Chapter 3

Multicarrier Interleave Division

Multiple Access
3.1 Introduction
MC-CDMA is robust to multipath propagation but its performance degrades rapidly as
the number of users increase due to multiple access interference. Various multiuser detec-
tion techniques have been proposed to combat MAI but their computational complexity
is prohibitive for practical implementations [74].
MC-IDMA is proposed to encounter the problems faced by IDMA and MC-CDMA
systems [75, 76]. Soon after [75, 76], several papers appeared on MC-IDMA [7780].
In [7579], uncoded MC-IDMA is considered and simple repetition code is used for
spreading each users signal, however as described in [55,56], IDMA and hence MC-IDMA
systems are spectrally efficient and outperform the CDMA and MC-CDMA systems only
if low rate turbo codes are used for bandwidth expansion. In [80], authors compared the
performance of a basic MC-IDMA system with other OFDM based multiple access tech-
niques such as MC-CDMA, OFDMA etc. While comparing MC-IDMA and MC-CDMA,
1
they used rate 2
convolutional codes in conjunction with repetition codes of length 16 in
MC-IDMA and in MC-CDMA, they used Walsh Hadamard codes of length 16 for number
of users up to 16. For number of users greater than 16, they used randomly generated
codes. As far as the number of users is less than or equal to 16, performance of both sys-
tems is same but as the number of users increase the performance of MC-CDMA degrades
rapidly due to non-orthogonality of random codes. In fact MC-CDMA doesnt support
users more than 16 due to its capacity limit. In [82], an MC-IDMA is considered with

37
low rate Concatenated Zigzag Hadamard (CZH) codes and its performance is compared
with MC-CDMA system which is considered as a reference. A simplified iterative detec-
tion algorithm is developed taking into account the complex nature of channel gains. An
MC-IDMA system like IDMA inherently uses BPSK or QPSK. In [84], implementations
of higher modulation schemes are suggested to enhance data rate. An MC-IDMA system
with adaptive subchannel allocation is presented in [83] which uses only those subcarriers
in the transmission which have largest fading amplitudes among all the subcarriers.
In this chapter, we consider an MC-IDMA system and combine the work presented
in [75,76,8284]. We analyze the iterative CBC detection scheme with multicarrier com-
munication in frequency selective multipath fading channels. It is shown that MC-IDMA
can combine all the energy scattered in frequency domain and reduces the complexity
J times as compared to IDMA where J is the number of multipaths. We exploit the
iterative CBC detection scheme with multicarrier communication in multipath fading
channels for uplink transmission. It is shown that MC-IDMA can combine all the energy
scattered in frequency domain and reduces the complexity approximately J times as com-
pared to IDMA, where J is the number of multipaths. We also compare the performance
of MC-IDMA and MC-CDMA with different coding schemes.

3.2 System Model

Consider an MC-IDMA system for uplink transmission shown in figure 3.1. There are
M active users, each transmitting QPSK symbols. The encoder (Enc) block consists of
an FEC and an optional spreader so that each symbol is represented by a sequence of
chips. A user specific interleaver interleaves the chip sequence of each user. Interleavers
disperse the coded sequence so that adjacent chips are uncorrelated. The interleaved data
is converted from serial to parallel form and is followed by Inverse Fast Fourier Transform
(IFFT) operation that modulates each chip on a different subcarrier. The modulated
data is then converted from parallel to serial form and cyclic prefixes are added to each

38
 a1 x1 s1
Enc C 1 IFFT CP h1

+
aM xM sM
Enc C M IFFT CP hM

1(x1)
^a u1 -11
1
Dec 2(x1) Gm
n
1
y(k) r(k)
SRE x FFT CP +
^a
1(xM)
uM -1M
M
Dec 2(xM)
M

Figure 3.1: Transmitter and iterative receiver structures for MC-IDMA system

symbol. The signal is then transmitted over a multipath multiple access channel. The
model of an MC-CDMA is similar to that of MC-IDMA and is shown in figure B.1 in
appendex B. The difference is that the role of spreader and interleaver for each user
is interchanged. The encoded signal of each user is spread by a user specific spreader
followed by an optional interleaver. Let us consider an MC-IDMA system. We can write
the m th users signal as [35]

Q1 L1 j
2(Ql+q)t
0
 
xm,i (Ql + q) p t iTs e ( )
0
X XX Ts
s (t) = (3.1)
t= q=0 l=0

where
0
Ts = QTs (3.2)

and QL=N=FFT size, {xm,i (1) , xm,i (2) , ..., xm,i (N )}, is the coded chip sequence of m th
user, i is the time index, is the length of cyclic prefix, L is the processing gain Ts = 1/
0
symbol rate, Ts is the symbol duration at subcarrier, Q is an integer greater than or equal
to one. When Q=1, one MC-IDMA symbol consists of more than one input symbols,
and in this case, frequency diversity is not achieved in an optimal way. In (3.1), p(t) is

