Intelligent information system for failure-resistant decoding of block turbo codes in distributed computer systems

Abstract

The paper addresses the problem of reducing the computational complexity of iterative decoding of block turbo codes in distributed heterogeneous computing systems. A four-layer intelligent information system (IIS) architecture is proposed, combining a heuristic modification of the Chase–Pyndiah algorithm based on probabilistic Bernoulli sampling, an SNR-oriented weighted-least-loaded (WLL) scheduler, and a task-level fault-tolerance mechanism without data loss. The first system layer estimates the signal-to-noise ratio and forms a priority queue of SNR-annotated frames. The second layer implements the WLL scheduler, routing computationally expensive frames to more powerful nodes according to their weighted current load. The third layer contains a pool of stateless decoders, and the fourth layer reorders results by frame sequence numbers. The heuristic modification replaces deterministic enumeration of test vectors with a tau-sample drawn from the a posteriori channel distribution, which adaptively reduces the number of calls to the hard-decision decoder depending on the current noise level. Simulation on codes with rates R = 0.697 and R = 0.254 at Eb/N0 ranging from 4.0 to 5.0 dB demonstrated simultaneous reduction of the frame error rate by 15–83 % and of the average Berlekamp–Massey decoder call count by 15–26 % compared to the baseline. System throughput grows almost linearly with the number of nodes: the utilisation of a heterogeneous eight-node cluster exceeds 89 %. When one node fails, all tasks are recovered without data loss with an overhead not exceeding 1.3 of the average single-frame decoding time. The proposed architecture targets satellite ground stations, sensor network gateways, and cloud data centre’s. Future research directions include online parameter adaptation via reinforcement learning and integration with graph neural decoders to further reduce the frame error rate.