Independent component analysis (ICA) is a signal processing technique used for unmixing of the mixed recorded signals. In wireless communication, ICA is mainly used in multiple input multiple output (MIMO) systems. Most of the existing work regarding the ICA applications in MIMO systems assumed static or quasi static wireless channels. Performance of the ICA algorithms degrades in case of time varying wireless channels and is further degraded if the data block lengths are reduced to get the quasi stationarity. In this paper, we propose an ICA based MIMO transceiver that performs well in time varying wireless channels, even for smaller data blocks. Simulation is performed over quadrature amplitude modulated (QAM) signals. Results show that the proposed transceiver system outperforms the existing MIMO system utilizing the FastICA and the OBAICA algorithms in both the transceiver systems for time varying wireless channels. Performance improvement is observed for different data blocks lengths and signal to noise ratios (SNRs).
1. Introduction
I
ndependent component analysis (ICA) is a technique of blind source separation (BSS) and extensively has been used in several systems due to its simplicity and applicability in practical scenarios
[1

3]
from the last two decades. Various applications of ICA have been reported in the literature e.g., vibration analysis
[4
,
5]
, robotics
[6]
, fault diagnosis
[7]
, biomedical signal processing
[8]
, speech processing
[9]
, and wireless communication systems
[10
,
11]
. In wireless communication, ICA is mainly used in multiple input multiple output (MIMO) systems. The MIMO system consists of multiple transmit and receive antennas that make it an ideal candidate for the ICA algorithms.
In general, the ICA algorithms assumed static or quasi static wireless channels
[1]
. Several ICA based communication systems are presented in the literature. We summarized a few important research articles as follows. A semiblind receiver structure is presented in
[12]
, that overcomes the phase and permutation indeterminacies in the ICA post processed data by using a minimum number of pilot bits attached with each transmit data block in a quasistatic scenario. Although, the authors in
[12]
have added a small number of bits with low computational cost but achieved good biterrorrate (BER) performance. Another MIMO based communication system is presented in
[13]
, that have improved bandwidth efficiency without considering the spatial coding. An ICA based low complexity orthogonal frequency division multiplexing (OFDM) based receiver structure for MIMO system is proposed in
[14]
with enhanced performance for large subcarriers.
 1.1 Motivation
Fast and accurate ICA algorithms are presented in the literature
[15]
for static and quasistatic wireless channels. Performance of these algorithms degrades if variations in the wireless channels are such that the quasi stationarity condition does not hold. However, the practical wireless channels are time varying and unmixing of the mixed received signals becomes a challenging work for ICA to blindly unmix them in a time varying scenario
[16]
. If we reduce the lengths of the processing data blocks to get the quasi stationarity, then the performance of the batch ICA algorithms further degrades. Performance degradation occurs due to the statistical parameters used in the update rules of the ICA algorithms. To the best of our knowledge, we are the first to dicuss the applications of the batch ICA algorithms in time varying wireless channels.
 1.2 Contribution
In this paper, we propose an ICA based MIMO transceiver for time varying wireless channels. The proposed transceiver system also performs well if variations in the channels are such that the quasi stationarity condition becomes invalid even for smaller data blocks. To the best of our knowledge, no such a transceiver is presented in the literature that efficiently utilized the existing ICA algorithms in a time varying scenario. The transmitter side of this transceiver system transmits multiple copies of each sample of the source data. The receiver receives mixtures of the transmitted source signals. The multiple mixed copies are averaged and then upsampled. The upsampled data is unmixed through the ICA algorithms and then down sampled to get back the original data blocks lengths.
Due to the use of samples copier (SC), samples averaging (SA) and up/down sampling operations used in the proposed transceiver system, we name it the MIMO SCSAUD transceiver system. The MIMO SCSAUD transceiver system improves the quality of the received signal in highly dynamic communication environment at the cost of reduced bandwidth efficiency due to the transmission of multiple copies of each source data sample. By highly dynamic or highly time varying we mean that the channels characteristics are changing for each transmitted sample. Moreover, we also propose the automatic values adjustment (AVA) technique in this paper. This technique controls the functions of the SC and the UD sampling operations blocks in such a way if there are no variations in the channel or the channel is quasi static then the SC block will copy each sample once. By this way the bandwidth efficiency will not be affected in quasi static channels condition. In short we can say that the bandwidth efficiency of the proposed transceiver becomes equal to the bandwidth efficiency of the existing MIMO system in quasi static channels conditions.
Performance of the proposed transceiver is evaluated over quadrature amplitude modulated (QAM) signals. The performance evaluation criterion is signal to interference ratio (SIR).
 1.3 Notations and Organization
The notations used in this paper are such that the capital size bold letters are used for matrices and the small size bold letters represent vectors. Symbols and abbreviations used in this paper are given in
Table 1
and
2
respectively.
List of Symbols
List of Abbreviations
Remaining paper is organized as follows. The signal model and overview of ICA is given in Section 2, Section 3 presents the proposed MIMO SCSAUD transceiver system. Section 4 illustrates the simulation results. Finally, conclusion is given in Section 5.
2. The Signal Model and Overview of ICA
We consider a multiuser MIMO wireless communication system with
M
_{T}
transmit and
M
_{R}
receive antennas. For the sake of simplicity, it is considered that
M
_{T}
=
M
_{R}
=
M
. The transmitted signals vectors are
s
_{1}
,
s
_{2}
,…,
s
_{M}
, where
s
_{m}
= [
s
_{m1}
,
s
_{m2}
,…,
s
_{mL}
] and the received mixed signals are
x
_{1}
,
x
_{2}
,…,
x
_{M}
. The system model for ICA signals processing is shown in
Fig. 1
. The ICA post process reconstructed signals are represented by
y
_{1}
,
y
_{2}
,…,
y
_{M}
. Where the post processed signals are the resultant unmixed signals of the ICA algorithms. The unmixed signals are the estimated resultant source signals transmitted by different antennas. The ICA algorithms unmix the received mixed signals in order to find the actual transmitted source signals. The received mixed signals can be modeled as follows
where,
X
is
M
×
L
mixed data matrix,
H
is
M
×
M
complex valued channel matrix and
S
is
M
×
L
matrix that contains the independent and nonGaussian source signals. Though the ICA research can be divided into the following categories