39
a pulse satisfying the following relations

0
1 t Ts
p (t) = (3.3)
0 otherwise

The channel is assumed to be frequency selective Rayleigh fading with impulse response
given by
J
X
h (t, ) = gj (t) (t j ) (3.4)
j=1

where t and are time and delay respectively, J is the number of multipaths, j is the
delay in the j th path, gj , the j th path gain which is independent Gaussian random variable
and is the Dirac delta function. The received signal that contains contribution from all
the users is given by

M Z
X
r (t) = sm (t ) hm (t, ) + n (t) (3.5)
m=1
Q1 M
L1 X  j 2(Ql+q)t
0

(t) xm,i (Ql + q) p t iTs e (Ts ) + n (t)
0
X XX
m
= zq,l (3.6)
t= q=0 l=0 m=1

m
where zq,l (t) is the received complex envelope at the (Ql + q)th subcarrier of m th user,
n (t) is AWGN and denotes the convolution operation. The received signal is sampled,
cyclic prefixes are removed and Fast Fourier Transform (FFT) coherently demodulates
it. The demodulated signal is given by

M
X
r (k) = zm (k) xm (k) + n (k) (3.7)
m=1

where zm and xm represent the channel gains and signal from m th user after cod-
ing/spreading respectively and are complex valued and n (k) are the samples of AWGN.
Let m th user be our desired user, we can rewrite (3.7) as

40
 M
X
r (k) = zm (k) xm (k) + zm0 (k) xm0 (k) + n (k) (3.8)
0
m =1
0
m 6=m
= zm (k) xm (k) + (k) (3.9)

where
M
X
m (k) = zm0 (k) xm0 (k) + n (k) (3.10)
0
m =1
0
m 6=m

is the signal from all the other users except the m th user plus noise. The received signal
given by (3.9) is multiplied by complex gains Gm (k) to combine the energy scattered in
the frequency domain. We choose Gm (k) a weighted complex conjugate of the received

envelope denoted by zm (k) i.e.

Gm (k) = wm (k) zm (k) , k (3.11)

The weighting factor wm (k) depends on the combining technique. Following are few
basic combining techniques with associated wm (k) [30, 35].

3.2.1 Maximum Ratio Combining

Maximum Ratio Combining (MRC) corrects the phase shifts and weighs the amplitudes of
the received signal according to the channel fading coefficients. The weighting coefficient
wm (k) is determined to be unity in this case.

wm (k) = 1 (3.12)

41
3.2.2 Equal Gain Combining

Equal gain Combining (EGC) corrects the phase shifts only and weighs all the subcarriers
equally. The associated wm (k) is

1
wm (k) = k (3.13)
|zm (k)|

3.2.3 Zero Forcing Combining

Zero Forcing Combining (ZFC) compensates the channel effects by inverting the channel
coefficients. ZF combining is associated with noise enhancements. The assigned wm (k)
is
1
wm (k) = k (3.14)
|zm (k)|2

3.2.4 Minimum Mean Square Error Combining

The Minimum Mean Square Error Combining (MMSEC) corrects the phase shifts and
the amplitudes of the received signal by taking into account the number of users and
SNR and resulting wm (k) can be written as

1
wm (k) = 2 LN0
k (3.15)
|zm (k)| + 2Eb M

where L is the processing gain, E b the bit energy and N0 is the noise variance. At the
receiver, the equalized signal can be written as

ym (k) = w (k) |zm (k)|2 xm (k) + w (k) zm

(k) m (k) (3.16)

where |zm (k)| represents the absolute value of zm (k) . If the number of users is large

and data from all users is independent, then by central limit theorem, w (k) zm (k) m (k)
can be approximated as Gaussian noise. The signal ym (k) goes to Signal Reliability
Estimator (SRE) where the LLRs are calculated for both real (inphase) and imaginary

42
(quadrature) components. Let the symbols < and = denote the real and imaginary
components respectively. The LLR for the real and imaginary components, respectively,
can be written as [55]

!
Pr x < <

m (k) = +1|y m (k)
1 x <
 
m (k) = log < < (k)]
(3.17)
Pr [xm (k) = 1|ym
!
Pr x = =

m (k) = +1|y m (k)
1 x =
 
m (k) = log (3.18)
Pr [x=
m (k) = 1|y =
m (k)]

It is shown in appendix A, in section A.1 that

<
2 ym (k) w (k) E [< {zm (k) m (k)}]
1 x <
 
m (k) = 2 |zm (k)| (k) (k)}]
(3.19)
V ar [< {zm m

and
=
2 ym (k) w (k) E [= {zm (k) m (k)}]
1 x =
 
m (k) = 2 |zm (k)| (k) (k)}]
(3.20)
V ar [= {zm m

The quantities E [< {zm (k) m (k)}], V ar [< {zm (k) m (k)}], E [= {zm (k) m (k)}] and

V ar [= {zm (k) m (k)}] are calculated in appendix A, in section A.2 and are given by

<
(k) E r< (k) + zm =
(k) E r= (k)
   
