We consider the simple additive white Gaussian noise (AWGN) channel. In his 1959 paper [1],
Claude E. Shannon showed that an optimal code is built by uniformly placing codewords
on an n-dimensional sphere. An upper and a lower bound for the word error rate
performance
of such a spherical code have been established by Shannon
on an AWGN channel [1] for finite n.
The main code parameters are its length n and its information rate
.
The length n is the number of real dimensions. The information rate
is expressed
in bits per real dimension. The spherical code is an ensemble of
points
uniformly placed on a sphere in
.
A quick review of Shannon results and its generalization to a Rayleigh fading channel
can be found in [7]. For
, the upper and lower bounds of
are superimposed. Hence, an accurate approximation for
is its lower bound
, the probability of a codeword being moved outside its cone of half-angle
.
Before you read [1] and [7], let me summarize all
the numerical evaluations by two formulas. The first one is used to find
from
n and
, the second one to evaluate
, where
. The two standard signal-to-noise ratios
are related by
.
For the cone half-angle, please use
 |
(1) |
For the word error rate
, please use
![$\displaystyle Q(\theta_0) \approx \frac{1}{\sqrt{n\pi}} \frac{1}{\sqrt{1+G^2} \...
...heta_0 \Big) \Big]^n}{\sqrt{\frac{2E_s}{N_0}}G \sin^2 \theta_0 - \cos \theta_0}$](img19.png) |
(2) |
The approximations are highly accurate in the above expressions. There is no need to compute
the exact integrals involving the solid angle and the error probability [1].
In the next section, I give a C program implementation for
. Sections 3 and 4
illustrate some numerical examples.
Joseph Jean Boutros
2006-11-11