Neural network based lossless coding schemes for telemetry data

This paper proposes new coding schemes based on neural networks for the compression of telemetry data. It is shown that neural network predictors can be used successfully in a two-stage lossless compression scheme. Single-layer perceptron, multi-layer perceptron and recurrent network models are investigated for this purpose. The proposed neural network based coding schemes are tested using different telemetry data files. For the encoder in the second stage, arithmetic and Huffman coding are employed. It is found that the performance of neural network based schemes is comparable and in some cases better than that of the methods using linear predictors such as FIR and lattice filters.