Implementation of Pilot Symbol Assisted Rateless

Codes to Mitigate Fading in Rayleigh

Fading Channels
Prema.G
1
, Amrutha.E
2
1
Professor,
2
PG Student
1, 2 Department of ECE
Mepco Schlenk Engineering College,
Sivakasi, India
[email protected]; [email protected]
Abstract The principal design problem encountered in wireless
communications is how to overcome the distortion of received
signals by a time-varying and frequency-selective propagation
path due to the phenomenon called fading. In this paper, a
channel coding method referred as pilot symbol assisted rateless
coding (PSAR) is considered to mitigate the effects of fading;
where a predetermined fraction of binary pilot symbols is
interspersed with the channel-coded bits at the channel coding
stage instead of multiplexing the pilots with the data symbols at
the modulation stage, as in classic pilot symbol assisted
modulation (PSAM). PSAR codes are a specific family of rateless
codes that are capable of modifying their code-rate as well as
their degree distribution based on the channel state information
(CSI), in an attempt to adapt to the time-varying nature of the
channel. The performance of pilot symbol assisted code aided
system is measured in Raleigh fading channels with the help of
two parameters, bit error rate (BER) and latency. Compared
with theoretical BER results, PSAM and a fixed rate (7,4) cyclic
code, the introduced PSAR codes is beneficial in terms of
significantly reducing the bit error rate. The results also suggest
that a significant reduction in the BER can be attained by PSAR
coding system, when compared to the PSAM scheme having the
same pilot overhead.

Index Terms Rateless codes, pilot symbol assisted rateless codes,
Rayleigh fading channel, pilot symbol assisted modulation, bit
error rate reduction.
I. INTRODUCTION
The concept of rateless codes (digital fountain codes) was
first introduced by Byers et al [1] in 1998 for information
distribution based on rateless error correcting codes for
erasure channels. Approximations to a digital fountain has
existed for some time, such as the Low-Density-Priority Code
(LDPC) code [2] and the Reed-Solomon (RS) code, which are
generated by loosening the requirements in various ways. LT
codes [3] are the first realization of a true digital fountain
code. The term rateless describes a channel code that does not
have principles and thus portraying their philosophical
differences, rateless codes do not fix their code-rate before
transmission. Rateless codes adapt to the channel by sending
additional code symbols if the receiver cannot decode the
transmitted symbols from that it already has. The symbol
length for the code can be arbitrary, from one-bit binary
symbols to general l-bit symbols. The number of encoding
symbols that can be generated from the data is potentially
limitless. If an original file can be described by k symbols,
then each coded packet can be independently generated, and
the content of the original file can be recovered from any
k + 0( kln
2
( k/ o) ) , packets with the probability of 1o, by
an average of 0( kln( k/ o) ) , symbol operations. Following
[4], Raptor codes were developed by Shokrollahi as rateless
codes with even smaller encoding and decoding complexities
than LT codes. Raptor codes make use of pre-coder before the
typical LT process.
In the available literature [4], [5], [6], and [7], rateless
codes are frequently employed in situations, where the
channel statistics are unknown to the transmitter and hence the
degree distribution of rateless codes is fixed; i.e. the degree
distribution used for coining the specific random degree for
each transmitted bit is time-invariant and thus channel
independent. Such rateless codes can only control the total the
number of bits transmitted, i.e. the code-rate, in order to cater
for the variations of the channel conditions encountered.
Addressing the shortcoming of the above discussed code, a
new rateless code called as pilot symbol assisted rateless
codes (PSAR) is introduced, where predetermined fraction of
binary pilot symbols is interspersed with the channel-coded
bits at the channel coding stage instead of multiplexing the
pilots with the data symbols at the modulation stage, as in
classic pilot symbol assisted modulation (PSAM).
The rest of the paper is organized as follows. Section II
details the PSA rateless code aided system model. A graph
based analysis of the pilot symbol assisted codes is given in
Section III. The Rayleigh channel model for modeling the
wireless channel is described in Section IV. Section V
presents the simulated results and discussions, followed by
conclusions in Section VI.
II. PILOT SYMBOL ASSISTED RATELESS CODE-AIDED
SYSTEM MODEL
The schematic of the architecture of pilot symbol assisted
rateless code-aided system is shown in Fig.1. For every
information bit sequence to be encoded at a specific
transmission instant t, the channel state information (CSI)
received via the feedback channel is exploited by what is
referred to as the degree distribution selector (DDS
T
).
Fig. 1 Pilot Symbol Assisted Code-Aided Rateless System Model

