5.5  Estimating Probability of Error for PAM

The goal of this section is to obtain an expression for the symbol error probability Pe for PAM on AWGN channels using ML. The symbols are assumed to have a uniform prior distribution and the optimal MAP criterion coincides with the ML.

The PAM constellation is assumed to have the same distance d among the neighbor symbols and all symbols are equiprobable. Listing 2.8 gives an example where d = 2. In general, the PAM symbols are: ± d 23d 25d 2 ,…,±(M−1)d 2 . In the case of PAM, the 2 symbols at the extrema of the constellation have only one neighbor, while the others have two. Hence, for a M-PAM with the symbols uniformly distributed with probability P(m) = 1∕M,∀m,

Pe = P(e) =m∈MP(m,e) =m∈MP(m)P(e|m) (5.4) = M − 2 M 2Q ( d ) + 2 MQ ( d ) (5.5) = 2 (1 − 1 M )Q ( d ), (5.6)

where M = 2b. It is useful to rewrite Eq. (5.6) in terms of SNR. Assuming matched filtering with Ep = 1, it has been shown that

SNRnMF = Ec σ2 ,
(5.7)

where the subscript nMF recalls that this SNR is obtained at the output of a matched filter when the (normalized) shaping pulse has unitary energy.

For PAM, Ec depends only on d and M. Hence, using Eq. (2.10), to express d in terms of Ec and M leads to

d = 12Ec M2 − 1
(5.8)

and, for PAM, one can write

d = 3 M2 − 1SNRnMF,
(5.9)

such that for PAM:

Pe = 2 (1 − 1 M )Q ( d ) (5.10) = 2 (1 − 1 M )Q ( 3 M2 − 1SNRnMF). (5.11)

Note that, for sufficiently high M (say M ≥ 32 or, equivalently, b ≥ 5)

Pe ≈ 2Q ( d ).

Estimating the bit error probability Pb is more evolved and the approximation of Eq. (2.4) is often adopted.

5.5.1  The union bound

Sometimes it is not necessary to have a precise value for Pe but only a bound on its maximum value in a given setup.

The union bound states that the probability of symbol error for the ML detector on AWGN with an M-point constellation is

Pe ≤ (M − 1)Q (d

min ),whered_ministheminimumdistancebetweenanypairofpointsoftheconstellation,i.e.d

min = min i≠j||mimj||,∀i,j.