A many-core CPU prototype of an MCAO and LTAO RTC for ELT-scale instruments

We propose a many-core CPU architecture for Extremely Large Telescope (ELT) scale adaptive optics (AO) real-time control (RTC) for the multi-conjugate AO (MCAO) and laser-tomographic AO (LTAO) modes. MCAO and LTAO differ from the more conventional single-conjugate (SCAO) mode by requiring more wavefront sensor (WFS) measurements and more deformable mirrors to achieve a wider field of correction, further increasing the computational requirements of ELT-scale AO. We demonstrate results of our CPU based AO RTC operating firstly in SCAO mode, using either Shack-Hartmann or Pyramid style WFS processing, and then in MCAO mode and in LTAO mode using the specifications of the proposed ELT instruments, MAORY and HARMONI. All results are gathered using a CPU based camera simulator utilising UDP packets to better demonstrate the pixel streaming and pipe-lining of the RTC software. We demonstrate the effects of switching parameters, streaming telemetry and implicit pseudo open-loop control (POLC) computation on the MCAO and LTAO modes. We achieve results of < 600$\mu$s latency with an ELT scale SCAO setup using Shack-Hartman processing and < 800$\mu$s latency with SCAO Pyramid WFS processing. We show that our MCAO and LTAO many core CPU architecture can achieve full system latencies of < 1000$\mu$s with jitters < 40$\mu$s RMS. We find that a CPU based AO RTC architecture has a good combination of performance, flexibility and maintainability for ELT-scale AO systems.
Comment: Accepted 2019 March 01 MNRAS. Received 2019 February 28 ; in original form 2018 December 14. 12 pages, 17 figures, 2 tables