DeepSHARQ: Hybrid Error Coding using Deep Learning

Abstract

Purpose: Cyber-Physical Systems operate under changing environments and on resource-constrained devices. Communication in these environments must use hybrid error coding, as pure pro- or reactiveschemes cannot always fulfill application demands or have suboptimal performance. However, findingoptimal coding configurations that fulfill application constraints—e.g., tolerate loss and delay—underchanging channel conditions is a computationally challenging task. Methods: Recently, the systems community has started addressing these sorts of problems usinghybrid decomposed solutions, i.e., algorithmic approaches for well-understood formalized parts ofthe problem and learning-based approaches for parts that must be estimated (either for reasons ofuncertainty or computational intractability). For DeepSHARQ, we revisit our own recent work andlimit the learning problem to block length prediction, the major contributor to inference time (andits variation) when searching for hybrid error coding configurations. The remaining parameters arefound algorithmically, and hence we make individual contributions with respect to finding close-to-optimal coding configurations in both of these areas—combining them into a hybrid solution. Results: DeepSHARQ applies block length regularization in order to reduce the neural networksin comparison to purely learning-based solutions. The hybrid solution is nearly optimal concerningthe channel efficiency of coding configurations it generates, as it is trained so deviations from theoptimum are upper-bound by a configurable percentage. Additionally, DeepSHARQ is capable ofreacting to channel changes in real-time, thereby enabling Cyber-Physical Systems even on resource-constrained platforms. Conclusion: Tightly integrating algorithmic and learning-based approaches allows DeepSHARQ toreact to channel changes faster and with a more predictable time than solutions that rely only oneither of the two approaches.