Abstract
To realize a high-precision link in space optical communication， we analyzed key factors that affect the accuracy of flat-panel detectors in terms of target positioning. The error of the centroid algorithm was analyzed from the mechanism and the necessity of satisfying the spatial lossless sampling condition was verified by using a simulation. The defined NU value served as an indicator in quantifying the nonuniformity of the detector. As the NU value increased linearly， the positioning error of the centroid continued to increase， whereas the speed reduced. When the NU value was 0.005， the maximum positioning error was 0.043 pixels. Considering that the light intensity of the target incident on the optical system varies constantly， the smaller the NU value， the closer the centroid is to the true position of the light spot. Furthermore， the pixel response of a typical complementary metal oxide semiconductor （CMOS） detector was experimentally tested under different illumination intensities. Based on the response curve， a mathematical model for the nonuniformity of the pixel response was established. It can be determined that the NU value fluctuates from 0.0045 to 0.0048 within the linear response range. The experimental results of the spot centroid positioning accuracy verify that the absolute positioning error is less than 0.05 pixels， which satisfies the requirements of high-precision links. Therefore， the effectiveness of the theory and simulation presented in this study can be validated.
Owing to several advantages， such as high speed， minimal terminal size requirements， and strong confidentiality， space optical communication has developed substantially in recent years
As accurate positioning of the target spot is essential for the detector in the space optical communication terminal， several researchers have explored this concept in recent years. A method was proposed to obtain the optical flow vector using image data considering the effect of background noise on a beacon in inter-satellite optical communication
However， the aforementioned studies are based on the assumption of the specific application backgrounds， focusing on methods to reduce errors in a certain aspect to improve the precision of detection. The factors that affect spot positioning have not been systematically analyzed from a global perspective. Moreover， the accuracy of point target positioning in the case of precise pointing in space optical communication has not been considered thus far. Precision is a prerequisite for ensuring accuracy because high precision does not necessarily result in high accuracy. Therefore， it is necessary to investigate the accuracy of the detector used in the optical communication terminal for target positioning.
The remainder of this paper is organized as follows. Section 1 systematically analyzes the factors that affect the positioning accuracy of the flat-panel detector in a communication terminal. Section 2 investigates the error of the centroid algorithm from the mechanism and verifies the necessity of satisfying the spatial lossless sampling condition using a simulation. In Section 3， we set the quantitative index of detector nonuniformity， quantitatively simulate the relationship between nonuniformity and positioning error， and propose correction algorithms. Section 4 presents the tested typical CMOS detector， which was used to establish a mathematical model for the nonuniformity of the pixel response and measure the positioning accuracy of the spot centroid. Section 5 discusses and summarizes the results.
The image data collected by the flat-panel detector in the communication terminal were post-processed， and the centroid of the light spot was extracted to obtain the position of the target. The closeness of the centroid measurement value of the target to the true value indicates the positioning accuracy. The general term used for the accuracy and precision of the measurement is target positioning accuracy. Various factors， such as the optical system at the front end of the detector， performance of the detector， circuit system around the detector， and algorithm for calculating the position of the center of mass， impact the accuracy of target positioning.
The positioning error of the target spot can be classified into two， namely systematic error and random error
| , | （1） |
where N represents the half-width of the centroid spot size， wherein the widths of the centroid spot in the horizontal and ordinate directions are 2N+1； S indicates the intensity of the target signal； and denotes noise. Considering the distribution characteristics of the noise signal， it can be expressed as：
| , | （2） |
where represents the shot noise of the pixel with coordinates ， denotes the readout noise， and indicates the dark current noise. Shot noise is classified as a type of Poisson distributed noise based on the form of energy distribution， whereas dark current and readout noises are classified as Gaussian distributed noise.
Poisson noise is a type of signal-dependent noise， which increases with the enhancement of the signal. Here， the average photon noise equals the square root of the number of incident photons. Gaussian noise is a type of additive white noise generated by the random thermal movement of electrons in photosensitive devices. Typically， the average level of Gaussian noise is retained irrespective of the light intensity. Therefore， the dark field image correction and multi-image single-pixel time-domain average processing can generally suppress random error
As random errors can be suppressed effectively， systematic errors are the primary errors that affect the positioning accuracy of the target spot. These include the error caused by the inherent nonuniformity of the flat-panel detector and the error generated by using the geometric center of the pixel rather than the true energy center of the pixel in the centroid algorithm. The former affects the sampling process of the spot in the flat-panel detector and distorts the image， whereas the latter affects the calculation of the target position.