(a) Equal number of transmit and receive antennas[14,16]: In this case we get the mixing coefficients in such a way that the mixing matrix becomes square. For square mixing matrix stable and efficient ICA algorithms exist like FastICA algorithms[25], OBAICA algorithm[16], and Infomax algorithm[26]etc.

(b) The number of transmitted signals greater than the number of received signals: This area of research in the ICA literature is called overcomplete ICA[17].

(c) The number of transmitted signals less than the number of received signals: This area of research in the field of ICA is known as the undercomplete ICA[18].

(d) More transmit signal and one receive antenna or sensor: This is a different area of research in the field of ICA signal processing, where we receive one mixed signal for further processing through ICA algorithms[19,20].
The existing MIMO transceiver system for ICA signals processing.
We use the square channel or mixing matrix in our analysis. As a future work, we would like to extend it for nonsquare case utilizing the techniques of nonsquare mixing scenario.
Equation (1) represents the noise free model of ICA. The noisy ICA model can be represented as
where,
N
is
M
×
L
matrix, which represents the complex valued additive white Gaussian noise with variance one and assumed independent to the source signals. The addition of noise also degrades the performance of the ICA algorithms. The role of the ICA algorithms is to estimate the unmixing matrix
W
, where
W
=
H^{1}
. After estimating the unmixing matrix the desired unmixed signals can be represented as follows
Once the unmixing matrix
W
is determined, then, the individual component signals can be separated from the mixed signals by using equation (3). Accuracy of the ICA algorithms depends on the criterion used for estimating the unmixing matrix.
In case of highly time varying channels, the coefficients of the channel matrix
H
changes for each incoming data sample and the unmixing becomes a challenging work for the ICA algorithms.
Assumptions used in this paper are given below

The source signals are mutually statistically independent. If different signals are related to different sources then the independence assumption becomes valid.

The source signals have nonGaussian distributions. The nonGaussian structure of the signals is also practical, because most of the signals from different physical processes have nonGaussian distribution. Moreover, the transmitted signals in digital modulation schemes like quadrature amplitude modulation (QAM), binary phase shift keying (BPSK), and quadrature phase shift keying (QPSK) are nonGaussian in nature as discussed in[2123], where we select finite alphabets from a finite alphabet set.

