Advanced optical-electronic neural network communication system

A hybrid optical-electronic convolutional neural network to demultiplex orbital angular momentum-coded signals is introduced. This system integrates Fourier optics convolution within a 4F optics setup, showcasing promising system efficiencies compared to the traditional electronic neural network.
Advanced optical-electronic neural network communication system

Share this post

Choose a social network to share with, or copy the shortened URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

A new paper in Nature Communications Physics shows the significant strides in optical computing and communications made by the CHIP Lab at the University of Florida. The team has presented innovative insights into the incorporation of machine learning within optical networks. This pioneering research introduces a novel hybrid optical-electronic convolutional neural network (CNN) system, conducted under the guidance of Hamed Dalir, Associate Professor in the Department of Electrical and Computer Engineering at the University of Florida. In the free-space optical communication field, this system incorporates a Fourier optics convolution to demultiplex multiplexed orbital angular momentum (OAM) beams. The study's innovative approach demonstrates significant improvement in the effectiveness of optical network accelerator systems and lays the foundation for future progress in optical communication technologies using OAM.

Ye et al. by delving into the Fourier optics principle, demonstrated the demultiplexing of the multiplexed OAM-coded signals under various atmospheric turbulent conditions, with the hybrid optical-electronic CNN  model offering a three times faster training period and a 65% reduction in computation in comparison to the traditional electronic CNN. The potential use of OAM beams in augmenting data transmission throughput and boosting system efficiency was underscored by Prof. Dalir, the principal investigator of the CHIP Lab. The results illustrate the capability of the hybrid optical-electronic CNN  accelerator to reduce the overall system latency and energy consumption in optical computing and communications. Moreover, this study provides fresh prospects for applications that need substantial data transfer rates and optimal system performance in GHz OAM-based free-space optical (FSO) communications.

The need for global data transmission has seen a substantial surge in the contemporary big data era. The increase in use may be ascribed to the requirements of advanced applications like ultra-high-definition live streaming, virtual meetings, immersive gaming, and real-time analytics. The increasing need for data burdens the bandwidth capabilities of existing networks since conventional wireless systems face significant limitations. In this context, FSO communication is considered a viable option for wireless data transmission. FSO communication employs wavelengths ranging from ultraviolet to infrared to facilitate data. Capitalizing on the expansive spectrum available in optical bands and the precise spatial confinement afforded by laser beams, FSO communication has demonstrated its capability to meet high-capacity optical data transmission demands.  

OAM represents a fundamental characteristic of electromagnetic waves, manifesting as a phase rotation in a vertical plane along the direction of propagation [1]. Specifically, when an electromagnetic wave exhibits OAM, the trajectory of a constant phase forms a helical pattern parallel to the propagation path. The optical beams carrying OAM have attracted attention in FSO communication, enhancing data transmission capacity through its orthogonality and the potential for an infinite number of topological charge states, paving the way for advancements in various high-capacity data transmission applications. 

The integration of machine learning in demultiplexing OAM beams has shown considerable promise.  Doster et al. introduced detecting the active OAM modes in a transmission link using a convolutional neural network [2] . Xiong et al. investigated the use of a cylindrical lens for CNN-based OAM modes recognition to demodulate OAM shift-keying (OAM-SK) signals [3]. However, the current mainstream convolutional neural networks introduce significant challenges. Notably, the convolutional layers in these networks are responsible for over 85% of the computational workload [4], leading to increased system latency and elevated power consumption, critical concerns in real-time applications, and energy-constrained environments. The system efficiency of these convolutional neural networks is hampered by the intensive computational demands of the convolutional operations, particularly when dealing with large datasets and deep network architectures. In response to these challenges, adopting 4f-based Fourier optics for convolution computation emerges as a groundbreaking advancement. This technique, leveraging the principles of Fourier optics within a 4f optical system, offers a novel approach to performing matrix multiplications. This paradigm shift not only addresses the latency and power issues associated with conventional convolutional neural networks but also sets a new benchmark for computational efficiency in optical data processing, heralding a new era in the synergy between machine learning and optics.

"We propose using machine learning methodology to demultiplex multiplexed OAM beams. Compared to the present conjugate mode sorting approach, this machine learning-based method offers the benefit of cost reductions in optical hardware and a more adaptable alignment precision needed for detection. The attainment of the intended power of the converted Gaussian beam into single-mode fiber requires meticulous alignment," explained Jiachi Ye, a doctoral candidate at the CHIP lab at the University of Florida who designed the system and led the experiment. "Alignment checking is a daily requirement based on my previous laboratory experience. We ran the test for the OAM-carrying beam modulated by the BERTScope at a 10GHz transmission ratio. It was at that time I realized the difficulty of reading the information using conjugate mode sorting-power budget is the issue that needs to be addressed when we talk about increasing the signal-to-noise ratio. Additionally, before our group decided to employ the single spatial light modulator (SLM) uploaded with the combined fork phase grating patterns to encode the OAM beams, I had considerable difficulties in aligning the whole free-space setup using two SLMs with a beam splitter to generate the co-propagating interference beams."