The optical system in space optical communication is equivalent to the Fraunhofer diffraction system. Therefore， the beacon spot that converges on the flat-panel detector is a diffracted spot with a diameter approximately equal to the Airy disk diameter of the optical system； the energy can be regarded as a Gaussian distribution. The actual centroid position of the target spot on the terminal detector， which is equivalent to the energy center of the image plane， can be expressed as
| , | （3） |
where x and y denote the horizontal and vertical coordinates on the image plane of the flat-panel detector， respectively； S indicates the spatial area occupied by the target spot； and represents the light intensity distribution function of the beacon spot on the detector plane.
| , | （4） |
where A denotes the number of pixels in the detector， indicates the geometric center coordinates of the pixels in the detector， and represents the sampling value of the light intensity distribution function of the kth effective pixel. As the pixel size of the detector cannot be infinitely small， the geometric center of the spot calculated using （4） cannot completely coincide with the energy center of the spot. Consequently， a systematic error occurs when the detector uses the centroid algorithm to locate the spot.
Using a flat-panel detector with a pixel interval of d to sample the spatial energy distribution of the diffracted spot results in a type of discrete sampling at equal intervals. To prevent the overlap of the sampled distributions and recover the energy distribution information of the original spot completely， the spatial sampling frequency 1/d of the detector must be greater than the spatial cut-off frequency of the modulation transfer function of the optical combination lens
| , | （5） |
According to （5）， the range of the diffracted spot diameter on the flat-panel detector can be deduced considering that the theoretical lossless sampling law is satisfied. As is approximately equal to the diameter of the Airy disk of the optical system，
| , | （6） |
combining （5） and （6） to obtain
| . | （7） |
Considering the aforementioned conclusions （section 2.2）， we performed a normalized simulation. Assuming that the detector is ideal for spot sampling and the point spread function of the optical system follows an ideal Gaussian distribution， the X- and Y-axis directions are consistent during the positioning of the target spot. Therefore， the positioning process only needs to be analyzed in the direction of the X-axis. The center of mass of the spot with as the energy center after convolution， sampling， and calculation is denoted as ［
| , | （8） |
where indicates the frequency domain representation of the received energy function of the pixel， denotes the frequency， and 1/d represents the spatial sampling frequency of the detector. When the Gaussian radius of the point spread function is ， can be expressed as
| . | （9） |
1/d was set as 1， and the Gaussian radius of the optical system was 0.3， 0.4， 0.5， and 1 pixel. The spot energy center initiates from 0 and moves 100 times in steps of 0.02 pixels， .
Fig. 1 Results of the normalized simulation
图1 归一化仿真的结果
The spot position calculated by the centroid algorithm is based on the true energy center ， and one pixel is a period that presents a sinusoidal distribution. When the Gaussian radius of the optical system was 0.3， 0.4， 0.5， and 1 pixel， the maximum positioning errors were 0.054， 0.014， 0.002， and 0 pixels， respectively. The positioning accuracy improves as increases. If the spatial lossless sampling of the detector is not satisfied， an error occurs when using the centroid algorithm to locate the light spot received by the detector. When the spatial sampling frequency 1/d of the detector is not less than the spatial cut-off frequency of the modulation transfer function of the optical system at the front of the detector， the error caused by the centroid algorithm can be ignored completely.
The photo response nonuniformity of the flat-panel detector is the inconsistent response of different pixels to the same incident radiation. This is associated with the variations in the quantum efficiency and other parameters of different pixels in the device manufacturing proces
| . | （10） |
To investigate the effect of the nonuniformity of the detector on the positioning accuracy of the spot centroid under the working conditions used in this study， we define the nonuniformity of the response NU as
| . | （11） |
NU is used as an index to quantify the corresponding nonuniformity of the pixel and simulate it to different nonuniformity images. The window size of the detector was set to 32 × 32 pixels， and the size of the light spot on the image was 5 × 5 pixels. The energy of the light spot conformed to the Gaussian distribution， and the response values of all pixels in the light spot area were within the linear response range. The response model parameters for each pixel within the window range were different.