The channel matrixHis assumed to be square. If channel matrix is square then estimation of the unmixing matrixWbecomes easy. This condition means that the number of the underlying component signals and the mixture signals is equal.
There are some indeterminacies in the ICA post processed data, as given below

Order of the ICA post processed signals changes for each incoming data block.

Phase and amplitude of the ICA post processed signals also change for each processing data block.
3. The Proposed MIMO SCSAUD transceiver system
In this section, the proposed transceiver system called the MIMO SCSAUD transceiver is presented. The MIMO SCSAUD transceiver improves the separation performance of the batch ICA algorithms in highly time varying channels. In case of highly time varying channels the quasi stationary condition becomes invalid even for smaller data blocks. The proposed transceiver system is given in
Fig. 2
.
The proposed MIMO SCSAUD transceiver system.
The proposed transceiver system first processes the source signals
s
_{1}
,
s
_{2}
,…,
s
_{M}
, through samples copier (SC) block that copies each sample
Q
times before transmission. The source signals are the information signals which we want to transmit. These signals may be speech or any other information signals, which we want to transmit. In order to generate these signals, we first generate random data in matlab using the rand (random) function and then modulate it using the qammod (quardrature amplitude modulation) function of matlab for further processing through the SC block. The transmitted source signals after the SC block are
s
^{'}
_{1}
,
s
^{'}
_{2}
,…,
s
^{'}
_{M}
, where
s
^{'}
_{m}
= [
s
^{'}
_{m1}
,
s
^{'}
_{m2}
,…,
s
^{'}
_{mLQ}
]. The SC process for each source data vector is demonstrated as follows
At the receiver side linearly mixed data vectors
x
_{1}
,
x
_{2}
,…,
x
_{M}
are received and passed through the samples separator (SS) and samples averaging (SA) blocks. The SS block separates the mixed received samples inserted by the SC block and the SA block calculates their average values. The signal processing of the SS and the SA blocks is shown in
Fig. 3
, where 1
D
represents shift right and 1
D
represents shift left by one sample. This figure shows that first the mixed data vector
x
(n) is decimated by integer value
Q
to discard the extra inserted samples, then delayed and down sampled
x
(n)
Q
times and then shift back all the decimated vectors to remove the delays, finally add them up and divide by
Q
to get the average values. This process is demonstrated mathematically as follows
where
x
^{1}
_{m1}
,
x
^{2}
_{m1}
, …,
x
^{Q}
_{m1}
represent the mixed received copies of samples that are transmitted
Q
times, where the transmitted source symbols are
s
^{1}
_{m1}
,
s
^{2}
_{m1}
, …,
s
^{Q}
_{m1}
.
The signal processing of the samples separator and samples averaging blocks.
In the next step, the mixed data vectors
x
^{'}
_{m}
are upsampled before unmixing. The upsampling (US) operation increases the data blocks lengths by using an integer factor
Z
. The US block generates (
Z
 1) samples between two adjacent samples. The upsampling is a two steps process, (1) zero insertion, and (2) the samples generation. The zero insertion process for a mixed received data block
x
^{'}
(n) can be represented as follows
where,
δ
is the unit impulse function.
After zero insertion the samples generation process is performed using the linear interpolator of
[24]
. The impulse response of the linear interpolator represented by
h
(n) and can be written as follows
The upsampled mixed data vectors are
x
^{↑}
_{m}
= [
x
^{↑}
_{m1}
,
x
^{↑}
_{m2}
, ….,
x
^{↑}
_{mZL}
], where
m
=1,2,…,
M
. The upsampled mixed data is unmixed by using the complex valued FastICA algorithm
[25]
. This algorithm is used as a bench mark to evaluate the effectiveness of the proposed transceiver system. The resultant unmixed data vectors are
y
^{↑}
_{1}
,
y
^{↑}
_{2}
, ….,
y
^{↑}
_{M}
. The unmixed data is then down sampled using the decimation parameter
Z
to get back the original data blocks lengths. The DS or decimator discards (
Z
1) number of samples and keeps the remaining samples. The decimated version of the data block
y
^{↑}
(n) can be represented as follows
The decimated data vectors are
y
_{1}
,
y
_{2}
, …,
y
_{M}
, where
y
_{m}
= [
y
_{m1}
,
y
_{m2}
,…,
y
_{mL}
].
The indeterminacies in the resultant reconstructed data can be resolved by transmitting few known symbols with each data block as performed in
[12]
.
Now, the important thing is how to select the optimal values of the key parameters
Z
and
Q
. For this purpose, we propose the automatic values adjustment (AVA) technique. This technique is responsible for selection of the desired values of
Z
and
Q
according to the channels conditions. The AVA technique is presented in
Fig. 4
. The AVA technique initially selects the SC parameter
Q
and the UD sampling parameter
Z
equal 1 and computes the signals to interference ratios (SIRs) of the resultant unmixed signals
y
_{1}
,
y
_{2}
, …,
y
_{M}
. The SIR represents the average ratio of the desired signal power to the power of the estimation error and is defined as follows
where i=1,2,...,
M
.
The automatic values adjustment (AVA) technique for the key parameters Q and Z.
SIR is used as a criteria to evaluate the results of the proposed and existing MIMO systems using the FastICA and OBAICA algorithms in both the transceivers. Basically, we are transmitting
M
number of signals utilizing
M
number of antennas. Equation (9) calculates the SIRs of all the transmitted data blocks. After calculating the SIRs we compute the average SIR (
SIR_{avg}
) as follows
Now compare the value of the
SIR_{avg}
with the predefined threshold μ, where this constant defines the quality of the resultant unmixed signals. If
SIR_{avg}
> μ, then keep the same values of
Q
and
Z
. In this scenario, the signal processing of the proposed transceiver system becomes equal to the signal processing of the existing MIMO system shown in
Fig. 1
. Moreover, if
SIR_{avg}
≤ μ, due to variations in the wireless channels, then the AVA technique will increase the values of
Q
and
Z
iteratively up to the desired limit. Furthermore, the signal processing steps of the proposed transceiver system with the AVA technique are summarized as follows