E [< {zm (k) m (k)}] = zm
|zm (k)|2 E x<
 
m (k) (3.21)

<

2
2
(k) V ar r< (k) + zm =
(k) V ar r= (k)
   
V ar [< {zm (k) m (k)}] = zm
< =
(k) m (k) |zm (k)|4 V ar x<
 
+2zm (k) zm m (k) (3.22)

<
(k) E r= (k) zm =
(k) E r< (k)
   
E [= {zm (k) m (k)}] = zm
|zm (k)|2 E x=
 
m (k) (3.23)

<

2
2
(k) V ar r= (k) + zm =
(k) V ar r< (k)
   
V ar [= {zm (k) m (k)}] = zm
< =
(k) m (k) |zm (k)|4 V ar x=
 
2zm (k) zm m (k) (3.24)

43
where X
< =
(k) V ar x<
   = 

(k) = zm (k) zm m (k) V ar xm (k) (3.25)
m

The quantities E r< (k) , E r= (k) , V ar r< (k) and V ar r= (k) from appendix are
       

 X <
E r< (k) = zm (k) E x< =
    = 
m (k) zm (k) E xm (k) (3.26)
m

 X <
E r= (k) = zm (k) E x= =
    < 
m (k) + zm (k) E xm (k) (3.27)
m
 X <
2
2
V ar r< (k) = zm (k) V ar x< =
 =
+ 2
   
m (k) + zm (k) V ar x m (k) (3.28)
m
 X =
2
2
V ar r= (k) = zm (k) V ar x< <
V ar x= 2
    
m (k) + zm (k) m (k) + (3.29)
m

where 2 is the noise variance. The quantities E [x (k)] and V ar [x (k)] are derived in
chapter 2 section 2.6.2 and are given by

E [x (k)] = tanh (2 [x (k)] /2) (3.30)

V ar [x (k)] = 1 (E [x (k)])2 (3.31)

Equations (3.19) to (3.29) are derived in appendix and are based on derivations outlined
in [55]. Equations (3.25) to (3.31) are used in the estimation of both real and imagi-
nary components and are evaluated only once for each chip. Since E [x (k)] and hence
V ar [x (k)] are calculated based on estimates from Dec block which are not available
during first iteration, these are initialized with zero and ones respectively for all k. The
SRE block is followed by an array of single user decoder or Dec blocks. The Dec blocks
carry out the APP decoding using deinterleaved output of the SRE as the input. Its
output is the extrinsic LLR, 2 [xm (k)]
 
Pr [xm (k) = +1|1 , C]
2 [xm (k)] = log 1 [xm (k)] (3.32)
Pr [xm (k) = 1|1 , C]

where C represents the constraints, 2 [xm (k)] is fedback to SRE, where means and vari-
ances are updated using (3.30) and (3.31). The algorithm is summarized in the next

44
section.

3.3 The CBC Detection Algorithm

Initialization
E x<
   =   <   = 
m (k) = E xm (k) = 0, V ar xm (k) = V ar xm (k) = 1, m, k

E r< (k) = E r= (k) = 0, m, k

   

V ar r< (k) = V ar r= (k) = m |Zm |2 + 2 , m, k

    P

(k) = 0
Main process
for i=1:iterations
for m=1:M
(SRE Block)
<
(k) E r< (k) + zm =
(k) E r= (k) |zm (k)|2 E x<
     
1. E [< {zm (k) m (k)}] = zm m (k)
<
(k) E r= (k) zm =
(k) E r< (k) |zm (k)|2 E x=
     
E [= {zm (k) m (k)}] = zm m (k)
<

2
2
(k) V ar r< (k) + zm =
(k) V ar r= (k)
   
2. V ar [< {zm (k) m (k)}] = zm
< =
(k) (k) |zm (k)|4 V ar x<
 
+2zm (k) zm m (k)

2
2
V ar [= {zm (k) m (k)}] = zm (k) V ar r= (k) + zm
< =
(k) V ar r< (k)
   

< =
(k) (k) |zm (k)|4 V ar x=
 
2zm (k) zm m (k)

3. Evaluate 1 x<
   = 
m (k) and 1 xm (k) using (3.19) and (3.20)

4. Deinterleave 1 x<
   = 
m (k) and 1 xm (k)

(Dec Block)
5. Evaluate 2 x<
   =   <   = 
m (k) and2 xm (k) using 1 xm (k) and 1 xm (k) based on

FEC equation (3.32) and interleave 2 x<

   = 
m (k) and 2 x m (k)
 <   <  <
 <  =
 = 

6. E r (k) = E r (k) zm (k) E xm (k) zm (k) E xm (k)
E r= (k) = E r= (k) zm <
(k) E x< =
       = 

m (k) + zm (k) E xm (k)

2
7. V ar r< (k) = V ar r< (k) zm <
(k) V ar x< =
V ar r= (k)
   
   
m (k) + zm (k)

2
2
V ar r= (k) = V ar r= (k) zm =
(k) V ar x< <
V ar r= (k)
       
m (k) + zm (k)