It is an ideal hypothetical condition that the detector is under uniformly distributed illumination. In the dark environment of the laboratory， using an integrating sphere as a light source can approximately achieve the above condition. The response gray level of the pixel measured by the flat-panel detector under this condition can be used to calculate the NU value to calibrate the nonuniformity. The response gray level of the pixel measured by the flat-panel detector under this condition can be used to calculate the NU value to calibrate the nonuniformity. If the above experimental conditions are not met， the gray images obtained cannot be used for nonuniformity calibration. That is， for the definition of NU value， the calculation is accurate only under approximately ideal conditions， otherwise， it is unreliable.
Fig. 2 Simulation results of spot positioning accuracy
图2 光斑定位精度的仿真结果
| . | （12） |
Fig. 3 Changes in the positioning error with nonuniformity
图3 定位误差随非均匀性的变化曲线
To satisfy the requirements for spot positioning accuracy in actual optical communication scenarios， the obtained results can be combined to determine the necessity of nonuniformity correction. Commonly used correction algorithms include two-point correction， piecewise linear correction
Although the detector must operate in the linear response range of the pixels in practical applications， the signal response amplitude of the pixel at the center of the spot is not a constant value owing to the varying intensity of the incident light. A simulation was performed considering a scenario where the actual position of the light spot remains constant with the light intensity changing from low to high. The energy of the spot was set to conform to the Gaussian distribution， and the true center position on the X-axis was 8. When the values of NU are 0.005， 0.01， 0.02， and 0.04， the simulation generates an image with a size of 16 × 16 pixels and a spot size of 5 × 5 pixels. The ideal amplitude of the pixel signal response at the center of the spot ranges from 30 to 120 with an interval of 10； the responses of other pixels were obtained using the settings identical to those described in section 3.1.
| . | （13） |
Fig. 4 The centroid position of the spot under different light intensities
图4 不同光强下光斑的质心位置
Equations （
In actual applications， the light intensity incident on the optical system should be maintained stable， and the reasonable working range of the flat-panel detector needs to be calibrated. Alternatively， a suitable nonuniformity correction algorithm can be used.
An integrating spherical light source was used to generate different illuminances in the dark environment of the laboratory and the response of multiple pixels was measured within the window range of the flat-panel detector. The window size of the detector was set to 256 × 256 pixels and 16 images were collected without light. The average value of the pixel response was set as the base. The camera collected 100 images continuously each time the brightness of the integrating spherical light source was adjusted； a total of 11 energy levels were collected.
Fig. 5 Response curves of different pixels under different light intensities
图5 不同像素在不同光强下的响应曲线
A polynomial fitting method was adopted to establish a mathematical model for the nonuniformity of the pixel response. The expression is
| . | （14） |
where represents the ideal pixel response， indicates the actual pixel response， and ， ， and denote the quadratic， gain， and bias coefficients， respectively. Based on the results of the statistical simulations， ， ， and in （14） are random numbers with mean values of ， 0.88 and 2.972， and standard deviations of ， 0.021， and 0.682， respectively. Furthermore， the calculated NU values were all between 0.0045 and 0.0048 in the linear response range of 30 to 120.
To test the positioning accuracy of the flat-panel detector and verify its impact on high-precision pointing in space optical communication， we constructed a test experimental device using the same parameters as described in section 3. The entire experimental system was placed on an air-floating platform. The beacon light emitted by the laser was passed through a collimator tube with an equivalent length of 10 m to generate parallel light， which converged on the detector through a camera lens with a focal length of 35 mm. The detector located the beacon light spot.
The pixel size d of the CMOS detector used in this experiment was 5.5 μm， the beacon light wavelength λ was 810 nm， the combined focal length f of the detection system was 35 mm， and the effective aperture D of the detection system was 4.38 mm， which satisfies （5）. This eliminates the systematic error of the algorithm in the centroid-positioning process.
The beacon light source was moved step-by-step using a high-precision electronically controlled microtranslation stage. The distance of each step was 78.5 μm， corresponding to 0.05 pixels on the flat-panel detector； a total of 40 steps were tested. The method mentioned in section 1 was used to perform a noise removal for the image obtained at each step.