The DDS
T
shown in Fig. 1 calculates the required coding
rate R
t
as well as the corresponding degree distributions. In the
design process as in [8], the bits in the input message
sequence and the corresponding code sequence is referred as
variable nodes and check nodes respectively. The number of
bits that are going to be selected to form a particular codeword
at a particular time instant is referred as degree and denoted
by d
c
. The pattern in which the input bits are selected to be
combined is given by the variable node distribution. The
number of times each input is being selected is denoted by
degree d
v
. The variable node distribution v
t
(x) is regular and is
defined by;
:
t
( x) = x
d

-1
(1)

The check node distribution
t
(x) is irregular and is
represented by means of a polynomial distribution defined by;

o
t
( x) = ` o
d
c
x
d
c
-1

d
c
d
t

= o
1
+ o
2
x + + o
d
c
x
d
c
-1
+ + o

c
x

c
-1
(2)
where the positive coefficients o
d
c
,d
c
d
t
denote the particular
fraction of intermediate bits (or check nodes) of degree d
c
and
D
c
= max(d
t
). The vector d
t
contains the range of (check)
degree values of the degree distribution. There are two
different categories of degree one bits in (2), as a result the
fraction o
1
can be rewritten as
o
1
= o
1
p
+ o
1
p
(3)
where o
1
p
and o
1
p
denote the fraction of degree-one nodes
corresponding to pilot bits and to information bits,
respectively.
A. Rateless Encoder
The rateless encoder of Fig. 1 maps a K - bit (input)
information sequence represented by a = [a
1
, a
2
,...,a
K
] into a
KR
t
- bit output sequence c = [c
1
, c
2
,...,c
K
|
R
t] by performing
the steps succinctly described below:
1) (Modified input sequence)Attach a predetermined pilot
bit sequence p = (p
1
, p
2
,...,p
Kp
), to the beginning of the
K - bit input stream a, so that the modified K - bit
input sequence becomes equal to a = [p a]. The
length of a is K= K+K
p
where K
p
is the length of the
pilot sequence. The pilot sequence length is calculated
using three parameters; input sequence length,
instantaneous code rate R
t
and pilot bit overhead o
1
p
. It
is denoted by K
p
and is given by

K
p
=
(K6
1

)
( R
t
-6
1

)
. (4)

