5

Code Division for Multiple

Access (CDMA)

5.1 Introduction to CDMA

CDMA stands for Code Division for Multiple Access and is considered a path-breaking
wireless technology due to its several superior properties. It was first employed in the 2nd
generation IS-95 cellular standard, which was predominantly used in North America, under
the brand name cdmaOne. It also forms the basis for several advanced 3rd Generation i.e., 3G
cellular standards such as Wideband CDMA (WCDMA), High-Speed Downlink Packet Access
(HSDPA), High Speed Uplink Packet Access (HSUPA), CDMA 2000, and 1x Evolution
Data Optimized (1xEV-DO). In order to understand the concepts in CDMA, it is critical to
understand the concept of multiple access. In conventional wired communication systems,
there is a dedicated wireline communication channel which is allocated exclusively to the
particular device such as a telephone, etc. However, in a wireless network, mobile phones and
other wireless-communication devices are required to share the common radio channel over
the air. This is shown in Figure 5.1. This is because the radio channel is common for all the
users/ devices and the available wireless frequency bands are limited. Thus, it is necessary to
device a mechanism for multiple users to access this common radio channel, which is termed
as a Multiple Access (MA) technology. Thus, multiple access is at the heart of modern wireless
technologies, especially 3G and 4G cellular technologies.
Several multiple-access technologies have been developed and employed for cellular
applications. In fact, each generation of cellular standards is characterized by a particular
multiple-access technology. For instance, the first generation, i.e., 1G cellular standards were
based on Frequency Division Multiple Access (FDMA). In FDMA, different users are alloted
different frequency bands. Thus, the users are multiplexed in the frequency domain and
120 Principles of Modern Wireless Communication Systems

Figure 5.1 Multiple access for wireless cellular networks

User 0 User 1 User 3

f
B B B

Figure 5.2 Frequency division for multiple access

they access the radio channel in their respective frequency bands of bandwidth B . This is
schematically shown in Figure 5.2. On the other hand, the second generation or 2G cellular
standards are based on digital Time Division for Multiple Access (TDMA) in which different
users are allocated different time slots of duration T for accessing the wireless channel.
Thus, the different users are multiplexed in the time domain as shown in Figure 5.3. These
technologies were replaced by CDMA in successive 3G wireless technologies. The motivation
and basic mechanism of CDMA is described in the next section.

User 0 User 1 User 2

T T T

Figure 5.3 Time division for multiple access

 Code Division for Multiple Access (CDMA) 121

5.2 Basic CDMA Mechanism

CDMA, as the name suggests, is a multiple-access technology based on code division. In other
words, different users are multiplexed using different codes. Consider a two-user scenario, i.e.,
two users accessing the radio channel simultaneously. Let a0 denote the symbol of the user 0,
while a1 denotes the transmit symbol corresponding to the user 1. Let the code c0 of the user 0
be given as c0 = [1, 1, 1, 1]. The above code c0 is of length N = 4 chips. Each element of the
code is termed as a chip. The transmitted signal x0 of the user 0 is then given by multiplying
the code c0 with the symbol a0 as

x0 = a0 × [1, 1, 1, 1]
= [a0 , a0 , a0 , a0 ] (5.1)

The structure of the above transmit signal x0 can be interpreted as follows. The symbol a0,
of the user 0, is multiplied by the code c0 to yield 4 chips x0 (i), 0 ≤ i ≤ N − 1. Similarly,
let the code c1 , given as c1 = [1, −1, −1, 1], correspond to the code of the user 1. Hence, the
sequence of chips corresponding to the user 1 transmission is given as

x1 = a1 × [1, −1, −1, 1]

= [a1 , −a1 , −a1 , a1 ] (5.2)

The signals x0 , x1 corresponding to users 1, 2 respectively are now summed to yield the net
signal x as

x = x1 + x2 = [(a0 + a1 ) , (a0 − a1 ) , (a0 − a1 ) , (a0 + a1 )] (5.3)

This sum, or composite, signal is then transmitted on the downlink from which each of the users
0, 1 detect their own signal. This is done as follows. User 1 correlates the received signal x with
his code c0 , i.e., basically multiplies each chip of the received signal x with the corresponding
chip of the code c0 = [1, 1, 1, 1] and sums across the chips as follows.

a0 + a1 a0 − a1 a0 − a1 a0 + a1
× 1 1 1 1
(5.4)
(a0 + a1 ) + (a0 − a1 ) + (a0 − a1 ) + (a0 + a1 ) = 4a0

Thus, the result of the above correlation is 4a0 , which is proportional to the transmitted symbol
a0 . Similarly, at the user 2, the received signal x is correlated with the chip sequence c1 =
122 Principles of Modern Wireless Communication Systems

[1, −1, −1, 1] of the user 1 as

a0 + a1 a0 − a1 a0 − a1 a0 + a1
× 1 −1 −1 1
(5.5)
(a0 + a1 ) − (a0 − a1 ) − (a0 − a1 ) + (a0 + a1 ) = 4a1

to yield 4a1, which is proportional to the transmitted symbol a1 of the user 1. Thus, unlike in
GSM or FDMA, in which the signals of different users are transmitted in different time slots
or frequency bands, in CDMA, all the signals of the different users are contained in the single
signal x over all time and frequency. However, in CDMA, the symbols of the different users
are combined using different codes. For instance, in the above example, the symbols a0 , a1 of
users 0, 1 are multiplied with codes c0 , c1 prior to transmission. Thus, the users of the different
signals are multiplexed over the common wireless channel employing different codes. Hence,
this is termed Code Divison for Multiple Access, i.e., multiple access based on different codes.
The key operations in CDMA can be summarized as follows.
1. Multiplying or modulation the symbols of the different users with the corresponding
assigned unique code, similar to the procedure illustrated in equations (5.1), (5.2).
2. Combining or adding the code-modulated signals of all the users to form the composite
signal as shown in Eq. (5.3), followed by subsequent transmission of the signal.
3. Finally, correlation of the composite received signal x at each user with the corresponding
code of the user to recover the transmitted symbol. This is described in Eqs (5.4), (5.5).

5.3 Fundamentals of CDMA Codes

In fact, from the example illustrated in the previous section, the astute reader will realize that
it is no accident by which we are able to recover the signals of users 0, 1. Computing the
correlation r01 of the user codes c0 , c1 yields
3
r01 = c0 (k) c1 (k)
k=0
 Code Division for Multiple Access (CDMA) 123

= 1 × 1 + 1 × (−1) + 1 × (−1) + 1 × 1
= 1 + (−1) + (−1) + 1
=0

Thus, since the correlation between the codes c0, c1 is zero, the codes are, in fact, orthogonal.
This is what helps us recover the symbols of the different users from the composite signal.
This is a key property of the codes employed in CDMA wireless systems, and a fundamental
principle on which the theory of CDMA is based.
Further, consider a fundamental property of the CDMA system arising because of the
employment of these codes. Let the symbol rate for the symbols a0 of the user 0 be 1 kbps.
Hence, the time period T per symbol is

1
T = = 1 ms
1 kbps

Hence, the corresponding bandwidth required for transmission is

1
B= = 1 kHz
T

However, now consider the transmission of the symbol a0 multiplied with the corresponding
code c0, i.e., a0 × [1, 1, 1, 1] = [a0 , a0 , a0 , a0]. Thus, for each symbol a0, one has to transmit
4 chips. Thus, to keep the symbol rate constant at 1 kbps, the time of each chip Tc has to be set
as Tc = 14 T = 0.25 ms. Thus, the bandwidth required for this system is

1 1
BCDMA = = = 4 kHz
Tc 0.25 ms

Thus, modulating with the code c0 of length N = 4, results in an increase of the required
bandwidth by a factor of N , i.e., from 1 kHz to 4 kHz. This is shown schematically in
Figure 5.4. Thus, it basically results in a spreading of the original signal bandwidth and, hence,
is termed a spreading code. Also, since the resulting signal occupies a large bandwidth, CDMA
systems are also termed spread spectrum or wideband systems.
Also, another interesting question the reader might be interested in is the following: How
many such orthogonals exist for a given spreading code length N ? The answer is there are N
such orthogonal codes. For instance, consider the case N = 4.
124 Principles of Modern Wireless Communication Systems

Figure 5.4 Spread spectrum communication

The different orthogonal spreading codes are

c0 = 1 1 1 1
c1 = 1 − 1 − 1 1
c2 = 1 − 1 1 − 1
c3 = 1 1 − 1 − 1

The reader can verify that the codes c0, c1 , c2 , c3 are orthogonal to each other. For example,
consider c1 , c2 . The correlation r12 between codes c1, c2 is
3
r12 = c0 (k) c1 (k)
k=0

= 1 × 1 + (−1) × (−1) + (−1) × 1 + 1 × br −1

= 1 + 1 + (−1) + (−1)
=0

This implies that given a spreading sequence length N , there exist N orthogonal codes and
hence, N users can be multiplexed together. This is important, since the bandwidth increases
by a factor of N due to transmission employing the codes as described earlier. However, it
is important to note that no inefficiency is introduced in the system because of the increase
in bandwidth, because this increase in bandwidth by a factor of N is compensated by the
parallel transmission of the signals corresponding to the N users over the same bandwidth.
Thus, the spectral efficiency of the system is not compromised. This is schematically illustrated
in Figure 5.5.
 Code Division for Multiple Access (CDMA) 125

a0 Code 0
a1 Code 1
N Symbols
a2 Code 2
a3 Code 3

a0 a1 a2 a3
T = NTc
T Symbols

Figure 5.5 Parallel transmission of N symbols over N codes in CDMA over

time interval T = N Tc (above) and comparison with transmission
of N symbols in time T = N Tc in a conventional single carrier or
time division system

5.4 Spreading Codes based on Pseudo-Noise (PN) Sequences

Consider the code c2 = [1, −1, 1, −1]. Observe that the code looks like a random sequence of
+1, −1, or a pseudo-noise (PN) sequence. This is so termed since it only resembles a noise
sequence, but is not actually a noise sequence. One method to generate such long spreading
codes based on PN sequences for a significantly large N is through the employment of a Linear
Feedback Shift Register (LFSR). This is described next.
Consider the shift register architecture shown in Figure 5.6, where the element D
represents delays. Thus, the digital circuit therein contains D = 4 delay elements or shift
registers. The input on the left is denoted by Xi , and the outputs of the different delays are
Xi−1 , Xi−2 , Xi−3 , Xi−4 . Let Xi−4 also denote the final output of the system. Also observe
that the xor Xi−4 ⊕ Xi−3 is fed back as Xi which is the input to the first shift register. Thus,
the governing equation of the circuit is

Xi = Xi−3 ⊕ Xi−4

which is a linear equation. Thus, since it implements a linear relation, with feedback and uses
delay elements or shift registers, such a circuit is also termed a Linear Feedback Shift Register
(LFSR) architecture. Since the next inpur, i.e., Xi depends on Xi−1, Xi−2, Xi−3, Xi−4, this
can also be thought of as the current state of the system. Consider initializing the system in the
state Xi−1 = 1, Xi−2 = 1, Xi−3 = 1, Xi−4 = 1. Thus, we have the corresponding Xi given
126 Principles of Modern Wireless Communication Systems

as

Xi = Xi−3 ⊕ Xi−4 = 1 ⊕ 1 = 0

This Xi becomes Xi−1 at the next instant and similarly, Xi−2, Xi−3 are shifted to the right as
Xi−3 , Xi−4 respectively. Continuing in this fashion, the entire sequence of state of the above
LFSR is summarized. It can be seen that the LFSR goes through the sequence of 15 states
1111, 0111, 0011, 0001, 1000, 0100, 0010, 1001, 1100, 0110, 1011, 0101, 1010, 1101, 1110,
before reentering the state 1111. Subsequently, the entire sequence of states repeats again.
Observe that this goes through 2D − 1 = 24 − 1 = 15 states. Also note that the maximum
number of possible states for D = 4 is 2D = 16. However, the LFSR can be seen to go
through all the possible states except one, which is the 0000 or the all-zero state.

