Performance evaluation of versal AI engines for radar signal processing

Abstract

This thesis aimed to evaluate the artificial intelligence engines found on the Versal VCK 190 evaluation board. To perform the evaluation, a non-homogeneous detector based on the generalised inner product was used. This algorithm was implemented in its entirety in MATLAB and on a graphics processing unit, which was used for comparison. For the Versal board however, only parts of the algorithm were implemented in the design. Further, the design was never run on the Versal board, instead, simulations were used to achieve the results presented in this thesis. This thesis found no real advantages of using the artificial intelligence engines for the parts of the generalised inner product algorithm that were implemented. For computations of matrices larger than 24 × 24, the graphics processing unit outperformed the artificial intelligence engines in terms of speed. Although this report did not find any real advantages for using the artificial intelligence engines, further studies are required, especially regarding the power consumption in comparison to the throughput, before any final conclusions can be drawn regarding the artificial intelligence engines.