Generate the signals which we want to transmit

Perform the SC operation by using equation (4)

Perform the SS & SA operations by using equation (5)

Perform the upsampling by using equation (6) and (7)

Perform unmixing using the Complex valued FastICA algorithm of[25]

Perform down sampling using equation (8)

Calculate the SIRs of each unmixed data block by using equation (9)

Calculate the SIRavgthrough using equation (10)

Define the value of μ

Compare the SIRavgwith μ, if SIRavg> μ then go to steps 1, 3, 5, and keep the same values of Z and Q, otherwise increase them up to the desired values.
Computational complexity of the FastICA algorithm utilizing the existing MIMO system shown in
Fig 1
is O(ML). The proposed transceiver system has the complexity of O(MLZQ). Though, the proposed transceiver system will increase the computational complexity as compared to the complexity of the existing MIMO system in time varying channels. However, the proposed system gives increased accuracy of the results in time varying scenario, which is shown in Section 4. Moreover, in case if the channel is quasi static then the complexity of the proposed transceiver becomes equal to the computational complexity of the existing MIMO system, which we shown in detail in the automatic values adjustment (AVA) technique in
Fig. 4
. In short, we can say that the computational complexity and bandwidth utilization of the proposed system increases if the channels become highly time varying, where the channels are varying for each transmitted source symbol.
4. Results and Discussion
In this section, the performance of the proposed transceiver system is simulated and discussed. A dual antenna MIMO transceiver system is considered for simulation purpose. The transmitter transmits quadrature amplitude modulated (QAM) signals with equal carrier frequencies and power. Signal to noise ratio (SNR) ranges from 3dB to 10dB, throughout the simulations. Additive white Gaussian noise (AWGN) channels with frequency flat fading are used. The receiver receives linear mixtures of the transmitted source signals.
We consider both the time varying and the quasi static wireless channels. In case of time varying channels the condition of the quasi stationarity becomes invalid even for smaller lengths of the processing data blocks and the coefficients of the channel matrix
H
are changing within the processing data blocks. The channel matrix used in our simulation is given below,
where coefficients
δ
_{1}
,
δ
_{2}
,
δ
_{3}
and
δ
_{4}
of the matrix
H
are complex valued random variables. We generate them randomly in our simulation for time varying as well as quasi static channels. In quasi static case, we keep the channel coefficients fixed during the processing data blocks, while in time varying scenario these coefficients are changed randomly for each transmitted data sample.
The performance measure used in our simulation is signal to interference ratio (SIR). Simulation is performed using Monte Carlo simulation for each input data block. The input data blocks lengths ranges from 20 to 200 samples.
Different parameters used in our simulation are summarized in
Table 3
.
Simulation parameters
In our simulations, first, we evaluate the performance of the proposed MIMO SCSAUD transceiver system utilizing the quasi static as well as the time varying wireless channels, without utilizing the automatic value adjustment (AVA) technique. SNR is kept constant with a value of 10dB. In this simulation setup, we keep the value of
Q
equal 1 to clearly observe the effect of up/down sampling parameter
Z
with the values ranges from 1 to 4. We consider the smaller values of Z to protect oversampling and also to reduce the computational complexity of the proposed transceiver system. For detailed discussion regarding the value of
Z
one can refer to
[24]
. Results are demonstrated in
Fig. 5
. Performance improvements can be observed from this figure for different values of
Z
even in time varying channels.