Xi Xi - 1 Xi - 2 Xi - 3 Xi - 4
D D D D

Feedback

Xi = Xi - 3 ≈ Xi - 4

Figure 5.6 Linear feedback shift register

Xi = Xi−3 ⊕ Xi−4 = 0 ⊕ 0 = 0

Further, observe that if the LFSR is initialized in the 0000 state, it continues in the 0000 state,
since the corresponding Xi is leading to the next state of 0000. Thus, the LFSR never gets out
of the all zero states! Therefore, it is desired that the LFSR never enter the all-zero state. Such
an LFSR circuit which goes through the maximum possible 2D − 1 states, without entering the
all-zero state is termed a maximum-length shift register circuit or maximum length LFSR. The
generated PN sequence is termed a maximum-length PN sequence. Thus, the maximum-length
PN sequence is of length 2D − 1. For instance, for the above LFSR, the maximum-length PN
sequence is the sequence of outputs Xi−4 given as

PN Sequence = 1 1 1 1 0 0 0 1 0 0 1 1 0 1 0
 Code Division for Multiple Access (CDMA) 127

We can map the bits 1, 0 to the BPSK symbols −1, +1 to get the modulated PN sequence,

PN sequence = −1 − 1 − 1 − 1 + 1 + 1 + 1 − 1 + 1 + 1 − 1 − 1 + 1 − 1 + 1 (5.6)

Next, we examine the properties of such PN sequences derived above.

5.4.1 Properties of PN Sequences

Property 1-Balance Property: Consider the BPSK-modulated PN sequence shown in
Eq. (5.6). As already described, the PN sequence is of maximal length 2D − 1 = 15
corresponding to D = 4. Counting the number of −1 and +1 chips in the sequence, it can
be seen that the number of −1s is one more than the number of +1s. This is termed the
balance property of the PN sequence. This fundamentally arisies from the noiselike properties
of PN sequences. If we are generating random noise of +1, −1 chips, with P (Xi = +1) =
P (Xi = −1) = 12 , we expect to find on an average that half the chips are +1 and the rest are
−1. In the above case, however, as the total number of chips is an odd number, i.e., 15, it is
not possible to have an exactly even number of +1, −1s. Hence, we observe that the number
of +1, −1s is close to half the total number, i.e., eight −1s and seven +1s. Thus, the balance
property basically supports the notion of a noiselike PN chip sequence.
Property 2-Run-Length Property: A run is defined as a string of continuous values. There
are a total of 8 runs in this PN sequences. For instance, the first run −1, −1, −1, −1 is a run
of length 4. Thus, there is one run of length 4. Similarly, there is one run +1, +1, +1 of length
3, and two runs of length 2, viz., −1, −1, +1, +1. Finally, it can also be seen that there are 4
runs of length 1, viz., two runs of +1 and two runs of −1. Thus, there are a total of 2(D−1) = 8
runs. Out of the 8 runs, it can be seen that 1, i.e., 18 of the runs are of length 3, 14 of the runs
are of length 2 and 12 of the runs are of length 1. This is termed the run-length property of
PN sequences and can be generalized as follows. Consider a maximal length PN sequence of
length 2D − 1. Out of the total number of runs in the sequence, 12 of the runs are of length 1, 14
of the runs are of length 2, 18 of the runs are of length 3, and so on. This is again in tune with
the noiselike properties of PN sequences. For instance, consider a random IID sequence of
+1, −1. In such a sequence, one would expect the average number of +1 or −1 to be half the
total chips. Further, the number of strings +1, +1 or −1, −1, i.e., runs of length two would
be expected to comprises 14 of the total runs. This arises since the probability of seeing two
128 Principles of Modern Wireless Communication Systems

consecutive +1, +1 symbols is

1 1 1
P (Xi = +1, Xi+1 = +1) = × =
2 2 4

Similarly, one can explain the fraction 18 corresponding to runs of length 3. Thus, this further
supports the noiselike properties of PN sequences.

Property 3-Correlation Property: The correlation property is one of the most important
properties of PN sequences. Consider again the BPSK chip sequence shown in Eq. (5.6) and
denote it by c0 (n). Let us now look at the correlation properties of this sequence. Consider the
correlation r00 (0), i.e., the correlation of the sequences c0 with itself (the meaning of the (0)
will become clear soon). This correlation is given as
N −1
1
r00 (0) = c0 (n) c0 (n)
N
i=0
N −1
1
= 1
N
i=0
1
= ×N = 1
N
Now, consider a circularly shifted version of the PN sequence, shifted by n◦ = 2. Let it be
denoted by c0 (n − 2). This circularly shifted sequence by 2 chips can be readily seen to be
given as

PN Sequence = −1 + 1, −1 − 1 − 1 − 1 + 1 + 1
+1 − 1 + 1 + 1 − 1 − 1 + 1 (5.7)

Let us denote the correlation between c0 (n) and c0 (n − 2) by r00 (2), where the (2) can now
be seen to represent a circular shift of 2. The correlation can be seen to be given as
N −1
1
r00 (2) = c0 (n) c0 (n − 2)
N
i=0
 Code Division for Multiple Access (CDMA) 129

1
= {(−1) × (−1) + (−1) × (1) + (−1) × (−1) + (−1) × (−1) + (1) × (−1) +
15

(1) × (−1) + (1) × (1) + (−1) × (1) + (1) × (1) + (1) × (−1) + (−1) × (1) +

(−1) × (1) + (1) × (−1) + (−1) × (−1) + (1) × (1)}

1
= (1 − 1 + 1 + 1 − 1 − 1 + 1 − 1 + 1 − 1 − 1 − 1 − 1 + 1 + 1)
15
1 1
= × (−1) = −
15 15
1
=−
N
In fact, one can compute the correlation for other such nonzero delays, and can demonstrate
the the correlation is always − N1 . This autocorrelation property of the PN sequence, i.e., of
the sequence with a delayed version of itself, is shown pictorially in Figure 5.7. Thus, it can
be seen that while the correlation of the sequence with itself corresponding to a lag of 0 is
1, for any other nonzero shift, it assumes a very low value of − N1 , which tends to the limit
0 as the spreading length N → ∞. This autocorrelation property of the PN sequences can be
summarized as follows.

Shift n0

-1/N

Figure 5.7 Autocorrelation of PN sequence

⎧
N −1 ⎨ 1 if n = 0
1 ◦
r00 (n◦ ) = =
N ⎩ − 1 otherwise
i=0 N

With this background, let us investigate the properties of random spreading sequences in the
next section.
130 Principles of Modern Wireless Communication Systems

5.5 Correlation Properties of Random CDMA Spreading Sequences

In the previous section, we have seen that CDMA spreading sequences can be chosen as PN
sequences, which have noiselike properties. In other words, one can choose a chip sequence
ck (i) , 0 ≤ i ≤ N − 1 for the user k such that P (ck (i) = +1) = P (ck (i) = −1) = 12 . Thus,
we have,

1 1
E {ck (i)} = × (+1) + (−1) = 0.
2 2

Further, another important aspect is to choose such sequences as containing Independent

Identically Distributed (IID) chips, i.e., satisfying the property

E {ck (i) ck (j)} = E {ck (i)} E {ck (j)} = 0 × 0 = 0

The above property implies that each chip ck (i) is uncorrelated with chip ck (j). Further, one
can choose independent sequences for different users, that is, to say

E {ck (i) cl (j)} = E {ck (i)} E {cl (j)}

Let us examine the correlation properties of such random spreading sequences. As before, let
r0 0 (k) denote the autocorrelation of the chip sequence of the user k = 0, corresponding to a
lag k = 0. This can be expressed as

N −1
1
r00 (k) = c0 (i) c0 (i − k)
N
i=0

The average or expected valued of r00 (k) can be seen to be given as

N −1
1
E {r00 (k)} = E c0 (i) c0 (i − k)
N
i=0

N −1
1
= E {c0 (i) c0 (i − k)}
N
i=0
 Code Division for Multiple Access (CDMA) 131

N −1
1
= E {c0 (i)} E {c0 (i − k)}
N
i=0

N −1
1
= 0=0
N
i=0

Thus, the average value or the expected value of the correlation E {r00 (k)} is zero for lags
k = 0. This is expected from the random properties of the spreading sequence. To compute the
2 (k) given as
variance of the autocorrelation r00 (k), consider r00
⎛ ⎞
N −1 N −1
2 1
r00 (k) = 2 c0 (i) c0 (i − k) ⎝ c0 (j) c0 (j − k)⎠
N
i=0 j=0

N −1 N −1
1
= 2 c0 (i) c0 (i − k) c0 (j) c0 (j − k)
N
i=0 j=0

Now, let us consider the quantity c0 (i) c0 (i − k) c0 (j) c0 (j − k). It can be seen that if i = j ,
the expected value of this quantity can be simplified as

E {c0 (i) c0 (i − k) c0 (j) c0 (j − k)} = E {c0 (i) c0 (i − k)} E {c0 (j) c0 (j − k)}

= E {c0 (i)} E {c0 (i − k)} E {c0 (j)} E {c0 (j − k)}

=0

On the other hand, if i = j , the same quantity can be simplified as

E {c0 (i) c0 (i − k) c0 (j) c0 (j − k)} = E {c0 (i) c0 (i − k) c0 (i) c0 (i − k)}

= E (c0 (i))2 E (c0 (i − k))2

= 1×1 = 1
2 (k) can be simplified as
Thus, the variance of r00 (k), i.e., E r00
N −1 N −1
2 1
E r00 (k) = 2 E {c0 (i) c0 (i − k) c0 (j) c0 (j − k)}
N
i=0 j=0
132 Principles of Modern Wireless Communication Systems

N −1
1
= E c20 (i) c20 (i − k)
N2
i=0

N −1
1 1
= 2 1= ×N
N N2
i=1

1
=
N
Thus, the variance or basically the power of r00 (k), the autocorrelation of the random CDMA
spreading sequence is E r00 2
(k) = N1 . Also, once again, the autocorrelation corresponding
to a lag of k = 0 can be readily seen to be given as
N −1
1
E {r00 (0)} = E c0 (i) c0 (i)
N
i=0

N −1
1
= E c20 (i)
N
i=0

N −1
1
= 1
N
i=0

1
= ×N = 1
N
Therefore, one can succinctly summarize the autocorrelation properties of the random
spreading sequence as follows. For k = 0, r00 (k) = 1. For k = 0, r00 (k) is a random variable
with E {r00 (k)} = 0 and variance E r002
(k) = N1 . Let us now examine the cross-correlation
properties of the random CDMA spreading sequences, i.e., the correlation between the
spreading sequences c0 (i) , 0 ≤ i ≤ N − 1 and c1 (j) , 0 ≤ j ≤ N − 1. We denote by r01 (k)
the cross-correlation between spreading sequences c0, c1 corresponding to a lag k as

N −1
1
r01 (k) = c0 (i) c1 (i − k)
N
i=0

Once again, the expected value for any lag k can be computed as
N −1
1
E {r01 (k)} = E c0 (i) c1 (i − k)
N
i=0
 Code Division for Multiple Access (CDMA) 133

N −1
1
= E {c0 (i) c1 (i − k)}
N
i=0

N −1
1
= E {c0 (i)} E {c1 (i − k)}
N
i=0

N −1
1
= 0×0=0
N
i=0
2
Further, the variance E r01 (k) for any delay k is given as
⎧ ⎛ ⎞⎫
⎨ N −1 N −1 ⎬
2 1
E r01 (k) = 2 E c0 (i) c1 (i − k) ⎝ c0 (j) c1 (j − k) ⎠
N ⎩ ⎭
i=0 j=0

N −1 N −1
1
= E {c0 (i) c1 (i − k) c0 (j) c1 (j − k)}
N2
i=0 j=0

N −1
1
= E c20 (i) E c21 (i − k)
N2
i=0

N −1
1
= 2 1
N
i=0

1 1
= 2
×N =
N N
where we have again used the fact E {c0 (i) c1 (i − k) c0 (j) c1 (j − k)} is nonzero only if
i = j in the above derivation. Thus, once again, it can be seen that the cross-correlation
r01 (k) between two random CDMA spreading sequences c0 c1 is a random variable
with E {r01 (k)} = 0 and variance E r01 2
(k) = N1 . Thus, unlike the codes introduced in
Section 5.3, these random spreading codes do not satisfy the definition of exact orthogonality.
However, they are approximately orthogonal, in that the average value of the correlation is
zero and the power in the correlation is proportional to N1 which tends to 0 as N → ∞.