Fig. 6 True position and centroid position of a moving target
图6 运动目标的实际位置和质心位置
To achieve high-precision links in space optical communication systems， we systematically analyzed the positioning error of the spot centroid of a flat-panel detector used in the communication terminal. The systematic error primarily originates from the centroid algorithm and the inherent nonuniformity of the area array detector. In this study， the centroid algorithm error was analyzed theoretically and the simulation results of the spatial non-destructive sampling conditions were investigated. The target spot diameter must be greater than 2.44 pixels. The ratio of the standard deviation of the actual response gray level to the average value of NU under uniform illumination was used to quantify the nonuniformity of the detector. The simulation results verify that the centroid positioning error continues to increase as NU increases linearly from 0； however， a speed reduction can be observed. This can be summarized as indicated in （12） of the general model of the flat-panel detector. Considering that the light intensity of the target incident on the optical system is changing constantly， the position of the center of mass is closer to the true position of the light spot with the decrease in NU value， and the quantitative relationship is presented in （13）. Finally， the response of pixels in a typical CMOS detector was experimentally tested with different light intensities， and a mathematical model for the nonuniformity of the pixel response was established through a polynomial fitting method. The NU value fluctuated from 0.0045 to 0.0048 within the linear response range. Additionally， the experiments were conducted by moving the target to test the centroid positioning accuracy of a flat-panel detector. The maximum error observed was 0.049 pixels， which is consistent with the theoretical analysis and simulation results of this study.
References
Bai S， Wang J， Zhang L. Development progress and trends of space optical communications ［J］. Laser & Optoelectronics Progress， 2015， 52（07）： 7-20. 10.3788/lop52.070001 [Baidu Scholar]
Lee S H. Pointing accuracy improvement using model-based noise reduction method ［C］//Free-Space Laser Communication Technologies， 16th ed.， 2002， Bellingham， WA， USA： SPIE-Int Soc Optical Engineering， 2002， 4635：65–71. 10.1117/12.464110 [Baidu Scholar]
Liu L. Laser communications in Space Ⅱ test and verification techniques on the ground ［J］. Chinese journal of lasers， 2007， 34（2）： 147–155. 10.1117/12.731949 [Baidu Scholar]
Chen M， Liu C， Rui D M， et al. Experimental results of atmospheric coherent optical communications with adaptive optics ［J］. Optics Communications， 2019， 434： 91-96. 10.1016/j.optcom.2018.10.013 [Baidu Scholar]
Komatu K， Kanda S， Hirako K， et al. Laser-beam acquisition and tracking system for ETS-VI laser-communication equipment （LCE） ［C］// Free-Space Laser Communication Technologies II， 1990， WA， USA： SPIE-Int Soc Optical Engineering， 1990， 1218： 96–107， DOI： 10.1117/12.18177. [Baidu Scholar]
Toyoshima M， Takizawa K， Kuri T， et al. Ground-to-OICETS laser communication experiments ［C］// the Conference on Free-Space Laser Communications VI， CA， USA： 6304： 63040B， DOI： 10.1117/12.693012. [Baidu Scholar]
Zhang F R， Ruan P， Han J F. Point ahead angle prediction based on Kalman filtering of optical axis pointing angle in satellite laser communication ［J］. Optical Review， 2020， 27（6）： 447-454. 10.1007/s10043-020-00608-6 [Baidu Scholar]
Toyoshima M， Takenaka H， Shoji Y， et al. Results of Kirari optical communication demonstration experiments with NICT optical ground station （KODEN） aiming for future classical and quantum communications in space ［J］. Acta Astronautica， 2012， 74： 40-49. 10.1016/j.actaastro.2011.12.020 [Baidu Scholar]
Zhang L， Wang J， Jia J， et al. Design and Performance of Fine Tracking System Based on CMOS for Quantum Communication ［J］. Chinese Journal of Lasers， 2011， 38（2）： 0205008. 10.3788/cjl201138.0205008 [Baidu Scholar]
