Abstract:
The cross-correlation of two-dimensional digital images is fundamental to solar adaptive optics
computations. It can be used in a simple tip-tilt correction system to identify the relative shift between consecutive images and correlating sub-aperture images of a Shack–Hartmann wave-front sensor. The typical frequency
of computation is about 1 kHz. While the software-based optimized cross-correlations may be sufficient when
a small number of sub-apertures are used in a wave-front sensor, hardware-accelerated (FPGA), correlations
may be required when a large number of sub-aperture images are involved, e.g., in the case of the proposed
National Large Solar Telescope in India. This paper presents SolarAccel: An FPGA-based acceleration of a
basic two-dimensional cross-correlation of two images. We accelerate the FPGA-based design by pipelining
the individual components of the cross-correlation process. We implemented our RTL logic on a few sets of
128 × 128 pixel images and 32 × 32 pixel images on a Xilinx Zynq Ultrascale+ MPSoC on the ZCU104
FPGA evaluation platform. SolarAccel performs a 2D FFT on a 128 × 128 image faster than existing work.
The cross-correlation on a 32 × 32 image is also faster than the existing work. This demonstrates that FPGA
acceleration is beneficial in solar adaptive optics applications.