5.6 Multi-User CDMA

Now let us analyze the performance of a multi-user CDMA system using the properties
of the spreading sequences described above. Let a0 , a1 denote the symbols of users 0, 1
134 Principles of Modern Wireless Communication Systems

respectively. Further, let the transmit powers of the users be denoted by E |a0 |2 = P0 and
E |a1 |2 = P1 . On the downlink, as described in Section 5.2, the signal x0 (n) of the user 0
is derived by modulating the symbol a0 with the spreading sequence c0 (n) as

x0 (n) = a0 c0 (n)

Similarly, the signal x1 (n) is given as x1 (n) = a1c1 (n). The net downlink multiplexed signal
x (n) is formed from the constituent signals x0 (n) , x1 (n) as

x (n) = x0 (n) + x1 (n)

Assuming a simplistic AWGN channel model to begin with, the received signal at the user 0 is
given in the presence of additive white Gaussian noise as

y (n) = x (n) + w (n)

= a0 c0 (n) + a1 c1 (n) + w (n)

where the noise w (n) is such that E {w (n)} = 0 and E |w (n)|2 = σn2 . Further, the
whiteness property of the noise implies that E {w (n1 ) w (n2 )} = 0 if n1 = n2 . Again, as
described already in Section 5.2, we correlate with the spreading code c0 of the user 0 to
recover the symbol of the user 0 as
N −1
1
d0 = y (n) c0 (n)
N
i=0

N −1 N −1 N −1
1 1 1
= a0 c0 (n) c0 (n) + a1 c1 (n) c0 (n) + w (n) c0 (n),
N N N
i=0 i=0 i=0
Desired user Interferer Noise
−1
where the component N1 N i=0 a0 c0 (n) c0 (n) corresponds to the desired user signal when
−1
considering decoding at the user 0, while the component N1 N i=0 a1 c1 (n) c0 (n), which arises
due to a1 constitutes the interference, and is also termed multi-user interference in the context
−1
of CDMA. The last component N1 N i=0 w (n) c0 (n) corresponds to the noise at the receiver.
Below, we analyze and derive the statistical properties of each of the components described
 Code Division for Multiple Access (CDMA) 135

above. We start with the signal of the desired user, which can be simplified as
N −1 N −1
1 1
a0 c0 (n) c0 (n) = a0 c0 (n) c0 (n)
N N
i=0 i=0

= a0 r00 (0) = a0 ,

where r00 (0) corresponds to the autocorrelation of spreading code c0 (n) of the user 0 for a
delay n◦ = 0, which was simplified above in Section 5.5. Hence, the desired signal power
is given as E |a0|2 = P0 . Next, we calculate the power in the multi-user interference
component. This component, denoted by I1 , i.e., interference from the user 1 can be simplified
as
N −1
1
I1 = a1 c1 (n) c0 (n)
N
i=0

N −1
1
= a1 c1 (n) c0 (n)
N
i=0

= a1 r01 (0)

Hence, the interference power E |I1 |2 can be simplified as

E |I1 |2 = E |a1 r01 (0)|2

= E |a1 |2 E |r01 (0)|2

1 P1
= P1 × = .
N N
Thus, unlike the previous cases described in Eqs (5.4) and (5.5), the interference from the user 1
is not exactly zero due to the approximate orthogonality of the random spreading codes as
described in Section 5.5. However, the interference power decays as N1 , which is significantly
small for large values of spreading length N . The noise power can be calculated as follows.
Let w̃0 denote the noise, i.e.,

N −1
1
w̃0 = w (n) c0 (n)
N
i=0
136 Principles of Modern Wireless Communication Systems

It can be readily seen that w̃0 is a linear combination of Gaussian noise components w (n) and
is, therefore, Gaussian in nature. Further, the expected or average value of w̃0 can be obtained
as
N −1
1
E {w̃0 } = E w (n) c0 (n)
N
i=0

N −1
1
= E {w (n)} c0 (n)
N
i=0

N −1
1
= 0 × c0 (n)
N
i=0

=0

Thus, the average value of the noise is 0. Also, the average power in the noise can be calculated
as
N −1 N −1
1
E |w̃0 |2 = E w (n) c0 (n) w (m) c0 (m)
N2
n=0 m=0

N −1 N −1
1
= 2 E {w (n) c0 (n) w (m) c0 (m)}
N
n=0 m=0

N −1
1
= E |w0 (n)|2 |c0 (n)|2
N2
n=0

N −1
1 1
= 2 σn2 = × N σn2
N n=0
N2

σn2
= (5.8)
N
In the above simplification, we have used the fact that the noise samples w (n) are uncorrelated.
Therefore, if n = m, we have

E {w (n) c0 (n) w (m) c0 (m)} = E {w (n)} E {w (m)} c0 (n) c0 (m)

= 0 × 0 × c0 (n) c0 (m) = 0
 Code Division for Multiple Access (CDMA) 137

Thus, the effective noise power is N1 σn2 . Hence, for the above CDMA scenario, one can define
the Signal-to-Interference-Noise power Ratio (SINR) as
Signal Power
SINR =
Interference Power + Noise Power
P0
= P1 σn
2

N + N

P0
=N× (5.9)
P1 + σn2
From the above expression, it can be clearly seen that the signal power is reduced at the receiver
not only due to the noise but also due to the interference. Hence, CDMA is an interference-
limited system, due which the interference power has to be managed for better performance.
Also, notice that there is a factor of N in the numerator, which arises because the interference
and noise power is suppressed by a factor of N . This is termed the spreading gain of the
CDMA system, which is equal to the spreading length.
The above expression in Eq. (5.9) for SINR in a CDMA network can be generalized to the
context of greater and two users. Consider K + 1 CDMA users 0, 1, ..., K transmitting with
powers P0 , P1 , ..., PK respectively on codes c0 (n) , c1 (n) , ..., cK (N ). Hence, the received
signal y (n) is given as

y (n) = a0 c0 (n) + a1 c1 (n) + . . . + aK cK (n) + w (n) ,

K
= ak ck (n) + w (n)
k=0

The corresponding SINR after for the user 0 after correlating and decoding with the spreading
code c0 (n) corresponding to the user 0 can be similarly obtained as
P0
SINR = P1 P2 PK σn
2

N + N + ... + N + N
P0
=N× K
(5.10)
k=0 Pk + σn2

5.7 Advantages of CDMA

In this section, we systematically investigate the advantages of CDMA-based cellular systems

over 1G FDMA and 2G TDMA based cellular systems.
138 Principles of Modern Wireless Communication Systems

5.7.1 Advantage 1: Jammer Margin

An important advantage of CDMA over conventional cellular systems is jammer suppression.
A jammer is basically a malicious user in a communication network who transmits with a
very high power to cause interference, thus leading to disruption of communication links. This
is shown schematically in Figure 5.8. Jammers are of significant concern, especially in the
context of highly secure communication systems such as those used for military and defense
purposes. The effect of jammer suppression in a CDMA system can be understood as follows.
Consider a communication system in which the signal x (n) of the power P is received in the
2
presence of additive white Gaussian noise w (n) of power σw . The baseband system model for
this communication system can be expressed as

y (n) = x (n) + w (n)

Figure 5.8 Disruption by jammer in wireless communication

Hence, the SNR at the receiver is SNR = σP2 . However, in the presence of a jamming signal
w

xj (n) of power Pj , the received signal y (n) is

y (n) = x (n) + xj (n) + w (n)

Thus, the jammer interferes with the signal reception and the signal-to-interference-noise
P
power ratio (SINR) can be calculated as SINR = Pj +σ 2 . Thus, the jammer has a significant
w

disruptive impact on the communication signal. Consider now a CDMA system in which the
transmitted signal x (n) is a spread-spectrum signal. As shown in the section above, the SINR
 Code Division for Multiple Access (CDMA) 139

for a CDMA scenario is given as

P
SINR = Pj σw
2 (5.11)
N + N

Thus, it can be seen that the jamming power Pj is suppressed by a factor of N . Moreover,
as the spreading factor N increases, the jammer suppression increases, minimizing the impact
of the jammer on the communication system. This is termed jammer suppression in CDMA
systems. Hence, CDMA which is inherently tolerant to jamming attacks is highly attractive for
defense applications. In fact, the earliest applications of CDMA were in the context of tactical
military secure communications, which were resistant to attacks by jammers. Only later were
the benefits of CDMA realized and applied in the context of civilian cellular networks. Also,
it is worthwhile noting that the gain of N in this context of jammer suppression is also termed
the jammer margin. Thus, the jammer margin is equal to N , i.e., the spreading length of the
CDMA codes.

5.7.2 Advantage 2: Graceful Degradation

Graceful degradation is another key property of CDMA-based wireless networks and as we
shall see soon, allows for much more efficient interference management, which ultimately
leads to universal frequency reuse and higher spectral efficiency. Consider the expression for
the SINR at the user 0 derived in Eq. (5.10). At this point, assume that another user, i.e., a user
with index K + 1 joins the network. Let PK+1 denote the corresponding transmission power
of this (K + 1)th user and aK+1 , cK+1 (n) denote his transmitted symbol and spreading code
respectively. The SINR of the user 0 now changes to
P0
SINR = P1 P2 PK PK+1 2
σn
N + N + ...+ N + N + N

P0
=N× K+1
k=0 Pk + σn2
Thus, the addition of a new user K + 1 with power PK+1 only causes an incremental
interference of PK+1
N at the user 0. Further, in general, at any user i = (K + 1), the additional
P
interference due to the introduction of this new user is K+1N . Therefore, the addition of the
new user K + 1 does not adversely affect any single user. Rather, the additional interference
caused by this new user is shared amongst all the existing users in the system leading to
interference distribution. This sharing of the interference by all the existing users leads to
140 Principles of Modern Wireless Communication Systems

a graceful degradation of the SINR at each user. This is termed the graceful degradation
property of CDMA systems. This idea of graceful degradation is key to understanding the big
advantage of CDMA networks, i.e., universal frequency reuse, which is described next.

5.7.3 Advantage 3: Universal Frequency Reuse

To understand the concept of universal frequency reuse, we have to begin by understanding
the frequency allocation in convention, i.e., 1G and 2G cellular systems. Consider a cellular
network organized into cells as shown in Figure 5.9. Consider two adjacent cells C0 , C1 shown
in the figure. Assume now that the same frequency f is allotted for transmission to users in
both C0 , C1 . Let x0 (n) with power P0 denote the signal of the user on the frequency f in the
cell 0, while x1 (n) with power P1 denotes the signal of the user in the cell 1. Since both the
signals are being transmitted on the identical frequency f , they will interfere with each other.
More specifically, the received signal y0 (n) at the user 0 is given as

y0 (n) = x0 (n) + x1 (n) + w (n)

Signal Interferer from C1 Noise

Figure 5.9 Intercell interference for the user 0 on the cell edge

Hence, the SINR at the user 0 is given as SINR = P1P+σ 0

2 . This is similar to the jamming
w

interference case described in Eq. (5.11). Thus, if the same frequency f is allocated in adjacent
cells, it will cause heavy interference and degradation of user SINR results from adjacent cell
 Code Division for Multiple Access (CDMA) 141

interference. Thus, in a typical 1G or 2G cellular network such as GSM, only a fraction of

the total available frequencies are allocated in each cell, carefully avoiding the allocation of
the same frequency in adjacent cells. For instance, as can be seen from the hexagonal-lattice-
based cellular structure in Figure 5.10, each hexagonal cell has 6 neighbours. Hence, to avoid
adjacent cell interference, any of the frequencies allocated to C0 cannot be allocated to its
neighbours C1 , C2 , ..., C6 . This holds true for all the cells in the network. Hence, only 71 of
the total available frequency bands can be allocated to each cell. This factor 17 is termed the
frequency-reuse factor of the cellular network. Thus, since only a fraction of the frequencies
are used in the cell, the total spectral efficiency is proportional to the frequency-reuse factor,
resulting in a rate which is 17 compared to that of using all the available bandwidth, since the
capacity is linearly related to bandwidth.

Figure 5.10 Grid or lattice of hexagonal cells

However, now consider the same scenario in the context of a CDMA network. Again,
assume that the same frequency f is allotted for transmission to users in both C0 , C1 . However,
let x0 (n) with power P0 is now transmitted on code c0 (n), while x1 (n) with power P1 is
transmitted in the cell 1 on the random code c1 (n). Hence, now similar to the jammer scenario
in a CDMA system, the interference caused by the user on the identical frequency f in the
adjacent cell is now reduced by a factor of N to PN1 . Therefore, the SINR is now given as,

P0
SINR = P1 2
σw
N + N

This is the result of graceful degradation described in the previous advantage in Section 5.7.2.
Thus, the interference of each user is limited to a fraction N1 of the interferer power. Hence,
basically the jammer margin in defence applications, can be used for adjacent cell interference
power suppression in modern cellular networks! This is a great advantage of CDMA, which
142 Principles of Modern Wireless Communication Systems

implies that the same frequency bands can be used in all cells across the network. Another way
of stating this is that the fraction of bands used in each cell is 1, i.e., all the bands. Therefore,
this is termed universal frequency reuse or equivalently, as a cellular network with frequency
reuse factor 1. Thus, compared to GSM, which uses only 17 of the frequency bands in each
cell, CDMA can use all the available frequency bands in each cell. This right away leads to an
increase of the spectral efficiency and resulting capacity by a factor of 7. Thus, CDMA-based
cellular networks have a much higher capacity compared to conventional 1G and 2G cellular
networks. This has led to a widespread adoption and embrace of CDMA-based technologies
for mobile communication.

