An adaptive optics real-time controller hardware architecture based on multi-core CPU and GPU

Abstract

The application discloses a kind of adaptive optics real-time controller hardware architecture based on multi-core CPU and GPU.Shack-Hartmann wavefront sensor sub-aperture slope calculation, wavefront recovery and wavefront control are completed in parallel by multi-core CPU and GPU using the real-time controller hardware architecture.The application uses general multi-core CPU and GPU as computing platform, after receiving image data in multi-core CPU, part of sub-aperture image data is transmitted to GPU by PCIE bus using a CPU core, and wavefront calculation task of the part of sub-aperture is completed by GPU;At the same time, the remaining CPU core is used to complete the remaining sub-aperture wavefront calculation task in parallel, to reduce the purpose of real-time controller calculation delay.The application uses multi-core CPU and GPU in parallel to complete wavefront calculation task, to reduce the effect of adaptive optics real-time controller calculation delay, to meet the real-time requirement of adaptive optics system to real-time controller, especially suitable for next-generation large-aperture ground-based telescope adaptive optics system real-time control.

Description

Technical Field

[0001] This invention relates to the field of adaptive optics, and in particular to a hardware architecture for a real-time adaptive optics controller based on a multi-core CPU and GPU. Background Technology

[0002] Due to the influence of atmospheric turbulence, adaptive optics (AO) has become an indispensable technology for achieving high-resolution imaging with large-aperture ground-based telescopes. The AO system detects wavefront aberrations using a Shack-Hartmann Wavefront Sensor (SH-WFS) and drives a deformable mirror (DM) to correct these aberrations based on the detection results. The Real-Time Controller (RTC) is the core of the AO system. By acquiring image array data from the SH-WFS detector camera, it performs sub-aperture slope calculations, wavefront reconstruction, and control operations, ultimately calculating a voltage control signal to control the deformable mirror for wavefront correction. Limited by the temporal coherence characteristics of atmospheric turbulence, the RTC needs to achieve a correction frequency of several kilohertz to effectively correct dynamic distortions. Therefore, the performance of the real-time controller has a significant impact on the AO system.

[0003] The performance of the controller depends on the delay and jitter in its wavefront calculation. Traditional adaptive optics real-time controllers typically employ hardware architectures using dedicated devices such as field-programmable gate arrays (FPGAs) and digital signal processors (DSPs) to meet the high frame rate and low latency requirements of AO systems (L. Kong, L. Zhu, L. Zhang, H. Bao and C. Rao, "Real-Time Controller Based on FPGA and DSP for Solar Ground Layer Adaptive Optics Prototype System at 1-m NVST," in IEEE Photonics Journal, vol. 9, no. 2, pp. 1-11, April 2017). With the continuous improvement of computer technology, multi-core CPUs and GPUs have been gradually applied in AO systems due to their programming flexibility and strong computing power (JENKINS, DAVID, RICHARD (2019) A Prototype Adaptive Optics Real-Time Control Architecture for Extremely Large Telescopes using Many-Core CPUs, Durham University). The continuous increase in telescope aperture and the emergence of complex AO systems, such as multi-layer conjugate adaptive optics systems, have led to a dramatic increase in the computational load of RTC. The computing power and memory bandwidth of multi-core CPUs are insufficient to meet the performance requirements of next-generation AO systems for RTC. Currently, GPU-based RTCs are mostly used in laboratory AO systems (Bitenc, U., Basden, AG, Dipper, NA et al. Suitability of GPUs for real-time control of large astronomical adaptive optics instruments. J Real-Time Image Proc 14, 743–751 (2018).). The main time consumption is in copying image data from multi-core CPUs to GPUs. At the same time, since the amount of image data is proportional to the square of the telescope aperture, GPU-based RTCs have not yet been widely used in ground-based telescopes.

[0004] To address the aforementioned issues, this invention proposes a hardware architecture for an adaptive optics real-time controller based on a multi-core CPU and GPU, reducing RTC computation latency to enable real-time correction of wavefront aberrations in the AO system. Summary of the Invention

[0005] The technical problem to be solved by this invention is: to address the above-mentioned problems and difficulties, a hardware architecture for an adaptive optics real-time controller based on a multi-core CPU and GPU is proposed, which uses a multi-core CPU and GPU to process the wavefront information of the AO system in parallel, thereby reducing the RTC calculation latency and meeting the real-time requirements of the AO system for RTC.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: an adaptive optics real-time controller based on a multi-core CPU and GPU, comprising a Shaker-Hartmann wavefront sensor camera acquisition card 1, a multi-core CPU 2, a GPU 3, and a signal output interface 4. Its real-time control process for the adaptive optics system is as follows: an input beam containing wavefront aberrations enters the Shaker-Hartmann wavefront sensor 1, and the resulting spot array image data is transmitted to the multi-core CPU 2. One core of the multi-core CPU 2 transmits a portion of the sub-aperture data to the GPU 3 via the PCIE bus. The GPU 3 completes the wavefront calculation task for this portion of the sub-aperture and, after the calculation is completed, transmits the obtained control signal back to the multi-core CPU 2 via the PCIE bus. Simultaneously, other cores of the multi-core CPU 2 complete the wavefront calculation task for the remaining sub-apertures. Finally, the multi-core CPU 2 summarizes the calculated control signals and outputs the control signal through the signal output interface 4, thus completing the real-time control of the adaptive optics system.

[0007] Furthermore, the Shaker-Hartmann wavefront sensor camera acquisition card 1 typically adopts an FPGA architecture, and its data transmission interface includes, but is not limited to, optical fiber, CameraLink, etc.

