Iteration Based Analysis of Straggling Channel Decoders for Network Function Virtualization

We consider the uplink of a cloud radio access network, in which baseband functionality is offloaded to the off-the-shelf servers in the cloud via network function virtualization (NFV). Specifically, we consider channel decoding of iteratively decodable channel codes as an important time-critical baseband operation. It has been shown that by reencoding the received channel codewords at the cloud with a linear NFV code, the impact of straggling decoders can be reduced. In our previous work the channel decoding operation at each server, after some probabilistic processing delay, was assumed to have converged to a codeword with only negligible probability of error. In contrast, in this work we leverage the fact that for iterative decoding we can trade-off latency versus decoding error by considering intermediate decoding results, i.e., after each decoder iteration. We derive an expression for the frame unavailability probability as a function of the decoding latency and thus of the number of completed iterations at all decoders. We show that for LDPC codes the latency is significantly reduced if we combine intermediate codeword estimates of the decoders at each iteration.