Soft methods of noise combating code decoding are widely used in modern information systems. It makes it possible to get the energy gain, which is especially important for mobile communication systems. Effectiveness of soft decoder’s construction depends on choice of the method of soft decision indices forming. Most of the research in this domain deal with the method of forming the soft decisions for the characters of code vectors by calculating the log-likelihood ratio. The restriction of this method consists in the real numbers notation of soft decisions. This paper presents the method of soft decisions forming based on a modification of the erasure channel. For this purpose there is used a wide interval of erasure and all the characters received outside of this interval are considered with a high probability to be received reliably. They get the highest reliability estimation. The characters in the erasure interval get the reliability indices in accordance with the linear dependence, which by-turn depends on the chosen maximum estimate of the soft decisions. The paper analyzes various approaches to the iterative transformations of soft decisions, which lead to increase the likelihood function for code vector.