We revisit the error correction scheme of real-valued signals when thecodeword is corrupted by gross errors on a fraction of entries and a smallnoise on all the entries. Combining the recent developments of approximatemessage passing and the spatially-coupled measurement matrix in compressedsensing we show that the error correction and its robustness towards noise canbe enhanced considerably. We discuss the performance in the large signal limitusing previous results on state evolution, as well as for finite size signalsthrough numerical simulations. Even for relatively small sizes, the approachproposed here outperforms convex-relaxation-based decoders.
展开▼
