Oversampled A / D Conversion using Alternate

In this paper, it is shown that oversampled A/D conversion methods can be substantially improved by using iterative postprocessing involving linear ltering as well as nonlinear operations. The only condition is that the quantizer produces a so-called convex code, which is the case for the simple and dithered A/D conversion , as well as for n th order single path modulation. The iterative algorithm, using alternate projection, searches for the intersection of a subspace (band limited signals) and a convex set (signals with same code). Convergence is guaranteed by convexity. We indicate practical algorithms for simple, dithered, 1 st and 2 nd order (single path) oversampled quantization. The improvements in SNR due to oversampling in these schemes which would normally be 3; 3; 9 and 15dB per octave of oversampling respectively, are increased to 6; 6; 12 and 18dB. That is an increase of 3dB per octave.