct estimation but includes mild
oversampling
L
2
norm Solution is never sparse approximately.

CONCLUSION & INFERENCES
>>Orthogonal matching pursuit algorithms are bandwidth
efficient , good convergence rate with shorter training sequences
but they dont have that much stable output as far as the
procedure moves on with the number of iterations in the time
varying channels, Thus , it creates little difficulty for the receiver
equalisers in estimation of the channel.
>>Whereas the least square classical approach works well
for high SNR signals with pretty good convergence rate when
modified by providing threshold to all the non-zero taps to
estimate the correct number. They are not as much bandwidth
efficient as OMP based equalisers are.
>>Cluster sparse estimation estimates smaller number of
channel freedom of degree & shorter training sequences. They
are spectrally efficient in wideband applications.
>>Approximate maximum likelihood estimation has been in
advancement now a days as it is modified with an iterative
approach by utilizing an initial estimate . widely used in the ofdm
systems.[15]
>>Compressive sensing is a very hot topic as it provides the
shortest training sequence along with minimum number of
samples for the sparse channels analysis. Since the correct
estimation using the Lo norm is not feasible to efficient
optimization thus showing a way for the convex optimization
techniques.Basis pursuit * lassos algorithm are known to work
well & are still under research to give more correct estimate.

ACKNOWLEDGEMENT
I wish to thank Dr. Angshul Majumdar to provide me this
opportunity to increase my bank of knowledge with such a new
field of signal processing.
REFERENCES
.[1] Sparse channel estimation based on the CS theory for UWB
systems by Eva Tagarona
[2]. F.Naini ,PVandergheyst , R.garbonval , ultrawide band width
time-hopping spread-spectrum Impulse radio for wireless multiple
access communications. IEEE International conference , speech &
signal processing ,2009.
[3]. Sparse channel estimation via matching pursuit with application to
equalization 2002 S.Cotter , ,B.rao .
[4] Wireless & Cellular systems Lect-11 , Dr. Ashok Bohara , IIIT
Delhi.
[5] Performance of OFDM systems using the least square channel
estimation in high speed railway environment by Z. chao sheng ,
Y. Fang , Shanghai University.
[6] Sparse Channel estimation with zero tap detection by C carbonelli ,
S.Vedantam , University of California.
[7] Fundamentals of digital communication by J ohn.G.Proakis. , 5
th

edition.
[8] Sparse channel estimation using orthogonal matching pursuit
algorithmby Gunes.Z , Abbas .Y , University of Ottawa.
[9] Compressive estimation of cluster sparse estimation channels by
G.gui , N.zheng , Nwang , 2011 , PER.
[10] Sparse signal recovery with OMP algorithm using sensing
measurement matrix. by G.gui , A.Mehbondiya , Q.wan, F.Adachi
, IEICE , Vol.8
[11] J . A tropp , A.C.gilbert , Signal recovery random measurement
via orthogonal matching pursuit , dec-2008 , IEEE trans. Inf
.theory.
[12] Application of compressive sensing to sparse channel estimation
by R.berger , Z.wang, S.Zhou , Carnegie Melan University
[13] Toeplitz compressive Sensing matrices with application to sparse
channel estimation by G.raz, J.Haupt , W.Bajwa , members IEEE.
[14] Introduction to compressed sensing , Mark.A , Marco.F , Y.eldar
[15] Pilot Assisted synchronization for wqireless ofdm channels over
fast time varying fading channels. , by S.kapoor , Y.huang ,
Vehicular Technology Conference.

Das könnte Ihnen auch gefallen