45
 < =
(k) V ar r< (k) V ar r= (k)
   

8. (k) = (k) zm (k) zm
9. E x<
   < 

m (k) = tanh 2 xm (k) /2

E x=
   = 

m (k) = tanh 2 xm (k) /2

2
10. V ar x<
   <
m (k) = 1 E xm (k)

2
V ar x=
   =
m (k) = 1 E xm (k)

11. E r< (k) = E r< (k) + zm <

(k) E x< =
       = 

m (k) zm (k) E x m (k)
 =   =  <
 <  =
 = 

E r (k) = E r (k) + zm (k) E xm (k) + zm (k) E xm (k)

2
12. V ar r< (k) = V ar r< (k) + zm <
(k) V ar x< =
V ar r= (k)
   
   
m (k) + zm (k)

2
2
V ar r= (k) = V ar r= (k) + zm =
(k) V ar x< <
V ar r= (k)
       
m (k) + zm (k)
< =
(k) V ar r< (k) V ar r= (k)
   

13. (k) = (k) + zm (k) zm
end
end

3.4 An Approximate CBC Detection Algorithm

The complexity of the CBC algorithm is drastically reduced if the hyperbolic function in
(3.30) is approximated by a clipping function and the variance (3.31) is set to zero [79,85].
i.e.

1 [xm (k)] 1 [xm (k)]  [+1, 1]
tanh (x) = (3.33)
1 otherwise
and
V ar [xm (k)] = 0 (3.34)
the equations (3.22),(3.25),(3.28),(3.29), reduces to

(k) = 0 (3.35)

V ar r< (k) = V ar r= (k) = 2

   
(3.36)

V ar [< {zm (k) m (k)}] = |zm (k)|4 2 (3.37)

46
3.5 Computational Complexity
The CBC detection algorithm for IDMA given in [55] in complex channels requires
LM (9J + 4) iter +L (M J + 1) /2 real additions M L (13J + 4) iter+M L (2J + 1) /2 real
multiplications M (2JL + 1) iter shifts and M iter/2 hyperbolic functions per coded
bit. The corresponding number of real additions and multiplications for MC-IDMA sys-
tem are 13M L iter + L (M + 1) /2 and 15M L iter + 3M L/2, respectively. Since there
is an FFT operation involved in MC-IDMA, and to have a fair comparison, the compu-
tational complexity of FFT operation must be taken into account. As shown in [86],
for an N -point FFT, there are N log2 N complex multiplications and additions and there
are 4 real multiplications and 2 real additions in a complex multiplication and two real
additions in a complex addition. So there are 4N log2 N real multiplications and additions
in an FFT operation. Since there are N chips and both real and imaginary components
are involved in an FFT operation, over all additions/subtractions and multiplications in
an MC-IDMA receiver become 2log2 N +17 and 2log2 N +33 respectively per coded bit per
chip per iteration. The complexity of DEC block is not considered here because it is the
same for both systems. The computational complexity of MC-CDMA detector is cal-
culated in appendix B. These considerations show that the complexities of MC-IDMA,
MC-CDMA and IDMA systems are O(LM ), O(LM 2 ) and O(LMJ ) respectively. The
number of operations/complexities of MC-IDMA, MC-CDMA and IDMA systems are
summarized in table 3.1 and are plotted as a function of iterations in figure 3.2. It is
obvious that MC-IDMA requires fewer number of operations as compared to IDMA and
MC-CDMA systems.

3.6 Adaptive Subchannel Allocation

In this section we propose adaptive subchannel allocation for MC-IDMA to enhance the
performance of the system. A lot of work has been done on subcarrier allocation in
OFDM systems [87, 88]. Instead of transmitting chips over each subcarrier, only those

47
 Table 3.1: Complexity comparison of MC-IDMA, MC-CDMA and IDMA systems.
Additions/ Multiplications/
Parameters Complexity
Subtractions Divisions
13 M L iter
15M L iter + 3M2 L
MC-IDMA +L (M + 1) /2 O (LM )
+2 log2 N + 33
+2 log2 N + 17
M 2 L iter
4M L (M + 1)
+2M L (2M + 1) M (4M 2 +12M +5)
MC-CDMA M (4M 2 3M 4) + O (LM 2 )
+ 6
6 +2 log2 N + 33
+2 log2 N + 17
M L (9J + 4) iter M L (13J + 4) iter
IDMA O (LM J)
+L (M J + 1) /2 +M L (2J + 1) /2

subcarriers are used which have largest fading amplitudes among all the subcarriers.
The receiver estimates the fading amplitudes of all the subcarriers and feeds back the
index information to the transmitter. Based on the index information, the transmitter
allocates users chips to the good conditioned subchannels. The receiver feeds back the
index information to the transmitter periodically by estimating the fading amplitudes
with the help of pilot signals transmitted with the data symbols. All the pilot symbols
are assumed to be 1. The pilot symbols are transmitted over all the subcarriers so that
the receiver can estimate all the subchannels. The allocation scheme is based on following
steps
1. Arrange the subchannel gains in descending order i.e.

zm (1) zm (2) ... zm (K) (3.38)

Let vm (k)th be the vector of indices of the gains vector of m th user given in
step 1.
2. Allocate k th chip of m th user to the vm (k)th subcarrier for all k.
The subcarriers are hopped periodically based on the channel conditions.