Results of the proposed MIMO SCSAUD transceiver utilizing the SC parameter Q=1, and changing the value of Z from 1 to 4, in quasi static as well as time varying channels conditions.
Secondly, we evaluate the performance of the proposed system using the AVA technique. The AVA technique is used with fixed value of
Q
equal 1, for variable blocks lengths. Other parameters of the simulation setup are kept same as in the above experiment. Simulation is performed over quasi static and time varying conditions with different values of the predefined threshold. In case of quasi static channels we use
μ
= 11, 8, 7, 6, and in case of time varying scenario we use
μ
= 8, 7, 6, 5 respectively. Results are given in
Fig. 6
. One can observe from this figure that once the value of the predefined threshold is set to any value, then the AVA technique updates the value of
Z
from 1 to the desired value depending upon the value of
μ
for different blocks lengths and channel conditions.
Performance of the proposed transceiver system utilizing the AVA technique with the value of Q equal 1, and Z changes from1 to 4 in both the quasi static and time varying scenarios.
In the third simulation setup, performance of the MIMO SCSAUD transceiver using the AVA technique with
Q
=1, and
μ
= 8, is compared with the FastICA
[25]
and the OBAICA
[16]
algorithms for different blocks lengths and fixed SNR of 10dB, in time varying scenario. Results are shown in
Fig. 7
. Performance improvement of the MIMO SCSAUD transceiver can be observed from this figure even at larger blocks lengths in time varying scenario. Actually, for smaller data blocks, the expectation operator used in the update rule of the algorithm fails to produce correct results. By increasing the data block length, the accuracy of the results also increases. However, in a time varying scenario, the batch ICA algorithms are unable to track the variations and at larger block length the variations increases. Due to this reason, in our simulation, first the accuracy of the results increases and then decreases. The important point is that the results of the proposed system are better at smaller as well as larger data blocks.
Performance comparison of the proposed system with the FastICA and OBAICA algorithms in time varying channels. The AVA technique is used in the proposed system with Q=1, and μ=8, where values of Z changes from 1 to 4.
Furthermore, we also compared the results for different SNRs utilizing a constant block length of 30 samples. Results are given in
Fig. 8
. This figure shows superior performance of the proposed system for different values of the SNRs and smaller data block length of 30 samples.
Performance of the proposed transceiver system in comparison with the existing MIMO system utilizing the FastICA and OBAICA algorithms in both the transceivers for a block length of 30 samples and SNRs ranges from 2 to 10dB.
As a cross check we compared the bit error rate (BER) performance of the proposed transceiver with the FastICA and OBAICA algorithms for data block length of 30 samples. Other parameters of the simulation setup are kept same as in the previous experiment. Results are demonstrated in
Fig. 9
where the proposed system shows superior performance in terms of BER for a block length of 30 samples and different SNRs. It must be noted that the value of
Q
=1, and
Z
varies from 1 to 4.
Bit error rate (BER) performance of the proposed system in comparison with the FastICA and OBAICA algorithms.
In the next simulation setup, we consider the proposed system with different values of
Q
and
Z
, where both the values range from 1 to 4. We manually adjust these parameters to clearly observe their effect over unmixing. The value of
Q
also kept smaller due to more bandwidth utilization in time varying scenario. Results are given in
Fig. 10
. In these results first we keep the value of
Q
=2 and change the value of
Z
from 1 to 4. Secondly, we keep
Q
=4 and change
Z
from 1 to 4, where more improvements in results observed while increasing the value of
