Bounds on a probability for the heavy tailed distribution and the probability of deficient decoding in sequential decoding

Although sequential decoding of convolutional codes gives a very small decoding error probability, the overall reliability is limited by the probability P_G of deficient decoding, the term introduced by Jelinek to refer to decoding failures caused mainly by buffer overflow. The number of computational efforts in sequential decoding has the Pareto distribution and it is this "heavy tailed" distribution that characterizes P_G. The heavy tailed distribution appears in many fields and buffer overflow is a typical example of the behaviors in which the heavy tailed distribution plays an important role. In this paper, we give a new bound on a probability in the tail of the heavy tailed distribution and, using the bound, prove the long-standing conjecture on P_G, that is, P_G ap constanttimes1/(sigma^rhoN^rho-1) for a large speed factor sigma of the decoder and for a large receive buffer size N whenever the coding rate R and rho satisfy E(rho)=rhoR for 0 les rho les 1