Optimizing Real-Time GPU Image Processing for Industrial MIMO Radar Sensors

The development of a new sensor system involves a multitude of different steps from the concept phase to the finished product. A number of new challenges arise when it comes to bringing new concepts into actual application. This paper focuses on the development of a new FMCW (Frequency Modulated Continuous Wave) MIMO (Multiple Input Multiple Output) radar sensor, specifically addressing the implementation and optimization of GPU-based (Graphics Processing Unit), real-time signal processing software. Various aspects of the software design are examined that must be considered during the transition from a theoretical concept to a real-time capable product. This paper introduces a modular 32-channel MIMO system, expandable to 192 channels, currently operating at a 200Hz frame rate. While the current performance is sufficient for many applications, an outlook is provided on achieving higher measurement rates in the future to facilitate inline sorting of black plastics.