Real number convolutional code correction and reliability calculations in fault-tolerant systems

Abstract

An efficient fault-detecting methodology, algorithm-based fault tolerance, may be extended to include error correction of the output data in a protected linear processing system by coupling a high-rate real convolutional code with a smoothed Kalman recursive estimation technique. A completely protected fault-tolerant linear processing system involving error correction is shown where it is guaranteed that no miscorrected data leave the configuration if at most one box-surrounded subsystem fails at a time. The real convolutional code dictates the comparable parity streams computed in two ways, forming the syndrome stream that is passed to the fixed-lag corrector when values exceed the threshold settings. The block processing and down sampling features of the convolutional code permit the overhead area to be from 20-50% of the main processing area.