Improving the Tolerance of Stochastic LDPC Decoders to Overclocking-Induced Timing Errors: A Tutorial and a Design Example

Abstract

Channel codes, such as low-density parity-check (LDPC) codes, may be employed in wireless communication schemes for correcting transmission errors. This tolerance to channel-induced transmission errors allows the communication schemes to achieve higher transmission throughputs, at the cost of requiring additional processing for performing LDPC decoding. However, this LDPC decoding operation is associated with a potentially inadequate processing throughput, which may constrain the attainable transmission throughput. In order to increase the processing throughput, the clock period may be reduced, albeit this is at the cost of potentially introducing timing errors. Previous research efforts have considered a paucity of solutions for mitigating the occurrence of timing errors in channel decoders, by employing additional circuitry for detecting and correcting these overclocking-induced timing errors. Against this background, in this paper, we demonstrate that stochastic LDPC decoders (LDPC-SDs) are capable of exploiting their inherent error correction capability for correcting not only transmission errors, but also timing errors, even without the requirement for additional circuitry. Motivated by this, we provide the first comprehensive tutorial on LDPC-SDs. We also propose a novel design flow for timing-error-tolerant LDPC decoders. We use this to develop a timing error model for LDPC-SDs and investigate how their overall error correction performance is affected by overclocking. Drawing upon our findings, we propose a modified LDPC-SD, having an improved timing error tolerance. In a particular practical scenario, this modification eliminates approximately 1-dB performance degradation that is suffered by an overclocked LDPC-SD without our modification, enabling the processing throughput to be increased by up to 69.4%, which is achieved without compromising the error correction capability or processing energy consumption of the LDPC-SD.