BER calculation

Vahid Meghdadi
reference: Wireless Communications by Andrea Goldsmith
January 2008

SER and BER over Gaussian channel

1.1

BER for BPSK modulation

In a BPSK system the received signal can be written as:

y =x+n

(1)

where x {A, A}, n CN (0, 2 ) and 2 = N0 . The real part of the above
equation is yre = x + nre where nre N (0, 2 /2) = N (0, N0 /2). In BPSK
constellation dmin = 2A and b is defined as Eb /N0 and sometimes it is called
SNR per bit. With this definition we have:
b :=

Eb
A2
d2
=
= min
N0
N0
4N0

(2)

So the bit error probability is:

Z

Pb = P {n > A} =

2 2 /2

2x2 /2

This equation can be simplified using Q-function as:

s


p 
2
dmin
d
min

=Q
Pb = Q
=Q
2b
2N0
2N0

(3)

(4)

where the Q function is defined as:

1
Q(x) =
2

1.2

x2
2

dx

(5)

BER for QPSK

QPSK modulation consists of two BPSK modulation on in-phase and quadrature

components of the signal. The corresponding constellation is presented on figure
1. The BER of each branch is the same as BPSK:
p 
Pb = Q
2b
(6)
1

Figure 1: QPSK constellation

The symbol probability of error (SER) is the probability of either branch
has a bit error:
p 
Ps = 1 [1 Q
2b ]2
(7)
Since the symbol energy is split between the two in-phase and quadrature components, s = 2b and we have:

Ps = 1 [1 Q ( s )]2

(8)

We can use the union bound to give an upper bound for SER of QPSK. Regarding figure 1, condition that the symbol zero is sent, the probability of error is
bounded by the sum of probabilities of 0 1, 0 2 and 0 3. We can write:
p
p
p
Ps Q(d01 / 2N0 ) + Q(d02 / 2N0 ) + Q(d03 / 2N0 )
(9)
p
p

= 2Q(A/ N0 ) + Q( 2A/ 2N0 )

(10)
Since s = 2b = A2 /N0 , we can write:
p

Ps 2Q( s ) + Q( 2s ) 3Q( s )

(11)

Using the tight approximation of Q function for z  0:

2
1
Q(z) ez /2
z 2

we obtain:
Ps

3
e0.5s
2s

(12)

(13)

Using Gray coding and assuming that for high signal to noise ratio the errors
occur only for the nearest neighbor, Pb can be approximated from Ps by Pb
Ps /2.

Figure 2: MPSK constellation

1.3

BER for MPSK signaling

For MPSK signaling we can calculate easily an approximation of SER using

nearest neighbor approximation. Using figure , the symbol error probability
can be approximated by:





p
2A sin M
dmin

Ps 2Q
2s sin(/M )
(14)
= 2Q
= 2Q
2N0
2N0
This approximation is only good for high SNR.

1.4

BER for QAM constellation

The SER for a rectangular M-QAM (16-QAM, 64-QAM, 256-QAM etc) with
size L = M 2 can be calculated by considering two M-PAM on in-phase and
quadrature components (see figure 3 for 16-QAM constellation). The error
probability of QAM symbol is obtained by the error probability of each branch
(M-PAM) and is given by:
Ps = 1

2 (sqrtM 1)
Q
1
sqrtM

3 s
M 1

!!2
(15)

If we use the nearest neighbor approximation for an M-QAM rectangular constellation, there are 4 nearest neighbors with distance dmin . So the SER for
high SNR can be approximated by:
c

(16)

In order to calculate the mean energy per transmitted symbol, it can be seen
that
M
1 X 2
Es =
A
(17)
M i=1 i

Figure 3: 16-QAM constellation

Modulation
BPSK
QPSK
MPSK
M-QAM

Ps (s )


Ps 2Q s

Ps 2Q 2s sin

q
3 s
Ps 4Q
M 1

Pb (b )


Pb = Q 2b

Pb Q 2b p
Pb log2 M Q
2 log2 M sin
2
q b

3 b log2 M
Pb log4 M Q
M 1

Table 1: Approximate symbol and bit error probabilities for coherent modulation
Using the fact that Ai = (ai + bi ) and ai and bi {2i 1 L} for i = 1, ..., L.
After some simple calculations we obtain:
Es =

L
d2min X
(2i 1 L)2
2L i=1

(18)

For example for 16-QAM and dmin = 2 the E s = 10. For 64-QAM and dmin = 2
the E s = 21.

1.5

conclusion

The approximations or exact values for SER has the following form:
p

Ps (s ) M Q
M s

(19)

where M and M depend on the type of approximation and the modulation

type. In the table 1 the values for M and M are semmerized for common
modulations.
We can also note that the bit error probability has the same form as for
SER. It is:
q

Pb (b )
M Q
M b
(20)
where
M = M / log2 M and M = M / log2 M .
Note: s = Es /N0 , b = Eb /N0 , b =
4

s
log2 M

and Pb

Ps
log2 M .

