Академический Документы
Профессиональный Документы
Культура Документы
A Report submitted
in Partial Fulfillment of the Requirements for the Degree of
B.Tech
in
Avionics
by
(SC10B115)
Krunal Kolhatkar
(SC10B134)
Department of Avionics
Indian Institute of Space Science and Technology
Thiruvananthapuram
May-2014
BONAFIDE CERTIFICATE
This is to certify that this project report entitled “Application of Statistical Esti-
mation Techniques to Various Problems in Design of OFDM based Communi-
cations Systems” submitted to Indian Institute of Space Science and Technology,
Thiruvananthapuram, is a bonafide record of work done by Chervi Vivek Sai and
Krunal Kolhatkar under my supervision from 06-01-2014 to 20-04-2014.
Dr. N.Selvaganesan
HOD
Dept. of Avionics
IIST
Place: Thiruvananthapuram
Date:
Declaration by Authors
This is to declare that this report has been written by us. No part of the report is
plagiarized from other sources. All information included from other sources have
been duly acknowledged. We aver that if any part of the report is found to be
plagiarized, we shall take full responsibility for it.
Krunal Kolhatkar
SC10B134
Place: Thiruvananthapuram
Date:
Acknowledgments
We express our sincere thanks to all concerned who have contributed either di-
rectly or indirectly for the successful completion of our project on OFDM Commu-
nication Systems.
We are highly indebted to our project guide Dr. R.Lakshminarayan for his sup-
port and constant supervision during the course of our work. We truly appreciate
and value his esteemed guidance and encouragement from the beginning till the end
of the project. We are indebted to him for having helped us shape the problem and
providing insight towards the solution.
We thank Lily Srujana for the guidance and support. We thank Dr. K. S. Das
Gupta, Director - IIST for providing this opportunity. We extend our sincere grati-
tude to Dr. N. Selvaganesan, HOD Avionics, for allowing us to take up this project.
We extend my sincere thanks to one and all of IIST family for helping us at the right
times.
Abstract
ABBREVIATIONS iv
List Of Figures v
1 INTRODUCTION 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Aim and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Basics of OFDM 3
2.1 Introduction to OFDM . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Multicarrier transmission . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 OFDM as a multicarrier transmission . . . . . . . . . . . . . . . . . 4
2.4 Basic principles of OFDM . . . . . . . . . . . . . . . . . . . . . . 6
2.4.1 Principle of orthogonality . . . . . . . . . . . . . . . . . . 6
2.5 Transmission Procedure . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5.1 OFDM modulation . . . . . . . . . . . . . . . . . . . . . . 7
2.5.2 Implementation using FFT . . . . . . . . . . . . . . . . . . 8
2.6 OFDM modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7 OFDM and the usage of guard intervals . . . . . . . . . . . . . . . 10
2.7.1 Effects of Multipath fading on OFDM Symbols . . . . . . . 10
2.7.2 Cyclic Extension of OFDM symbols . . . . . . . . . . . . . 11
2.7.3 Cyclic Prefix . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.7.4 Zero Padding . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
i
3 FADING CHANNEL 13
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Simulation Algorithms of Flat Fading Channel . . . . . . . . . . . . 15
3.2.1 Modified Smiths Algorithm . . . . . . . . . . . . . . . . . 16
3.3 Auto regressive Model for Fading Channel . . . . . . . . . . . . . . 18
3.3.1 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.2 AR modelling . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4.1 Simulation of Flat Fading Channels by Single IDFT Algo-
rithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 CHANNEL ESTIMATION 23
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Pilot Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.1 Block Type . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.2 Comb Type . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 System definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4 Estimation techniques . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4.1 Least Square Estimator . . . . . . . . . . . . . . . . . . . . 26
4.4.2 MMSE channel estimation . . . . . . . . . . . . . . . . . . 26
4.5 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
ii
6 OFDM RECEIVER PARAMETERS ESTIMATION 40
6.1 Time and Frequency Offset Estimation . . . . . . . . . . . . . . . . 40
6.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.1.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . 41
6.1.3 ML estimation . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2 SNR and Noise power Estimation . . . . . . . . . . . . . . . . . . 46
6.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.2.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2.3 Noise power Estimation . . . . . . . . . . . . . . . . . . . 48
6.3 Optimum Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.3.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . 51
REFERENCES 56
APPENDIX 58
iii
ABBREVIATIONS
iv
List of Figures
v
6.2 Shows the signals that generate the ML estimates (N = 400, L =
100, ε = .4 and SNR=15 dB). . . . . . . . . . . . . . . . . . . . . . 45
6.3 Shows the estimator mean-squared error as a function of L is estimated 45
6.4 Representation of OFDM frequency channel response and noise
spectrum. Spectrum for both white and colored noise is shown. . . . 47
6.5 Shows the Mean Squared error of Noise power estimation . . . . . . 49
6.6 Shows the SNR estimated for each symbol. . . . . . . . . . . . . . 50
6.7 Shows the BER performance of the BPSK when Noise Power is fed
in Kalman filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.8 Complete baseband transmission model . . . . . . . . . . . . . . . 52
6.9 Structure of an OFDM receiver. . . . . . . . . . . . . . . . . . . . 53
6.10 BER plot of BPSK symbol over fading channel. . . . . . . . . . . . 54
6.11 MSE plot for channel estimation. . . . . . . . . . . . . . . . . . . . 54
vi
Chapter 1
INTRODUCTION
1.1 Background
In a basic wireless communication system, the data is modulated on to a single
carrier frequency. The total available bandwidth is totally occupied by each symbol.
But transmission rates are severely limited in a single carrier environment due to
Inter Symbol Interference (ISI). ISI occurs when the symbol time is comparable to
the channel delay spread. This limits the transmission rate in a single carrier system.
Hence we require multicarrier transmission schemes.
