Optical computing holds promise for high-speed,energy-efficient information processing,with diffractive optical networks emerging as a flexible platform for implementing task-specific transformations.A challenge,howev...Optical computing holds promise for high-speed,energy-efficient information processing,with diffractive optical networks emerging as a flexible platform for implementing task-specific transformations.A challenge,however,is the effective optimization and alignment of the diffractive layers,which is hindered by the difficulty of accurately modeling physical systems with their inherent hardware imperfections,noise,and misalignments.While existing in situ optimization methods offer the advantage of direct training on the physical system without explicit system modeling,they are often limited by slow convergence and unstable performance due to inefficient use of limited measurement data.Here,we introduce a model-free reinforcement learning approach utilizing Proximal Policy Optimization(PPO)for the in situ training of diffractive optical processors.PPO efficiently reuses in situ measurement data and constrains policy updates to ensure more stable and faster convergence.We validated our method through both simulations and experiments across a range of in situ learning tasks,including targeted energy focusing through a random diffuser,image generation,aberration correction,and optical image classification,demonstrating in each task better convergence and performance.Our strategy operates directly on the physical system and naturally accounts for unknown real-world imperfections,eliminating the need for prior system knowledge or modeling.By enabling faster and more accurate training under realistic experimental constraints,this in situ reinforcement learning approach could offer a scalable framework for various optical and physical systems governed by complex,feedback-driven dynamics.展开更多
China Geodetic Coordinate System 2000(CGCS2000),as the formal national coordinate reference frame,has been used for 20 years.The coordinates of all Global Navigation Satellite System(GNSS)stations in China need referr...China Geodetic Coordinate System 2000(CGCS2000),as the formal national coordinate reference frame,has been used for 20 years.The coordinates of all Global Navigation Satellite System(GNSS)stations in China need referring to this system.To this end,the first step is to align the coordinates of all stations,usually included in a regional GNSS network,with a given International Terrestrial Reference Frame(ITRF),then these coordinates are corrected to the CGCS2000 in consideration of plate movement.For a better alignment result,regional control stations are needed and their coordinates were estimated from the combination of constraint-free normal equation systems provided by several International GNSS Service(IGS)analysis centers.The effect in using these refined coordinates,which deter-mine a regional coordinate datum,on the alignment result should be evaluated by the coordinate corrections of the regional control stations to the regional coordinate datum,i.e.smaller corrections mean better alignments of the two associated frames.The test results show that the refined coordinates are more accurate than the ones calculated from the station’s velocity,and are well aligned with the ITRF2005.Moreover,for obtaining the coordinates of GNSS stations in an updated CGCS2000 frame,a gridded linear velocity field based on the estimated velocities at 1025 CGCS2000 stations was generated for China's Mainland using the optimal interpolation method,the inverse distance weighting,which is selected from five interpolation methods.The overall precisions of the constructed velocity field at all stations in the East(E)and,North(N)directions are 0.78 mm/a and 0.95 mm/a,respectively.For evaluating the accuracy of the updated CGCS2000 frame,monthly solutions for the coordinates of some CGCS2000 CORS stations in the ITRF2014 during the period from 2000.0 to 2018 were obtained and the Root Mean Square(RMS)of the differences between the coordinates corrected to the CGCS2000 and the known coordinates at these stations are about 2-3 cm.展开更多
We propose and demonstrate free-space optical data links based on coaxial sidelobe-modified optical vortices（CSMOVs）. In contrast to the optical communication systems based on amplitude, frequency, or phase detectio...We propose and demonstrate free-space optical data links based on coaxial sidelobe-modified optical vortices（CSMOVs）. In contrast to the optical communication systems based on amplitude, frequency, or phase detection, the proposed scheme uses the radii ratio between the principle ring and the first sidelobe of the CSMOV.Therefore, the demand of stringent alignment and/or accurate phase matching is released. We design and optimize a composite computer-generated hologram to generate a CSMOV with four topological charges（TCs）.Extracted from the images captured by a CCD camera, the radii ratio between the principle ring and the first sidelobe of different TCs are consistent with the theoretical values.展开更多