5.7.4 Multipath Diversity and Rake Receiver

Another important advantage of CDMA is its ability to achieve diversity gain via multipath
scatter components. This is termed multipath diversity and is achieved through coherent
combining of the multipath-signal components employing a rake receiver. Consider a
multipath frequency-selective channel with several delayed signal paths. We have seen in
earlier chapters that such a multipath frequency-selective channel can be modelled as a Finite
Impulse Response (FIR) channel filter with channel taps h (0) , h (1) , . . . , h (L − 1). The
received symbol y (n) can be expressed as

y (n) = h (0) x (n) + h (1) x (n − 1) + . . . + h (L − 1) x (n − L + 1) + w (n)

L−1
= h (l) x (n − l) + w (n)
l=0

Recall that this represents a frequency-selective or intersymbol interference-limited channel

since the current output y (n) depends not only on the current symbol x (n), but also L − 1
previous input symbols x (n) , x (n − 1) , . . . , x (n − L + 1). Consider now a CDMA signal
in which x (n) = a0c0 (n). To simplify this illustration, we currently consider a single user and
this can be readily extended to multi-user scenarios as will be seen in the subsequent sections.
Substituting this in the expression for y (n) above, the received signal across a frequency-
selective channel in a CDMA system is given as
L−1
y (n) = h (l) a0 c0 (n − l) + w (n)
l=0
 Code Division for Multiple Access (CDMA) 143

As done previously, let us correlate with c0 (n) to recover the symbol corresponding to the user
0. This operation can be expressed as
N −1
1
d (0) = y (n) c0 (n)
N
n=0

N −1 L−1
1
= h (l) a0 c0 (n − l) + w (n) c0 (n)
N
n=0 l=0

N −1 L−1 N −1
1 1
= h (l) a0 c0 (n − l) c0 (n) + w (n) c0 (n)
N N
n=0 l=0 n=0
w̃0
2
Recall from the analysis in Eq. (5.8) that the noise w̃0 is Gaussian of power E w̃0 2 = σNw .
The first term in the above expression can be split into two components: one corresponding to
l = 0 and the other corresponding to l == 0. Simplifying, we have
N −1 L−1 N −1
1 1
d (0) = h (0) a0 c0 (n) c0 (n) + h (l) a0 c0 (n − l) c0 (n) + w̃0
N N
n=0 l=1 n=0

N −1 L−1 N −1
1 1
= h (0) a0 c0 (n) c0 (n) + a0 h (l) c0 (n − l) c0 (n) + w̃0
N N
n=0 l=1 n=0

L−1
= h (0) a0 r00 (0) + a0 h (l) r00 (l) + w̃0
l=1

At this point, we will employ the following simplifying approximation. Consider the quantity
r00 (l). As demonstrated in Section 5.5, for l = 0, r00 (l) is a random variable with mean 0
and power N1 . Observe that the power tends to 0 as N → ∞. Hence, for large values of N ,
r00 (l) ≈ 0. Employing this approximation in the above expression, and noting that r00 (0) =
1, we have

d (0) = h (0) a0 + w̃0

Observe now a very interesting property. Even though there is intersymbol interference in
the above channel, we are able to extract the signal corresponding to h (0), i.e., delay 0 by
correlation with the spreading code c0 (n). This is once again because of the approximate
orthogonality property that was introduced in the previous sections. Moreover, interestingly,
144 Principles of Modern Wireless Communication Systems

one can repeat this process by individually correlating with delayed versions of the spreading
sequence c0 (n − v) , 1 ≤ v ≤ L − 1 to extract the multipath components corresponding to
h (1) , h (2) , . . . , h (L − 1). Thus, correlating with c0 (n − v), the resulting statistic d (v) can
be simplified as,
N −1
1
d (v) = y (n) c0 (n − v)
N n=0

N −1 L−1
1
= h (l) a0 c0 (n − l) + w (n) c0 (n − v)
N
n=0 l=0

N −1 L−1 N −1
1 1
= h (l) a0 c0 (n − l) c0 (n − v) + w (n) c0 (n − v)
N N
n=0 l=0 n=0
w̃v

w̃2
Once again, it can be easily seen that the noise w̃v is Gaussian with variance E |w̃v |2 = Nv .
Further, splitting the first term into two components corresponding to l = v and l = v , one can
derive the expression for d (v) , 1 ≤ v ≤ L − 1 as
N −1
1
d (v) = h (v) a0 c0 (n − v) c0 (n − v)
N
n=0

L−1 N −1
1
+ h (l) a0 c0 (n − l) c0 (n − v) + w̃v
N
l=0, l=v n=0

N −1
1
= h (0) a0 c0 (n − v) c0 (n − v)
N
n=0

L−1 N −1
1
+ a0 h (l) c0 (n − l) c0 (n − v) + w̃v
N n=0
l=0, l=v

L−1
= h (v) a0 r00 (0) + a0 h (l) r00 (l − v) + w̃v
l=1

≈ h (v) a0 + w̃v

where we have again employed the approximation r00 (l − v) ≈ 0, l = v in the above

simplification. Now, one can process that extracted components d (0) , d (1) , . . . , d (L − 1)
 Code Division for Multiple Access (CDMA) 145

as follows. Employing vector notation, the components can be expressed as

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
d (0) h (0) w̃0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ d (1) ⎥ ⎢ h (1) ⎥ ⎢ w̃ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ 1 ⎥
⎢ .. ⎥=⎢ .. ⎥ a0 + ⎢ . ⎥
⎢ . ⎥ ⎢ . ⎥ ⎢ .. ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
d (L − 1) h (L − 1) w̃L−1
d h w̃

It can be readily seen that the above system is now similar to the multiple receive antenna
system, i.e., receive diversity system with channel coefficients h (0) , h (1) , . . . , h (L − 1).
The model can be, therefore, be succinctly expressed in vector notation as

d = ha0 + w̃

h
Hence, the optimal combiner is the Maximum Ratio Combiner (MRC) given by h . Also
2
observe that the power of each noise component w̃v is E |w̃v | = 1. Denoting the symbol
power E |a0 |2 by P and combining the observation vector d with the MRC yields the SNR
2
h P
SNR = σ 2
w
N
! P
= N × |h (0)|2 + |h (1)|2 + . . . + |h (L − 1)|2 2
σw
2
h

Observe that the above expression is similar to the SNR for the multiple receive
antenna system, in that there is a factor h 2 in the numerator, where h is the vector
of frequency-selective channel coefficients h = [h (0) , h (1) , . . . , h (L − 1)]T . This is in
addition to the spreading gain factor N . Thus, this is equivalent to the performance of a system
with diversity order L. This is the essence of multipath diversity and CDMA is able to exploit
this multipath diversity by correlating with the spreading code c0 (n − v) corresponding to
different lags 0 ≤ v ≤ L − 1, and combining the individual components employing MRC.
This receiver structure in CDMA is termed the rake receiver and the diversity gain thus
achieved is termed multipath diversity since it is extracted from the multipath components.
This multipath diversity arising out of rake combining is a unique feature of CDMA and
146 Principles of Modern Wireless Communication Systems

significantly improves its performance over wireless channels because of the higher diversity
order of decoding.

5.8 CDMA Near–Far Problem and Power Control

In this section, we introduce a unique aspect of the CDMA systems which is termed the near–
far problem. Recall from Eq. (5.9) that the SINR at the user 0 of a 2 user CDMA system is
given as,

P0
SINR = P1 2
σn
(5.12)
N + N

Consider, now a scenario where the user 1 is much closer to the base station than the user 0 as

Figure 5.11 Near-far problem in CDMA networks

√
shown in Figure 5.11. Specifically, let d0 = N d1. Let the transmitted power of each user at
the base station be denoted by PT . Since the radiated electromagnetic power decays at the rate
of d2 in free space, we have,
PT PT
P1 = 2
=
d0 N d21

PT
P0 =
d21
 7
Orthogonal Frequency-Division
Multiplexing

7.1 Introduction

Orthogonal Frequency-Division Multiplexing (OFDM) forms the basis for 4G, i.e., Fourth
Generation wireless communication systems. OFDM is used in 4G wireless cellular standards
such as Long-Term Evolution (LTE) and WiMAX (Worldwide Interoperability for Microwave
Access). OFDM is a key broadband wireless technology which supports data rates in excess of
100 Mbps. Similarly, the wireless local area (LAN) standards such as 802.11 a/g/n are based
on OFDM. Next we describe multicarrier transmission, which is the motivation and key idea
behind OFDM.

7.2 Motivation and Multicarrier Basics

Consider a bandwidth B = 2W available for communication, where W is the one-sided

bandwidth, or, in other words, the maximum frequency. For a single carrier communication
system, the symbol time T is given as

1
T =
B

1
basically implying that symbols can be transmitted at intervals of B seconds each. Therefore,
the symbol rate is given as

1
Rate = =B (7.1)
1/B
 Orthogonal Frequency-Division Multiplexing 231

Such a system is termed a single-carrier communication system. In such a system, a single

carrier is employed for the entire baseband bandwidth of B . Therefore, roughly speaking, the
symbols are transmitted as symbol X(0) from 0 ≤ t < T , symbol X(1) from T ≤ t < 2T ,
and so on, i.e., roughly one symbol transmitted every T = B1 seconds.
Consider now dividing the total bandwidth B into N sub-bands of bandwidth B/N each as
shown in Figure 7.1. Each subcarrier can now be represented by a subcarrier. Therefore, the
B B
subcarriers are placed at . . . , − N , 0, N , . . ., as shown in the figure. For instance, consider the
bandwidth B = 256 kHz with N = 64 subcarriers. The bandwidth per sub-band is equal to
256
64 = 4 kHz, which is also the frequency spacing between the subcarriers. We now implement
a multi-carrier transmission system as follows. Consider the ith subcarrier at the frequency
B
fi = i N , with − N2 − 1 ≤ i ≤ N2 . Let Xi denote the data transmitted on the ith subcarrier.
Then, the signal si (t) corresponding to the ith subcarrier is given as

B
si (t) = Xiej2πfi t = Xiej2πi N t

where fi is the ith subcarrier centre frequency, as described above, and ej2πfi t is the ith
subcarrier. The above equation shows the data modulation process over the ith subcarrier.
The N different data symbols Xi are modulated over the N different subcarriers with centre
frequencies fi . Hence, there are a total of N data streams. Next we illustrate the scheme for
multicarrier transmission.
Subcarrier
Sub-band spacing N subcarriers

-(N/2-1)B/N -B/N B/N 2B/N (N/2)B/N

Total bandwidth B
Figure 7.1 Multi-carrier concept
232 Principles of Modern Wireless Communication Systems

7.2.1 Multicarrier Transmission

Consider now the different modulated signals si (t) corresponding to the N different
subcarriers. These signals are then superposed at the transmitted to form the composite signal
s (t) given as

s (t) = si (t)
i

= Xi ej2πfi t
i
B
= Xi ej2πi N t (7.2)
i

This composite signal s (t) is then transmitted over the wireless channels. Thus, N different
data streams are transmitted over N subcarriers in parallel in this multicarrier system. At the
receivers, the individual data streams have then to be isolated from the composite signal s (t)
above. This is accomplished as follows. Consider the signal y (t) received as

y (t) = s (t) = Xi ej2πfi t

i

For simplicity, to illustrate the demodulation procedure at the receiver, we have assumed noise
to be absent above. We will consider the general case of a noisy received signal later. From the
expression for the composite signal s (t) in Eq. (7.2), it can be readily seen that the expression
on the right-hand side is indeed the Fourier series respresentation s (t), corresponding to the
fundamental frequency f0 = (B) N and the various Xi representing the Fourier coefficients.
B
Indeed, all the frequencies i N are multiples of the fundamental frequency f0 = T10 = N B
.
Therefore, to extract Xl , which is the Fourier coefficient corresponding to the frequency
fl = lf0 , one needs to follow the procedure similar to compute the Fourier series as
N
T0
j2πfl t
∗ B B B B
f0 y (t) e dt = Xi ej2πi N t e−j2πl N t dt
0 N 0 i

N
B B
= Xi ej2π(i−l)f0 t dt
N 0
i
 Orthogonal Frequency-Division Multiplexing 233