In multicarrier transmission scheme total bandwidth available to the system is
divided into a series of frequency sub-bands. The data is then sent in parallel
through these sub-bands. Orthogonal Frequency Division Multiplexing (OFDM)
is a multicarrier scheme that improves the bandwidth efficiency of the system. Ba-
sic idea behind OFDM is to divide the available spectrum into several orthogonal
subcarriers. By doing so the spectrum of subcarriers overlap thereby increasing
the bandwidth efficiency and also, each narrowband sub-channels experience al-
most flat fading thereby increasing the reliability. A way to minimize Inter Block
Interference (IBI) in OFDM is to use Cyclic Prefix (CP). OFDM modulation and de-
modulation can be easily implemented using Inverse Fast Fourier Transform (IFFT)
and Fast Fourier Transform (FFT) techniques respectively.
Channel estimation and equalization is an integral part of OFDM system de-
sign. For proper decoding of the signal at the receiver accurate channel state in-
1
formation is necessary. Often channel state information is obtained by transmitting
known training sequence as pilot symbols. A conventional communication system
receiver, first obtains the channel estimates at the of the pilot symbols using various
algorithms like LS, MMSE, Kalman filtering, etc, which are then interpolated and
used for decoding the received signal.
2
Chapter 2
Basics of OFDM
3
2.2 Multicarrier transmission
The idea of multicarrier transmission was introduced as the data rates were severely
limited due to ISI. Basic approach behind multicarrier transmission is to split the
data streams in to K smaller sub streams and transmit these data sub-streams in to
adjacent carrier.
4
Figure 2.2: Carrier spectrum of OFDM and FDM
In practice Discrete Fourier Transforms (DFT) and Inverse DFT (IDFT) are used
to make OFDM system easier to implement. Practically DFT and IDFT can be
implemented using FFT and IFFT respectively. IFFT is used to assign the OFDM
symbols to their respective orthogonal subcarriers, so by taking N point IFFT of the
signal we can assign a stream of N OFDM symbols to N orthogonal sub carriers.
5
Figure 2.4: OFDM System implemented using IDFT/FFT pair
OFDM is a special case FDM where O stands for orthogonal. In OFDM, the sub car-
riers are orthogonal to each other unlike any other multicarrier transmission scheme
which is necessary for proper decoding of data at the receiver.
Mathematically orthogonality between sub carriers can be proved by fol- lowing
steps given below, We will consider the time-limited complex exponential signals
[e j2π fk t ]N−1
k=0 which represent the different subcarriers at f k = k/Tsym in the OFDM
signal where 0 ≤ t ≤ Tsym . These signals will be orthogonal if the integral of the
product of the first signal and the complex conjugate of the second signal, for their
fundamental period is zero, that is,
Z T Z T
1 j2π fk t − j2π fi t 1
e e dt = e j2π(k/Tsym )t e j2π(−i/Tsym )t dt
Tsym 0 Tsym 0
Z T
1
= e j2π(k−i)/Tsymt
Tsym 0 (2.1)
= 1 ∨ integer k = i
0 otherwise
Above condition is also true for the discrete samples with the sampling instances at
t = nTs = nTsym /N, n = 0, 1, 2, ...N − 1. it can be proved using the same given steps
above.Thus we have proved that sub carriers in OFDM are orthogonal to each other
as orthogonality is an essential condition for the OFDM signal to be free of Inter
Carrier Interference (ICI).
6
The Orthogonal part of OFDM suggest that there exist a special kind of relationship
between the sub carriers of the system.
The orthogonality property of the subcarriers allows the individual carriers to
overlap and still be immune to inter carrier interference (ICI). In the receiver, de-
modulators integrate the received signal over one symbol period to obtain the raw
data. The integration process results in zero contribution from all other sub carriers
even though there will be various delayed versions of the signal. In other words or-
thogonal sub carriers are uncorrelated from each other and are linearly independent.
The OFDM transmitter first modulates the input data bits into blocks of N symbols.
In Figure 2.5 Xl [k] denotes the symbol correspondong to the kth subcarrier in the l th
block. The symbol time for each symbol within the block gets extended to NTs as
data is sent in parallel. So after OFDM modulation, an OFDM block of time period
Tsym = NTs is formed. Then the baseband OFDM signals in the continuous time
domain can be expressed as:
∞ N−1
xl (t) = ∑ ∑ Xl [K]e j2π fk (t−lTsym)
l=0 k=0
In OFDM systems once the available bandwidth is known we have to decide on the
7
number of subcarriers or the subcarrier spacing. Subcarrier spacing is based on the
environment the system has to operate in, including maximum Doppler shift and
time delay spread of the channel. Once the subcarrier spacing has been decided, the
number of subcarrier can be estimated based on the total transmission bandwidth
available.
Carrier Spacing should be selected such that the carriers remain orthogonal to each
other. Orthogonal signals can be easily separated by using different correlation
techniques at the receiver.
Practically OFDM modulation can be easily implemented using inverse fast Fourier
transform technique. IFFT processing decreases the complexity and increases the
computational efficiency of OFDM system.
The continuous-time baseband signal can be sampled at the time instant, t =
lTsym + nTs where Ts = Tsym /N and fk = k/Tsym which gives the following discrete-
time OFDM symbols
N−1
xl (n) = ∑ Xl [K]e j2πkn/N
k=0
IFFT operation on continuous time OFDM signal yields the exact same result as
given above and it is also simple and easier to implement. IFFT operation also de-
creases the complexity in the transmitter, as instead of using bank of modulators
we can simply use N point IFFT operation. Thus the sequence xn , n = 0, 1, 2..,
N-1 is the size-N Inverse Discrete Fourier Transform (IDFT) of the block of mod-
8
ulation symbols X[0], X[1], ...., X[N-1]. Thus OFDM modulation can easily be
implemented using IDFT processing followed by digital to analog conversion, as
illustrated in Figure 2.6. In particular, by selecting the IDFT size N equal to 2m
for some integer m, the OFDM modulation can be easily implemented by means of
implementation efficient radix-2 inverse fast Fourier transform (IFFT) processing.
