| Summary: | We develop source-aided channel decoding techniques and apply them for image
transmission using the channel forward error correction codes of the IS-95 Code Division
Multiple Access (CDMA) standard. The source is modeled as a first order Markov model
with states that correspond directly to the source coder codewords. The model is used to
form a MAP version of the Viterbi algorithm for decoding convolutional codes. For the
case of a two-state Markov model, the generalization of the Viterbi algorithm involves
only a modification of the branch metric; while for N-state Markov models, a technique
called trellis merging is also implemented to keep the decoding complexity low. An
iterative model recovery technique is developed which allows the receiver to recover the
source model without any a priori information. Simulating these techniques for the case of
the two-bit DPCM encoded Lenna image, we find a coding gain over an Additive White
Gaussian Noise (AWGN) channel of approximately 1.1 dB at a BER of 10⁻⁴ for both the
forward and reverse link of the IS-95 standard.
We go on to develop a turbo version of the IS-95 reverse link decoder. This
involves implementing a "soft-in/soft-out" version of the component decoders, and
introducing an iterative decoding procedure. The coding gain found by this turbo enhancement
is 0.75 dB at a BER of 10⁻³. A Markov model-aided version of the turbo reverse link
decoder is then developed by migrating the techniques used in the Viterbi algorithm to the
soft-in/soft-out decoders. The ensuing Markov model aided turbo decoder has a two-level
iterative structure due to the fact that there are source model recovering iterations and
turbo iterations. Three different architectures for the two-level iterative structure are
proposed and compared. Although all three methods provide similar coding gain over an
AWGN channel, they differ in speed of convergence and implementation complexity. For
the case of transmission of the two-bit DPCM encoded Lenna image, the coding gain over
an AWGN channel at the final iteration is 1.6 dB at a BER of 10⁻². === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|
|---|
