In this thesis, the joint source-channel coding method, channel optimized scalar quantization, is applied to real-valued, correlated data. The data is sent over the orthogonal multiple access channel, with non-binary noisy discrete channels with memory as the two sub-channels. Three different schemes are compared for this system: in the first scheme encoding and decoding are performed independently, in the second scheme encoding is done independently and joint decoding is carried out, and the third scheme is with jointly optimized encoders and joint decoding. The goal is to derive optimality conditions that will result in a lower end-to-end distortion. To this end, necessary optimality conditions for the two latter schemes are fully derived and implemented for the bivariate Gaussian and bivariate Laplacian distributions of varying correlation.;The first and second methods are then further compared, by implementing them for an image transmission system. Here the images are first processed with the 2 dimensional discrete cosine transform, and then encoded using channel optimized scalar quantization. At the decoder, two different methods are used, the independent and joint decoder.;In addition to comparing the different coding methods, various channels characteristics are exploited. For example, the non-binary noisy discrete channel can be used to model memory and the 2q-ary output allows for performance improvement via soft-decision decoding. It is observed that by taking the source correlation into consideration, significant signal-to-distortion ratio gains can be achieved. For example, the highest gain incurred from the third scheme is when the bivariate Gaussian is compressed at rate 2, where the gain in signal-to-distortion ratio due to source correlation is 10.90 dB.
展开▼