Z ∞ N−1
1 1
Z ∞
Yl [K] = yl [t]e j2π fk (t−lTsym )
dt = ∑ Xl [i]e j2pi fi (t−lTsym ) e− j2π fk (t−lTsym ) dt
Tsym −∞ Tsym −∞ k=0
N−1 Z Tsym
1
= ∑ Xl [i] Tsym e j2π( fi − fk )(t−lTsym )
k=0 0
= Xl [K]
(2.2)
In the above set of equations effect of channel and noise were not taken into account.
However in multicarrier transmission any corruption due to frequency selective ra-
dio channel may lead to loss of orthogonality between sub carriers and thus lead to
interference between adjacent sub carriers. To take care of the effects of frequency
selective channels cyclic-prefix insertion is used.
9
Figure 2.7: OFDM demodulation
If there are multiple paths from the transmitter to the receiver either due to reflec-
tions or obstructions, we can get fading effects. In such case, the signal reaches
the receiver following many different routes, each a copy of the transmitted signal.
Each of these rays has a slightly different delay and slightly different gain. The
time delays result in phase shifts which added to main signal component causes the
signal quality to be degraded.
In fading, the reflected signals that are delayed add to the main signal and cause
either gains in the signal strength or deep fades. And by deep fades, we mean that
the signal is nearly wiped out. The signal level is so small that the receiver cannot
decide what was there.
Delay Spread is defined as maximum time delay that can occur to the signal during
the transmission through the channel in that environment. Delay spread and multi-
path propagation causes Inter Symbol Interference (ISI) in the signal. Based on the
delay spread of the channel’s comparision to the symbol period there are two cases:
• If the symbol period is very large compared to the delay spread of the channel
the distortion caused is negligible.
• If the symbol period is comparable to the delay spread then distortion occurs.
10
2.7.2 Cyclic Extension of OFDM symbols
To eliminate ISI and multipath fading a guard interval is used in OFDM symbols.
Two techniques that can be used to form a guard interval are:
2. Zero Padding
Cyclic Extension is to extend the OFDM symbol by copying the last or first few
samples of the symbol and adding it as a prefix or suffix respectively to the symbol.
We use cyclic prefix as the extension in all the simulations.
Cyclic Prefix is the technique used to make OFDM symbols insensitive to time
dispersion in multipath channel. As we can see in Figure 2.8, the last part of the
OFDM symbol is copied into the starting of the OFDM symbol.
In zero padding instead of making the cyclic extension of OFDM symbols we insert
zeros equal to the length of the guard interval at the beginning of each OFDM
symbol. The only condition to be followed by the guard interval is that the time
interval of number of zeros should be greater than the maximum delay spread of
the channel that is, the ISI effect of one OFDM symbol on the symbol should be
confined within the guard interval.
11
Figure 2.9: OFDM Symbol with Zero Padding [1]
2.8 Summary
OFDM is a multicarrier transmission scheme which is bandwidth efficient as it uses
orthogonal subcarriers and also reliable as it converts a wide band into a number of
narrow band channels. The use of IFFT and FFT blocks reduces the complexity of
the system, compared to a bank of modulators. Another problem associated with
the OFDM system is ISI, due to the delay spread of the channel which is taken care
of by either zero padding or by using a cyclic extension.
12
Chapter 3
FADING CHANNEL
3.1 Introduction
In wireless mobile communications, due to reflections and scattering by obstruc-
tions such as buildings, automobiles etc. in the path between the transmitter and
receiver the transmitted signal arrives in multiple paths. This is called multipath
propagation. Each of the transmitted signal which follows different paths from the
transmitter to the receiver undergo different attenuations and have different delays
when it reaches the receiver. Due to these multipath components the strength of the
received signal randomly fluctuates and this phenomenon is called multipath fading.
Due to the difference in relative path lengths between individual paths, multiple
copies of the transmitted signal arrive at different times at the receiver which makes
signal to spread in time domain. Spreading of the signal depends upon the maxi-
mum excess delay of the channel τm , which is defined as time duration between the
first and last received components of the signal at the receiver or in other words we
can say τm is the maximum delay a signal can experience during its course in the
channel. An ideal channel should have zero excess delay.
Based on excess delay of the channel τm and symbol duration Tsym fading can
be characterised as follows
1. Flat Fading
13
When maximum excess delay of the channel is less than symbol period i.e. τm <
Tsym maximum delay the channel can offer will always be less than one symbol pe-
riod so all the multipath components of the symbol arrive during the same signalling
interval and there will not be any significant interference between adjacent symbols
at the receiver. Such a channel is called Flat Fading Channel.
Now if τm > Tsym , the channel exhibits frequency selective fading where excess
delay by the channel can exceed one symbol period. In such a case, some of the
multipath components of nth transmitted symbol can arrive during (n + 1)th symbol
duration. This results in Inter Symbol Interference (ISI). In frequency selective
channel the gain of the channel is different for different frequency components.
Until now we have assumed that there is no relative motion between transmitter
and receiver. Now considering that possibility, propagation paths can change with
the position of the transmitter and receiver which can result in change in channel
characteristics. Since the channel characteristics are dependent on the position of
transmitter and receiver which varies with time because of their movement , so
mobile channel can be characterised as time variant channel.
In the frequency domain, Doppler spread characterizes the effect of time varying
nature of the channel on the received signal. The Doppler shift in the frequency
of each of the multipath waves arises due to the motion of transmitter and receiver.
The amount of Doppler shift will be different for each of the multipath waves as
the angle of arrival of each wave is randomly distributed. This results in spectral
broadening of the signal. The amount of spectral broadening is expressed in terms
of Doppler spread. The Doppler spread, fd , is the maximum Doppler shift and is
given by,
v
fd =
λ
where v is the velocity of relative motion between transmitter and receiver in me-
ters/second and λ is the wavelength of the signal in meters.
