![]() ![]() The features extracted from the last block are connected to the classification part comprised of two fully connected (FC) layers. Then the size of the activated feature maps is reduced by half through a max pooling layer, which allows the model to detect the weak variance coming from fractional mode separation 41. The purpose of the nonlinear activation function is to introduce non-linearity into the output, which allows the network to learn complex data and provide an accurate prediction 33. Then the extracted feature maps are processed by a nonlinear activation function. Multiple convolution kernels in the convolutional layer detect various local features of input images 32. For this reason, 4 and 5 block models were taken into account, and the latter showed the lower validation loss despite similar training times. Therefore, a model causes inefficiently large number of parameters if the size of feature maps is not scaled down properly. The number of parameters in the first fully connected (FC) layer is proportional to the pixel size of input feature maps. In this experiment, the number of blocks was determined considering the number of trainable parameters and model performance. The part of the feature extraction is constructed by 5 blocks, and each block consists of a convolutional layer and a max pooling layer. The designed CNN model comprises two parts for feature extraction and classification, as depicted in Fig. Designed network structure and dataset preparation The former separates the desired first diffraction order, and the latter compensates for the phase deformation caused by the spatial inhomogeneity of the SLM. ![]() Meanwhile, a blazed phase grating and a surface correction hologram, made by a combination of Zernike polynomials, are additionally included in Eq. ( 1). ![]() Light beams with a helical wavefront originating from an azimuthal phase \( = 0.10\) (8 modes) and OAM modes from \(l = 1.03\) to \(l = 1.96\) with \(\Delta l = 0.03\) (32 modes). Light orbital angular momentum (OAM), which is one of the spatial degrees of freedom (DoF), has been proposed as a potential solution to overcome the limitation 5, 6. Despite significant improvements in the capacity by the use of wavelength- and polarization-division multiplexing techniques and the application of multilevel modulation formats 1, 2, 3, 4, exponentially growing data traffic is facing bandwidth crunch. Since the beginning of innovations in information technology, such as the Internet of Things, big data, cloud computing, and artificial intelligence, demand for high-capacity communication systems has been explosively growing. This research will present a new approach to realizing higher data rates for advanced optical communication systems. Moreover, the proposed scheme successfully achieves image transmission despite its densely packed mode space. Our results show that the trained model can simultaneously recognize two independent DoF without any mode sorter and precisely detect small differences between fractional modes. Data encoded by switching phase holograms is efficiently decoded by a deep-learning classifier that only requires the intensity profile of transmitted modes. Spatial modes of Bessel-Gaussian beams separated by fractional intervals are employed to represent 8-bit symbols. We propose a data transmission system using fractional mode encoding and deep-learning decoding. Structured light with spatial degrees of freedom (DoF) is considered a potential solution to address the unprecedented demand for data traffic, but there is a limit to effectively improving the communication capacity by its integer quantization. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |