Speckle-based deep learning approach for classification of orbital angular momentum modes

Proposal to correct aberration and turbulence effects in the propagation of Laguerre-Gaussian modes

Laguerre-Gaussian (LG) modes are known as carriers of orbital angular momentum (OAM) and, for this reason, such modes have potential applications in optical communications. In this work, we present a study of the effects of aberration and turbulence on LG modes and propose a correction for these effects using a spatial light modulator. The aberrations are introduced by a phase mask obtained through a combination of Zernike polynomials. A scaling factor in the corrective phase mask enables us to optimize the recovery of the transverse structure of the LG beam, opening, to our knowledge, a new investigative avenue on aberration and turbulence mitigation. Numerical simulations and experiments are presented with good agreement.

Machine-learning-assisted orbital angular momentum recognition using nanostructures

The recognition of orbital angular momentum (OAM) in light beams holds significant importance in optical communication. The majority of current OAM recognition techniques are highly sensitive to stringent alignment issues. The speckle-based OAM recognition method reported in J. Opt. Soc. Am. A39, 759 (2022)JOAOD61084-752910.1364/JOSAA.446352 is alignment-free in the transverse direction of light propagation and has been shown to operate successfully in the far-field region using macrostructures. This study introduces a proof-of-concept for speckle-learned OAM recognition with nanostructures, relaxing the strict alignment requirements in both the transverse and along the direction of light propagation. When the OAM beam interacts with random inhomogeneities at micron and/or nanoscale, it generates an OAM speckle field. Initially, a comprehensive examination of the dynamic evolution of OAM speckle fields, ranging from near field to far field, has been conducted using a ground glass diffuser, featuring random phase inhomogeneities at the micron scale. Subsequently, the investigation proceeds to randomly grown ZnO nanosheets on an aluminum substrate. To achieve rapid and precise OAM recognition, a tailored three-layer CNN is trained and tested on OAM speckle fields ranging from near field to far field to attain an accuracy surpassing 92%. This research expands the technique's applicability, enabling recognition of OAM across near-field to far-field regimes, while leveraging micro- to nanostructures.

Machine learning-based classification of structured light modes under turbulence and eavesdropping effects

This paper considers the classification of multiplexed structured light modes, aiming to bolster communication reliability and data transfer rates, particularly in challenging scenarios marked by turbulence and potential eavesdropping. An experimental free-space optic (FSO) system is established to transmit 16 modes [8-ary Laguerre Gaussian (LG) and 8-ary superposition LG (Mux-LG) mode patterns] over a 3-m FSO channel, accounting for interception threats and turbulence effects. To the best of authors' knowledge, this paper is the first to consider both factors concurrently. We propose four machine/deep learning algorithms-artificial neural network, support vector machine, 1D convolutional neural network, and 2D convolutional neural network-for classification purposes. By fusing the outputs of these methods, we achieve promising classification results exceeding 92%, 81%, and 69% in cases of weak, moderate, and strong turbulence, respectively. Structured light modes exhibit significant potential for a variety of real-world applications where reliable and high-capacity data transmission is crucial.

1D speckle-learned structured light recognition

In this Letter, we introduce a novel, to the best of our knowledge, structured light recognition technique based on the 1D speckle information to reduce the computational cost. Compared to the 2D speckle-based recognition [J. Opt. Soc. Am. A39, 759 (2022)10.1364/JOSAA.446352], the proposed 1D speckle-based method utilizes only a 1D array (1×n pixels) of the structured light speckle pattern image (n × n pixels). This drastically reduces the computational cost, since the required data is reduced by a factor of 1/n. A custom-designed 1D convolutional neural network (1D-CNN) with only 2.4 k learnable parameters is trained and tested on 1D structured light speckle arrays for fast and accurate recognition. A comparative study is carried out between 2D speckle-based and 1D speckle-based array recognition techniques comparing the data size, training time, and accuracy. For a proof-of-concept for the 1D speckle-based structured light recognition, we have established a 3-bit free-space communication channel by employing structured light-shift keying. The trained 1D CNN has successfully decoded the encoded 3-bit gray image with an accuracy of 94%. Additionally, our technique demonstrates robust performance under noise variation showcasing its deployment in practical cost-effective real-world applications.

Vision transformers motivating superior OAM mode recognition in optical communications

Orbital angular momentum (OAM) has recently obtained tremendous research interest in free-space optical communications (FSO). During signal transmission within the free-space link, atmospheric turbulence (AT) poses a significant challenge as it diminishes the signal strength and introduce intermodal crosstalk, significantly reducing OAM mode detection accuracy. This issue directly impacts the performance of OAM-based communication systems and leads to a reduction in received information. To address this critical bottleneck of low mode recognition accuracy in OAM-based FSO-communications, a deep learning method based on vision transformers (ViT) is proposed for what we believe is for the first time. Designed carefully by numerous experts, the advanced self-attention mechanism of ViT captures more global information from the input image. To train the model, pretraining on a large dataset, named IMAGENET is conducted. Subsequently, we performed fine-tuning on our specific dataset, consisting of OAM beams that have undergone varying AT strengths. The computer simulation shows that based on ViT method, the multiple OAM modes can be recognized with a high accuracy (nearly 100%) under weak-to-moderate turbulence and with almost 98% accuracy even under long transmission distance with strong turbulence (C N2=1×10-14). Our findings highlight that leveraging ViT enables robust detection of complex OAM beams, mitigating the adverse effects caused by atmospheric turbulence.

Speckle-based structured light shift-keying for non-line-of-sight optical communication

We report an experimental proof of concept for speckle-based one-to-three non-line-of-sight (NLOS) free space optical (FSO) communication channels employing structured light shift-keying. A 3-bit gray image of resolution 100×100 pixels is encoded in Laguerre-Gaussian or Hermite-Gaussian beams and decoded using their respective intensity speckle patterns via trained 1D convolutional neural network. We have achieved an average classification accuracy of 96% and 93% using L G ml and H G pq beams, respectively, among all three channels. It demonstrates the directional independence and broadcasting capability of speckle-based decoding (SBD) in FSO communication using structured light. Further, we have extended the study from 2D to 1D SBD in one-to-three NLOS FSO communication channels to decrease the computational cost and to emphasize the importance of the 1D SBD approach.

Classifying beams carrying orbital angular momentum with machine learning: tutorial

This tutorial discusses optical communication systems that propagate light carrying orbital angular momentum through random media and use machine learning (aka artificial intelligence) to classify the distorted images of the received alphabet symbols. We assume the reader is familiar with either optics or machine learning but is likely not an expert in both. We review select works on machine learning applications in various optics areas with a focus on beams that carry orbital angular momentum. We then discuss optical experimental design, including generating Laguerre-Gaussian beams, creating and characterizing optical turbulence, and engineering considerations when capturing the images at the receiver. We then provide an accessible primer on convolutional neural networks, a machine learning technique that has proved effective at image classification. We conclude with a set of best practices for the field and provide an example code and a benchmark dataset for researchers looking to try out these techniques.