Abstract:Convolutional neural network (CNN), as a model of deep learning (DL), has been widely applied to the field of computer vision as well as optical communication. In this paper, we focus on the adaptive demodulation technique in orbital angular momentum (OAM) free-space optical (FSO) communication system with the improved CNN. In order to achieve adaptive demodulation under free-space turbulence channel, the traditional CNN with our preliminary optimization methods has been firstly demonstrated. Then, in view of … Show more
“…The results demonstrate that despite the strong AT level and the consequent collapse of the field structure, the ATANN can learn information on the unperturbed field pattern by discovering intrinsic local features that compose each fractional OAM mode from collected data. Additionally, we think that the relatively low accuracy at , described above, can be improved with a hybrid training set considering various turbulence levels 23 . Figure 7 Generalization ability under unknown turbulence environments.…”
Section: Resultsmentioning
confidence: 99%
“…All tests presented here were conducted at the same turbulence level ( ). The recognition accuracy of integer OAM beams was less than 90% despite their wide mode spacing, which is because the size feature for identification, such as radius, is weakened by AT 23 . On the other hand, the accuracy for both fractional schemes was measured to be over 99%.…”
Section: Resultsmentioning
confidence: 99%
“…Deep-learning models directly recognize transmitted spatial modes without any optical mode sorter for extracting phase information 18 , 20 and provide reliable performance against distorting factors such as optical misalignment 18 , 24 , 25 . Adaptive demodulation 21 – 23 and turbulence correction 26 , 27 for integer vortex beams have been studied in depth by pioneering researchers. Meanwhile, the previous studies on fractional OAM modes focused on high-resolution recognition and the capability to increase communication capacity 17 , 18 .…”
Since the great success of optical communications utilizing orbital angular momentum (OAM), increasing the number of addressable spatial modes in the given physical resources has always been an important yet challenging problem. The recent improvement in measurement resolution through deep-learning techniques has demonstrated the possibility of high-capacity free-space optical communications based on fractional OAM modes. However, due to a tiny gap between adjacent modes, such systems are highly susceptible to external perturbations such as atmospheric turbulence (AT). Here, we propose an AT adaptive neural network (ATANN) and study high-resolution recognition of fractional OAM modes in the presence of turbulence. We perform simulations of fractional OAM beams propagating through a 1-km optical turbulence channel and analyze the effects of turbulence strength, OAM mode interval, and signal noise on the recognition performance of the ATANN. The recognition of multiplexed fractional modes is also investigated to demonstrate the feasibility of high-dimensional data transmission in the proposed deep-learning-based system. Our results show that the proposed model can predict transmitted modes with high accuracy and high resolution despite the collapse of structured fields due to AT and provide stable performance over a wide SNR range.
“…The results demonstrate that despite the strong AT level and the consequent collapse of the field structure, the ATANN can learn information on the unperturbed field pattern by discovering intrinsic local features that compose each fractional OAM mode from collected data. Additionally, we think that the relatively low accuracy at , described above, can be improved with a hybrid training set considering various turbulence levels 23 . Figure 7 Generalization ability under unknown turbulence environments.…”
Section: Resultsmentioning
confidence: 99%
“…All tests presented here were conducted at the same turbulence level ( ). The recognition accuracy of integer OAM beams was less than 90% despite their wide mode spacing, which is because the size feature for identification, such as radius, is weakened by AT 23 . On the other hand, the accuracy for both fractional schemes was measured to be over 99%.…”
Section: Resultsmentioning
confidence: 99%
“…Deep-learning models directly recognize transmitted spatial modes without any optical mode sorter for extracting phase information 18 , 20 and provide reliable performance against distorting factors such as optical misalignment 18 , 24 , 25 . Adaptive demodulation 21 – 23 and turbulence correction 26 , 27 for integer vortex beams have been studied in depth by pioneering researchers. Meanwhile, the previous studies on fractional OAM modes focused on high-resolution recognition and the capability to increase communication capacity 17 , 18 .…”
Since the great success of optical communications utilizing orbital angular momentum (OAM), increasing the number of addressable spatial modes in the given physical resources has always been an important yet challenging problem. The recent improvement in measurement resolution through deep-learning techniques has demonstrated the possibility of high-capacity free-space optical communications based on fractional OAM modes. However, due to a tiny gap between adjacent modes, such systems are highly susceptible to external perturbations such as atmospheric turbulence (AT). Here, we propose an AT adaptive neural network (ATANN) and study high-resolution recognition of fractional OAM modes in the presence of turbulence. We perform simulations of fractional OAM beams propagating through a 1-km optical turbulence channel and analyze the effects of turbulence strength, OAM mode interval, and signal noise on the recognition performance of the ATANN. The recognition of multiplexed fractional modes is also investigated to demonstrate the feasibility of high-dimensional data transmission in the proposed deep-learning-based system. Our results show that the proposed model can predict transmitted modes with high accuracy and high resolution despite the collapse of structured fields due to AT and provide stable performance over a wide SNR range.
“…By learning sufficient features, the CNN model can support robust recognition of four mode bases with high efficiency even the system suffers disturbance. [58,59] 3) High recognition efficiency might be also achieved by first training the CNN to reconstruct the intensity images of mode bases and then recognize the four mode bases based on the recovered intensity images. By improving the architecture of the CNN model (e.g., U-net), the intensity images of four mode bases can be reconstructed from the distorted intensity images recorded by the camera or PD array.…”
Section: Resultsmentioning
confidence: 99%
“…[58] Adaptive demodulation technique was investigated in OAM free-space optical communication system with the improved CNN. [59] In OAM shift keying communication system, [60] the bit rate and bit-error ratio performance were effectively improved with the help of the adaptive and intelligent demodulation. The recognition accuracy is often promoted by deepening the number of network layers.…”
In fiber-optic communications using diverse spatial modes for sustainable capacity scaling, the intelligent recognition of different mode bases is of great importance to enhance the flexiblity and compatibility of mode management. Here a convolutional neural network (CNN) model is introduced to recognize the four mode bases with the azimuthal index = 5, namely the LP 5,1 mode group, the linearly and circularly polarized OAM ±5,1 mode group, and the vector EH 4,1 or HE 6,1 mode group in a ring-core fiber. A camera is first used to capture intensity profiles of mode bases as training and testing data sets of the neural network. The CNN-based deep learning successfully recognizes different mode bases with an overall recognition rate of close to 100%. Furthermore, an alternative compact and cost-effective approach is considered toward practical applications by replacing the camera with a photodetector (PD) array for intelligent mode bases recognition. A 1 × 5 PD array can perfectly recognize different mode bases with a recognition rate of close to 100%. Even a 1 × 2 PD array with only two PDs can obtain a high recognition rate of close to 93.3%. The demonstrations may open up new perspectives for deep learning enabled robust and intelligent optical communications exploiting spatial modes.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