48
 6
x 10
4
No. of Multiplications
No. of Additions
3.5
MCCDMA
IDMA
3 MCIDMA
No. of Operations
2.5

1.5

0.5

0
1 2 3 4 5 6 7 8 9 10
No. of Iterations

Figure 3.2: Complexity comparison of MC-IDMA, IDMA and MC-CDMA systems.

3.7 Multicode MC-IDMA

In this section multicode communication is considered for MC-IDMA system to accom-
modate higher modulations. A multirate CDMA scheme in AWGN channels is considered
in [89]. In multicode transmission each user transmits his data by choosing one sequence
from an orthogonal code-set [90]. In a variant of multicode scheme where a user needs
U times the basic data rate, the user converts his signal stream using a serial to paral-
lel converter into U basic rate streams, encodes each with a different code, modulates
them with a different modulator and super imposes them before transmission [91, 92].
A Multicode-Multicarrier-Direct Sequence-CDMA (MC-MC-DS-CDMA) system is ana-
lyzed in [93]. The data stream of a user with data rate U is first multiplexed into U
different serial streams with a base data rate and each serial stream is treated as an in-
dividual or effective user. Each of the U serial streams is then converted into P parallel
sub-streams and spread by the same spreading code with constant spreading factor. The
parallel substreams from all the other effective users are combined and are transmitted

49
 U-ary data symbol
Code set aU

G1
G2
MC-IDMA Transmitted signal
transmitter of mth user
GU

(a)

Matched filter
bank
G1
V1
^a
1
Decision W1
unit
GU
G1
^a Decision W2 V2
2
unit MC-IDMA Received
GU receiver signal

G1
VM
^a Decision WM
M
unit
GU

(b)

Figure 3.3: MC-MC-IDMA system model: (a) transmitter (b) receiver.

using P orthogonal subcarriers. Since each data stream is treated as an effective users
data, the system experiences more interference per user as the data rate is increased.
In [94], an adaptive multicarrier-CDMA system is presented in which the symbols from
serial to parallel converted substreams are copied into F branches. The symbols on each
branch are multiplied by the corresponding bit of assigned spreading code. The parallel
signals from all the effective users are combined correspondingly before transmission.
To support variable data rates, a Multicode-Multicarrier IDMA (MC-MC-IDMA)
system is presented in this section. The U -ary data is first multiplexed by U orthogonal
codes. Each bit of the code is then sent to MC-IDMA transmitter. The transmitter of
the proposed system is depicted in figure 3.3a. It uses a set of U codes called the code-
set for U -ary modulation. If there are M active users, each user has the same code-set
which represents an information symbol of log2 U bits. The size of the code-set depends
on the required data rate. In the normal case, the size is 2, i.e. there are two sequences

50
in the set, one to represent a 0 and one to represent a 1. When the data rate is to
be made log2 U times the standard data rate, the sequence set is made of size U and
each group of log2 U bits is mapped to one of code sequences of length U. As shown in
3.3a, a U -ary symbol selects one of the U pre-mapped code sequences for transmission.
Each code sequence has a time domain spreading ratio U/ log2 U. At the receiver, the soft
output of iterative detector is passed through the bank of match filters which correlates
the soft output with the code set and the code with maximum correlation is selected as
the desired code. Let Vm be the detector output for m th user and G=[G1 , G2 , ..., GU ]
be the code set, then
T


j = arg max Gu Vm (3.39)
1uU

and
m = Gj
a (3.40)

3.8 Performance Analysis

To analyze the performance of the proposed system, we follow the strategy outlined
in [55]. Setting w (k) = 1, we can write (3.19) using (3.16) as

|zm (k)|2 x<

m (k) + < {zm (k) m (k)}

E [< {zm (k) m (k)}]
1 x < = 2 |zm (k)|2
 
m (k) (k) (k)}]
(3.41)
V ar [< {zm m

In the above equation, |zm (k)|2 x<

m (k) and < {zm (k) m (k)} E [< {zm (k) m (k)}] rep-

resent signal and noise components respectively. Since x<

m (k) = 1, then the signal

power is |zm (k)|4 . The noise power after soft cancellation is

E (< {zm (k) m (k)} E [< {zm (k) m (k)}]) = V ar [< {zm (k) m (k)}] (3.42)

51
Thus the chip SNR for 1 x<
 
m (k) for m th user is

c |zm (k)|4
snrm (k) = (k) (k)}]
(3.43)
V ar [< {zm m

Let us suppose that

V ar x<
   = 
m (k) = V ar xm (k) = V ar [xm (k)] (3.44)

then (3.28) and (3.29) give

V ar r< (k) V ar r= (k)

