Accelerating FPGA Implementation of Neural Network Controllers via 32-bit Fixed-Point Design for Real-Time Control
Chanakya Dinesh Hingu, Xingang Fu, Rajab Challoo, Jiang Lu, Xiaokun Yang, Letu Qingge · 2023
This paper focuses on using a 32-bit fixed-point design instead of a 32-bit floating-point representation to implement a neural network controller for the real-time control of a solar inverter by utilizing a Field Programmable Gate Array (FPGA) for faster calculation in real-time environments. Two designs of the tanh (tangent hyperbolic) activation function were implemented using both LookUp Table (LUT) and COordinate Rotation DIgital Computer (CORDIC)-based approaches. The MATLAB/Simulink HDL coder toolbox was used in the design. The complete neural network controller was implemented on Intel Altera Cyclone V boards to verify output, resource requirements, and memory allocation on the hardware. Further, the error-bound analysis of the LUT and CORDIC-based approaches was conducted. The accuracy verification shows that the 32-bit fixed-point design has achieved at least 4 decimal accurate digits compared to the actual double format MATLAB calculation. This demonstrates the feasibility and effectiveness of the proposed neural network controller design as a real-time controller.