ON THE ERROR EXPONENT OF CONVOLUTIONALLY CODED ARQ

Forney's exponent is effectively the best one for the block-coded automatic repeat-request systems (ARQ) and it has been conjectured that a counterpart of Forney's exponent is attainable by convolutional codes with reasonable decoder complexity. In this paper, it is show that the scheme of Yamamoto and Itoh, which is known to be one of the simplest convolutionally coded ARQ schemes, actually gives a better performance than ever expected and that it has error exponent quite close to the conjectured counterpart of Forney's exponent at relatively low rates. It is further shown that the likelihood ratio test used by Yamamoto and Itoh gives Forney's exponent in the case of block-coded ARQ.