The received signal can be expressed as sum of in-phase and quadrature-phase com-
ponents and for large number of paths between transmitter and receiver, these com-
ponents can be approximated as Gaussian random variables with zero mean and the
envelope of the received signal can be shown to have Rayleigh distribution [6]. The
14
probability density function (pdf) of the envelope of the received signal R(t) is given
by
r2
−
r
e 2σ2 , if r ≥ 0
fR (r) = σ2
0, if r < 0
In frequency domain power spectrum of received fading signal has the U shaped
bandlimited form:
1
, if | f − fc | ≤ fd
r
f − fc 2
S( f ) = π fd 1 − ( f )
d
0,
elsewhere
r(τ) = J0 (2π fd τ)
where J0 (.) is the zeroth order Bessel function of the first kind, and τ is the separa-
tion between observation times in seconds.
15
sequence having autocorrelation se- quence,
We know that magnitude of a zero mean complex Gaussian sequence with uncorre-
lated real and imaginary parts gives a sequence having Rayleigh pdf.
Modified filter FM [k] sequence ensures that output of a single IDFT operation will
give uncorrelated real and imaginary parts, hence there is no need of second branch
(IDFT operation). This results in significant reduction in computational complexity
of smiths algorithm.
The modified filter FM [k] which is used to shape the random Gaussian variables to
16
get desired U shaped doppler power spectrum is
0 k=0
v
u r 1
u
k = 1, 2, ..., km − 1
k 2
u
2( 1 − (
t
) )
N fm
r
km π km − 1
[ − arctan( √ )] k = km
2km − 1
2 2
FM [k] =
0 k = km + 1, ...., N − km − 1
r
km π km − 1
[ − arctan( √ k = N − km
)]
2 2 2km − 1
v
u r 1
u
k = N − km + 1, ...., N − 2, N − 1
k 2
u
2( 1 − (
t
) )
N fm
17
3.3 Auto regressive Model for Fading Channel
The IDFT technique discussed above is a high quality and efficient fading genera-
tor. But the disadvantage of this method is that all the samples are generated in a
single FFT operation. The memory constraint restrains us from using this method to
generate a large number of samples. In this section we describe the use of a general
auto regressive (AR) modeling approach for the generation of correlated Rayleigh
samples.
3.3.1 Generation
As discussed before, the theoretical power spectral density of either the in- phase
or quadrature phase part of the received fading signal has the U shaped bandlimited
form:
1
, if | f − fc | ≤ fd
r
f − fc 2
S( f ) = π fd 1 − ( f )
d
0,
elsewhere
Here fd is the maximum Doppler shift in Hertz. The equivalent normalized continu-
ous time autocorrelation of the received signal is R(τ) = J0 (2π fd τ), here J0 (.) is the
zeroth order Bessel function of the first kind. For the discrete time scenario, the-
oretically, the generated in-phase and quadrature Gaussian processes should each
have the autocorrelation sequence: R(n) = J0 (2π fm |n|), Here, fm = fd /Fs is the
maximum Doppler shift normalized by the sampling rate Fs . Also the in-phase
and quadrature processes must be independent and each must have zero mean for
Rayleigh fading.
3.3.2 AR modelling
A complex p-th order AR process can be generated by the time domain recursion
p
x[n] = − ∑ ak x[n − k] + w[n]
k=1
18
Where w[n] is a complex white Gaussian noise process with uncorrelated real and
imaginary parts. For Rayleigh modelling w[n] should have zero mean. The AR
model parameters that need to be found are the coefficients a1 , a2 , ...., a p and the
variance σ2p of w[n]. The Power Spectral Density of this model has the rational
form given by
σ2p
S( f ) = p
|1 + ∑k=1 ak exp(−2π f k)|2
Although the theoretical spectrum is not rational, an AR model of sufficiently high
order can closely approximate the theoretical spectrum.
− ∑ p am Rxx [k − m]
if k ≥ 1
m=1
Rxx [k] =
− ∑ p am Rxx [−m] + σ2
if k = 0
m=1 p
19
3.4 Results
20
Figure 3.2 and 3.3 show the results when two different fade rates, 10Hz and 100Hz,
are simulated using the single IDFT method. For each simulation, the time varying
envelope of the fading signal has been plotted.
Inference
As the doppler frequency increases the rate at which the magnitude of the fade vari-
able changes also increases which is expected as the relative velocity increases and
hence more variation in the channel.
Figure 3.4 and 3.5 show the results when two different fade rates, 10Hzand 100Hz,
are simulated using the single IDFT method. For each simulation, the PDF of the
fading signal has been plotted.
21
Figure 3.5: PDF of the Fading Signal, fd = 100 Hz
Inference
The PDF plots shows that the generated fading signal is Rayleigh distributed, thus
validating this model for Rayleigh fading simulation.
22
Chapter 4
CHANNEL ESTIMATION
4.1 Introduction
In an OFDM system, the transmitter modulates the message data into PSK/QAM
symbols, performs IFFT on the symbols to convert them into time-domain signals,
and sends them out through a channel. The received signal usually suffers distortion
by the channel between transmitter and receiver. In order to recover the transmitted
bits, the channel effect must be estimated and compensated in the receiver.
In an OFDM system each subcarrier can be considered as separate channel. The re-
ceived signal can be defined as a product of transmitted signal and channel response,
thus transmitted signal can be recovered by estimating the channel response at each
subcarrier in the receiver due to the orthogonality of the subcarriers. Pilot symbols
are those symbols known to both transmitter and receiver, which are used to es-
timate the channel at those particular locations which can then be extended using
various interpolation techniques to estimate the channel response of the subcarriers
between pilot tones. Computational complexity of the technique and time-variation
of the channel should be considered when choosing a particular technique for chan-
nel estimation
23
4.2 Pilot Structure
In a block type pilot structure shown in Figure 4.1 the pilot signal is assigned to a
particular block that is, the data at all the subcarriers is known to the receiver at a
particular time interval periodically. Using these pilots, a time-domain interpolation
is performed to estimate the channel. During the time period St for which pilots are
known, it is to be noted that the channel’s variation with time is constant in that
period.