   
(3.45)
X
= |zm (k)|2 V ar [xm (k)] + 2 (3.46)
m

and (3.25) becomes

m (k) = 0 (3.47)
Substituting (3.44) to (3.25) into (3.22)

2 X
<
|zm0 (k)|2 V ar [xm0 ] + 2


V ar [< {zm (k) m (k)}] = zm (k)
m0

2 X
=
|zm0 (k)|2 V ar [xm0 ] + 2


+ zm0 (k)
m0
4
|zm (k)| V ar [xm (k)]
X
= |zm (k)|2 |zm0 (k)|2 V ar [xm0 ] + 2


m0
4
|zm (k)| V ar [xm (k)] (3.48)
X
= |zm (k)|2 |zm0 (k)|2 V ar [xm0 ] + 2 (3.49)
m06=m

similarly
X

V ar [= {zm (k) m (k)}] = |zm (k)|2 |zm0 (k)|2 V ar [xm0 ] + 2 (3.50)
m06=m

c
so the chip snrm (3.43) becomes

c |zm (k)|2
snrm (k) = P 2 (3.51)
2
m06=m |zm0 (k)| V ar [xm0 ] +

52
The bit snrm of m th user is given by averaging the chip snrm
c
(k) over k
K
X
c
snrm = snrm (k) (3.52)
k=1

The evaluation of (3.51) for each chip is computationally prohibitive and we can replace
|zm (k)|2 and V ar [xm (k)] by their averages Zm and Vxm respectively over all the chips
i.e
K1
1 X
Zm = |zm (k)|2 (3.53)
K k=0

and
K1
1 X
Vxm = V ar [xm (k)] (3.54)
2K k=0

where the factor 1/2 comes due to the average taken out for both real and imaginary
components. Using (3.53) and (3.54) in (3.51), the average snr for m th user can be
approximated as
Zm
snrm P 2
(3.55)
m06=m Zm0 Vxm0 +

The average variance Vx in (3.55) is a function of snr [55]

Vxm = f (snrm ) (3.56)

similarly the BER is also a function of snr [55]

BERm = g (snrm ) (3.57)

In general, there is no closed form solution for f (.) and g (.) . These functions are obtained
by simulating a single user APP decoder for repetition codes of length 64 in AWGN
channels [55] and are plotted in figure 3.4. Polynomial fitting strategy of fifth order is
applied to obtain following empirical relations for f (.) and g (.) for specified values of

53
 Table 3.2: Fitting Coeffients
Coefficients Values Coefficients Values
a0 0.13526 b0 0.4295
a1 18.64461 b1 24.4869
a2 9.15476 b2 112.49394
a3 94.69643 b3 651.74424
a4 419.21277 b4 1813.06859
a5 367.98638 b5 1995.83658

1
10

0
10

1
10

2
Variance or BER

10

3
10

4
10

5 Simulation
10
Polynomial Fit
6 Variance=f(SNR)
10 BER=g(SNR)

7
10
30 25 20 15 10 5 0 5
SNR (dB)

Figure 3.4: The variance and BER as functions of the SNR of a single user APP decoder.

spreading and snr, s.

f (snr) = 10(a0 a1 snr+a2 snr a3 snr +a4 snr a5 snr )

2 3 4 5
(3.58)

g (snr) = 10(b0 b1 snr+b2 snr b3 snr +b4 snr +b5 snr )

2 3 4 5
(3.59)

The numerical values for the coefficients as and bs are given in table 3.2, using 64
point FFT and E b /N 0 = -2,5 dB to 0 dB. Once the expressions for f (.) and g (.) are
obtained, the performance assessment process can be started using (3.55). In Let i be
the iteration index, combining (3.51) and (3.56), the snr evaluation can be tracked by

54
following recursive process

(i) Zm
snrm =P (3.60)
m06=m Zm0 f (snrm )(i1) + 2

In AWGN channels Zm = 1. for all m then

(i) 1
snrm =P (i1)
(3.61)
m06=m f (snrm ) + 2

The performance evaluation algorithm can be summarized as

Initialization
Set Vm = f (snrm ) = 1, m
V = m Zm + 2
P

Main process
for i=1:iterations
for m=1:M
V = V Zm Vm
snrm = Zm /V
Evaluate Vm = f (snrm ) using (3.58)
V = V + Zm Vm
Evaluate BERm = g (snrm ) using (3.59)
end
end
P
BER = m BERm

3.9 Simulation Results

In this section, simulation results are provided to illustrate the performance of MC-
IDMA system in frequency selective channels. The parameters used in the simulations
are summarized in table 3.3. It can be seen from table 3.3 that the bandwidth of each

