A pipelined adaptive differential vector quantizer for low-power speech coding applications

A fine-grain pipelined adaptive differential vector quantizer architecture is proposed for low-power speech coding applications. The pipelined architecture is developed by employing the relaxed look-ahead technique. The hardware overhead due to pipelining is only the pipelining latches. Simulations with speech sampled at 8 kHz show that, for a vector dimension of 8, the degradation in the signal-to-noise ratio (SNR) due to pipelining is negligible. Furthermore, this degradation is independent of the level of pipelining. Thus the proposed architecture is attractive from an integrated circuit (IC) implementation point of view.<
>