Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
August 16 - August 21
I. CHANNEL
MODEL
As in [l],the frequency hopping channel is modeled by a discrete time-varying AWGN channel. The channel adds zero
mean white Gaussian noise with time varying variance ( U : ) .
We assume perfect symbol synchronization and perfect power
control. Without loss of generality, we also assume that the
transmitted symbols dt E (1, -1). Hence, the log-likelihood
ratio at time t (assuming prior knowledge of the channel variance) is
Lt
Evaggelos Geraniotis
Dept. of Electrical Engineering
Institute for Systems Research
University of Maryland
College Park, MD 20742
evagelosOeng.umd.edu
the noise variance to be used in calculating the log-likelihood
ratio at time k is given by
1
4 e 8
=n-1
[POt(Yt
l<t<n,t#k
(2)
where c = 2(1 - 2 ~ 0 6 ) ~2 is a constant added to unbias the
estimator; and
pot = [eq+%,t) I ] -
(3)
in which LeZt is the extrinsic information provided by the
previous decoding step.
When the receiver does not know which received symbols
are corrupted by a constant variance noise, or when the noise
variance changes sufficiently fast such that each symbol is affected by a noise sample with different variance, we need to
find a robust estimator for the log-likelihood ratio such that
the highly corrupted symbols do not affect their surrounding
symbols. The generalized maximum likelihood ratio test is
used for obtaining this robust estimator
2Yt
4
11. EST~MATION
SCHEMES
In this section, we briefly describe three log-likelihood ratio
estimation schemes. The first scheme attempts to take advantage of the iterative structure of the Turbo decoder in order
to improve the channel variance estimate after each decoding step. In order to use this technique, the variance of the
channel must be constant over more than one symbol, and
the receiver must have a priori knowIedge of which symbols
were corrupted by noise vector with constant variance. The
extrinsic information supplied by the previous decoding step
is used as a priori probability in the variance estimator.
Let pot be the probability that (dt = -l), and {yl,yz, ......y,}
are the received symbols which are known to be corrupted by
a constant variance noise vector. A suboptimum estimate of
0-7803-5000-6/O8/$10.00
0 1998 IEEE.
+ 11
(5)
+ 11
IYt - 11
The third scheme is the simplest and is only used for comparison purposes. The log-likelihood ratio used is equal to the
channel output yt.
Upper bounds for the BER achieved by the three schemes
were obtained following [Z]. The reader is referred to [3] for
the detailed numerical results obtained through both analysis
and simulation.
REFERENCES
[I] E. Geraniotis, Multiple Access Capability of Requency
Hopped Spread Spectrum Revisited, IEEE Pans. on Communications, pp. 1066-1077,Jul 1990.
[2] D. Divsalar, S. Dolinar, and F. Pollara, Transfer function
bounds on the performance of turbo codes, Telecom. and Data
Acquisition Progress Report 42-122, Jet Propulsion Laboratory,
August 1995.
Gamal, and E. Geraniotis, Turbo Codes with Channel Estimation and Dynamic Power Allocation for Anti-Jam
SFHISSMA, submitted to MILCOM 98, October 1998.
[3] H. El
457