In a comb-type pilot arrangement as shown in Figure 4.2 pilot symbols are dis-
tributed uniformly among subcarriers which are known throughout the transmission
time which are interpolated in frequency domain to estimate the channel. Similar
to the block type the period of the pilot tones S f should be such that the variation in
the channel with time is constant in that period
24
Figure 4.2: Comb-type pilot arrangement
X[0] 0 ··· 0
X[1] · · ·
0 0
X = .. .. ... .
. . ..
0 0 · · · X[N − 1]
Where X[k] denotes a pilot tone at kth subcarrier, k = 0, 1, 2,....,N 1. Given that
channel gain is H[k] for each subcarrier k, the received signal Y[k] can be defined as
Y [0] X[0] 0 · · · 0 H[0] Z[0]
Y [1] 0 X[1] · · ·
0 H[1] Z[1]
= +
.. .. .. .. .. ..
..
.
. . . . . .
Y [N − 1] 0 0 · · · X[N − 1] H[N − 1] Z[N − 1]
where H is a channel vector given as H = [H[0], H[1], ....., H[N − 1]]T and Z is a
noise vector given as Z = [Z[0], Z[1], ....., Z[N1]]T with E[Z[k]] = 0 and Var[Z[k]] =
σ2Z , k = 0, 1, 2, ...., N − 1.
25
4.4.1 Least Square Estimator
Least square (LS) channel estimation methods finds the channel estimate H in such
a way that the following cost function is minimized:
= Y H Y −Y H X Ĥ − Ĥ H X H Y + Ĥ H X H X Ĥ
∂J(Ĥ)
= −(X H Y ) + (X H X Ĥ) = 0
∂Ĥ H
Lets consider the LS estimate of the channel for each subcarrier as ĤLS [k], k = 0, 1,
2.....N-1. Since X is a diagonal matrix because of no ICI, the LS estimate ĤLS can
be written for each subcarrier as
Y [k]
ĤLS = , k = 0, 1, 2, ....., N − 1
X[k]
For a Gaussian and uncorrelated channel with impulse response given by h and
channel noise, the MMSE estimate of h is given by,
where,
26
Ryy = E[y, yH ] = XFT Rhh FTH X H + σ2n In
are the cross-covariance matrix between h and y and the auto-covariance matrix of
y. And Rhh is the auto-covariance matrix of h and σ2n denotes the noise variance.
Now in this method we are using a comb type pilot arrangement. So the X matrix
that is used is a diagonal matrix with the known N p pilot symbols.
FT is a truncated DFT matrix that we are using to reduce the complexity of the
system by bringing down the number of complex multiplications. It has the first Ct
columns of the original DFT matrix F and rows corresponding to the pilot locations.
27
4.5 Result
Bit error rate (BER) associated with the LS and MMSE channel estimation methods
at various SNR
Inference
Figure 4.5 shows that MMSE out performs LS for the entire SNR range and also the
performance of both estimators improves as SNR increases.The performance of the
MMSE estimator is better than that of the LS estimator due to the known covariance
matrix of the channel.
28
Chapter 5
5.1 Introduction
MMSE and LS estimators described in previous sections assumes channel to be
wide-sense stationary random process. But in reality, this assumption is incorrect
because the channel can be a non-stationary process. Kalman filter is an adaptive
filter which recursively calculates the channel estimates and hence ttrack the exact
value of the channel matrix as well.
In this chapter, the Kalman filtering problem is defined first. Then this problem
is extended to suit the channel estimation problem in OFDM systems. It ends
with simulation results which shows the improvement of Kalman filtering over least
square estimation.
Discrete time linear systems are usually described in a state variable form given by:
x j = ax j + bu j
29
where the state x j is a scalar, a and b are constants and the input u j is a scalar; j
represents the time index.
Now if the system gets corrupted by noise, the system definition changes to:
x j = ax j + bu j + w j
where, w is white noise with zero mean and covariance Q and is uncorrelated with
the input. Now if the signal x is measured, and the measured value is z,
z j = hx j + v j
The measured value z depends on the current value of x and on the gain h. Ad-
ditionally, the measurement has its own noise v associated with it. The noise v is
white noise source with zero mean and covariance R that is uncorrelated with the
input or with the noise w. The two noise sources are independent of each other and
independent of the input. The Kalman filter problem is to filter z to get an estimate
of x which minimizes the effect of both w and v.
A simple solution to the problem is to recreate the system model to get the estimate
ˆ j ). But this approach has two inherent problems. The first is
of x j (given by (x)
that there is no correction. If we dont know the quantities a, b or h exactly (or the
initial value x0 ), the estimate will not track the exact value of x. Secondly, we dont
compensate for the addition of the noise sources (w and v). In this approach we will
first make a prediction regarding x at time j based on the information available till
j-1. This is called the a priori estimate and is given by:
We use this a priori estimate to predict an estimate for the output, ẑ j . The differ-
ence between this estimated output and the actual output is called the residual, or
innovation.
Residual = z j − ẑ j = z j − hx̂−j
30
If the residual is small, it generally means we have a good estimate; if it is large the
estimate is not so good. We can use this information to refine our estimate of x j ;
we call this new estimate the a posteriori estimate, x̂ j . If the residual is small, so
is the correction to the estimate. As the residual grows, so does the correction. The
entire system can be modeled as:
The only task now is to find the quantity k that is used to refine our estimate, and it
is this process that is at the heart of Kalman filtering.
