Implementation of linear predictive and Huffman coding algorithms for real time data compression

Digital signal processing techniques are used extensively in data acquisition and processing systems. These systems generate large amounts of digital data, most of which must be archived for subsequent analysis. With today's higher sampling and processing rates, this results in very large storage requirements for the acquisition system. To effectively decrease these storage requirements, data compression techniques are applied to the data stream before storage. A recent study has shown that a combination of linear predictive coding and Huffman coding offers a significantly improved method of data compression for instrumentation data. The purpose of this thesis is to adapt these techniques to perform data compression of instrumentation signals in real time, using a dedicated digital signal processor - the Texas Instruments TMS320C40. Here, real time means that the input data rate is not interrupted or changed as a result of the compression process. In this thesis, existing linear predictive compression algorithms are adapted and optimized for use in a real time signal processing environment. The original FORTRAN-based compression code is first ported to C for use with the TMS320C40. In this development, the code is verified and various optimization techniques are applied. The special hardware features of the TMS320C40 are exploited to gain every possible performance advantage. Finally, select portions of the code are benchmarked using Texas Instruments' TMS320C40 simulator. The result of this work is a prediction of the compression code's real time performance as a function of both the processor's clock rate and data throughput.