2) (Degree selection) Randomly choose a degree d
c
from
the degree distribution o
t
(x) o
1
p
, calculated by the
degree distribution selector based upon the received
CSI.
3) (Input bit/s selection) Randomly select the d
c
number
of bits from a having the least number of connections
(selections) up to the current transmission instant. The
selection of depends on the check node distribution.
From the d
c
values given by o
t
(x), any one value is
chosen each time to form an encoded symbol.
4) (Intermediate bit calculation) The value of each of the
intermediate bit b
i
b is calculated by means of
combining the d
c
input bits selected during the previous
step using modulo-2 addition. This is done by
organizing the input sequence into a specific data
structure named Tanner graph.
5) (Modified intermediate bit selection) After calculating
the value of each of the bits in b, the same pilot bit
sequence p is again attached to the beginning of the
intermediate bit sequence b generated in the previous
step in order to create b = [p b]. The length of this
sequence will be KR
t
.
6) (Codeword bit calculation) Determine the value of the
encoded bit c
i
c, i = 1,,KR
t
, by calculating the
values of c
1
= b
1
, c
i
= b
i
c
i
1, for i=2,,KR
t
,
where, bi b and the pilot bits in c correspond to the
bits ci c with i = 1, . . . , Kp.
B. Pilot Position Interleaver (
p
)
As shown in Fig. 1, the codeword c is then interleaved by
the pilot position interleaver p, which will position a pair of
pilots every (n 1) data bits, where n denotes the pilot
spacing. This process is similar to that described in [8], which
represents the effective sampling of the channels complex
valued envelope at a rate that is higher than the Nyquist rate
and thus allowing the receiver to extract the channel
attenuation as well as phase rotation estimates for each bit.
The data bits are separated by means of a pair of pilot bits
(instead of a single pilot) since the channels between transmit
and receive antennas have to be estimated. For clarity, the
pictorial representation of the rateless encoder is given in
Fig.2.
C. Modulator and Demodulator
The interleaved codeword p (c) is then modulated using
binary phase shift keying (BPSK). In BPSK modulation,
individual data bits are used to control the phase of the carrier.
During each bit interval, the modulator shifts the carrier to one
of two possible phases, which are 180 degrees or radians
apart. The modulated codeword is transmitted through the
Rayleigh channel model which will be described in the
Section II. In BPSK demodulation, the demodulator multiplies
the modulated signal by the recovered carrier, and this product
is integrated over a bit interval. The probability of bit error for
BPSK is obtained as
P
b
= 0|2
L
b
N
0
1 =
1
2
crc |
L
b
N
0
1 (5)

where Q(.) is the Q function, erfc(.) is the complementary
error function, E
b
is the energy per bit and N
0
is the noise
spectral density.
D. Channel Estimator
The channel is estimated with the help of known pilot
symbols inserted into the transmitted data stream. It
potentially achieves an improved bit error ratio (BER)
performance, at the expense of an unavoidable reduction of
the effective throughput due to the associated pilot overhead.
E. Pilot Position Deinterleaver ( L
p
-1
)
The demodulated output from the demodulator is given as
input to the pilot position deinterleaver. The function of pilot
position deinterleaver is to extract the pilot symbols that are
interspersed among the code word. The pilot spacing is a
priori knowledge stored at both the transmitter and receiver.
The output of the deinterleaver is c which is given as input to
the rateless decoder.
F. Rateless Decoder
An ideal PSAR code should exhibit the following
properties:
Given a message that would require k packets of length l
for transmission, this code would allow the source to generate
an infinite supply of encoded packets of length l. The decoder
can then reconstruct the original message once any k encoded
packets of the infinite supply of encoded packets are received.
Moreover, the complexity of the reconstruction process should
be linear in k. Before starting with the decoding process, it is
also implicitly assumed that there is another subsidiary DDS
located at the receiver, namely DDS
R
(DDS receiver). It
calculates the check node and distributions o
t
(x) and u
t
(x)
respectively based on the estimated CSIR and then passes
these distributions to the rateless decoder. The decoding
process for rateless codes is based on belief propagation
algorithm [3]. A step by step procedure of the algorithm is as
follows.
1) Locate a self-contained symbol, i.e. a so-called degree-
one input symbol which is not combined with any
other.
2) The decoder will then add (modulo-2) the value of this
symbol to all the encoded symbols relying on it and
then removes the corresponding modulo-2 connections.
3) Repeat step 1 and 2 until the message is recovered.