There are two errors which are of interest in the above problem definition. An a
priori error ê−j , and an a posteriori error e j .
ê−j = x j − x̂−j
e j = x j − x̂ j
p−j = E(e−j )2
p j = E(e j )2
where the operator E represents the expected or average value. A Kalman filter
minimizes the a posteriori variance p j by suitably choosing the value of k.
The value for k is calculated to be,
hp−j
k=
h2 p−j + R
However, there is still a problem because this expression needs a value for the a
priori covariance which in turn requires a knowledge of the system variable x j .
Therefore our next task will be to come up with an estimate for the a priori covari-
31
ance.
But before that lets examine this equation for k. First note that the constant, k,
changes at every iteration.
Next, and more significantly, we can examine what happens as each of the three
terms in equation are varied.
• If the a priori error is very large (so that the measurement noise term R in the
denominator is unimportant) then k = 1/h. This, in effect, tells us to throw
away the a priori estimate and use the current (measured) value of the output
to estimate the state.
Now, the a priori and a posteriori covariance terms are given by:
p−j = a2 p j−1 + Q
p j = p−j (1 − hk)
Any Kalman filter operation begins with a system description consisting of gains a,
b and h. The state is x, the input to the system is u, and the output is z. The time
index is given by j.
x j = ax j + bu j + w j
z j = hx j + v j
The process has two steps, a predictor step and a corrector step.
In the predictor step, an a priori estimate of the state(x̂−j ) and the apriori covariance(p−j
32
) is calculated.
In the corrector step the Kalman gain(k) is calculated and used to re- fine the a priori
estimate to give us the a posteriori estimate of the state(x̂ j ).
Then we can calculate the a posteriori covariance which will be used in the next
iteration.
p
hn,τ = ∑ ak hn−k,τ + wn,τ
k=1
In the next section we will discuss how to extend this concept to a MIMO system
with an m tap channel.
For the general system the state equation for the Kalman filter is given by:
Hn = AHn−1 +Wn
33
where,
A1 0 ··· 0 b hn,0 bwn,0
··· 0
1 A1 b h bwn,1
Hn = n,1
A=
.. .. .. .
B =
.. ..
Wn =
..
. ..
. . . . .
0 0 · · · A1 b hn,m bwn,m
Yn = Xn Hn +Vn = Xn F 0 hn +Vn
Yn = Xn F 0CHn +Vn
34
Where Yn corresponds to the received signals at the pilot positions. So, Yn ∈ CNp ×1 .
Yn is constructed by first creating x j . x j is a diagonal matrix with its diagonal filled
with the symbols transmitted at the pilot positions.
C is constructed by diagonally extending c, m times. Here c = [10 . . . 0]1×p . So
C ∈ Rm×pm .
Now the Kalman filtering equations are given by:
Ĥn|n−1 = AĤn−1|n−1
Pn|n−1 = APn−1|n−1 AH + Q
Pn|n = (I − Kn Xn F 0C)Pn|n−1
The Kalman filter requires an initial estimate of the state and an initial error
covariance matrix to start the recursive process. The initial state estimate is given
by an LS estimation to fill the matrix Ĥn−1|n−1 in the above equations. Pn−1|n−1 is
initialized as an identity matrix.
35
5.4 Results
Simulation Parameters
The number of subcarriers used are 128. Length of the Cyclic prefix is 16. The
channel length is 5 and the power delay profile (PDP) of the channel = e(−n/5) for
n= 0,1,...,4.The number of pilot symbols used are 32 and the number of symbols in
each OFDM alphabet are 100.
• BER vs SNR plot for different modulation schemes in a Rayleigh fading chan-
nel and equalization using a Kalman filter. Figure 5.1 below shows the BER at
various SNR for BPSK, QPSK, QAM 16 and QAM 64 modulation schemes.
As we know from theory BPSK has the least BER followed by QPSK, QAM
16 and QAM 64 respectively which is also shown by the simulated results.
Figure 5.1 also shows that BER performance is better as SNR increases.
36
• BER vs SNR plot for BPSK and QAM 64 (theoretical and simulated results)
in an AWGN (additive white Gaussian noise) channel
Figure 5.2 shows the theoretical and simulated results for both BPSK and
QAM 64 in an AWGN channel which are almost equal and follow the water-
fall curve. BER decreases with SNR.
37
• BER vs SNR plot for BPSK symbols in an OFDM system in a fading and
AWGN channel
Figure 5.3: BER vs SNR for BPSK in Rayleigh Fading and AWGN channel
Figure 5.3 shows that the BER performance is better in a AWGN channel than
that of a fading channel
38
• Mean Squared Error (MSE) Performance of LS, MMSE and Kalman estima-
tors
Figure 5.4: MSE vs SNR of LS, MMSE and Kalman estimation techniques
Mean square error associated with the LS, MMSE and Kalman channel esti-
mation methods at various SNR is plotted for BPSK modulated symbols
We see in figure 5.4 that Kalman out performs MMSE and LS for the entire
SNR range. Also as the SNR increases, the estimates by all three estimators
become more accurate.
39
Chapter 6
6.1.1 Introduction
40
already contain sufficient information to perform synchronization. Our novel al-
gorithm exploits the cyclic prefix preceding the OFDM symbols, thus reducing the
need for pilots.
Now, consider the transmitted signal s(k) This is the DFT of the data symbols
xk , which we assume are independent. Hence, s(k) is a linear combination of in-
dependent, identically distributed random variables. If the number of subcarriers is
sufficiently large, we know from the central limit theorem that s(k) approximates
a complex Gaussian process whose real and imaginary parts are independent. This
41
Figure 6.1: Structure of OFDM symbol with cyclicly extended symbol s(k).
process, however, is not white since the appearance of a cyclic prefix yields a corre-
lation between some pairs of samples that are spaced N samples apart. Hence, r(k)
is not a white process either, but because of its probabilistic structure, it contains
information about the time offset θ and carrier frequency offset ε.This is the cru-
cial observation that offers the opportunity for joint estimation of these parameters
based on r(k).
