Real-time wavefront processors for the next generation of adaptive optics systems: a design and analysis

Abstract

ABSTRACT Adaptive optics (AO) systems currently under investigation will require at least two orders of magitude increase in thenumber of actuators, which in turn translates to effectively a 10 4 increase in compute latency. Since the performance ofan AO system invariably improves as the compute latency decreases, it is important to study how today’s computersystems will scale to address this expected increase in actuator utilization. This paper answers this question bycharacterizing the performance of a single deformable mirror (DM) Shack-Hartmann natural guide star AO systemimplemented on the present-generation digital signal processor (DSP) TMS320C6701 from Texas Instruments. Wederive the compute latency of such a system in terms of a few basic parameters, such as the number of DM actuators, thenumber of data channels used to read out the camera pixels, the number of DSPs, the available memory bandwidth, aswell as the inter-processor communication (IPC) bandwidth and the pixel transfer rate. We show how the results wouldscale for future systems that utilizes multiple DMs and guide stars. We demonstrate that the principal performancebottleneck of such a system is the available memory bandwidth of the processors and to lesser extent the IPC bandwidth.This paper concludes with suggestions for mitigating this bottleneck.Keywords: Adaptive optics, wavefront processors, TMS320C6701