Study of Neural Network Based Coding Schemes for Telemetry Data

This project evaluated the performance of various coding schemes used for compression of multiple channel telemetry data. Lossless transmission and latency have been important considerations in addition to the compression ratio. Several schemes including linear predictive coding and entropy coding schemes such as adaptive Huffman coding and adaptive arithmeitc coding were studied and their performance compared with non-linear prediction schemes using neural networks. Flight data obtained from ISRO was used for testing the schemes. Run length encoding of individual channels along with a type of random multiplexing is found to offer good results with compression ratios of around 10 for reasonable latency.