The decoding process is simplified with the help of tanner
graph based analysis.
III. GRAPH BASED ANALYSIS OF THE PILOT SYMBOL ASSISTED
CODES
A tanner graph representation of the PSAR code is
provided in Fig.3, which shows an unbalanced tripartite graph
G consisting of the finite set of vertices V and the finite set of
edges E. The vertices set I can be further divided into three
disjoint sets representing the variable nodes, the check nodes
and the parity nodes. The variable (information) nodes would
then correspond to a; the check (intermediate) nodes are
represented by b whilst the parity nodes relate to the PSAR
encoded codeword bits c. Given the graph G, G (v) will then
denote the set of vertices adjacent to the vertex v V. The
degrees d
v
and d
c
d
t
correspond to the discrete values
assumed by the variable node distribution and the check node
distribution, respectively.
Fig. 2 Pictorial Representation of the Rateless Encoder
Input Sequence: a (K-bits)

Modified Input Sequence: a (K -bits)

Intermediate Sequence: b (K
c
K
p
-bits)

Modified Intermediate Sequence: b (K
c
-bits)

Codeword Sequence: c (K
c
-bits)

Pilot position Interleaving: (c)

Fig. 3 A tripartite graph representation of pilot symbol assisted rateless code
PSAR codes also possess pilot nodes and pilot edges.
Formally, the pilot variable nodes are pi a, where i = 1,.,K
p

and have known information. The pilot check nodes are p
i

b, where i = 1,...,K
p
, are the degree-one check nodes
connected by a single edge to the pilot variable nodes. The
output contains the pilot parity nodes are the received code
words c
i
c, where i = 1,...,K
p
.
It is important to note from the Fig. 3, that in order to
ensure the initialization of the iterative decoding convergence,
the pilot edges sprouting from the K
p
pilot variable nodes are
not only associated with the pilot check nodes, but are also
involved in other parity-check equations containing higher-
degree check nodes. The messages passed over the pilot edges
are perfectly known, since they originate from nodes having
predetermined values. The three different pilot nodes as well
as the pilot edges are explicitly marked in Fig. 3. The
encoding and decoding operations of the pilot symbol assisted
rateless codes is explained in the sections above. Next the
channel model is discussed.
IV. RAYLEIGH CHANNEL MODEL
The modulated signal is subjected to Rayleigh fading and
Additive White Gaussian Noise (AWGN). Since Rayleigh
fading is a multiplicative process, the transmitted signal is
multiplied with the Rayleigh fading signal. A single-user
transmission system employing two single input single output
(SISO) antennas, each at transmission side and reception side
is considered. The time-complex baseband-equivalent SISO
channel model used is given by

y(t) = H(t)x(t) + n(t) (6)

where x(t), y(t) and n(t) are vectors corresponding to the
transmitted signal, received signal and noise. The time variant
SISO channel matrix H(t) contains elements corresponding to
the channel gains of a Rayleigh-fading process [9]. In this
model, the variability of the wireless channel over time is
reflected in its autocorrelation function (ACF). This second-
order statistic generally depends on the propagation geometry,
the velocity of the mobile, and the antenna characteristics. A
common assumption is that the propagation path consists of a
two dimensional isotropic scattering with a vertical monopole
antenna at the receiver [10]. In this case, the theoretical power
spectral density (PSD) associated with either the in-phase or
quadrature portion of the received fading signal has the well-
known U-shaped bandlimited form and is given by S(f).

S( ) =
1
n]
d
1- |
]
]
d
1
2
, | |
d
0, clscwcrc
(7)

where f
d
is the maximum Doppler frequency in Hertz, given
by f
d
= v/ , v is the mobile speed and is the wavelength of
the received carrier wave. The corresponding normalized
(unit variance) continuous-time autocorrelation of the received
signal under these conditions is given by R[n] =J
0
(2f
d
),
where - represents the correlation lag, and [
0
() represents
the zero-order Bessel function of the first kind. The complex
additive white Gaussian noise (AWGN) is represented by the
vector n(t) cJ(0,N
0
), where N
0
denotes the two
dimensional noise variance.
V. RESULTS AND DISCUSSIONS
In this section, the performance of pilot symbol assisted
rateless codes is evaluated. The PSAR codes utilize the
specifications of Universal Mobile Telecommunication
System (UMTS) standard. The parameters are listed in the
Table I. The performance is measured in terms of average bit
error rate and latency.
A. Bit Error Rate (BER)
Bit error rate is the ratio of number of received bits that
have been altered while passing through the channel to the
total number of transferred bits during a studied time interval.
BER has been calculated by taking average of 10 individual
runs.
B. Latency
Latency is a measure of time. It is defined as the total time
measured from the start of the encoding until the end of
decoding. Although rateless codes are well suited for channels
with varying error rate, their use often causes data overhead
and delays.
The performance of the PSA codes is compared with a
fixed rate (7,4) cyclic code and classic pilot symbol assisted
modulation (PSAM) [10].
TABLE I
SIMULATION PARAMETERS FROM UMTS STANDARD