Q
from 2 to 4 for larger blocks lengths in time varying scenario.
Results of the proposed MIMO SCSAUD transceiver system for different values of Q and Z without using the AVA technique.
In
Fig. 11
, we demonstrate the results of the proposed system with the AVA technique. Different values of the predefined threshold
μ
are used which are 5, 7, and 9. SNR is kept constant with the value of 10dB. The channels are time varying. The AVA technique continuously checks the value of
SIR
_{avg}
in comparison with the predefined threshold
μ
i.e, for
μ
=5 the
SIR
_{av}
is greater than 5 for initial block length of 20 samples, so the values of
Q
and
Z
remained 1. In case of
μ
=7, and 9, the values of
Q
and
Z
are increased from 1 to the desired values, because the
SIR
_{avg}
is less than 7 or 9 for block length of 20 samples. In this simulation more performance improvements are observed for
μ
=7, and 9, which is the main contribution of the proposed system. Moreover, the proposed system is also evaluated for different SNRs, with the data block length of 100 samples. Other parameters are same as in the above experiment. Results are given in
Fig. 12
, which shows performance improvements for larger value of
μ
.
Results of the proposed MIMO SCSAUD transceiver system utilizing the AVA technique for μ=5, 7, and 9.
Results of the proposed transceiver system utilizing the AVA technique with μ=5, 7, and 9 for data block length of 100 samples and SNRs ranges from 2 to 10dB.
Finally, we demonstrate the performance of the MIMO SCSAUD transceiver (with
μ
=7) in comparison with the existing MIMO system utilizing the FastICA and OBAICA algorithms in both the transceivers for time varying scenario. Results are tabulated in
Table 4
. Results shown in this table are such that the SIR performance of the FastICA algorithm degrades for increased length of the processing data blocks while performance of the proposed transceiver further improves. This is the main achievement of the proposed transceiver system.
The SIR performance comparison of the proposed transceiver system with the FastICA and the OBAICA algorithms in time varying scenario.
The SIR performance comparison of the proposed transceiver system with the FastICA and the OBAICA algorithms in time varying scenario.
The results for different SNRs with block length of 100 are given in
Fig. 12
. This figure also shows superior performance of the proposed MIMO SCSAUD transceiver system for time varying wireless scenario.
5. Conclusion
In most of the ICA algorithms implemented in wireless communication systems, the channels are considered static or quasi static. Its use in time varying channels to blindly estimate the original signal from the mixed received signals is limited and inefficient. In this paper, the MIMO SCSAUD transceiver system is proposed that performs well in time varying scenario. The channels are assumed highly time varying with the channel characteristics changing for each transmitted sample. We use the complex valued FastICA algorithm as a benchmark to evaluate the effectiveness of the proposed system. Simulation is performed over QAM modulated signals for various SNRs and input data block lengths. Performance of the MIMO SCSAUD transceiver is compared with the OBAICA algorithm and the FastICA algorithm. Results show that the proposed transceiver outperforms the OBAICA and FastICA algorithms in highly time varying wireless channels.
As a future work, we would like to extend the proposed approach for nonsquare case utilizing the techniques of nonsquare mixing scenario.
BIO
Zahooruddin, is a faculty member in the department of Electrical Engineering, COMSATS Institute of Information Technology. He received an Industrial Electronics Engineering degree from NED University, Pakistan, an MS Degree in Electronics Engineering from Air University Pakistan in 2002 and 2008 respectively. He is currently doing their PhD in Signal Processing and Communication from COMSATS Institute of Information Technology Pakistan. His current research is algorithms and applications of Independent Components Analysis in wireless Communication systems.