[0008] Furthermore, the number of sub-apertures processed by the multi-core CPU2 and GPU3 depends on the performance of the multi-core CPU2 and GPU3 respectively, and the computation time of the two is made similar through sub-aperture allocation.

[0009] Furthermore, the wavefront calculation task includes sub-aperture image preprocessing, slope calculation, wavefront restoration, and wavefront control.

[0010] Compared with the prior art, the present invention has the following advantages:

[0011] This invention proposes an adaptive optics real-time controller based on a multi-core CPU and GPU. It uses a multi-core CPU and GPU to process the wavefront information of the adaptive optics system in parallel. On the one hand, it can reduce the bandwidth requirements of the multi-core CPU and GPU, and on the other hand, it can reduce the computational latency of the real-time controller, thus meeting the real-time requirements of the adaptive optics system. Attached Figure Description

[0012] Figure 1 This is a schematic diagram illustrating the control process of the adaptive optics system according to the present invention;

[0013] Figure 2 This is the wavefront calculation timing diagram of the present invention;

[0014] Figure 3 The results are real-time controller performance tests using a multi-core CPU alone, a GPU alone, and both a multi-core CPU and a GPU simultaneously.

[0015] The labels in the figure are as follows: 1 is the Shaker-Hartmann wavefront sensor camera acquisition card, 2 is the multi-core CPU, 3 is the GPU, and 4 is the signal output interface. Detailed Implementation

[0016] The following detailed description, in conjunction with the accompanying drawings, provides a specific implementation of the adaptive optics real-time controller based on a multi-core CPU and GPU proposed in this invention.

[0017] like Figure 1 As shown, a beam containing aberrations enters the Shaker-Hartmann wavefront sensor, and the resulting sub-aperture image data is transmitted to the multi-core CPU 2 via camera acquisition card 1. Core 0 then transmits a portion of the sub-aperture image data to GPU 3 via the PCIe bus. GPU 3 performs wavefront calculations for this portion of the sub-aperture and transmits the calculated control signals back to the multi-core CPU 2 via PCIe. Simultaneously, Cores 1–CoreN process the remaining sub-aperture data to complete the wavefront calculations. After both multi-core CPU 2 and GPU 3 have completed their calculations, multi-core CPU 2 outputs control signals to the deformable mirror via signal transmission interface 4, controlling the deformable mirror to correct the wavefront aberrations.

[0018] Figure 2 The diagram illustrates the timing of an adaptive optics real-time controller based on a multi-core CPU2 and GPU3. After receiving a frame of image data from camera acquisition card 1, the cores Core1–CoreN of the multi-core CPU2 begin wavefront calculations for their assigned sub-apertures. Simultaneously, Core0 transmits the remaining sub-aperture image data to GPU3 via the PCIe bus. After the image data transmission is complete, GPU3 performs the wavefront calculations for that portion of the sub-aperture. The multi-core CPU2's computational tasks and GPU3's image reception and computation can be performed in parallel, thereby reducing the real-time controller's computational latency.

[0019] Figure 3The image shows the simulation results of the real-time controller computation latency using this architecture. The input simulation image size is 1024*1024 pixels, containing 1024 sub-apertures, each sub-aperture being 32*32 pixels in size. The slope calculation method in the simulation is frequency domain cross-correlation. The computation latency statistics for each frame start from the receipt of image data by the multi-core CPU2 and end with the output voltage control signal, for a total of 10,000 frames. Under the above simulation input, the average computation latency using only the multi-core CPU2 is 918µs, the average computation latency using only the GPU3 is 882µs, and the average computation latency using both the multi-core CPU2 and GPU3 is 688µs.

[0020] The parts of this invention not described in detail are well-known in the field.

Claims

1. A multi-core CPU and GPU based adaptive optics real-time controller hardware device, characterized by: The application relates to a Shack-Hartmann wavefront sensor camera acquisition card (1), a multi-core CPU (2), a GPU (3) and a signal output interface (4), which realize the real-time control process of an adaptive optical system as follows: an input light beam containing a wavefront aberration enters a Shack-Hartmann wavefront sensor, the obtained spot array image data is transmitted to the multi-core CPU (2) through the camera acquisition card (1), a core of the multi-core CPU (2) transmits part of sub-aperture data to the GPU (3) through a PCIE bus, the wavefront calculation task of the part of sub-aperture is completed by the GPU (3), the obtained control signal is transmitted back to the multi-core CPU (2) through the PCIE bus after the calculation is completed, meanwhile, the wavefront calculation task of the remaining sub-aperture is completed by other cores of the multi-core CPU (2), finally, the obtained control signal is collected by the multi-core CPU (2), and the control signal is output through the signal output interface (4), so that the real-time control of the adaptive optical system is completed.

2. The adaptive optics real-time controller hardware device based on multi-core CPU and GPU of claim 1, wherein: The Shack-Hartmann wavefront sensor camera acquisition card (1) adopts an FPGA architecture, and a data transmission interface of the Shack-Hartmann wavefront sensor camera acquisition card (1) comprises an optical fiber and a CameraLink.

3. The adaptive optics real-time controller hardware device based on multi-core CPU and GPU of claim 1, wherein: The number of sub-apertures processed by the multi-core CPU (2) and the GPU (3) respectively depends on the performance of the multi-core CPU (2) and the GPU (3), and the calculation time of the two is similar through sub-aperture distribution.

4. The multi-core CPU and GPU based adaptive optics real-time controller hardware apparatus according to claim 1, wherein: The wavefront calculation task comprises sub-aperture image preprocessing, slope calculation, wavefront recovery and wavefront control.

Images