Scribd
Upload
2 views

Extracted pages from WirelessComm_Chp1-16_March32020_Part5

The document discusses the relationship between bit error probability (Pb) and bit energy (γb) using approximations for Gray encoding and high SNR. It details the error probability for differential modulation, highlighting that differential modulation performs approximately 3 dB worse than coherent modulation. Additionally, it introduces an alternate representation of the Q-function to simplify error probability calculations in AWGN channels.

Uploaded by

yogeti9322
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
2 views

Extracted pages from WirelessComm_Chp1-16_March32020_Part5

The document discusses the relationship between bit error probability (Pb) and bit energy (γb) using approximations for Gray encoding and high SNR. It details the error probability for differential modulation, highlighting that differential modulation performs approximately 3 dB worse than coherent modulation. Additionally, it introduces an alternate representation of the Q-function to simplify error probability calculations in AWGN channels.

Uploaded by

yogeti9322
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Performance specifications are generally more concerned with the bit error probability Pb as a function of the

bit energy γb . To convert from Ps to Pb and from γs to γb we use the approximations (6.3) and (6.2), which assume
Gray encoding and high SNR. Using these approximations in (6.32) yields a simple formula for Pb as a function
of γb :
q
Pb (γb) = α̂M Q( β̂M γb ), (6.33)

where α̂M = αM / log2 M and β̂M = (log2 M )βM for αM and βM in (6.32). This conversion is used in what
follows to obtain Pb versus γb from the general form of Ps versus γs in (6.32).

6.1.7 Error Probability for Differential Modulation


The probability of error for differential modulation is based on the phase difference associated with the phase
comparator input of Figure 5.20. Specifically, the phase comparator extracts the phase of

r(k)r∗(k − 1) =A2 ej(θ(k)−θ(k−1)) + Aej(θ(k)+φ0 ) n∗ (k − 1)


+ Ae−j(θ(k−1)+φ0 ) n(k) + n(k)n∗ (k − 1) (6.34)

in order to determine the transmitted symbol. By symmetry we can assume a given phase difference when com-
puting the error probability. Assuming then a phase difference of zero, θ(k) − θ(k − 1) = 0, yields

r(k)r∗(k − 1) = A2 + Aej(θ(k)+φ0 )n∗ (k − 1) + Ae−j(θ(k−1)+φ0 ) n(k) + n(k)n∗ (k − 1). (6.35)

Next we define new random variables

ñ(k) = n(k)e−j(θ(k−1)+φ0 ) and ñ(k − 1) = n(k − 1)e−j(θ(k)+φ0 ) ,


which have the same statistics as n(k) and n(k − 1). Then

r(k)r∗(k − 1) = A2 + A(ñ∗ (k − 1) + ñ(k)) + ñ(k)ñ∗ (k − 1). (6.36)


There are three terms in (6.36): the first term, with the desired phase difference of zero; and the second and third
terms, which contribute noise. At reasonable SNRs the third noise term is much smaller than the second, so we
neglect it. Dividing the remaining terms by A yields

z̃ = A + Re{ñ∗ (k − 1) + ñ(k)} + j Im{ñ∗ (k − 1) + ñ(k)}. (6.37)


Let us define x = Re{z̃} and y = Im{z̃}. The phase of z̃ is then given by

θz̃ = tan−1 y/x. (6.38)


Given that the phase difference was zero, an error occurs if |θz̃ | ≥ π/M . Determining p(|θz̃ | ≥ π/M ) is identical
to the case of coherent PSK except that, by (6.37), we have two noise terms instead of one and so the noise power
is twice that of the coherent case. This will lead to a performance of differential modulation that is roughly 3 dB
worse than that of coherent modulation.
In DPSK modulation we need only consider the in-phase branch of Figure 5.20 when making a decision, so we
set x = Re{z̃} in our analysis. In particular, assuming a zero is transmitted, if x = A + Re{ñ∗ (k − 1) + ñ(k)} < 0
then a decision error is made. This probability can be obtained by finding the characteristic or moment generating

193
function for x, taking the inverse Laplace transform to get the distribution of x, and then integrating over the
decision region x < 0. This technique is quite general and can be applied to a wide variety of different modulation
and detection types in both AWGN and fading [9, Chap. 1.1]: we will use it later to compute the average probability
of symbol error for linear modulations in fading both with and without diversity. In DPSK the characteristic
function for x is obtained using the general quadratic form of complex Gaussian random variables [1, Apx. B; 12,
Apx. B], and the resulting bit error probability is given by
1
Pb = e−γb . (6.39)
2
The error probability of DQPSK is harder to obtain than that of DPSK, since the phase difference between two
adjacent symbols depends on two information bits instead of one. An elegant analysis of the resulting error prob-
ability is obtained in [1, Apx. B] as a special case of the probability analysis for quadratic functions of complex
Gaussian random variables. This yields, under Gray coding, the bit error probability [1, Eqn. 4.5-66]
 
1 −(a2 + b2 )
Pb = Q1 (a, b) − exp I0 (ab), (6.40)
2 2
where Q1 (a, b) is the Marcum Q function defined as
Z ∞  
4 −(a2 + z 2 )
Q1 (a, b) = z exp I0 (az)dz, (6.41)
b 2
q √ q √
for I0 (x) the modified Bessel function of the first kind and zeroth order, a = 2γb(1 − .5), and b = 2γb(1 + .5).

6.2 Alternate Q-Function Representation


In (6.32) we saw that Ps for many coherent modulation techniques in AWGN is approximated in terms of the
Gaussian Q-function. Recall that Q(z) is defined as the probability that a Gaussian random variable X with mean
0 and variance 1 exceeds the value z:
Z ∞
1 2
Q(z) = p(X ≥ z) = √ e−x /2 dx. (6.42)
z 2π
The Q-function is not that easy to work with since the argument z is in the lower limit of the integrand, the
integrand has infinite range, and the exponential function in the integral doesn’t lead to a closed-form solution.
In 1991 an alternate representation of the Q-function was obtained by Craig [13]. The alternate form is given
by
Z π/2  
1 −z 2
Q(z) = exp dφ, z > 0. (6.43)
π 0 2 sin2 φ
This representation can also be deduced from the work of Weinstein [14] or Pawula et al. [5]. In this alternate
form, the integrand is over a finite range that is independent of the function argument z, and the integral is Gaussian
with respect to z. These features will prove important in using the alternate representation to derive average error
probability in fading.
Craig’s motivation for deriving the alternate representation was to simplify the probability of error calculation
for AWGN channels. In particular, we can write the probability of bit error for BPSK using the alternate form as
p  1 Z π/2  
−γb
Pb = Q 2γb = exp dφ. (6.44)
π 0 sin2 φ

194

You might also like