Ayaz Ahmad obtained his B.Sc. in Electrical Engineering from University of Engineering and Technology, Peshawar, Pakistan in 2006. In 2008, he received Master degree in Wireless Communication Systems from Ecole Superieure d'Electricite (Supélec), GifsurYvette, France. He obtained his Ph.D. degree in Telecommunication from Supélec, France in 2011. Since June 2012, he is serving as an Assistant Professor in the Department of Electrical Engineering, COMSATS Institute of Information Technology, Wah Campus, Pakistan. His research interests include resource optimization in wireless communication systems and signal processing for wireless systems and energy management in smart grid.
Muhammad Iqbal, got B.Sc. Electrical Engineering degree in 1999 from University of Engineering and Technology, Lahore and MS Telecommunication Engineering from University of Engineering and Technology, Peshawar. After completing PhD he joined COMSATS Institute of Information Technology and working as Assistant Professor in the department of Electrical Engineering. His research interest is signal and information processing in wireless communication.
Nadir Shah is working as an Assistant Professor at COMSATS Institute of Information Technology, Pakistan. He received PhD from SinoGerman Joint Software Institute, Beihang University, Beijing, China. He earned M.S. in Computer Science from International Islamic University, Islamabad, Pakistan in 2007. He earned B.Sc and M.Sc in Computer Science from Peshawar University, Peshawar, Pakistan in 2002 and 2005 respectively. He was a Lecturer at Department of Computer Science, COMSATS Institute of Information Technology, Abbottabad, Pakistan between August 2007 and June 2008. His current research interests include Computer Networks, Distributed Systems, and Network Security. He has published several research papers in international journals/conferences (like ACM Computing Surveys, IEEE Communication Letters etc). He has been serving asa reviewer for several journals/conferences (e.g. ICC, Computer Networks Elsevier, TheComputer Journal etc.).
Stone J. V.
2004
“Independent component analysis, a tutorial introduction,”
MIT Cambridge Press
https://mitpress.mit.edu/books/independentcomponentanalysis
Ge Z.
,
Song Z.
2014
“Ensemble independent component regression models and soft sensing application,”
Elsevier NeuroImaging, Chemometrics and Intelligent Laboratory Systems
130
115 
122
DOI : 10.1016/j.chemolab.2013.09.009
Khorshidia G. S.
,
Douauda G.
,
Beckmannb C. F.
,
Glasserd M. F.
,
Griffantia L.
,
Smitha S. M.
2014
“Automatic denoising of functional MRI data combining independent component analysis and hierarchical fusion of classifiers,”
Elsevier NeuroImaging
90
449 
468
DOI : 10.1016/j.neuroimage.2013.11.046
Razaghi H.
,
Saatchi R.
,
Offiah A.
,
Burke D.
,
Bishop N.
,
Gautam S.
2013
“Assessing material densities by vibration analysis and independent component analysis,”
Malaysian Journal of Fundamental and Applied Sciences
http://www.academia.edu/3716978/Assessing_Material_Densities_by_Vibration_Analysis_And_Independent_Component_Analysis
9
(3)
123 
128
Li X.
,
Li W.
,
Sun Y.
,
Zhao H.
“Blind source separation of vibration signal of electric traction experiment system,”
in Proc. of International Conference on Intelligent Control and Information Processing (ICICIP2010)
2010
93 
96
Saruwatari H.
,
Mori Y.
,
Takatani T.
,
Ukai S.
,
Shikano K.
,
Hiekata T.
,
Morita T.
“Twostage blind source separation based on ICA and binary masking for realtime robot audition system,”
in Proc. of IEEE/RSJ International Conference on Intelligent Robots and Systems
2005
2303 
2308
Lou W.
,
Shi G.
,
Zhang J.
“Research and application of ICA technique in fault diagnosis for equipments,”
in Proc. of IEEE International Conference on Intelligent Computing and Intelligent Systems
2009
