Designing and implementing Machine Learning Algorithms for advanced communications using FPGAs

Communications systems can obtain substantial benefits from increased intelligence. Improvements to communications include increased spectral situational awareness, spectral optimization, and robust operation in dynamic and demanding communications environments. Furthermore, complex communication systems require a high degree of autonomous intelligence to optimize performance under such varying conditions. Machine Learning Algorithms provide a means to increase the intrinsic intelligence of wideband communication systems. This paper considers the use of Machine Learning Algorithms to increase the intelligence of communication systems. Specifically, the focus of this paper is to sense and learn the communication environment in real-time and optimize system parameters to maximize end-to-end performance. Communications systems have existing adaptive capabilities in many subsystems such as equalization. The focus in this paper is top level system intelligence by learning from the environment, and based on the system capabilities determine an optimal mode in the solution space in real-time. Furthermore, the goal of this paper is to consider implementation of Machine Learning Algorithms using FPGAs. Design data for implementing Machine Learning Algorithms using FPGAs is provided in the paper as well as reference circuits for implementation. Finally, an example implementation of a Machine Learning Algorithm for intelligent communications is provided based on implementation in a Xilinx UltraScale FPGA.