Identifying Error-causing Patterns in Sequences for Comprehensive Receiver Validation

Verification of the receiver physical layer is a crucial phase in the development and deployment of high-speed serial communication technologies. With data rates doubling every 3-4 years, ensuring reliable receiver operation becomes increasingly challenging. One of the key problems is the identification of data patterns that are especially difficult for the receiver and have highest risk to trigger errors. Such patterns may have complex structure, they are not necessary continuous sequences of symbols and can span over hundreds of symbols. Extracting error-causing patterns manually is time-consuming and error-prone. Therefore, there is a need for a systematic approach that can efficiently identify error-causing patterns to support comprehensive testing of high-speed receivers. The SHAP (SHapley Additive exPlanations) algorithm, offers a promising method for discovering these patterns with machine learning (ML). This work presents a novel application using SHAP algorithm for pattern detection in data sequences intended for testing of high-speed receivers. We employ three different base models with SHAP to capture the underlying structure of the data, and then leverage the SHAP algorithm to assess the importance of individual or small grouped sequences. The algorithms are tested on four different batches of data. These were obtained from the Industrial Partner of the Project START (Self-Adapting Tool for Automated Receiver Testing). The START project aims to develop an automated tool for above-compliance testing. All the results discussed in this work are obtained from the START project.