Lifting the Fundamental Cone and Enumerating the Pseudocodewords of a Parity-Check Code

Abstract

The performance of message-passing iterative decoding and linear programming decoding depends on the Tanner graph representation of the code. If the underlying graph contains cycles, then such algorithms could produce a noncodeword output. The study of pseudocodewords aims to explain this noncodeword output. We examine the structure of the pseudocodewords and show that there is a one-to-one correspondence between graph cover pseudocodewords and integer points in a lifted fundamental cone. This gives a simple proof that the generating function of the pseudocodewords for a general parity-check code is rational (a fact first proved by Li, Lu, and Wang (Lecture Notes in Computer Science, vol. 5557, 2009) via other methods). Our approach yields algorithms for producing this generating function and provides tools for studying the irreducible pseudocodewords. Specifically, Barvinok's algorithm and the Barvinok-Woods projection algorithm are applied, and irreducible pseudocodewords are found via a Hilbert basis for the lifted fundamental cone.