A hybrid FEC-ARQ communication system using sequential decoding

Abstract

This paper presents the results of a study which aimes at defining an error correcting scheme matched to the troposcatter (Rayleigh) channel, which would permit to insure very high reliability links for digital data transmissions.

Forward Error Correction (FEC) and interleaving methods are not compliant to the high variability of the fades durations, and insuring an excellent quality by these means is quite impossible, even with very efficient, complex and expensive codes associated with deep interleavers.

Usual Automatic Repeat reQuest (ARQ) systems are not satisfactory either, because of three reasons at least. First, an error detecting code would be overwhelmed and the retransmissions would be too numerous ; moreover, an error burst may be undetected and the residual BER can be relatively high. Secondly, each retransmission request increases the delay which may become incompatible with telephony applications or even diverge. In the third place, ARQ systems usualy assume that the feedback channel is noiseless ; in our applications the backward channel is also a Rayleigh channel, and the acknowledgement may sometimes be erased by fades.

We propose a hybrid ARQ+FEC system using a long constraint length convolutional code and the sequential decoding algorithm (Stack or MultiStack Algorithm), associated with a modified Go-Back-N ARQ protocol, able to cope with a two-way noisy channel. The data are sent in a continuous stream of blocks, according to an adaptive frame structure permitting to catch up the transmission delay due to the repeat requests. The protocol control is achieved by a modified HDLC structure.

The features of this system will be given for a particular implementation example : for a data rate of 1 Mbit/s and a rate 1/2 convolutional code of constraint length 35, the effective throughput rate is about 1/4, and a zero-error transmission is assured with a bounded transmission delay (<250 ms) as soon as the average SNR is better than 1 dB in threefold diversity.