Performance Peculiarities of Viterbi Decoder in Mathworks Simulink, GNU Radio and Other Systems with Likewise Implementation

TL;DR

This paper investigates how specific implementation details of the Viterbi decoder in various systems can cause decoding performance to depend on input message composition, contrary to theoretical expectations.

Contribution

It identifies that certain Add-Compare-Select units in Viterbi decoders cause performance variations based on input message distribution, highlighting a potential issue in commercial systems.

Findings

Performance depends on input message composition in some implementations

Identifies specific implementation features affecting decoding efficiency

Raises concerns about variability in commercial communication equipment

Abstract

The performance of convolutional codes decoding by the Viterbi algorithm should not depend on the particular distribution of zeros and ones in the input messages, as they are linear. However, it was identified that specific implementations of Add-Compare-Select unit for the Viterbi Algorithm demonstrate the decoding performance that depends on proportion of elements in the input message. It is conjectured that the modern commercial hard- and software defined communication equipment may also feature similar implementation and as such their decoding performance could also vary.