N N
B B B B
= Xl dt + Xiej2π(i−l)f0 t dt
N 0 N 0
i=l
i=l

N
B B
= Xl + Xi ej2π(i−l)f0 t dt
N 0
i=l
=0

= Xl

T
where we have used the fact that 0 0 ej2π(i−l)f0 t dt = 0 for i = l , since this is basically
integrating a sinusoid of frequency (i − l) f0 , which is a multiple of the fundamental frequency
f0 over the period T0 . Therefore, since there are an integer number of cycles of the sinusoid
of frequency (i − l) f0 , this integral is 0. In fact, this basically implies that the different
sinusoids ej2πif0 t and ej2πlf0 t are orthogonal. It is this key property of orthogonality which
helps extract the different streams Xi modulated over the different subcarriers. This property
of orthogonality can be summarized as
⎧
N/B ⎨0 i=l
B
j2π(i−l) N t
e =
0 ⎩N i= l
B

Therefore, all the subcarriers other than the l th subcarrier are orthogonal to the l th
∗
subcarrier. Further, observe that multiplying with ej2πfl t and integrating is basically
coherent demodulation, i.e., demodulation with the carrier matched to the subcarrier frequency
B
fl = l N . Thus, Xl , the data modulated on the different subcarriers, can be conveniently
recovered by coherently demoudulating with each of the subcarriers corresponding to
l = − N2 − 1 , . . . , N2 . The above scheme of transmission on multiple orthogonal subcarriers
and the associated data recovery at the receiver is termed MultiCarrier Modulation (MCM).
Also observe that the window of time associated with detection of this multicarrier signal
is N 1
B = f0 = T0 , which is basically the time period of integration. Hence, MCM basically
transmits N symbols using N subcarriers in a time period of N B . The symbol rate is, therefore,
N
N/B = B . Thus, the overall symbol rate in single carrier vs multicarrier systems is unchanged.
The transmitter and receiver block schematics for this MCM system are shown in figures 7.2
and 7.3 respectively.
234 Principles of Modern Wireless Communication Systems

Composite
signal

Bank
S/P Summer
of
Symbols Demux + To channel
modulators

Serial to parallel i th data stream is Sum all the

conversion because modulated onto subcarriers
we are transmitting N the i th subcarrier
symbols in parallel

Figure 7.2 Multicarrier modulation transmitter

N Information
symbols

Bank
of
Repeater P/S
correlators
y(t) Mux Serail symbol
or
from channel demodulators stream

Coherent Parallel to serial

demodulation multiplexing
with exp (-j2pflt)

Figure 7.3 Multi-carrier modulation receiver

It is very important now to note the following fact. Observe from Eq. (7.1) and the above
rate for an MCM system. It is clear that the symbol rate in both these systems is exactly
identical, i.e., B . The single-carrier system transmits each symbol in time B1 , while the
MCM system transmits N symbols in parallel in time N B . What then is the advantage of
an MCM system over the single-carrier system? To understand this, consider an example
with a transmission bandwidth of B = 1.024 MHz, i.e., 1024 kHz. As seen in an earlier
chapter, notice that this bandwidth B is much greater than the coherence bandwidth Bc
which is typically around 250 kHz, i.e., Bc ≈ 250 kHz. Therefore, since the transmission
bandwidth B >> Bc , the single-carrier system experiences frequency-selective fading and
inter-symbol interference. However, consider an OFDM system with employs N = 256
subcarriers in the same bandwidth. The bandwidth per subcarrier is Bs = 1024
256 = 4 kHz. It can
be readily seen that the subcarrier bandwidth of 4 kHz is significantly lower than the coherence
 Orthogonal Frequency-Division Multiplexing 235

B
bandwidth of 250 kHz. Thus, since N << Bc , each subcarrier experiences flat fading.
Hence, there is no inter-symbol interference in the data transmitted on any of the subcarriers.
Thus, the most critical and key benefit of this MCM system is that through parallel transmission
using multiple narrowband subcarriers, it eliminates the Inter-Symbol interference (ISI), thus
avoiding distortion of the received symbols.
However, the above MCM system suffers from a significant bottleneck. Implementing the
bank of N modulators and N demodulators with closely spaced subcarrier frequencies is an
extremely challenging task. This was solved by the key idea proposed by Weinstein and Ebert
in 1972, in the paper titled "Data Transmission by Frequency Division Multiplexing using the
Discrete Fourier Transform". Both of them were engineers at Bell Telephone Laboratories.
Their idea can be described as follows. Consider the MCM transmt signal s (t). Observe that it
is band-limited to the bandwidth B (total bandwidth). Therefore, the Nyquist sampling rate is
B and the associated sampling time is Ts = B1 . Consider now the composite MCM sigal given
in Eq. (7.2). The uth sample at time instant uTs = Bu is given as
B u
s (uTs ) = x (u) = Xi ej2πi N B
i

iu
x (u) = Xi ej2π N
i
DFT

Observe from the expression above that the sample x (u) is basically the In-
verse Discrete Fourier Transform (IDFT) coefficient of the information symbols
X (0) , X (1) , . . . , X (N − 1) at the uth time point. Thus, the Inverse Fast Fourier Transform
(IFFT) can be conveniently employed to generate the sample MCM signal. This scheme of
generating the composite transmit signal through IDFT was proposed by Weinstein and Ebert
in 1971. Thus, it drastically reduces the complexity of implementing an OFDM system since
it eliminates the need for the bank of modulators corresponding to the different subcarrier
frequencies. This technique, where the MCM signal is generated by employing the IFFT
operation is termed Orthogonal Frequency Division Multiplexing, or OFDM. At the receiver,
to recover the information symbols, one can correspondingly employ an FFT operation.
Schematic figures of the OFDM transmitter and receiver with the IFFT and FFT blocks are
given in figures 7.4 and 7.5 respectively.
236 Principles of Modern Wireless Communication Systems

Figure 7.4 OFDM transmitter schematic with IFFT

Figure 7.5 OFDM receiver schematic with FFT

7.2.2 Cyclic Prefix in OFDM

In this section, we explain the concept of cyclic prefix, which is an important component
of an OFDM system. Consider a frequency-selective channel modelled with channel taps
h (0) , h (1) , . . . , h (L − 1). Thus, the received symbol y at a given time instant n can be
expressed as

y (n) = h (0) x (n) + h (1) x (n − 1) + . . . + h (L − 1) x (n − L + 1),

ISI component

from which it can be seen that the received symbol y (n) at the time instant n experiences inter
symbol interference from the previous L − 1 transmitted symbols. Consider now two OFDM
symbols as follows. Let x (0) , x (1) , . . . , x (N − 1) denote the IFFT samples of the modu-
lated symbols X (0) , X (1) , . . . , X (N − 1), while x̃ (0) , x̃ (1) , . . . , x̃ (N − 1) denote the
 Orthogonal Frequency-Division Multiplexing 237

IFFT samples of the previous modulated symbol block X̃ (0) , X̃ (1) , . . . , X̃ (N − 1). Thus,
the samples corresponding to these two blocks of OFDM symbols are transmitted sequentially
as

x̃ (0) , x̃ (1) , . . . , x̃ (N − 1) , x (0) , x (1) , . . . , x (N − 1)

Previous block Current block

Now, consider the received symbol y (0) corresponding to the transmission of x (0). This can
be expressed as

y (0) = h (0) x (0) + h (1) x̃ (N − 1) + . . . + h (L − 1) x̃ (N − L + 1) .

ISI from previous OFDM symbol

It can be seen from the above equation that the received symbol y (0) experiences inter-symbol
interference from x̃ (N − 1) , x̃ (N − 2) , . . . , x̃ (N − (L − 1)). Thus, there is inter-OFDM
symbol interference in this new OFDM system. The initial samples of the current OFDM
symbol block are being subject to interference from the N − 1 samples of the previous OFDM
block. This is shown in Figure 7.6. Similarly, the received symbol y (1) is given as

y (1) = h (0) x (1) + h (1) x (0) h (2) x̃ (N − 1) + . . . + h (L − 1) x̃ (N − L + 2),

ISI from previous OFDM symbol

which can again be seen to experience inter-OFDM symbol interference from the previous
OFDM block symbols x̃ (N − 1) , x̃ (N − 2) , . . . , x̃ (N − L + 2). Let us now consider a
modified transmission scheme as follows. To each transmitted OFDM sample stream, we pad
the last Lc symbols to make the transmitted stream as follows.
x̃ (0) , x̃ (1) , . . . , x̃ (N − 1) , x (N − Lc ) , x (N − Lc + 1) , . . . , x (N − 1) x (0) , x (1) , . . . , x (N − 1) .
| {z }| {z }| {z }
Previous block Cyclic prefix Current block

Initial samples, of subject to inter OFDM symbol interference

OFDM symbol size = N/B samples

Figure 7.6 Inter-OFDM symbol interference

238 Principles of Modern Wireless Communication Systems

Observe that we are prefixing the transmitted sample block x (0) , x (1) , . . . , x (N − 1) of
the current block with the Lc samples x (N − Lc ) , x (N − Lc + 1) , . . . , x (N − 1). Further,
this prefix is cyclic in nature, since the same samples from the end of the block are being cycled
towards the beginning. Therefore, this is known as the cyclic prefix and is an important aspect
of OFDM systems. Consider now the received symbol corresponding to x (0). This is given as

y (0) = h (0) x (0) + h (1) x (N − 1) + . . . + h (L − 1) x (N − L + 1)

ISI from same OFDM symbol

The inter-symbol interference can be seen to now be from

x (N − 1) , x (N − 2) , . . . , x (N − L + 1), if Lc ≥ L − 1. Thus, with the cyclic prefix
of appropriate length, i.e., Lc ≥ L − 1, inter-OFDM symbol interference can be avoided and
inter-symbol interference is restricted to samples from the same OFDM symbol. Therefore,
the samples y (0) , y (1) , . . . , y (N − 1) are given as

y (0) = h (0) x (0) + h (1) x (N − 1) + . . . + h (L − 1) x (N − L + 1)

y (1) = h (0) x (1) + h (1) x (0) + . . . + h (L − 1) x (N − L + 2)

..
.
y (N − 1) = h (0) x (N − 1) + h (1) x (N − 2) + . . . + h (L − 1) x (N − L)

It can now be clearly seen that the output y (n) is a circular convolution between the channel
filter h (n) and the input x (n).
This can, therefore, be expressed as

[y (0) , y (1) , . . . , y (N − 1)] = [h (0) , h (1) , . . . , h (L − 1) , 0, . . . , 0] ∗N [x (0) ,

x (1) , . . . , x (N − 1)]

where ∗N denotes circular convolution of modulo N . Therefore, the output y can be written as

y = h ∗N x

Therefore, taking the DFT of y (n) at the output, we have

Y (k) = H (k) X (k) , 0 ≤ k ≤ N − 1 (7.3)

 Orthogonal Frequency-Division Multiplexing 239

where Y (k) , 0 ≤ k ≤ N − 1, denotes the N -point DFT of y (n). Similarly, X (k) denotes
the N -point DFT of x (n). Further, observe that the samples x (n) have been generated as the
IDFT of X (n). Therefore, the DFT of the samples x (n) yields back the original transmitted
symbols X (n). The coefficients H (k) denotes the DFT of the zero-padded channel filter,

h (0) , h (1) , . . . , h (L − 1) , 0, . . . , 0 . (7.4)

(N −L)

Thus, observe that Eq. (7.3) represents the flat-fading channel across the kth subcarrier in
the OFDM system. The quantity Y (k) represents the output symbol, while H (k) denotes
the equivalent flat-fading channel coefficient. This holds true for each subcarrier k, i.e., for
0 ≤ k ≤ N − 1. Thus, the frequency-selective fading channel is converted into a group of
narrowband flat-fading channels, one channel across each subcarrier. Observe that if a single
carrier system was used, and the symbols X (0) , X (1) , . . . , X (N − 1) were transmitted
directly then the received symbol y (n) would be given as

y (n) = h (0) X (n) + h (1) X (n − 1) + . . . + h (L − 1) X (n − L + 1)

Each symbol X (n) would experience inter-symbol interference of L − 1 past symbols.