Parameter Value
Operating band I
Uplink frequency 1920 1980 MHz
Downlink frequency 2110 2170 MHz
User data rate 15 Kbps
Maximum Doppler frequency 30 Hz
The BER is calculated over a range of signal to noise ratio
(SNR). The number of information bits for the rateless code,
K, was set to 256, 1024, 10,000 and 12,000 bits and fraction
of pilot bits is set to 0.05. The rate of the cyclic code is fixed
to 4/7, where the message is divided into equal blocks of
length 4 and then coded into a block of length 7. For PSAR
coding system the coderates chosen as per the SNR values is
given in Table II. The variation of pilot sequence length with
rate and message size for the PSAR coding method is given in
Table III. In Table IV the variation of code word lengths for
different rate and message sizes in PSAR coding method is
given also the code word lengths for cyclic codes is given.
TABLE II
SNR AND CODERATE SELECTION
Signal to noise ratio
(dB)
Coderate
-5 to 5 0.6667
6 to 10 0.75
11 to 15 0.8333
16 to 20 0.8571
20 to 30 0.8750

TABLE III
VARIATION OF PILOT SEQUENCE LENGTH WITH RATE AND MESSAGE SIZE
FOR PSAR CODING METHOD

Message
size
K bits
Pilot sequence length K
p
bits
Rate : (n,k)
(3,2) (4,3) (6,5) (7,6) (8,7)
256 20 18 16 15 15
1024 83 73 65 63 62
10000 810 714 638 619 606
12000 972 857 765 743 727

TABLE IV
VARIATION OF CODEWORD LENGTH FOR PSAR CODES AND FIXED RATE
CYCLIC CODES

Message
size
K bits
Rate : (n,k)
PSAR Code length K
c
bits
Cyclic
code
(3,2) (4,3) (6,5) (7,6) (8,7) (7,4)
256 340 338 328 327 327 896
1024 1355 1329 1313 1311 1310 28672
10000 13170 12962 12798 12755 12734 70000
12000 15804 15553 15357 15311 15279 84000
For the message lengths in Table III the bit error rates
achieved over a range of signal to ratio are shown in fig. 4,
fig. 5, fig. 6, and fig. 7. The comparison of latency of PSAR
codes with cyclic code and PSAM is given in Fig.8.
Fig. 4 Average BER comparison for message size K = 256 bits
Fig. 5 Average BER comparison for message size K = 1024 bits

Fig. 6 Average BER comparison for message size K = 10000 bits
Fig. 7 Average BER comparison for message size K = 12000 bits
Fig. 8 Performance comparison in terms of latency
Thus from the BER curves given in Fig 4 7, it can be
noted that, for any given message length the presented pilot
symbol assisted codes gives less BER than the fixed rate code
and pilot symbol assisted modulation technique. It is also
observed that the curve for pilot symbol assisted codes is
much lower than the fixed rate cyclic code. For pilot symbol
assisted modulation, the pilot overhead is considered as 10%
of the message size. The difference between PSAR codes and
PSAM coding technique is that the pilot symbols are
embedded into the code word at the channel coding stage
itself in PSAR. Whereas in PSAM, the pilot symbols are
appended to the beginning of the code word only at the
modulation stage.
Overall there is a 30% decrease in BER of the proposed
PSAR coding method when compared with the other two
methods. From the Fig. 8, it is observed that in terms of
latency the PSAM method is performing better than that of
PSAR and fixed rate code. But the difference between PSAR
and PSAM is very little. This small delay in computations for
PSAR can be compensated by its capacity of channel
estimation and BER performance.


