On error-rate characteristics of oversampled A/D conversion

Accuracy of simple A/D conversion can be improved by refining either sampling interval or quantisation step. Although the former is more convenient for implementation reasons, the latter seems to be more efficient. When the quantization step tends to zero for a fixed sampling interval, the expected value of the quantization error decays exponentially as a function of the bit-rate B, E(e(t)/sup 2/)=O(exp(-2/spl alpha/B)). On the other hand, if the oversampling ratio is increased for a fixed quantization step the error behaves as E(e(t)/sup 2/)=O(1/B). We show that this inferior error-rate performance of oversampled A/D conversion is a consequence of inadequate reconstruction and encoding, and propose an efficient scheme for lossless encoding of quantized samples. With this encoding scheme the error of oversampled A/D conversion decays exponentially with the bit-rate.