Therefore, using this novel scheme of OFDM, we have been able to totally eliminate
the inter-symbol interference arising out of the frequency-selective nature of the channel.
The set of parallel flat-fading channels can be summarized by the expressions

Y (0) = H (0) X (0)

Y (1) = H (1) X (1)

..
.
Y (N − 1) = H (n − 1) X (N − 1)

This conversion of the frequency-selective wideband channel into N narrowband flat-fading

channels is shown schematically in Figure 7.7. Also, the modified transmitter and receiver
schematics with the blocks corresponding to the cyclic prefix are given in Figures 7.8 and
7.9 respectively. Now, considering the noise at the receiver, the received symbol Y (k) can be
expressed as

Y (k) = H (k) X (k) + N (k) (7.5)

240 Principles of Modern Wireless Communication Systems

Figure 7.7 OFDM parallel subchannels

where N (k) denotes the noise across the kth subcarrier. A simple detection scheme for
X (k) is to use the zero-forcing detector for the subcarrier as

1 N (k)
X̂ (k) = Y (k) = X (k) +
H (k) H (k)
Ñ (k)

Further, for a simplistic BPSK or QPSK-modulated transmission, the coherent or matched

filter detector can be simply obtained by multiplying with H ∗ (k), i.e., the complex conjugate
of H (k) as

H ∗ (k) Y (k) = |H (k)|2 X (k) + H ∗ (k) N (k)

N (k)

X(0) x(0) Samples at

rate B

S/P IFFT P/S Add

Symbols Demux N Mux CP To channel

X(N-1) x(N-1)
Serial to parallel N = # of Block for addition of
conversion Subcarriers cyclic prefix (CP)

Figure 7.8 OFDM transmitter schematic with CP

 Orthogonal Frequency-Division Multiplexing 241

Also, one can employ the MMSE detector as

H ∗ (k)
X̂MMSE (k) = Y (k)
|H (k)|2 + σn2

The above equation gives the MMSE receiver across the kth subcarrier in this OFDM system.
N information
symbols

x(0) X(0)

Samples CP S/P N-pt P/S

from channel removal demux FFT mux

x(N-1) X(N-1)
Removes CP symbols To demodulate Parallel to serial
subject to inter OFDM symbol on each multiplexing
symbol interference subcarrier

Figure 7.9 OFDM receiver schematic with CP

7.2.3 Impact of Cyclic Prefix on Data Rate

Consider the transmitted samples x (n) with the cyclic prefix, as given below.

x (N − Lc ) , x (N − Lc + 1) , . . . , x (N − 1) x (0) , x (1) , . . . , x (N − 1) .
Cyclic prefix Current block

The minimum required length of the cyclic prefix is L − 1 as was described in the previous
section. Also, observe that L − 1 is the delay spread of the wireless channel. Therefore, it fol-
lows that the length of the cyclic prefix should be greater than the delay spread of the channel.
However, since the samples in the tail, i.e., x (N − Lc ) , x (N − Lc + 1) , . . . , x (N − 1) are
simply repeated in the beginning, they do not constitute any additional information. Hence, the
effect of the addition of a long CP is lost in the throughput of the system. More specifically,
242 Principles of Modern Wireless Communication Systems

the loss in efficiency can be calculated as

Cyclic prefix
Loss in effciency =
Total OFDM symbol length
L−1
=
N +L−1
L−1
=
N +L−1
However, as the block length N becomes very large, we have

L−1
lim →0
N →∞ N +L−1

Thus, the loss in throughput approaches 0 as the number of subcarriers N increases, for a fixed
length of the delay spread L. Also observe that as the number of subcarriers N increases, the
symbol time N B increases as shown in Figure 7.10. Increasing N results in increasing OFDM
symbol time, thus restricting the ISI to a small fraction of the OFDM symbol block, i.e., the
L
fraction N is progressively smaller. However, as the block length N increases, the decoding
delay at the receiver also increases as one has to wait for arrival of the entire block of N samples
before it can be demodulated. Hence, there is a trade-off for increasing N vs decoding delay.
Now, we present another intuitive framework to understand the effect of various parameters.
As we have said previously, the duration of the cyclic prefix has to be greater than the delay
spread.

Figure 7.10 Inter OFDM symbol interference with increasing OFDM symbol time
 Orthogonal Frequency-Division Multiplexing 243

Therefore, we need

Lc × T s ≥ T d

where Ts denotes the sample time and Td denotes the delay spread. Also, the sample time
Ts = B1 , where B is the total bandwidth of the system and Td = B1c , where Bc is the coherence
bandwidth of the system. The above condition implies
Td
Lc ≥ ,
Ts
B
=
Bs
Combining this with the earlier condition that N >> Lc for efficiency in terms of the effective
data rate, we have

B
N >> Lc ≥ .
Bc

B
This can also be recast as Bc >> N . Interestingly, this is the same condition for frequency flat
B
fading across each subcarrier since this implies that the subcarrier bandwidth N is required to
be much less than the coherence bandwidth Bc . Thus, an appropriately designed OFDM system
converts a frequency-selective fading channel into a set of parallel narrowband flat-fading
channels across the subcarriers. The example next illustrates an OFDM system design example.

7.3 OFDM Example

In this section, we consider a practical WiMAX example to illustrate the impact of the various
parameters in the design of a complete OFDM system. As already stated in the beginning,
WiMAX, which stands for Worldwide Interoperability for Microwave Access, is a prominent
4G wireless standard. The total number of subcarriers N = 256, with a bandwidth of 15.625
kHz per subcarrier. Therefore,
B
= 15.625 kHz
N

⇒ B = N × 15.625 = 256 × 15.625

= 4 MHz
244 Principles of Modern Wireless Communication Systems

Also, observe that the subcarrier bandwidth is less than the coherence bandwidth, i.e.,
Bs = 15.625 kHz << Bc = 250 kHz. Therefore, each subcarrier experiences frequency flat
fading. The OFDM symbol time without CP is
N 256
= = 64 μs.
B 4 × 106
The raw OFDM symbol time, corresponding to the N = 256 IFFT samples, is 64 μs. WiMAX
employs a cyclic prefix which is 12.5% of the symbol time. Therefore, the duration of the cylic
prefix is

Duration of cyclic prefix = 12.5% of symbol time

12.5
= × 64μs,
100

= 8 μs

Thus, the total transmitted OFDM symbol duration with cyclic prefix is 64 μs + 8 μs = 72 μs.
Also, the number of samples in the CP is
CP duration
# Samples in CP =
Sample time
8 μs
=
1/B

= 8 μs × 4 × 106

= 32

Thus, the length of the cyclic prefix Lc = 32 samples and the total number of samples is
256 + 32 = 288. This break-up of the OFDM symbol in terms of the regular samples and the
cyclic prefix is shown in Figure 7.11. Finally, the loss in spectral efficiency is
32
Loss in spectral efficiency =
288
8 μs
=
72 μs

= 11.1%

This is the loss in spectral efficiency arising because of the addition of the cyclic prefix.
 Orthogonal Frequency-Division Multiplexing 245

Figure 7.11 WiMAX OFDM symbol with cyclic prefix

7.4 Bit-Error Rate (BER) for OFDM

Consider the OFDM subcarrier system model given in Eq. (7.5), i.e.,

Y (k) = H (k) X (k) + N (k) (7.6)

where N (k) is the subcarrier noise obtained from the FFT of the noise samples at the output
of the receiver as

N −1
km
N (k) = n (m) e−j2π N

m=0

where N is the number of subcarriers, and n (0) , n (1) , . . . , n (N − 1) are additive noise
samples for each of the output samples y (0) , y (1) , . . . , y (N − 1). We now deduce the
statistical properties of these noise samples N (k), which are required to characterize the
BER performance of the OFDM system. Firstly, observe that the noise N (k) is the linear
combination of Gaussian noise samples n (0) , n (1) , . . . , n (N − 1). Hence, it is Gaussian in
nature. Further, the mean or expected value of N (k) is given as
N −1
km
E {N (k)} = E n (m) e−j2π N

m=0

N −1
km
= E {n (m)} e−j2π N

m=0
246 Principles of Modern Wireless Communication Systems

2
Further, the variance σN of the noise sample N (k) is given as
2
σN = E |N (k)|2

N −1 N −1 ∗
−j2π km −j2π kl
=E n (m) e N n (l) e N

m=0 l=0

N −1 N −1
k
=E n (m) n∗ (l) e−j2π(m−l) N (7.7)
m=0 l=0

Observe that since the noise samples n (m) are independent identically distributed Gaussian
of variance σn2 , it follows that E {n (m) n∗ (l)} = 0 if m = l and σn2 if m = l . Therefore, the
above expression for the noise variance can be simplified as
N −1 N −1
k
2
σN = E {n (m) n∗ (l)} e−j2π(m−l) N
m=0 l=0

N −1
= σn2
m=0

= N σn2

Further, let us assume that each of the channel taps h (0) , h (1) , . . . , h (L − 1) is Rayleigh
fading in nature, i.e., has a complex symmetric Gaussian distribution of mean 0 and variance
1. Therefore, the channel coefficient across the kth subcarrier is given as
N −1
km
H (k) = h (m) e−j2π N

m=0

As can be seen, H (k) is a linear combination of Gaussian random variables h (k) , 0 ≤ k ≤

L − 1. Therefore, H (k) is indeed complex Gaussian i.e. has a Rayleigh fading envelope.
Further, since each h (k) is zero mean, H (k) also has mean zero. Further, identical to the
development of the noise variance above, it follows on similar lines that the channel power
 Orthogonal Frequency-Division Multiplexing 247

gain E |H (k)|2 is
2
E |H (k)|2 = E
km
m = 0L−1 h (m) e−j2π N

L−1
2
E |h (m)|2
km
= e−j2π N

m=0

=L

Therefore, the system model in Eq. (7.6) represents a standard Rayleigh fading channel of
power gain L with receiver noise N (k) of variance N σn2 . Therefore, the average SNR is NLP2 .
σn
Hence, the BER is given by the standard expression for that of a Rayleigh fading wireless
channel as
⎛ ⎞
LP
1⎝ N σn
2
BEROFDM = 1− LP
⎠
2 2 + N σ2
n

7.5 MIMO-OFDM

MIMO-OFDM is a combination of the Multiple-Input Multiple-Output (MIMO) wireless

technology with that of OFDM, to further increase the rate in broadband multi-antenna wireless
systems. Similar to OFDM, MIMO-OFDM converts a frequency-selective MIMO channel into
multiple parallel flat fading MIMO channels. Hence, MIMO-OFDM significantly simplifies
baseband receive processing by eliminating the need for a complex MIMO equalizer. We have
already seen that the frequency-selective SISO channel is modelled as an FIR channel filter,
with the output y (n) at time instant n given as
L−1
y (n) = h (l) x (n − l) + w (n) ,
l=0

= h (0) x (n) + h (1) x (n − 1) + . . . + h (L − 1) x (n − L + 1) +w (n)

ISI from previous symbols
248 Principles of Modern Wireless Communication Systems

Figure 7.12 MIMO OFDM transmitter schematic

where w (n) denotes the noise. Hence, a MIMO frequency-selective channel can be modelled
as a MIMO FIR filter, which can be described as
L−1
y (n) = H (l) x (n − l) + w (n)
l=0

= H (0) x (n) + H (1) x (n − 1) + . . . + H (L − 1) x (n − L + 1) +w (n)

ISI from previous symbol vectors

Therefore, the symbol vector y (n) at the time instant n is affected by inter-symbol vector inter-
ference from x (n − 1) , x (n − 2) , . . . , x (n − L + 1). This is an L-tap frequency-selective
MIMO channel. As can be seen, in a MIMO frequency-selective channel, the interference
occurs between current and previous transmit symbol vectors. In a MIMO-OFDM system,
one needs to perform the IFFT operation at each transmit antenna. The schematic figures
showing clearly the processing at the transmitter and receiver of the MIMO-OFDM system
are shown in figures 7.12 and 7.13 respectively. Hence, employing MIMO-OFDM, the MIMO
frequency-selective channel can be converted into a set of parallel flat-fading MIMO channels.
 Orthogonal Frequency-Division Multiplexing 249

These can be described as,

ỹ (0) = H̃ (0) x̃ (0)

ỹ (1) = H̃ (1) x̃ (1)

..
.
ỹ (N − 1) = H̃ (N − 1) x̃ (N − 1)

The model across the kth subcarrier is ỹ (k) = H̃ (k) x̃ (k), where ỹ (k) and x̃ (k) are the
received and transmitted symbol vectors corresponding to the kth subcarrier, and H̃ (k) is
the flat-fading channel matrix corresponding to the subcarrier k. Each of the received vectors
ỹ (0) , ỹ (1) , . . . , ỹ (N − 1) can be processed by a simple MIMO zero-forcing receiver or a
MIMO-MMSE receiver for detection of the vectors x̃ (0) , x̃ (1) , . . . , x̃ (N − 1). The zero-
forcing MIMO receiver is given as
†
ˆ ZF (k) = H̃ (k)
x̃ ỹ (k)

−1
= H̃H (k) H̃ (k) H̃H (k) ỹ (k)

Also, the MMSE receiver for the subcarrier k of the MIMO-OFDM system is given as
†
ˆ MMSE (k) = H̃ (k)
x̃ ỹ (k)