VI. CONCLUSION
In this paper, a framework for pilot symbol assisted rateless
codes is proposed in which the channel coding components
exploit the knowledge of CSI to mitigate fading. From the
simulation results, it can therefore be concluded that, it is
more beneficial to appropriately intersperse pilots at the
channel code stage as in the presented PSAR codes rather than
at the modulation stage as in classic PSAM, since the pilot bits
are not only useful for channel estimation but also for
enhancing the decoders performance. From another point-of-
view, it can also be regarded that the pilot symbol assisted
rateless codes are specifically designed for a channel that is
affected by Rayleigh fading. From the results it is observed
that as the message size increases, the latency also increases
due to the increase in computations for encoding and
decoding. But this can be compensated with reduce in bit error
rate. Thus, the PSAR codes are demonstrated to be a suitable
technique to be used for a Rayleigh fading channel to
compensate the fading effects through effective coding
scheme.
The pilot symbol assisted rateless code-aided system is a
single user single-input single-output (SISO) transmission
scheme. As a future work, the performance of the system over
Rayleigh fading channel can be improved when a multi-input
multi-output transmission (MIMO) scheme is used. Also the
bandwidth of the UMTS standard can be further utilized with
a multi user system employing orthogonal frequency division
multiplexing (OFDM).
REFERENCES
[1] J.w. Byers, M. Luby, M. Mitzenmachert, and A. Rege, A Digital
Fountain Approach to Reliable Distribution of Bulk Data, ACM
SIGCOMM Computer Communication Review, Vol. 28, no. 4, pp.
56-67, 1998.
[2] M. Kwon, Y. H. Lee, M.-K. Oh, and H. D.-J. Park, Iterative channel
estimation and LDPC decoding with encoded pilots," IEEE Trans. Veh.
Technol., vol. 57, pp. 273-285, Jan. 2008.
[3] Michael Luby, LT Codes. Proc. IEEE Symposium on the
Foundations of Computer Science (FOCS): 271280, 2002.
[4] N.Bonello, S. Chen, L. Hanzo, and R. Zhang, Reconfigurable rateless
codes, in IEEE Transactions on Wireless Commun., Vol. 8, no. 11, pp.
5592-5600, 2009.
[5] Uri Erez, Mitchell D. Trott, and Gregory W. Wornell, Rateless Coding
and Perfect Rate-Compatible Codes for Gaussian Channels, ISIT, pp.
528 532, July, 2006.
[6] Michael Luby, M. Mitzenmacher, M. A. Shokrollahi, and D. A.
Spielman, Efficient erasure correcting codes, IEEE Trans. Inf. Theory,
vol. 47, pp. 569-584, Feb. 2001.
[7] R. Palanki and J. S. Yedidia, Rateless codes on noisy channels," Proc.
IEEE International Symp. Inf. Theory, Chicago, IL, USA, p. 37, June
2004.
[8] Nicholas Bonello, Sheng Chen, Du Yang, and, Lajos Hanzo, and
Generalized MIMO Transmit Preprocessing Using Pilot Symbol
Assisted Rateless Codes, IEEE Transactions on wireless comm., vol. 9,
no. 2, pp. 754 763, Feb 2010.
[9] Kareem E. Baddour, and Norman C. Beaulieu, Autoregressive
Modeling for Fading Channel Simulation, IEEE Transactions on
Wireless Communictions, vol. 4, no. 4, pp. 1650 1662, July 2005.
[10] James K. Cavers, An analysis of Pilot Symbol Assisted Modulation for
Rayleigh Fading Channels, IEEE Transactions on Veh. Tecchnol., vol.
40, no. 4, pp. 686 693, November, 1991.