6.1.3 ML estimation
42
tions,the log-likelihood function can be written as
where f (.) denotes the probability density function of the variables in its argu-
ment.The product Πk f (r(k)) in equation (6.2) is independent of θ (since the product
is over all ) and ε (since the density f (r(k)) is rationally invariant).Since the ML
estimation of θ and ε is the argument maximizing Λ(θ, ε) Under the assumption that
r is a jointly Gaussian vector,(6.2) is shown in the Appendix to be
m+L+1
γ(m) , ∑ r(k)r? (k + N) (6.4)
k=m
m+L+1
1
φ(m) , ∑ r(k)2 + r(k + N)2 (6.5)
2 k=m
and
? (k + N)}
E{r(k)r
ρ , p
E{|r(k)|2 }E{|r(k + N)|2 }
σ2s SNR
= 2 2
= (6.6)
σs + σn SNR+1
is the magnitude of the correlation coefficient between r(k) and r(k + N) The first
term in (6.3) is the weighted magnitude of γ(θ) which is a sum L of consecutive
correlations between pairs of samples spaced N samples apart.The weighting factor
depends on the frequency offset.The term φ(θ) is an energy term, independent of
the frequency offset ε. The maximization of the log-likelihood function can be
43
performed in two steps :
max Λ(θ, ε) = max max Λ(θ, ε) = max Λ(θ, ε̂ML (θ)), (6.7)
(θ, ε) θ ε θ
The maximum with respect to the frequency offset ε is obtained when the cosine
term in (6.3) equals one. This yields the ML estimation of ε
1
ε̂ ML (θ) = − ∠γ(θ) + n (6.8)
2π
where n is an integer.By the periodicity of the cosine function, several maxima are
found.We assume that an acquisition, or rough estimate, of the frequency offset has
been performed and that |ε| < 1/2 ; thus n =0. Since cos(2πε̂ ML θ + ∠γ(θ)) = 1,
the log-likelihood function of θ becomes
Only two quantities affect the log-likelihood function (and thus the performance
of the estimator): the number of samples in the cyclic prefix L and the correlation
coefficient ρ given by the SNR.The quantity γ(θ) provides the estimates of ε and
θ.Its magnitude, which is compensated by an energy term, peaks at time instant
θ̂ ML ,while its phase at this time instant is proportional to ε̂ ML .
44
Simulation
Figure 6.2: Shows the signals that generate the ML estimates (N = 400, L = 100, ε =
.4 and SNR=15 dB).
45
Conclusion:
6.2.1 Introduction
Signal-to-noise ratio (SNR) is broadly defined as the ratio of the desired signal
power to the noise power. SNR estimation indicates the reliability of the link be-
tween the transmitter and receiver. In adaptive system design, SNR estimation is
commonly used for measuring the quality of the channel.In many SNR estimation
techniques, the noise is assumed to be white and Gaussian distributed. However, in
wireless communication systems, noise is often caused by a strong interferer, which
is colored in nature. Color of the noise is defined as the variation in power spectral
density in the frequency domain.Of particular interest is OFDM based multi-carrier
modulation systems, where the channel bandwidth is wide and the interference is
not constant over the whole band. It is very likely that there is variation of spectral
content over the OFDM sub-carriers i.e. some part of the spectrum is affected more
by the interferer than the other parts. Figure 6.4 shows OFDM frequency spectrum
and two types of noise over this spectrum, colored and white.
46
Figure 6.4: Representation of OFDM frequency channel response and noise spec-
trum. Spectrum for both white and colored noise is shown.
Most commonly used approach for noise power estimation in OFDM systems
is based on finding the difference between the noisy received sample in frequency
domain and the best hypothesis of the noiseless received sample. Calculation of
the received sample hypothesis requires channel state information for each car-
rier.Assuming that the noise is white and Gaussian distributed. In order to get the
long term estimates, the instantaneous SNR estimates are averaged over the whole
OFDM band by taking the mean of all the estimates over all the sub-carriers.
An OFDM based system model is used. Time domain samples of an OFDM symbol
can be obtained from frequency domain symbols as
where Sm (k) is the symbol that is transmitted on k-th subcarrier of the m-th OFDM
symbol, and N is the number of sub-carriers. After the addition of cyclic prefix
and D/A conversion, the signal is passed through the mobile radio channel.In this
Simulation, we assume the channel to be constant over an OFDM symbol.At the
receiver, the signal is received along with noise. The noise power is assumed to be
varying across OFDM sub-carriers as well as in time. After synchronization, down
sampling, and removing the cyclic prefix, the simplified received baseband model
47
of the samples can be formulated as
L−1
ym (n) = ∑ xm(n − l)hm(l) + zm(n) (6.13)
l=0
where L is the number of channel taps, zm (n) is the noise sample which is white
Gaussian noise.After taking DFT of the OFDM symbols, the received samples in
frequency domain can be shown as
where Hm (k) and Zm (k) are DFT of hm (l) and zm (n) respectively.
In this method we use the virtual carriers i.e its does not contain any data. Virtual
carriers are used for improvement of the estimation of Noise power.We have used
the pilot symbol for best estimate of the channel.With this estimate of the chan-
nel,noise power is calculated. The channel estimates in frequency domain can be
obtained using OFDM training symbols, or by transmitting regularly spaced pilot
symbols in between the data symbols and by employing frequency domain interpo-
lation. We assume transmission of training OFDM symbols. Using the knowledge
of the training symbols, channel frequency response can be estimated as
Ym (k)
Hls (K) = (6.16)
Xm (k)
N(k) = Ym (k) − Hls (k)Xm (k) (6.17)
48
where k belongs to the pilot locations of OFDM symbol, Y (k) is the received data at
pilots and X(k) transmitted data at pilot locations. So N(k) is the noise estimated at
the pilot locations.After collecting the noise samples in the pilot locations ,extrapo-
lated over the whole symbol.After that collecting the data from the virtual carriers
and combining it with the N(k). In conventional noise power estimation algorithms
the absolute square of the instantaneous noise samples are averaged over all OFDM
symbols, providing an averaged noise power estimate.
