A computation compression technique for linear algorithms
High computation requirements for digital signal processing (DSP) have prohibited many real time DSP implementations. This dissertation presents a technique for reducing the number of computations that are required to implement three classes of DSP algorithms, linear, homomorphic, and point process homomorphic algorithms. The technique, the CCT, is based on grouping the data to be processed into uniformly sized blocks. Each possible data block is processed with the DSP algorithm and stored in memory. In order to limit the number of blocks that must be stored, lossy data compression schemes, such as vector quantization (VQ), are used. The CCT is shown to reduce computation time by up to two orders of magnitude or more.
The advantages of CCT are fewer and simpler computations. The disadvantage is the increase in memory size. The memory required for the CCT is large, growing exponentially as the block size increases for a constant data compression rate. Techniques for reducing the memory size are therefore described. The CCT is easily performed in parallel, so architectures for CCT implementation are described.