Fast software polar decoders

Abstract

Among error-correcting codes, polar codes are the first to provably achieve channel capacity with an explicit construction. In this work, we present software implementations of a polar decoder that leverage the capabilities of modern general-purpose processors to achieve an information throughput in excess of 200 Mbps, a throughput well suited for software-defined-radio applications. We also show that, for a similar error-correction performance, the throughput of polar decoders both surpasses that of LDPC decoders targeting general-purpose processors and is competitive with that of state-of-the-art software LDPC decoders running on graphic processing units.

Autogenerating software polar decoders

“…In this section, we compare the proposed software polar decoder with the fastest software polar decoders in literature [6] and to some hardware polar decoders. For that purpose, all software were compiled using the C++ compiler from GCC 4.8 using the flags "-march=native -funroll-loops -Ofast".…”

“…In this subsection we compare the information throughput and latency of decoders generated with the proposed framework against the single-precision floating point version of the decoders presented in [6]. To the best of our knowledge, the SIMD decoders of [6] are the fastest software polar decoders in the literature.…”

“…To the best of our knowledge, the SIMD decoders of [6] are the fastest software polar decoders in the literature. Table I shows that our proposed decoder has a lower latency and greater throughput for all polar codes and every code rate.…”

“…Recently, software polar decoders have been shown to have high throughput when running on modern CPUs [6]. These decoders implement the fast simplified successive-cancellation (Fast-SSC) algorithm introduced in [5], exploiting CPU parallelism by using single-instruction multiple-data (SIMD) instructions.…”

“…The decoders of [6] could be configured at runtime to decode any polar code of a given length that is set at compile time. This flexibility comes at the cost of extra control logic and indirection in function calls, reducing the speed of the decoder.…”