Efficient Computation of Convolutional Decoder Reliability Without a CRC

The reliability-output Viterbi algoritlun (ROVA) of Raghavan and Rawn computes the probability that the codeword selected by Viterbi decoding is in error, allowing unreliable decoding to be identified without the overhead of a cyclic redundancy check (CRC). ROVA can be used as a stopping criterion for variable length (VL) codes with feedback. Separately, Polyanskiy et al. proposed accumulated information density (AID) as a stopping criterion for VL codes for computation of random coding bounds. This paper compares the accuracy and complexity of ROVA and AID. It turns out that All) is far less accurate than ROVA. This paper proposes codeword information density (CID), which modifies AID to improve its accuracy and leads to a lower complexity implementation of ROVA. The paper concludes with an analytical expression for the random variable describing the correct decoding probability computed by ROVA and uses this expression to characterize how the probabilities of correct decoding, undetected error, and negative acknowledgement behave as a function of the selected threshold for reliable decoding.