55
 Table 3.3: Simulation parameters.
Parameter Value
Channel bandwidth 20 MHz
Root Mean Square (RMS)
200ns
delayspread (rms )
Coherent bandwidth of the channel 1/(2rms ) = 786KHz
Information portion of the symbol 3.2 s
Length of Cyclic Prefix 0.8 s
Total symbol duration 4.0 s
Data rate 20 Kbits/s
Chip rate 32 MChips/s
Subcarrier spacing 312.5 KHz
Sampling rate 20 MHz
FFT size 64
Combining MRC, i.e. w (k) = 1, k
Three path frequency
channel type
selective channel

subchannel (312.5KHz) is much less than the coherence bandwidth of the channel and
each subchannel observes frequency selective fading as flat fading. An MC-CDMA sys-
tem is considered as a reference. It uses a Parallel Interference Cancellation detector.
The block diagram of such a detector is shown in appendix B. Figure 3.5 shows the
performance comparison of an uncoded MC-CDMA and MC-IDMA system with CBC
detector. We see that the performance of MC-CDMA PIC detector degrades as the num-
ber of users increase and for 64 users, the MC-CDMA system fails to distinguish users.
Injection of more power only slightly improves the performance. MC-CDMA systems are
therefor interference limited systems. MC-IDMA system, on the other hand outperforms
the MC-CDMA system by supporting 112 users at 18 dB E b /N 0 . When M =96 or 112,
BER curves fall down abruptly with minor increase in SNR. MC-IDMA can serve even
more users by slightly increasing the transmitted power which shows that MC-IDMA
systems are power limited and not interference limited as the MC-CDMA systems. In
these simulations, MC-CDMA uses Walsh Hadamard codes of length 64 when the num-
ber of users is less than or equal to the 64 and Pseudo Noise (PN) sequences when the

56
 0
10

1
10

2
10
BER

3
10
M=1
M=32
4 M=64
10
M=96
M=112
5 MCIDMA
10
MCCDMAPIC

0 5 10 15 20
Eb/N0(dB)

Figure 3.5: Average BER performance of uncoded MC-IDMA and MC-CDMA systems
in frequency selective multipath fading channels for various number of users.

number of users is more than 64. In the case of MC-IDMA, the input symbols are spread
by a balanced sequence of alternating +1 and -1 of length 64. With block length of
256, the interleaver length (block lengthspread length) equals 18384. The interleavers
used in these simulations are chosen at random. No attempt is made to optimize the
interleaver, as literature [95] indicates that an interleaver picked at random is enough.
To test this assumption, we repeated the results with different interleavers, all of which
were chosen at random. The simulated performance results with the different interleavers
were virtually identical.
Figures 3.6 and 3.7 compares the results of an uncoded MC-IDMA obtained by sim-
ulation and evolution in AWGN and multipath channels respectively. We observed that
the results agree quite well upto the number of users slightly greater than the length of
spreading sequence. For larger number of users the results obtained by evaluation de-
grade drastically. In figure 3.8, performance is compared using Concatenated Zigzag (CZ)
codes [71]. The input symbols to the MC-IDMA systems are first spread with alternating

57
 0
10

1
10

2
10
BER

3
10
M=1
4 M=32
10
M=64
M=96
5
10 M=112
Simulation
Evolution
6
10
0 2 4 6 8 10 12 14 16
Eb/N0(dB)

Figure 3.6: Comparison of perfromance of MC-IDMA system obtained by simulation and

evolution in AWGN channels using length 64 repetition code with QPSK modulation.

+1 and -1 sequences of length 32. MC-CDMA input symbols are spread with WH codes
of length 32 when the number of users is less than 32 and PN sequences of same length
1
for more than 32 users. The spread sequences are encoded with rate 2
CZ codes with
four constituent encoders so that code rate of both systems is 1/64. The performance
of both systems is nearly the same as for as the number of users is less than or equal to
the spread length of 32. MC-IDMA system outperforms the MC-CDMA system when
the number of users exceed the spread length because both systems utilized same coding
schemes and detection algorithms with the only difference in the spreading codes. MC-
IDMA uses same spreading code for each user and different interleaver, the combination
of same spreading code and distinct interleaver can be thought as a different spreading
code for each user although not orthogonal to other users codes. This is justified because
even the orthogonal codes no longer remain orthogonal due to channel impairments. The
interference originated due to this non-orthogonality, is removed by interference rejection
capability of SRE block. However the PIC detector used in MC-CDMA, cannot remove

58
 0
10

1
10

2
10
BER

3
10
M=1
4
M=32
10 M=64
M=96
5 M=112
10
Simulation
Evolution
6
10
0 5 10 15 20
Eb/N0(dB)

Figure 3.7: Comparison of perfromance of MC-IDMA system obtained by simulation and

evolution in multipath channels using length 64 repetition code with QPSK modulation.