Simulation
Figure 6.5: Shows the Mean Squared error of Noise power estimation
49
Figure 6.6: Shows the SNR estimated for each symbol.
Figure 6.7: Shows the BER performance of the BPSK when Noise Power is fed in
Kalman filter.
50
Conclusion:
In Fig 6.5 shows the mean square error in estimation of noise power which is very
low. So the estimation is significantly better.in Fig 6.6 shows the SNR estimation
for each OFDM symbol which estimated using the noise power estimation results.In
the Fig 6.7 shows the estimated noise is given to the kalman filter.Theoretical and
estimated variances is given and checked the BER performance which is close to
each other hence above algorithm is good for estimating the white noise.
6.3.1 Introduction
51
We will further assume that the channel taps are uncorrelated with respect to each
other and can be modeled as a wide-sense stationary process where the average
energy of the total channel energy is normalized to one, and the delays τi are taken
to be constant for the time of interest. Assuming the receiver filter is flat within the
transmitter bandwidth, the receiver input signal is
After removing the guard interval for further receiver processing, the lth received
OFDM symbol is represented by N samples.Demodulation of the subcarriers via a
fast Fourier transform (FFT) yields the received data symbols.For the moment, we
assume the channel to be constant during the transmission of one OFDM symbol
denoted by hi (l). In a real-world passband transmission system, the following pa-
rameters cause disturbances in the receiver.The sampling time T 0 at the receiver can
no longer be assumed to be identical with the transmitter time T .The same holds for
the carrier frequency oscillators used for modulating and demodulating the signal.
Assuming a small frequency offset relative to the transmission bandwidth, the fre-
quency difference between transmitter and receiver oscillators can be modeled as a
time-variant phase offset Θ(t) at the receiver.
All of the effects mentioned are incorporated in the equivalent system model
depicted in Fig.1.8
52
Figure 6.9: Structure of an OFDM receiver.
Fig.6.9 shows the basic structure of an OFDM receiver. Depending on the avail-
ability of training data, the parameters are estimated before or after demodulation
via the FFT. Following downconversion, the received signal should be sampled
same rate as that of transmitter.Timing offset is( determined with ML estimation)
corrected after that frequency correction performed.After guard removal, blocks of
N samples pertaining to one OFDM symbol are processed in the FFT.The demod-
ulated subcarrier samples and estimated channel samples (coherent detection) are
then passed to an appropriate outer receiver for further processing.
Simulation
We have taken the N=500 OFDM sybols o which are send over 128 subcarrier over
a fading channel.After performing the all the corrections and estimation of chan-
nel.Bit Error Ratr is calculate for BPSK data.
53
Figure 6.10: BER plot of BPSK symbol over fading channel.
Conclusion
We have performed all the corrections in the receiver side and achieved the Bit Error
Rate and MSE for different SNR which is decreasing as the SNR decreases.As in
54
Fig 6.10 and 6.11 shows that maximum amount of performance loss in Estimated
parameters with respect to Actual parameters is .078 dB at 20 dB SNR.
55
REFERENCES
1. Yong Soo Cho, Jaekwon Kim, Won Young Yang, and Chung-Gu Kang, MIMO-
OFDM Wireless Communications with MATLAB, in John Wiley & Sons
(Asia) Pte Ltd, 2010, pp. 129-193
6. Greg Welch, and Gary Bishop, An Introduction to the Kalman Filter, in SIG-
GRAPH, 2008
7. Jan-Jaap van de Beek, Ove Edfors, Magnus Sandell, Sarah Kate Wilson and
Per Ola Borjesson, On Channel Estimation in OFDM Systems, in Vehicular
Technology Conference , IEEE 45th vol. 2 , 1995, pp. 815-819
8. H. Arsalan and S. Reddy, Noise Power and SNR Estimation for OFDM-based
Wireless Communication Systems, in IASTED International Conference on
Wireless and Optical Communications , Jul 2003, pp. 389394
56
9. M. Speth , S. Fechtel , G. Fock and H. Meyr ,Optimum receiver design for
broadband systems using OFDM-part I , IEEE Trans. Commun., vol. 47,
pp.1668 -1677 1999
10. Jan-Jaap van de Beek, ML estimation of time and frequency offset in OFDM
systems, IEEE Trans. Sig. Proc. Letters, vol. 45, pp.1800-1806, July 1997.
12. Yun Wu, and Hanwen Luo, Channel Estimation for MIMO OFDM Systems
in Non sample spaced Multipath Channels in Congress on Image and Signal
Processing, 2008, pp. 88-92
13. Meher Jain, Rohit Kurien,Channel Estimation in STBC OFDM Systems, re-
port submitted to IIST
57
APPENDIX
θ+L−1
f (r(k),r(k+N))
Λ(θ, ε) = ∑ log f (r(k)) f (r(k+N)) , (6.20)
k=θ
where ρ is the magnitude of the correlation coefficient between r(k) and r(k + N) as
defined in Eq(6.6).The denominator of (6.20) consists of two 1-D complex Gaussian
distribution
|r(k)|2
exp − σ2 +σ2
s n
f (r(k)) = (6.22)
π(σ2s + σ2n )
and the log-likelihood function (6.19), after some algebraic manipulations, becomes
where γ(m) and φ(m) are defined in Eq (6.4) and (6.5) and c1 and c2 are con-
stants,independent of θ and ε.Since the maximizing argument of Λ(θ, ε) is inde-
pendent of the constants c1 ans c2 .
58