Combination and selection of machine learning algorithms in GNSS architecture design for concurrent executions with HIL testing

As machine learning (ML) continuing to gain popularity, ML-assisted Global Navigation Satellite System (GNSS) receivers facilitate the performance of Autonomous Systems (AS) navigation solutions. However, selections of ML is often a trade-off in practice where empirical knowledge is taken to alleviate complexities. Therefore, this paper explores decision-making solutions for maximising determined hardware performance under quantitative and qualitative considerations. This work proposes Algorithm Selection and Matching with Fuzzy Analytic Hierarchy Process (ASM-FAHP) that maps multiple trade-off concerns into a Multi-Criteria Decision-Making (MCDM) problem. The ASM-FAHP firstly searches all the possible alternatives to find possible combinations with hardware resource limitations taken into account. Afterwards, ASM-FAHP prioritizes the most significant candidate by constructing a hierarchical structure with several attributes and scoring with fuzzy numbers. Hereby, the most suitable ML combinations are determined by calculating synthesised fuzzy weights per each alternative. The performance of the ML combination is evaluated by concurrently executing it on resource-constrained hardware, specifically the Jetson Nano board. The ML models are trained and tested using high-fidelity synthetic datasets produced from Spirent GSS7000 simulator and SimGen while connected to hardware-in-the-loop (HIL). It has been discovered that when approaching hardware limits, the selected combination of machine learning algorithms makes full use of memory resources but sacrifices processing speed.