engleski

Reduced Complexity LSF Vector Quantization with Switched-Adaptive Prediction

A modification of a classical Predictive Vector Quantization (PVQ) technique with switched-adaptive prediction for line spectrum frequencies (LSF) quantization is proposed in this paper, enabling significant reduction in complexity. Lower complexity is achieved through use of higher number of switched prediction matrices but with reduced number of their non-zero elements. The structures of such matrices and optimal matrix elements are obtained to maximize the quantizer closed-loop prediction gain. A comparison of the proposed quantizer to the ones with full prediction matrices as well as to the quantizer incorporating diagonal matrices is given. The effectiveness of the proposed approach is shown and the trade-off between complexity and quality of the quantizer is analyzed.

vector linear prediction; sparse matrices; complexity reduction; LSF quantization

nije evidentirano