Embedded multicore architectures for LDPC decoding

Recently, the development of Low-Density Parity-Check (LDPC) decoding solutions has been proposed for a vast set of architectures, ranging from dedicated hardware to fully programmable ones (e.g. Cell/B.E. and graphics processing units). In this paper we propose efficient embedded programmable multicore architectures for achieving real-time LDPC decoding. The proposed multicore architectures allow to exploit data parallelism by decoding in parallel multiple codewords on the provided cores, with enough local memory capacity to store all data corresponding to the Tanner graph. Therefore, with this distributed memory and local computing approach, only a single shared bus is required to communicate the codewords. The proposed class of architectures can be prototyped on field programmable gate arrays or implemented on application specific integrated circuits, and it is validated by using the popular Cell processor, which relates very closely with the one here proposed. Finally, we discuss the related art of dedicated and programmable LDPC decoders, and discuss the advantages and disadvantages regarding the proposed solution.