Gain scheduled neural network tuned pi feedback control system for the lansce accelerator

The current LANSCE LLRF system is an analog PI Feedback control system which achieves the amplitude and phase error within 1% and 1 degree. The feedback system receives the cavity amplitude and phase and the crosstalk between the amplitude and the phase paths is significant. We propose an In-phase (I) and Quadrature (Q) based feedback control system which easily decouples the crosstalk of I and Q channels. A gain scheduled PI feedback controller with optimally generated set point trajectory reduces the transient peak of the feedback controller and hence reduce the fatigue of the RF amplifier chain. An additional feature of the controller is the Neural network based self-tuning PI feedback, where the neural network tunes the feedback gains to minimize the errors in the least squares sense. The proposed control system is implemented with Altera Stratix II FPGA. The control system is modeled with DSP Builder and automatically generates HDL. Altera SOPC Builder is used for the hardware integration of the DSP Builder model, memories, peripherals, and 32 bit NIOS II embedded processor. NIOS II processor equipped with real time operating system communicates with the host computer via Ethernet, uploads data, computes parameters, and downloads parameters. The proposed control system is tested with the low power test-stand for the robustness of the algorithm.