engleski

Implementation of Sparse Matrix Arithmetic on a DSP Processor

The paper presents a method for sparse matrix multiplication on a DSP processor. Its high efficiency is a consequence of the proposed pseudo-random data memory access and parallelism of the multifunctional instructions of a DSP. Sparse matrix multiplication is implemented as linear expanded DSP code automatically generated by specially designed program. The method is applied to predictive vector quantization of Line Spectrum Frequencies vectors used in speech coding. It will be shown that the obtained reduction in computational complexity and fixed storage requirements is between two and three-fold.

sparse matrix; random access; DSP; speech coding; LSF quantization; predictive vector quantization

nije evidentirano