Method of improving of soft multithreshold decoder's performance for self-orthogonal codes is considered. A modified way of selecting weights coefficients on initial iterations of multithreshold decoding (MTD) at high noise is described. Some simulated results obtained by using these two decoding techniques (the base and modified ways ) are presented as well.Comparison of the results showed that the change in the computing of weights coefficients can significantly improve decoding parameters of a convolutional code with soft MTD. It's shown that MTD is in many cases as effective as an optimum decoder. MTD implementation complexity is also discussed.MTD is characterized with a very small number of operations, soft versions of these decoders can correct in channels with high enough noise level streams of the data in high-speed communication systems