Mixed-Integer ADMM Decoding for LDPC Codes

In this paper, we study the use of the alternating method of multipliers (ADMM) to decode LDPC codes at moderate blocklengths (100 to 1000 bits). Placing ADMM decoding in a mixed-integer optimization framework, this allows us to leverage an exploration-exploitation paradigm to realize further error-correction performance for the same budget of total number of iterations. We demonstrate our ideas with numerical simulations for the (155,64) rate-0.4 Tanner Code and the (480,360) rate-3/4 code from the IEEE 802.22 WRAN standard.