Using Machine Learning to Mitigate Single-Event Upsets in RF Circuits and Systems

The present article applies the $k$ -nearest neighbors ( $k$ -NN) machine learning (ML) algorithm to detect and correct single-event upsets (SEUs). In particular, this work focuses on SEUs resulting from single-event transients in RF systems carrying modulated data. Pulsed-laser measurements were performed, and the resulting SEUs were used to train separate $k$ -NN algorithms to detect and correct these upsets. The results show that the algorithm correctly classified data into “upset” and “no upset” 99.2% of the time. In addition, the number of symbol upsets was reduced by 30%. Some of the challenges to implementing and deploying this error correction approach in flight systems are discussed. The results of the present work demonstrate the potential benefits of applying ML techniques to the field of radiation effects.