1.6

Appendix

In this appendix the reference curve for AWGN channel is presented in figure
4. As we expected , the results for BPQK and QPSK are the same.
Gaussian Channel

10

BPSK
QPSK
8PSK
16QAM

10

10

BER

10

10

10

10

10

10

10

12

14

16

18

Eb/N0 (dB)

Figure 4: BER over AWGN channel for BPSK, QPSK, 8PSK and 16QAM
The following matlab program illustrates the BER calculations for BPSK
over an AWGN channel.
%BPSK BER
const=[1 -1];
size=100000;
iter_max=1000;
EbN0_min=0;
EbN0_max=10;
SNR=[];BER=[];
for EbN0 = EbN0_min:EbN0_max
EbN0_lin=10.^(0.1*EbN0);
noise_var=0.5/(EbN0_lin); % s^2=N0/2
iter = 0;
err = 0;
while (iter <iter_max && err <100),
bits=randsrc(1,size,[0 1]);
s=const(bits+1);
x = s + sqrt(noise_var)*randn(1,size);
bit_hat=(-sign(x)+1)/2;
err = err + sum(bits ~= bit_hat);
iter = iter + 1;
5

end
SNR =[SNR EbN0];
BER = [BER err/(size*iter)];
end
semilogy(SNR,BER);grid;xlabel(E_bN_0);ylabel(BER);
title(BPSK over AWGN channel);
The following program uses some advanced functions of matlab to evaluate the
symbol error rate for QPSK modulation:
M = 4; % Alphabet size
EbN0_min=0;EbN0_max=10;step=2;
SNR=[];SER=[];
for EbN0 = EbN0_min:step:EbN0_max
SNR_dB=EbN0 + 3; %for QPSK Eb/N0=0.5*Es/N0=0.5*SNR
x = randint(1000000,1,M);
y=modulate(modem.qammod(M),x);
ynoisy = awgn(y,SNR_dB,measured);
z=demodulate(modem.qamdemod(M),ynoisy);
[num,rt]= symerr(x,z);
SNR=[SNR EbN0];
SER=[SER rt];
end;
semilogy(SNR,SER);grid;titel(Symbol error rate for QPSK over AWGN);
xlabel(E_b/N_0);ylabel(SER);

SER and BER over fading channel

2.1

PDF-based approach for binary signal

A fading channel can be considered as an AWGN with a variable gain. The

gain itself is considered as a RV with a given pdf . So the average BER can
be calculated by averaging BER for instantaneous SNR over the distribution of
SNR:
Z
Pb (E) =

Pb (E|)p ()d
0

The BER is expressed by a Q-function as seen in previous chapter:

Z p
Pb (E) =
Q( 2g)p ()d

(21)

where g = 1 for the case of coherent BPSK.

Example 1. Rayleigh fading channel with coherent detection:
The received signal in a Rayleigh fading channel is of the form:
y = hx + w
1 Digital

Communication over Fading Channel by Simon and Alouini

(22)

where h is the channel attenuation with normal distribution h CN (0, 1) and

n is a white additive noise w CN (0, N0 ). The coherent receiver constructs
the following metric from the received signal:
h y = |h|2 x + h w

(23)

Using BPSK modulation and since the information are real, only the real part
of the equation is of interest. So the following sufficient statistic is used for
decision at the receiver.
 
h
y = |h|x + n
(24)
<
|h|
The noise n has the same statistics as <w because h /|h| = exp(j) with
uniformly distributed in (0, ), therefore n CN (0, N0 /2). This equation
shows that we have a normal AWGN channel with the signal scaled by |h|.
The bit error probability as seen before for this case, given h, will be:
p

Pb = Q
2|h|2 b
Now, we compute the SER by averaging this BER over the distribution of h.
Since h is complex Gaussian, the distribution of r = |h|2 will be exponential
with:
Pr (r)

=
=

d
P (h2r + h2i < r)
dr
!
Z 2 Z r
d
1
x2
e
xdxd
dr
21/2
0
0

d
1 er
dr
= er U (r)
=

Therefore the signal-to-noise-ratio distribution = |h|2 b will be:

p () =

1 /b
e
b

The error probability can be calculated by:

Z p
Z
Pb =
Q( 2)p ()d =
0

p
1
Q( 2) e/b d
b

Using the following form of Q-function and MGF function, the integral can eb
calculated.
Z
1 /2
x2
Q(x) =
exp(
)d
0
2 sin2


r
1
b
pb =
1
2
1 + b
7

Example 2. Consider a SIMO system with L receive antennas. Each branch

has a SNR
PL per bit of l and therefore the SNR at the output of MRC combiner
is t = l=1 l . Suppose a Rayleigh channel, the pdf of SNR for each channel
will be (supposing i.i.d. channels):
pl (l ) =

1 l /
e

At the output of combiner, the SNR follows the distribution of chi-square (or
gamma) with L degrees of freedom:
1
L1 et /
(L 1)!
L t

pt (t ) =

The average probability can be calculated using the integration by part and
resulting in the following formula:

Pb (E) =

2.2
2.2.1

1
2

L L1
X
l=0

L1+l
l



1+
2

l

MGF-based approach
Binary PSK

We can use the other representation of Q-function to simplify the calculations.

Z

Q(x) =
x

 2


Z
y
1 /2
x2
1
exp
dy =
exp
d
2
0
2 sin2
2

Therefore the equation (5) can be written as:

Pb (E|{l }L
l=1 ) =

/2

exp(
0

gt
1
)d =

sin2

Z
0

L
/2 Y

exp(

l=1

gl
)d (25)
sin2

This form of Q-function is more convenient because it allows us to average

first over the individual distributions of l and then perform the integral over
.
Z Z Z
L
Y
Pb (E) =
...
Pb ({l }L
)
pl (l )d1 d2 ...dL
(26)
l=1
0

l=1

Using (25) in (26) and changing the order of integration gives:

Pb (E) =

Z
0

L
/2 Y
l=1


M l

g
sin2


d

(27)

2.2.2

MPSK

For MPSK signaling the SER given all the SNRs is:
Ps (E|{l }L
l=1 ) =

Ps (E|{l }L
l=1 )

1
=

(M 1)/M



gt
exp 2
d
sin

L
(M 1)/M Y


gl
exp 2
d
sin
l=1

where g = sin2 (/m).

(28)

(29)