Abstract: A path metric used in Viterbi decoding can be calculated by adding a slightly attenuated value of a previously calculated path metric to a current branch metric value. The attenuation of the previous path metric may be implemented by multiplying the previous path metric by a constant, where the constant is typically greater than 0.9 and is less than 1. Thus, the maximum path metric value is inherently bounded and is an exponentially-weighted approximation of a moving average of selected previous branch metrics. This eliminates the need for a normalization function to prevent overflow and simplifies the hardware necessary for calculating the path metric.

Abstract: Method for randomly detecting a unique word comprising M bits in a predetermined number of segments within a frame containing L bits of data, where M and L are integers greater than one and wherein M is less than L, includes steps for (a) randomly sampling the data to thereby obtain the segments, (b) comparing a first segment of the segments with the first portion of an actual unique word so as to permit determination of a match between the first segment and the first portion of the actual unique word, (c) when a match is found between the first segment and the respective portion of the actual unique word, comparing a next segment of the segments with the next portion of the actual unique word, (d) when a match is found between the next segment and the next portion of the actual unique word, repeating step (c) until all of the segments are matched with corresponding portions of the actual unique word, and (e) when the comparison of any segment does not result in an appropriate match, returning to step (a) to

Abstract: A modified Type I decoder generates analog syndrome bits from a selected combination of analog signals representing bits yet to be decoded and digital signals representing already decoded information bits. A selected combination of analog syndrome bits is used to generate a weighted majority decision by analog summation. The weighted majority decision is compared to a variable threshold signal which is proportional to the analog magnitude of the bit being decoded. The decision signal which results from this comparison is used to correct the digital value of the bit being decoded. The required number of analog multipliers is reduced by coupling a plurality of multipliers in series and storing the outputs from different analog multipliers in respective analog storage devices.

Abstract: Reliability weighting is provided in the input and/or feedback paths of an analog threshold decoder to achieve additional coding gain. Examples of the reliability weighting function for a bipolar signalling system (.+-.A volts), a receive voltage v and Gaussian noise having a variance .sigma..sup.2 are R=tanh (vA/.sigma..sup.2), R=tanh (1.6 v/A), and a linear transformation having voltage clamps at .+-.1 volt. In the latter case, the gain of the linear transformation in the feedback weighting function may be n times that of the input weighting function, where n is the number of estimates involves in the output.

Abstract: A decoder for correcting and decoding convolutionally-coded digital data utilizes probability information. Convolutionally-coded information and parity sequences are first derived from a demodulator in analog form and then sent to analog shift registers which store the sequences. The shift registers are tapped according to the particular code, and analog multiplications of the tapped shift register outputs are then performed, with sign inversion where necessary, to produce either independent estimates of the output or independent parity checks on a single output estimate. In the former case, the estimates are then summed to a single analog output which represents a binary 1 if positive and a binary 0 if negative. In the latter case, the parity checks are summed to a single analog value which represents disagreement with the single output estimate if positive and agreement with it if negative.