Zhang L， Dai J， Li C， et al. Design and in-orbit test of a high accuracy pointing method in satellite-to-ground quantum communication ［J］. Optics Express， 2020， 28（6）： 8291-8307. 10.1364/oe.387011 [Baidu Scholar]
Wang Q， Yu S Y， Tan L Y， et al. Approach for recognizing and tracking beacon in inter-satellite optical communication based on optical flow method ［J］. Optics Express， 2018， 26（21）： 28080-28090. 10.1364/oe.26.028080 [Baidu Scholar]
Bashir M S， Bell M R. Optical Beam Position Estimation in Free-Space Optical Communication ［J］. IEEE Transactions on Aerospace and Electronic Systems， 2016， 52（6）： 2896-2905. 10.1109/taes.2016.150567 [Baidu Scholar]
Ye T， Zhou F Q. Autonomous space target recognition and tracking approach using star sensors based on a Kalman filter ［J］. Applied Optics， 2015， 54（11）：3455–3469. 10.1364/ao.54.003455 [Baidu Scholar]
Cao Y， Li B， Li H， et al. High-Accuracy Star Sensor Centroid Algorithm Based on Star Image Resampling ［J］. Acta Optica Sinica， 2019， 39（7）： 0712003. 10.3788/aos201939.0712003 [Baidu Scholar]
Liu F， Wang X， Zhang L. Measurement method of random errors in spot target detection by flat-panel detector ［J］. Acta Optica Sinica， 2021， 41（4）： 0404001. 10.3788/aos202141.0404001 [Baidu Scholar]
Jiang L， Zhang L， Zhang X， et al. Compensation for star centroid systematic error of star trackers ［J］. Chinese Journal of Lasers， 2015， 42（3）： 0314001-0314011. 10.3788/cjl201542.0314001 [Baidu Scholar]
Sun H， Xu S， Sun S， et al. Research on Evaluation Method of Optical Imaging Sensors' Photon Response Non-Uniformity Noise，” ［J］. Laser & Optoelectronics Progress， 2015， 52（4）： 042302. 10.3788/lop52.042302 [Baidu Scholar]
Chen S J， Zhang L， Wu J C， et al. Realization and optimization of fine tracking system of free space laser communication ［J］. Journal Of Infrared And Millimeter Waves， 2018， 37（1）： 35–46. [Baidu Scholar]
Hu X， Mao X， Wu Y， et al. Pixel frequency error compensation method based on sub-pixel coordinates ［J］. Infrared and Laser engineering， 2017， 46（7）： 0717006. 10.3788/irla201746.0717006 [Baidu Scholar]
Yao P P， Tu B H， Xu S L， et al. Non-uniformity calibration method of space-borne area CCD for directional polarimetric camera ［J］. Optics Express， 2021， 29（3）： 3309-3326. 10.1364/oe.410768 [Baidu Scholar]
Jian X Z， Lu R Z， Guo Q， et al. Single image non-uniformity correction using compressive sensing ［J］. Infrared Physics & Technology， 2016， 76： 360-364. 10.1016/j.infrared.2016.03.015 [Baidu Scholar]
Cheng K H， Zhou H X， Qin H L， et al. An improved non-uniformity correction algorithm and its GPU parallel implementation ［J］. Infrared Physics & Technology， 2018， 90： 156-163. 10.1016/j.infrared.2018.03.005 [Baidu Scholar]
Yang J， Messinger D W， Dube R R， et al. Fixed Pattern Noise Pixel-wise Linear Correction for Crime Scene Imaging CMOS Sensor ［C］//Algorithms and Technologies for Multispectral， WA， USA： SPIE- Int Soc Optical Engineering， 2017， 10198： 1019802， DOI： 10.1117/12.2262035. [Baidu Scholar]
Li J， Mahmoodi A， Joseph D. Using Polynomials to Simplify Fixed Pattern Noise and Photometric Correction of Logarithmic CMOS Image Sensors ［J］. Sensors， 2015， 15（10）： 26331-26352. 10.3390/s151026331 [Baidu Scholar]
Boutemedjet A， Deng C W， Zhao B J. Edge-Aware Unidirectional Total Variation Model for Stripe Non-Uniformity Correction ［J］. Sensors， 2018， 18（4）： 1164. 10.3390/s18041164 [Baidu Scholar]
Lu C H. Stripe non-uniformity correction of infrared images using parameter estimation ［J］. Infrared Physics & Technology， 2020， 107： 103313. 10.1016/j.infrared.2020.103313 [Baidu Scholar]
He Z W， Cao Y P， Dong Y F， et al. Single-image-based nonuniformity correction of uncooled long-wave infrared detectors： a deep-learning approach ［J］. Applied Optics， 2018， 57（18）： D155-D164. 10.1364/ao.57.00d155 [Baidu Scholar]
Bao J Y， Xing F， Sun T， et al. CMOS imager non-uniformity response correction-based high-accuracy spot target localization ［J］. Applied Optics， 2019， 58（16）： 4560-4568. 10.1364/ao.58.004560 [Baidu Scholar]