文摘Optical computing holds promise for high-speed,energy-efficient information processing,with diffractive optical networks emerging as a flexible platform for implementing task-specific transformations.A challenge,however,is the effective optimization and alignment of the diffractive layers,which is hindered by the difficulty of accurately modeling physical systems with their inherent hardware imperfections,noise,and misalignments.While existing in situ optimization methods offer the advantage of direct training on the physical system without explicit system modeling,they are often limited by slow convergence and unstable performance due to inefficient use of limited measurement data.Here,we introduce a model-free reinforcement learning approach utilizing Proximal Policy Optimization(PPO)for the in situ training of diffractive optical processors.PPO efficiently reuses in situ measurement data and constrains policy updates to ensure more stable and faster convergence.We validated our method through both simulations and experiments across a range of in situ learning tasks,including targeted energy focusing through a random diffuser,image generation,aberration correction,and optical image classification,demonstrating in each task better convergence and performance.Our strategy operates directly on the physical system and naturally accounts for unknown real-world imperfections,eliminating the need for prior system knowledge or modeling.By enabling faster and more accurate training under realistic experimental constraints,this in situ reinforcement learning approach could offer a scalable framework for various optical and physical systems governed by complex,feedback-driven dynamics.
基金the National Key Research and Development Program of China(2016YFB0501405)Natural Resources Innovation Platform Construction and Capacity Improvement(A19090).
文摘China Geodetic Coordinate System 2000(CGCS2000),as the formal national coordinate reference frame,has been used for 20 years.The coordinates of all Global Navigation Satellite System(GNSS)stations in China need referring to this system.To this end,the first step is to align the coordinates of all stations,usually included in a regional GNSS network,with a given International Terrestrial Reference Frame(ITRF),then these coordinates are corrected to the CGCS2000 in consideration of plate movement.For a better alignment result,regional control stations are needed and their coordinates were estimated from the combination of constraint-free normal equation systems provided by several International GNSS Service(IGS)analysis centers.The effect in using these refined coordinates,which deter-mine a regional coordinate datum,on the alignment result should be evaluated by the coordinate corrections of the regional control stations to the regional coordinate datum,i.e.smaller corrections mean better alignments of the two associated frames.The test results show that the refined coordinates are more accurate than the ones calculated from the station’s velocity,and are well aligned with the ITRF2005.Moreover,for obtaining the coordinates of GNSS stations in an updated CGCS2000 frame,a gridded linear velocity field based on the estimated velocities at 1025 CGCS2000 stations was generated for China's Mainland using the optimal interpolation method,the inverse distance weighting,which is selected from five interpolation methods.The overall precisions of the constructed velocity field at all stations in the East(E)and,North(N)directions are 0.78 mm/a and 0.95 mm/a,respectively.For evaluating the accuracy of the updated CGCS2000 frame,monthly solutions for the coordinates of some CGCS2000 CORS stations in the ITRF2014 during the period from 2000.0 to 2018 were obtained and the Root Mean Square(RMS)of the differences between the coordinates corrected to the CGCS2000 and the known coordinates at these stations are about 2-3 cm.
基金partially supported by the National Nature Science Foundation of China (Nos. 61138003, 61427819, and 61405121)the Natural Science Foundation of SZU (No. 201454)the startup funding of SZU (Nos. 000011, 000075)
文摘We propose and demonstrate free-space optical data links based on coaxial sidelobe-modified optical vortices（CSMOVs）. In contrast to the optical communication systems based on amplitude, frequency, or phase detection, the proposed scheme uses the radii ratio between the principle ring and the first sidelobe of the CSMOV.Therefore, the demand of stringent alignment and/or accurate phase matching is released. We design and optimize a composite computer-generated hologram to generate a CSMOV with four topological charges（TCs）.Extracted from the images captured by a CCD camera, the radii ratio between the principle ring and the first sidelobe of different TCs are consistent with the theoretical values.
