GPFFC: Wideband Signal Channelization Algorithm Leveraging GPU Parallel Technology

To meet the requirements of real-time processing of wideband or ultra-wideband signals from large aperture radio telescopes, we designed and implemented a GPU-based parallel Polyphase Filter bank Fast Channelization algorithm (GPFFC). We studied the processes of the pulsar baseband data unpacking, channelization, and packing. GPU channelization technology has been optimized. By using the CUDA multidimensional thread model, tasks are subdivided into multiple small blocks to improve parallel computing efficiency. The data processing was optimized based on CUDA asynchronous processing to reduce GPU idle time, which improved the parallel computing efficiency, achieving an average speedup of approximately 1.4× compared to the non-optimized implementation. The pulsar J0437-4715 baseband data recorded by the CASPSR backend of the Parkes 64-meter radio telescope were used for testing. The pulsar baseband signal was subdivided into subbands, and the data from each was then packaged. Finally, standard pulsar software DSPSR was used to process the baseband data of multiple subbands, and the pulse profiles of multiple subbands were merged into wideband pulses, which were consistent with the pulse profiles of the raw baseband data in phase and signal-to-noise ratio, verifying the effectiveness of the GPFFC. It processes 400 MHz dual-polarized data in 5.8 s with 8-bit quantization over an 8 s observation time, meeting the real-time processing requirements of pulsar baseband data.