interference and. In MC-CDMA, we can neither change the spreading codes because
they are optimally used here for user separation nor afford any further addition of re-
dundancy in the form of FEC due to imposed data rate and bandwidth constraints and
consequently, cannot further improve its performance. However in MC-IDMA, we can
add redundancy without any such constraint because users are identified by interleavers
and we are free to choose any mechanism of spreading. If we devote the entire spreading
to FEC, which will provide additional coding gain not present in MC-CDMA due to
constraints already discussed. The simulation results shown in figure 3.9 illustrate the
BER performance improvement of MC-IDMA system as compared to MC-CDMA. In
1
these simulations we used same rate 2
CZ codes in MC-CDMA as used in figure 3.8.
In MC-IDMA low rate systematic ZH codes [72, 73] with Hadamard codeword sets of
length 16 (r =4), are employed. The code rate of ZH codes is 1/4. To make the code
rate 1/64, the resulting codes are spread by a repetition code of length 16. The code
rate of both systems is still 1/64, but we get 4 to 5 dB performance improvement in

59
 0
10

1
10

2
10
BER

3
10

4
10
M=1
M=32
5
10 M=64
MCIDMA
MCCDMA
6
10
0 2 4 6 8 10
Eb/N0(dB)

Figure 3.8: Average BER performance of concatenated zigzag coded MC-IDMAand MC-
CDMA systems in frequency selective multipath fading channels as a function of E b /N 0
for various number of users. The lengths of spreading sequences is 32 in both systems.

MC-IDMA system. Figure 3.10 shows the convergence of the iterative algorithm as a
function of iterations for various values of E b /N 0 for 32 users. We observed that for low
values of E b /N 0 , the increase of iterations has no effect on BER, however for sufficiently
high value of Eb /N0 , the BER decreases rapidly as the number of iterations is increased
and the algorithm converges after about twelve iterations.
Figures 3.11 and 3.12 show the impact of interleaver length on the BER performance
of MC-IDMA system. The interleaver length can be varied either by varying the block
length or the spread length. It can be observed from figures 3.11 and 3.12 that in both
cases, larger interleaver sizes improve the performance between the threshold and the
single user. Figure 3.13 and 3.14 shows the performance of approximate model presented
in the chapter in AWGN and multipath channels respectively. The approximate model
performs well as long as the number of users is small, but as the number of users grow,
the performance degrades. Figure 3.15 shows the performance of the system for various
modulation schemes for the case of single user. It is observed that we can double the

60
 0
10

1
10

2
10
BER

3
10

4
10
M=1
M=32
5
10 M=64
MCIDMA
6
MCCDMA
10
0 1 2 3 4 5 6 7 8
Eb/N0(dB)

Figure 3.9: Average BER performance comparison of concatenated zigzag Hadamard

coded MC-IDMA system and zigzag coded MC-CDMA system in frequency selective
multipath fading channels as a function of E b /N 0 for various number of users.

data rate with only a performance loss of 1 dB.

3.10 Conclusions
The principle of MC-IDMA system is described in this chapter. It has been shown that
MC-IDMA has excellent BER performance for higher number of users, it has reduced
complexity and is more robust to multipath fading effects and at relatively higher E b /N 0
the performance of uncoded system approaches to that of a single user. It is shown that
MC-IDMA system outperforms the MC-CDMA system when zigzag Hadamard codes
are used to spread the signal which provides additional coding gain not present in MC-
CDMA system, because redundancy cannot be added in it arbitrarily. The performance
of the system is analyzed and it is shown that analytical performance agrees with that
obtained by simulations. We conclude that MC-IDMA supports a large number of users
with a simple receiver and shows excellent BER performance.

61
 0
10

1
10

2
10
BER

3
10

4 Eb/N0=6 dB
10
Eb/N0=12 dB

5 Eb/N0=13 dB
10
Eb/N0=14 dB
E /N =15dB
6
b 0
10
2 4 6 8 10 12 14 16 18
Iterations

Figure 3.10: BER as a function iterations for various values of E b /N 0 . The number of
users is equal to 32.

0
10

1
10

2
10
BER

3
10

4
10 Spread length=8
Spread length=16
Spread length=32

1 2 3 4 5 6
Block Length

Figure 3.11: BER as a function of block length for various values of spread length.

62
 1
BER 10

2
10 Block length=1
Block length=32
Block length=64

3 3.5 4 4.5 5 5.5 6

log2(Spread Length )

Figure 3.12: BER as a function of spread length for various values of block length.

0
10

1
10

2
10
BER

3
10

M=1
4
10 M=32

M=64
5
10 M=96

M=112
6
10
0 5 10 15 20
E /N (dB)
b 0

Figure 3.13: BER performance of uncoded MC-IDMA system using approximate model
in AWGN channels.

63
 0
10

1
10

2
10
BER

3
10

M=1
4
10 M=32

M=64
5
10 M=96

M=112
6
10
0 5 10 15 20
Eb/N0 (dB)

Figure 3.14: BER performance of uncoded MC-IDMA system using approximate model
in multipath channels.

0
10
QPSK
16 QAM
64 QAM
1
10 256 QAM

2
10
BER

3
10

4
10

5
10
0 5 10 15 20
Eb/N0 (dB)

Figure 3.15: BER performance of multicode MC-IMDA system for various modulation
schemes in frequency selective multipath channels for a single user.

64