Software-Based Fault Tolerance with Arithmetic Coding for Floating-Points

Arithmetic coding, also known as encoded or coded processing, is a software-based fault-tolerance approach that detects hardware faults at the software level by exploiting information redundancy. Existing arithmetic coding focuses on integer arithmetic. However, the increasing complexity of safety functions, e.g. from industrial robotics, requires floating-point arithmetic. This paper shows the challenges of applying existing integer codes to floating-point numbers. We identify and compare existing approaches from literature and new approaches in this paper. The comparison includes an analysis of performance, fault detection capability, hardware reliability, constraints on algorithm development, and accuracy.