−1
= Pd Pd H̃H (k) H̃ (k) + σw
2
I H̃H (k) ỹ (k)

where Pd denotes the data power. The channel matrices H̃ (0) , H̃ (1) , . . . , H̃ (N − 1)
corresponding to the OFDM subcarriers are given as follows. Let hu,v (k) , h̃u,v (k) denote
the (u, v)th entries of the matrices H (k) , H̃ (k) respectively. Then, h̃u,v (k) is given as the
N -point DFT of the zero-padded coefficients

hu,v (0) , hu,v (1) , . . . , hu,v (L − 1) , 0, . . . 0

(N −L)

In effect, the channel matrix H̃ (k) is the kth frequency point corresponding to the FFT of the
zero padded channel matrices [H (0) , H (1) . . . , H (L − 1) , 0r×t , . . . , 0r×t ].
250 Principles of Modern Wireless Communication Systems

MIMO detection
for 1st subcarrier

1st receive
antenna CP S/P N pt MIMO P/S
removal demux FFT detection mux

2st receive CP S/P N pt MIMO P/S

antenna removal demux FFT detection mux

MIMO detection P/S

for 2st subcarrier demux

rth receive
antenna CP S/P N pt MIMO P/S
removal demux FFT detection mux

Figure 7.13 MIMO-OFDM receiver schematic

7.6 Effect of Frequency Offset in OFDM

OFDM divides the available wideband amongst a set of orthogonal overlapping subcarriers.
Hence, the presence of a carrier-frequency offset can introduce severe distortion in an OFDM
system, as it results in a loss of orthogonality amongst the subcarriers. Hence, the presence
of a carrier-frequency offset introduces Inter-Carrier Interference (ICI) in OFDM systems. In
this section, we characterize the effect of frequency offset on the performance of the OFDM
system. Consider a frequency offset Δf such that

Δf
=
B/N

where denotes the normalized frequency offset, normalized with respect to the subcarrier
bandwidth B/N . Corresponding to the frequency offset , the baseband received samples y (n)
are given as

N
2
1 k+
y (n) = X (k) H (k) ej2πn N + w (n) (7.8)
N
k=− N2
 Orthogonal Frequency-Division Multiplexing 251

To verify the above equation, set = 0. The above equation then reduces to

N
2
1 k
y (n) = X (k) H (k) ej2πn N + w (n)
N
k=− N
2

Performing now the FFT of y (0) , y (1) , . . . , y (N − 1) at the receiver, Y (l), which
corresponds to the symbol received on the l th subcarrier, is given as
1 nl
Y (l) = y (n) e−j2π N
N n

N
2
1 k nl 1 nl
= X (k) H (k) ej2πn N e−j2π N + w (n) e−j2π N
N n k=− N
N n
2

W (l)
N
2
1 k−l
= X (k) H (k) ej2πn N + W (l)
N n k=− N
2

N
2
1 k−l
= X (l) H (l) + X (k) H (k) ej2πn N +W (l)
N n
k=− N2 , k=l
0

= X (l) H (l) + W (l)

k−l
where we have used the fact that n ej2πn N = 0 if k = l . Thus, in the absence of a carrier-
frequency offset, i.e., = 0, the system in Eq. (7.8) reduces to the earlier flat-fading OFDM
system across each subcarrier, i.e.,

Y (l) = X (l) H (l) + W (l)

Now consider the received symbols y (n) in the presence of a carrier-frequency offset

N
2
1 k+
y (n) = X (k) H (k) ej2πn N + w (n)
N
k=− N2
252 Principles of Modern Wireless Communication Systems

Therefore, the demodulated symbol Y (l) in the presence of a carrier-frequency offset is given
as
1 nl
Y (l) = y (n) e−j2π N
N n

N
2
1 k−l+ 1 nl
= X (k) H (k) ej2πn N + w (n) e−j2π N
N n k=− N
N n
2

W (l)
N
2
1 1 k−l+
= X (l) H (l) ej2πn N + X (k) H (k) ej2πn N + W (l)
N n
N
k= −N
2
, k=l n

= X (l) H (l) + W (l)

To simplify the above expression, we will use the result below

N −1 Nθ
sin
ejθn = 2
θ
ej φ̃
n=0
sin 2

where ej φ̃ is a phase factor, which does not affect the power at the output. Using the relation
above, the expression for Y (l) can be simplified as
N
2
sin (π ) 1 j φ̃l sin (π )
Y (l) = H (l) X (l) e + H (k) X (k) ej φ̃kl +W (l)
sin πN N N sin π l−N
k+
k=− N
2
, k=l
Desired signal
Inter-carrier interferenceIl

The SINR is given as

Signal power
SINR =
Interference + Noise power
Signal power
=
E |Il |2 + σn2

The desired signal power can be calculated as

! "2
2 2 sin (π )
Signal power = E |H (l)| E |X (l)|
N sin πN
 Orthogonal Frequency-Division Multiplexing 253

For a large value of N , i.e., number of subcarriers, we have

π π
lim sin ≈
N →∞ N N
π π
⇒ N sin
→N =π
N N
Hence, the signal power for a large number of subcarriers N is given as
! "
sin (π ) 2
Signal Power = E |H (l)|2 P
π
! "2
2 sin (π )
= P |H |
π

where P = E |X (l)|2 is the power of the transmitted data symbols and

|H |2 = E |H (l)|2 denotes the average channel-power gain across each subcarrier.
The interference power is given as,

k= N2 2
2 2 2 sin (π )
E |Il | = E |X (l)| E |H (l)|
N sin π l−N
k+
k=− N2 , k=l

Setting k − l = u and letting N → ∞, we have

∞ ! "2
2 2 sin π
E |Il | = P |H | u
N sin π N
u=−∞,u=0
∞ ! "2
2 2 1
= P |H | (sin π ) u
N sin π N
u=−∞,u=0

2θ
Employing the inequality sin θ ≥ π, we have
πu 2πu/N 2u
sin ≥ =
N π N
πu
⇒ N sin ≥ 2u (7.9)
N
254 Principles of Modern Wireless Communication Systems

The interference power E |Il |2 can, therefore, be approximated as

∞ ! "2
1
E |Il |2 ≤ P |H |2 (sin π )2
2u
u=−∞, u=0

∞ "2
!
2 2 1
= P |H | 2 (sin π )
2u
u=1
∞ ! "2
1 2 2 1
= P |H | (sin π )
2 u=1
u
π2
6

π2
= P |H |2 sin2 π
12

= 0.822P |H |2 sin2 π

Hence, the SINR in the presence of carrier-frequency offset of is given as

2
P |H |2 sin π
π
SINR = 2 2
(7.10)
0.822P |H | sin π + σn2

Example 7.1 illustrates the effect of ICI in reducing the SINR at the output of the OFDM
receiver.

EXAMPLE 7.1

Consider |H |2 = 1, and data power P = 10 dB with noise power σn2 = 0 dB. Derive the
SNR/SINR with and without a carrier-frquency offset of = 5% = 0.05 in a WiMAX
system.

Solution: If the carrier-frequency offset = 0, the SNR at the receiver is given as

P |H |2
SNR = = 10 dB
σn2
 Orthogonal Frequency-Division Multiplexing 255

Considering now a carrier-frequency offset of = 0.05 and employing the expression in

Eq. (7.10), the SINR is given as
2
10 × sinπ0.05
π0.05
SINR = = 8.25
0.82210 × sin2 (π0.05) + 1
Therefore, the reduction in SINR is a factor of 1.75 or, basically, 17.5%. Further, the WiMAX
subcarrier bandwidth is 15.625 kHz. The absolution value of the frequency offset Δf , is
therefore,

Δf = Bs = 0.05 × 15.625 ≈ 0.78kHz

Further, considering a typical 4G carrier frequency of 2.4 GHz, the carrier-frequency offset as
a fraction of the centre frequency is
0.78 × 103 1
9
= × 10−6
2.4 × 10 3
This is also termed as 0.33 ppm or parts per million with respect to the carrier frequency since
1
1 × 10−6 = 1000000 is equivalent to 1 part per million.

7.7 OFDM–Peak-to-Average Power Ratio (PAPR)

The Peak-to-Average Power Ratio (PAPR) is a critical problem in OFDM systems, which needs
to be handled effectively in order to limit the distortion at the receiver. Consider a non-OFDM
or single-carrier system with BPSK modulated symbols. For example, let the symbol stream
x (0) , x (1) , x (2) , . . . be given as +a, −a, +a, . . . and so on. The power in each symbol
equals a2 . Further, also observe that this is the peak power at any given instant of time.
Therefore, we have

Peak power = Average power = E |x (k)|2 = a2

256 Principles of Modern Wireless Communication Systems

Thus, since the peak and average power are equal, the peak-to-average power ratio, or PAPR,
is given as
Peak power
PAPR =
Average power

a2
=
a2
= 1 = 0 dB

X(0) x(0)

IFFT
N

X(N-1) x(N-1)
N = # of
subcarriers

Figure 7.14 OFDM subcarrier loading

The above relation clearly shows that there is no significant deviation of the instantaneous
power level from the mean power level. Now, consider an OFDM system in which the different
information symbols X (0) , X (1) , X (2) , . . . given by +a, −a, +a, . . . for instance, are
loaded onto the subcarriers. This is shown schematically in Figure 7.14. The actual samples
transmitted over the wireless channel are x (0) , x (1) , x (2) , . . . , x (N − 1), which are the
IFFT samples of the information symbols X (0) , X (1) , X (2) , . . . , X (N − 1). Consider
the kth IFFT sample x (k) given as

N −1
1 ki
x (k) = X (i) ej2π N
N
i=0

where X (i) denotes the information symbols. The average power in the symbols is given as

Average power = E |x (k)|2

 Orthogonal Frequency-Division Multiplexing 257

N −1
1 2
E |X (i)|2 E
ki
= ej2π N
N
i=0
1

N −1
1
= E |X (i)|2
N2
i=0

N −1
1
= a2
N2
i=0

1 2 a2
= a N =
N2 N
As can be seen from the above equation, the average power of transmission is aN . The peak
2

power can be found as follows. Observe that the peak of the OFDM sample arises for all
symbols X (i) = +a or X (i) = −a. This can be verified as follows.
N −1
1 ki
|x (k)| = X (i) ej2π N
N
i=0

N −1
1 ki
≤ X (i) ej2π N
N
i=0

N −1
1 ki
= |X (i)| ej2π N
N
i=0 a
1

N −1
1
= a
N
i=0

=a

Therefore, the peak power is given as a2 . Hence, the peak-to-average power ratio in an OFDM
system is given as

a2
OFDM PAPR = =N
a2 /N
258 Principles of Modern Wireless Communication Systems

OFDM CCDF of PAPR for Various Subcarriers N

1

0.9

0.8

0.7
Probability (PAPR > x)

0.6

0.5

0.4

0.3

0.2

0.1

4 5 6 7 8 9 10 11 12
PAPR in dB
Figure 7.15 OFDM PAPR for various number of subcarriers N

From the above expression, it can be seen that the peak-to-average power ratio in an OFDM
system is N , which is significantly higher compared to that of the single-carrier system, which
is 1. Further, interestingly, this PAPR rises with N , i.e., the number of subcarriers. Larger the
number of subcarriers, larger is the PAPR. This high PAPR of the OFDM arises because of the
IFFT operation. The data symbols across the subcarriers can add up to produce a high peak
valued signal as seen above. For instance, in an OFDM system with 512 subcarriers and BPSK
modulation, the PAPR at the output can be as high as 10 dB. The PAPR of an OFDM system
is characterized using the CCDF, i.e., the complementary cumulative distribution function.
The CCDF F X (x) of a random variable X is given as the probability that X > x, expressed as

F X (x) = Pr (X > x)
 Orthogonal Frequency-Division Multiplexing 259

Naturally, the CCDF is related to the CDF, i.e., cumulative distribution function FX (x) of X as

FX (x) = Pr (X ≤ x)

= 1 − Pr (X > x)

= 1 − FX (x)

The CCDF of the PAPR then shows the probability that the PAPR, which is a random quantity,
exceeds a particular threshold. A plot of the CCDF of the PAPR for various values of N , the
total number of subcarriers, is shown in Figure 7.15.
The impact of PAPR on the OFDM system hardware can be understood as follows. Every
communication system has a receiver amplifier, which serves to amplify the amplitude of the
receive signal, in order to boost its strength. However, the characteristic of the amplifier is
linear only for a limited amplitude range of the signal. Typically, the amplifier operates around
a bias point, as shown in Figure 7.16, which is roughly around the average power of the signal.
As long as the signal amplitude is restricted to the dynamic range of the amplifier around this
bias point, for which the amplifier characteristic is linear, there is no noninear distortion at the
output. However, in the case of OFDM, since the peak power deviates significantly from the
average power, there is a high chance that the signal crosses into the voltage region outside the

Figure 7.16 Nonlinear amplifier characteristic

260 Principles of Modern Wireless Communication Systems

dynamic range of the amplifier, thus resulting in a noninear distortion of the received signal.
This nonlinear effect, arising out of amplifier saturation, leads to loss of orthogonality of the
subcarriers and inter-carrier interference. The net result is a poor decoding performance and a
rise in the bit-error rate. A slightly modified OFDM technique, which can significantly reduce
the PAPR, is SC-FDMA, which is described next.