vol. 4
310 
313
Mehrabian H.
,
Chopra R.
,
Martel A.
2013
“Calculation of intravascular signal in dynamic contrast enhancedmri using adaptive complex independent component analysis,”
IEEE Transactions on Medical Imaging
32
(4)
699 
710
DOI : 10.1109/TMI.2012.2233747
Uddin Z.
,
Alam F.
2010
“Hardware implementation of blind source separation of speech signals using independent component analysis,”
International Journal of Electrical and Computer Sciences
http://www.ijens.org/1006012525%20IJECSIJENS.pdf
10
(1)
85 
86
Obradovic D.
,
Madhu N.
,
Szabo A.
,
Wong C. S.
“Independent component analysis for semiblind signal separation in MIMO mobile frequency selective communication channels,”
in Proc. Of IEEE International Joint Conference on Neural Networks
2004
vol. 1
25 
29
Gupta M.
,
Santhanam B.
“ICA based blind adaptive MAI suppression in DSCDMA systems,”
in Proc. of IEEE Signal Processing Education Workshop
2004
201 
205
Jiang Y.
,
Zhu X.
,
Lim E.
,
Dong L.
,
Huang Y.
2011
“Lowcomplexity independent component analysis based semiblind receiver for wireless multiple input multipleoutput systems,”
International journal of design, analysis and tools for circuits and systems
http://ijdatics.distributedthought.com/current_issues/IJDATICS_02_02_02.pdf
2
(2)
91 
98
Curnew S.
,
Ilow J.
“Blind signal separation in MIMO OFDM systems using ICA and fractional sampling,”
in Proc. of International Symposium on Signals, Systems and Electronics
2007
67 
70
Sarperi L.
,
Zhu X.
,
Nandi A.
2007
“Blind OFDM receiver based on independent component analysis for multipleinput multipleoutput systems,”
IEEE Transactions on Wireless Communications
6
(11)
4079 
4089
DOI : 10.1109/TWC.2007.060191
Tan X.
,
Zhang X.
,
Ye J.
2014
“Adaptive step size natural gradient ICA algorithm with weighted orthogonalization,”
Journal of Circuits systems and signal processing
33
(1)
211 
221
DOI : 10.1007/s0003401396241
Mikhael W. B.
,
Yang T.
2006
“A gradientbased optimum block adaptation ICA technique for interference suppression in highly dynamic communication channels,”
EURASIP Journal on Applied Signal Processing
2006
1 
10
DOI : 10.1155/ASP/2006/84057
Davies Mike
,
Mitianoudis Nikolaos
“Simple mixture model for sparse over complete ICA,”
in Proc. of IEE Vision, Image and Signal Processing
2004
vol. 151, no. 1
35 
43
Jordan Francis R.
,
Bach Michael I
“Blind onemicrophone speech separation: A spectral learning approach,”
in Proc. of International Conference on Advances in Neural Information Processing Systems
2004
http://www.cs.berkeley.edu/~jordan/papers/speechseparation.pdf
vol. 17
65 
72
Kristjansson Trausti
,
Attias Hagai
,
Hershey John
“Single microphone source separation using high resolution signal reconstruction,”
in Proc. of IEEE International Conference on. Acoustics, Speech, and Signal Processing
2004
vol. 2
814 
817
Zhao X.
,
Davies M.
2010
“Codingassisted blind mimo separation and decoding,”
IEEE Transactions on Vehicular Technology
59
(9)
4408 
4417
DOI : 10.1109/TVT.2010.2066588
Giorgetti A.
,
Chiani M.
2005
“Influence of fading on the Gaussian approximation for BPSK and QPSK with asynchronous cochannel interference,”
IEEE Transaction on Wireless Communication
4
(2)
384 
389
DOI : 10.1109/TWC.2004.843036
Lozano Angel
,
Tulino Antonia Maria
,
Verdú Sergio
2006
“Optimum power allocation for parallel Gaussian channels with arbitrary input distributions,”
IEEE Transactions on Information Theory
52
(7)
3033 
3051
DOI : 10.1109/TIT.2006.876220
Openheim A. V.
,
Schafer R. W.
,
Buck J. R.
2006
Discrete time signal processing
2nd, Ed.
Pearson education signal processing series
Bingham E.
,
Hyvarinen A.
2000
“A fast fixed point algorithm for independent component analysis of complex valued signals,”
International Journal of Neural Systems
10
(1)
1 
8
DOI : 10.1142/S0129065700000028
Jin Q.
,
Liang G.
“A new algorithm of infomax for small numbers of sound signal separation,”
in Proc. of International conference on artificial intelligence and education (ICAIE)
2010
159 
162