The UC Berkeley Combinatorics Seminar
Spring 2005
March 28, 2005, Time: 2:10-3:00, Location: 939 Evans





Geometric perspective of error correcting codes

Roman Vershynin

UC Davis




Abstract


Error correcting codes are used in modern technology to protect information from errors. An encoder transforms an n-letter word X (over some some alphabet F) into an m-letter word Y with m>n. The decoder must be able to recover X even if r letters of Y are corrupted in any way. Known constructions of algorithmically efficient error correcting codes involve deep combinatorial and algebraic methods. I will talk on a different approach to error correcting codes, through the asymptotic convex geometry (geometric functional analysis). The main ideais that most of orthogonal transformations from R^n to R^m form good error correcting codes over reals (with an efficient decoder). These codes are robust, so quantization leads to good codes over finite alphabets. The intuition for this method has been developed from experimental and theoretical work in signal processing (sparsest representations of signals). This a joint work with Mark Rudelson.