The miniaturized calcium imaging is an emerging neural recording technique that can monitor neural activity at large scale at a specific brain region of a rat or mice. It has been widely used in the study of brain functions in experimental neuroscientific research. Most calcium-image analysis pipelines operate offline, which incurs long processing latency thus are hard to be used for closed-loop feedback stimulation targeting certain neural circuits. In this paper, we propose our FPGA-based design that enables real-time calcium image processing and position decoding for closed-loop feedback applications. Our design can perform real-time calcium image motion correction, enhancement, and fast trace extraction based on predefined cell contours and tiles. With that, we evaluated a variety of machine learning methods to decode positions from the extracted traces. Our proposed design and implementation can achieve position decoding with less than 1 ms latency under 300 MHz on FPGA for a variety of mainstream 1-photon miniscope sensors. We benchmarked the position decoding accuracy on open-sourced datasets collected from six different rats, and we show that by taking advantage of the ordinal encoding in the decoding task, we can consistently improve decoding accuracy without any overhead on hardware implementation and runtime across the subjects.