7.8 SC-FDMA

SC-FDMA, which stands for Single-Carrier Frequency Division for Multiple Access, can
be employed to reduce the peak-to-average power ratio in an OFDM system. Consider the
following hypothetical modification of the OFDM transmitter, shown in Figure 7.17, by the
insertion of an N -point FFT block before the N -point IFFT block. It can then be seen that the
FFT and the IFFT cancel the effect of each other and the net output is the exact input symbol
stream, i.e., corresponding to a single-carrier system. This drastically reduces the PAPR, since,
as seen previously, the PAPR of a single-carrier system is 0 dB. However, instead of using an
N -point FFT, one can use an M -point FFT, where M < N , to reduce the PAPR, while still
retaining the properties of the OFDM system. This proposed SC-FDMA schematic is shown in
Figure 7.18. Hence, introduction of the M -point FFT in SC-FDMA significantly reduces the
PAPR of the system. This is the central principle of SC-FDMA.
Transmitted
Symbols

Symbol S/P N pt N pt P/S Add

Stream Demux FFT IFFT Mux CP

Figure 7.17 Hypothetical modification of OFDM transmitter

Transmitted
symbols

Symbol S/P M pt Subcarrier N pt P/S Add

stream demux FFT mapping IFFT mux CP

2 new blocks in SC-FDMA

transmitter schematic
Figure 7.18 SC-FDMA transmitter schematic
 Orthogonal Frequency-Division Multiplexing 261

7.8.1 SC-FDMA Receiver

The SC-FDMA receiver schematic is shown in Figure 7.19. The SC-FDMA receiver
incorporates two new blocks compared to the OFDM receiver. The purpose of these additional
blocks can be described as follows. After the N -point FFT operation at the receiver, the signals
are equalized across all the subcarrires, to remove the effect of the fading-channel coefficient
across the subcarriers. Following the above operation, they are demapped from the subcarriers,
which are N in number, to the original FFT block size of M . Finally, the M -point FFT is
performed on these samples to generate the symbol stream.

Equalizer
CP S/P N pt M pt P/S Symbol
and
removal demux FFT IFFT mux stream
demapping

2 new blocks in SC-FDMA

receiver schematic
Figure 7.19 SC-FDMA receiver schematic

7.8.2 Subcarrier Mapping in SC-FDMA

Subcarrier mapping, in which the M samples at the output of the M -point FFT are mapped
to the N subcarriers, is a key operation in SC-FDMA, a block representation of which
can be seen in the SC-FDMA transmitter schematic in Figure 7.18. The various possible
SC-FDMA subcarrier mappings are illustrated through the following example. Consider
M = 4 SC-FDMA symbols and N = 12 subcarriers. Let x (0) , x (1) , x (2) , x (3) denote
the symbols and X (0) , X (1) , X (2) , X (3) denote the corresponding M = 4-point FFT
samples which are to be loaded onto the subcarriers. Let the number of subcarriers be N = 12.
In Interleaved FDMA (IFDMA) shown in Figure 7.20, the samples X (i) are interleaved with
zeros. In LFDMA, which is also employed in the uplink of the 4G mobile standard LTE,
the samples are loaded as a block onto the subcarriers, with appropriate zero padding. This
is shown in Figure 7.20. Post this subcarrier mapping, the rest of the procedure prior to
transmission proceeds as shown in the SC-FDMA transmitter block.
262 Principles of Modern Wireless Communication Systems

Interleaving Zeroes

I-FDMA

X(0) 0 0 X(1) 0 0 X(2) 0 0 X(3) 0 0

X(0) X(1) X(2) X(3) 0 0 0 0 0 0 0 0

L-FDMA
Employed in LTE Uplink Zero-Padding

Figure 7.20 SC-FDMA various subcarrier mapping schemes

1. Consider a three-tap wireless channel [h (0) , 0, h (2)] (i.e., h (1) = 0) with each tap a unif-
norm Rayleigh fading channel coefficient. Employ the WSSUS channel assumption. Let an
OFDM system with 256 subcarriers in the 2.1 GHz band be implemented over this channel
with delay spread Td = 8 μ s. The DFT operation at the receiver is given as

N −1
1 2πlk
F (k) = √ f (l) e−j N

N l=0

(a) With a bare minimum cyclic-prefix duration, the system has a useful bit rate of 8.6486
Mbps for QPSK transmission. What is the bandwidth of the system?
(b) For a cyclic prefix of 12 μs duration, what is the loss in spectral efficiency for this
system?
(c) Consider the channel coefficients H (0) and H (64) at the 0th and 64th subcarriers
respectively. What is their joint distribution?
(d) It is not difficult to see that every OFDM system must satisfy a very important constraint
that the wireless channel is constant during the transmission of the total OFDM symbol
(i.e., useful plus cyclic). Assuming a system with 20% loss of spectral efficiency and
an extreme vehicle moving at 200 km/h, what is the maximum possible number of
subcarriers?
 Orthogonal Frequency-Division Multiplexing 263

2. OFDM System Design Consider an OFDM system with N = 256 subcarriers over a
bandwidth of B = 5 MHz. Let the corresponding frequency selective fading channel have
an impulse response with 3 multipath components at delays of [0, 0.40, 1.0] μs, with each
component of −3 dB average power. Noise power at the receiver is σv2 = 3 dB. Assume
that the IFFT and FFT operations are given respectively as

N −1 N −1
1 j2πkn/N
x (n) = X (k) e , X (k) = x (n) e−j2πkn/N
N
k=0 n=0

In that case, the system model after FFT at the receiver becomes Y (k) = H (k) X (k)
+ V (k), where V (k) is the FFT of the AWGN. Answer the following questions.
(a) Describe the time-domain model of the above frequency-selective channel.
(b) What is the minimum number of samples required in the cyclic prefix?
(c) What is the duration of this minimum cyclic prefix?
(d) If the actual cyclic prefix employed is twice the minimum length required with QPSK
modulated subcarriers, what is the effective bit rate of the OFDM system?
(e) What is the bit-error rate across each subcarrier if the total transmit power of 70 dB is
distributed equally across the subcarriers?
(f) What is the reduction in SNR across each subcarrier in the presence of a 5% carrier-
frequency offset relative to the subcarrier bandwidth?
3. OFDM System Design Consider a mobile OFDM profile with N = 512 subcarriers over a
bandwidth of B = 10MHz. Let the corresponding frequency-selective fading channel have
an impulse response with 4 multipath components at delays of 0 μs, 0.60 μs, 1.2 μs, 2.4 μs,
with each component of average power- −3 dB. Noise power at the receiver is σn2 =
3 dB per sample (i.e., before FFT). Assume that the IFFT and FFT operations are given
respectively as,

N −1 N −1
1
x (n) = X (k) ej2πkn/N X (k) = x (n) e−j2πkn/N
N
k=0 n=0

In that case, the system model after FFT at the receiver becomes Y (k) = H (k) X (k) +
W (k), where W (k) is the FFT of the AWGN. Answer the following questions.
(a) Describe the time-domain model of the above frequency-selective channel, i.e., what
264 Principles of Modern Wireless Communication Systems

is the length of the channel impulse response filter and what is the number of nonzero
channel taps?
(b) What is the minimum number of samples required in the cyclic prefix in this system?
(c) What is the duration of this minimum cyclic prefix?
(d) If the actual cyclic prefix employed is three times the minimum length required with
QPSK-modulated subcarriers, what is the effective bit rate of the OFDM system?
(e) What is the QPSK bit-error rate across each subcarrier if the total transmit power of
80 dB is distributed equally across the subcarriers?
(f) What is the reduction in SNR across each subcarrier in the presence of an 8% carrier-
frequency offset relative to the subcarrier bandwidth?
4. Consider an OFDM system with total passband bandwidth B = 5 MHz with N = 512
subcarriers. The channel has a maximum delay spread of Td = 4 μs. Answer the questions
that follow.
(a) What is the symbol time of a corresponding single-carrier system?
(b) What is the sample time of the OFDM system and the raw symbol time without the
cyclic prefix?
(c) What is the minimum number of samples required in the cyclic prefix?
(d) If the length of the cylic prefix is twice the required minimum calculated above, what
is the total OFDM symbol time?
(e) What is the loss in efficiency due to the overhead of the cyclic prefix?
(f) If the modulation employed is 16-QAM, what is the effective bit rate of the above
OFDM system?
(g) At a carrier frequency of fc = 2.4 GHz, what is the maximum possible velocity of a
mobile for the system to be able to function?
5. Alamouti Coded OFDM Consider an Alamouti-coded OFDM system with two transmit
and one receive antennas. Consider symbols X1 (k) , X2 (k) loaded onto transmit antennas
1,2 respectively on the kth subcarrier of the OFDM system in the first OFDM symbol.
Consider a frequency-selective channel with channel taps h (0) , h (1) , . . . , h (L − 1).
Answer the questions that follow.
(a) Describe the operation of the Alamouti-coded OFDM system above, i.e., the
transmission on each subcarrier and the decoding operation at the receiver.
(b) If the transmit power is P per subcarrier, noise power σn2 , find the instantaneous receive
SNR expression for each symbol.
 Orthogonal Frequency-Division Multiplexing 265

(c) Considering the channel taps to be IID Rayleigh random variables of average power
unity, derive the average Symbol Error Rate (SER) expressions for BPSK and 16-PSK
modulation.
(d) Compute the average SER above for both BPSK and 16-PSK for P = 35 dB, number
of subcarriers N = 128, L = 5, and noise power σn2 = −3 dB.
(e) Considering the channel taps to be independent Rayleigh random variables of average
power σl2, 0 ≤ l ≤ L − 1, derive the average SER expressions for BPSK and 16-PSK
modulation.
(f) Compute the average SER above for both BPSK and 16-PSK for P = 35 dB, number
of subcarriers N = 128, σl2 = l dB for 0 ≤ l ≤ 4 and noise power σn2 = −3 dB.
6. Consider a multipath channel with L i.i.d Rayleigh faded taps h(i), 0 ≤ i ≤ L − 1, each
distributed as the symmetric complex Gaussian CN (0, 1). Let H(u), H(v) denote the
complex channel coefficients corresponding to subcarriers u, v respectively in an OFDM
system with N subcarriers and bandwidth B . Consider uncorrelated scattering and answer
the questions that follow.
(a) Express H(u), H(v) in terms of the channel taps.
(b) Compute the auto-correlation R (0) corresponding to the the complex channel
coefficient across each subcarrier.
(c) Derive the exact expression for the correlation between the subcarrier coefficients
H(u), H(v).
(d) Derive a suitable bound for the correlation between the subcarrier coefficients
H(u), H(v) in terms of |u − v | and N .
(e) Derive an expression for the number of subcarriers per coherence bandwidth Bc in the
above system. Denote this by Nc.
(f) Let |u − v | = αNc . Derive a bound for the correlation between the subcarrier
coefficients H(u), H(v) as a function of R (0) and α. From this, deduce the rate at
which the correlation is decreasing as a function of α.
7. Consider an OFDM system with N = 4 subcarriers and L = 2 tap wireless channel with
channel coefficients h(0), h(1). Let noise power per subcarrier (i.e., after the FFT at the
receiver) be σ 2 = 3 dB. Let the total transmit power be PT . Answer the questions that
follow.
1
(a) Consider h(0) = h(1) = 2 and derive the complex channel coefficients across all the
subcarriers.
266 Principles of Modern Wireless Communication Systems

(b) Consider h(0) = h(1) = 12 , PT = 18 dB with equal power allocation and derive the
instantaneous BER across each subcarrier for BPSK modulation.
(c) Consider h(0) = h(1) = 12 , PT = 18 dB and derive the optimal power allocation for
each subcarrier.
(d) Consider coefficients h(0), h(1) to be Rayleigh fading i.i.d. with average power unity
and derive the average BER as a function of PT , σ 2 for BPSK modulation with total
power PT and equal power allocation across subcarriers.