GPU Spectrometer for the Total Power Array

The rapid advancement in GPU (Graphic Processing Unit) technology, driven by the growing needs of artificial intelligence, has paved the way for significant improvements in astronomical data processing. The launch of the NVIDIA Grace Hopper superchip is a major breakthrough. It speeds up the transfer of data from antennas to GPU memory and greatly increases processing power. In light of these advancements, we propose the development of a new Total Power GPU Spectrometer (TPGS) to replace the existing ACA Spectrometer (ACAS) for the upcoming WSU phase. The TPGS is designed to produce full polarization products by processing digitized signals from each of the four TP antennas. It will play a crucial role in the calibration of pointing and focus for the antennas, thanks to its ability to generate cross-correlation products. Utilizing a FFX (Filter-FFT-Multiplication) architecture, the TPGS methodically processes data on a per spectral window basis, accommodating digitized signals across a vast 20 GHz bandwidth for each polarization, sideband, and antenna. This system uses a polyphase filter to isolate a specific spectral window, and then applies a Fast Fourier Transform (FFT) to the filtered signal to calculate auto-correlation. This design's flexibility in selecting spectral window bandwidths and FFT sizes ensures it meets the varied bandwidths and spectral resolutions for each spectral window.