Early prediction of packet errors in FEC-encoded systems with very few decoding iterations

Abstract

Modern Forward Error Correction (FEC) techniques can involve significant processing delays associated to FEC decoding. So-called centralized RAN (C-RAN) networks can be impaired when FEC processing is centralized, particularly if large bandwidths are foreseen (like in mm-wave systems), because of the tight delay requirements imposed by Hybrid Automatic Repeat Request (HArQ). Decoupling HARQ from FEC decoding has the potential advantage of relaxing the transport network requirements by predicting errors prior to performing FEC decoding. In this paper we propose a technique to predict packet errors in FEC-encoded systems based on statistical analysis of soft metrics after very few FEC decoding iterations. Simulations confirm the suitability of the proposed technique for 256-bit Turbo encoded packets with rate 1/3 and two decoding iterations, by defining appropriate uncertainty regions outside which the probability of false decisions is bounded. Analysis of other block sizes and encoding rates suggests the existence of an optimum packet size and number of iterations as functions of the encoding rate for error prediction.