Stochastic Computing Improves the Timing-Error Tolerance and Latency of Turbo Decoders: Design Guidelines and Tradeoffs

Stochastic computing has been recently proposed for the hardware implementation of both low-density parity-check (LDPC) decoders and turbo decoders, which facilitate near-optimal error correction capabilities in wireless communication applications. Previous contributions have demonstrated that stochastic LDPC decoders offer an attractive tradeoff between their error correction capabilities, hardware performance, and timing-error tolerance. Motivated by this, we propose a pair of stochastic turbo decoder (STD) designs having significantly enhanced timing-error tolerance and significantly reduced processing latency. Moreover, we characterize the tradeoffs between chip area, energy efficiency, latency, throughput, and error correction capabilities of both the timing-error-tolerant STD and of the reduced-latency STD. We demonstrate that our proposed timing-error-tolerant STD operated at 1.20 V, with a clock period of 2.2 ns and in the presence of a three-standard deviation power supply variation of 7%, exhibits an unimpaired performance, compared with the state-of-the-art STD, operated at 1.20 V and 4 ns and with no power supply variations. This corresponds processing throughput improvement by a factor of 2.42 and energy consumption reduction by a factor of 4. Finally, we demonstrate that our proposed reduced-latency STD has a processing latency that is an order of magnitude lower than that of the state-of-the-art STD. This is despite reducing the chip area by a factor of 4, increasing the processing throughput by a factor of 65, while consuming only 0.005 times the energy of the state-of-the-art STD, when using binary phase-shift keying for communication over an additive white Gaussian noise channel having E_b/N₀ = 3 dB.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.