"In our comparative analysis, the proposed optical-electronic CNN demonstrated a slight reduction in demultiplexing accuracy compared to the electronic CNN. The discrepancy arises from the optical-electronic CNN dataset, which contains more noise and distortion due to experimental setup errors. Furthermore, the limited dataset size and the similarity between consecutive topological charge numbers in OAM modes present challenges in distinguishing between the modes." Added Dr. Haoyan Kang, postdoctoral researcher at the CHIP lab who led the image processing and data analysis of the study. "The training time comparison highlights the optical-electronic CNN's efficiency, especially with increasing dataset class numbers, showing the optical-electronic CNN's adeptness at handling large datasets crucial for OAM-based FSO communication. "

"This marks a significant step forward in enhancing the performance and applicability of OAM-based communications in free-space optical systems, as researchers will leverage the findings for integrated optical-electronic CNNs," emphasized Prof. Hao Wang, Assistant Research Professor in the Department of Electrical and Computer Engineering at the University of Florida. "Our work unveils the new interesting behavior in demultiplexing OAM beams. Looking ahead, our strategic direction focuses on overcoming the performance limitations imposed by conventional SLMs and cameras within our OAM-based optical CNN systems. To this end, we are exploring the deployment of advanced metasurfaces coupled with developing custom-designed, high-speed SLMs. This innovative approach aims to significantly enhance the modulation efficiency, speed, and resolution of our optical components. "

The metasurfaces, comprised of subwavelength antenna arrays, are significant in the generation and processing of OAM-coded signal arrays. Ye and Dalir claim that integrating metalens significantly improves data transmission efficiency and reduces power consumption, hence facilitating the advancement of optical-electronic CNN systems. The utilization of metalens-based technology represents a notable advancement in the progression of optical communication systems characterized by high throughput and energy efficiency. The advancement exhibits significant promise in increasing the transmission throughput and power efficiency of the OAM-based FSO communication system. 

"We are moving towards OAM-coded signal processing arrays using fabricated metalens and ITO-based SLM to complete massively parallel Fourier optics CNN as our next endeavor," said Prof. Chandraman Patil, the Assistant Research Professor at the Department of Electrical and Computer Engineering at the University of Florida. "This approach enables the power of massively parallel Fourier optics convolution solutions and has great potential to extend in telecommunication and mid-infrared applications exploring chaotic secured communication protocols and systems."

According to Dalir and Ye, an integrated optical-electronic communication system running at a system frequency of 350 GHz has been proposed. The present work has the potential to significantly impact the domain of optical computing and communication. Fourier optics will provide extensive parallel convolution, hence broadening the scope of applications for optical manipulation using OAM, microscopic particles in optical tweezers, encrypted communication, and introducing new prospects in these domains.


1. Allen, L., Beijersbergen, M. W., Spreeuw, R. J. C., & Woerdman, J. P. (1992). Orbital angular momentum of light and the transformation of Laguerre-Gaussian laser modes. Physical review A45(11), 8185.

2. Doster, T., & Watnik, A. T. (2017). Machine learning approach to OAM beam demultiplexing via convolutional neural networks. Applied optics56(12), 3386-3396.

3. Xiong, W., Chen, J., Wang, P., Wang, X., Huang, Z., He, Y., ... & Chen, S. (2023). Robust neural network-assisted conjugate orbital angular momentum mode demodulation for modulation communication. Optics & Laser Technology159, 109013.

4. Li, H., Kadav, A., Durdanovic, I., Samet, H., & Graf, H. P. (2016). Pruning filters for efficient convnets. arXiv preprint arXiv:1608.08710.

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Electronics Design and Verification
Technology and Engineering > Electrical and Electronic Engineering > Electronic Circuits and Systems > Electronics Design and Verification

Related Collections

With collections, you can get published faster and increase your visibility.

Synthetic dimensions for topological and quantum phases

This focus collection aims to present the most recent advances and future directions in exploring novel physics of topological and quantum phases with synthetic dimensions.

Publishing Model: Open Access

Deadline: Mar 21, 2024

Interactive Active Matter: crosstalk and interfaces between distinct active systems

In this focus issue, we bring together interdisciplinary research at the interface of physics, microbiology, stem cell biology, and mechanobiology, to present the most recent advances, outstanding challenges, and future directions in studying interacting active materials.

Publishing Model: Open Access

Deadline: Jun 05, 2024