Research On Real-time Processor Based On GPU For Performance Evaluation Of Adaptive Optics System

The adaptive optics technique is now widely implemented in ground-basedastronomical telescope for real-time detection and compensation of wavefrontaberration caused by atmospheric turbulence effects.During the operation of an AOsystem, it is of great importance to calculate the turbulence parameters as well as theperformance index of the system, which not only lead to a better understanding of thesystem execution in different atmospheric conditions, but also provide an objectiveevaluation on the correction capability of AO system. Moreover, based on thereal-time assessment, the AO system is able to adjust its control parameters to theexternal conditions in order to achieve the optimum result.In this paper, a real-timeprocessor based on GPU and CPU is proposed to implement the on-line calculation ofparameters for both atmospheric turbulence and system performance are studied.The traditional and common method of performance evaluation is based on anexternal far-field measurement system, through which the far-field image is obtaineddirectly and processed to get the performance indexedin an off-line way.This methodhas a small amount of computation and is simple to execute. Yet it is not able tocalculate parameters corresponding to the system in a time shaft.The on-line method,however, is able to do the real-time estimation with heavy computations and big datasize. Therefore, to implement the on-line calculation, the real-time processor is quitedemanding in computational capability.The computational complexity and requirement of the delay is analyzed beforeprogram design. Since there is a contradiction between the heavy computation burdenand the real-time processing requirement, a parallel method is presented to decomposethe whole task into smaller ones in system level and data level. Some algorithms,which are of high complexity, are also modified to parallel programming.Compared with the existed processing platform overseas and domestic, a real-timeprocessor based on the CPU and GPU heterogeneous architecture is proposed, aimingto achieve a better performance, especially shorten the time spent. Based on theanalysis of GPU and CPU programming features, the parallel program of parameterscalculation of an AO system is successfully mapped in the hardware and software.The real-time processor of performance evaluation is finally equipped for the 127-element adaptive optics system, including both the turbulence parametersestimation and performance indexes calculation. According to the results, it turned outthat the processor is of high-precision and short computational delay, which proposeda new method in the on-line parameters calculations of an adaptive optics system.