Hardware realization of a multistage speech waveform vector quantizater

In this paper, we report on a working multistage speech waveform VQ system based on mostly LSTTL devices without any special purpose processors. The main features of the system are as follows. First, we use companding AD/DA converters to overcome the inadequate linear 8-bit word length while maintaining the whole system efficient. Secondly, full pipeline structure and table lookup for the squaring operations make the full search speed of our system depend only on the speed of memories. Lastly, multi-stage scheme is used to achieve higher vector dimension and codebook size. It turns out that the hardware coder can match the computer simulation very well, the SNR value of which achieves about 15 dB at 2 bits/sample and 12 dB at 1 bit/sample.