Online Bunch-by-Bunch Position Measurement System for a Storage Ring Based on Machine Learning and Direct Sampling

Abstract: Bunch-by-bunch (BxB) beam position measurement is an important tool for synchrotron radiation light sources, especially the new-generation diffraction-limited storage rings, for studying beam instabilities and improving light source performance. The BxB BPM processor is currently a key research direction in beam instrumentation; however, no mature processor is yet available for large-scale engineering applications. Here, a BxB BPM processor based on direct RF sampling and machine learning is proposed and implemented at the Shanghai Synchrotron Radiation Facility (SSRF). It samples the BPM signal at 2.5 GSPS. A convolutional neural network (CNN) is developed on the field-programmable gate arrays (FPGA) of the processor to identify the sampling-delay automatically, while fully connected neural networks (FCNNs) on the FPGA predict the amplitudes of the four BPM electrodes. The processor can also update the machine learning weight factors online, enabling online applications adaptable to different accelerator facilities and BPMs at different locations. Experiments at SSRF demonstrated a BxB transverse position resolution of 10 μm, comparable to the best of other BxB BPM measurement methods. Moreover, the processor features real-time online measurement, a simple structure, and lower cost, making it suitable for large-scale engineering deployment. It thus provides a practical and feasible solution for BxB BPM measurement.