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
 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

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 =
[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
 = 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.
 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.
 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
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
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
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
 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.
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
 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
 N −1
1
= 2 E c20 (i) c20 (i − k)
N
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
 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
= 2 E {c0 (i) c1 (i − k) c0 (j) c1 (j − k)}
N
i=0 j=0

N −1
1
= 2 E c20 (i) E c21 (i − k)
N
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
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
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
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
2 1
E |w̃0 | =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 = 2
× N σn2
N n=0
N

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
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.
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
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
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
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
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
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,
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)
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
observe that the power of each noise component w̃v is E |w̃v |2 = 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 = σw
2

N
! P
2 2 2
= N × |h (0)| + |h (1)| + . . . + |h (L − 1)| 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
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