Gradient Flow Decoding for LDPC Codes

Abstract: The power consumption of the integrated circuit is becoming a significant burden, particularly for large-scale signal processing tasks requiring high throughput. The decoding process of LDPC codes is such a heavy signal processing task that demands power efficiency and higher decoding throughput. A promising approach to reducing both power and latency of a decoding process is to use an analog circuit instead of a digital circuit. This paper investigates a continuous-time gradient flow-based approach for decoding LDPC codes, which employs a potential energy function similar to the objective function used in the gradient descent bit flipping (GDBF) algorithm. We experimentally demonstrate that the decoding performance of the gradient flow decoding is comparable to that of the multi-bit mode GDBF algorithm. Since an analog circuit of the gradient flow decoding requires only analog arithmetic operations and an integrator, future advancements in programmable analog integrated circuits may make practical implementation feasible.