Versal AI Engines for On-Board Data Compression: Implementation and Performance Analysis of CCSDS 121 and CCSDS 123

The rapid increase in data generated by Earth observation and deep-space missions, coupled with the limited capacity of downlink channels, has made data compression a crucial task in Payload Data Handling Units. This paper evaluates an efficient implementation of the CCSDS 121.0-B-3 and CCSDS 123.0-B-2 compression algorithms on the AMD Versal Adaptive Intelligent Engines. Both compression standards, designed for general data and hyperspectral image compression, are mapped to the Versal’s many-core architecture to assess performance in terms of data throughput and resource efficiency. Targeting the Versal AI Cores series, the algorithms are adapted to vectorized processing in a pipelined manner. The results demonstrate that the AI Engines can achieve high throughput, with both compression algorithms reaching up to 8.8 Gbit/s while utilizing only half of the AI Engine array. Challenges encountered during the implementation, including vectorization, memory constraints, and dependency management within the CCSDS 123 algorithm, are analyzed and discussed. Although the AI Engines demonstrate a strong performance for certain tasks, the results indicate that further improvements could be achieved by combining the AI Engines with Programmable Logic.