論文

Optical eigenvalue communication is a promising technique for overcoming the Kerr nonlinear limit in optical communication systems. The optical eigenvalue associated with the nonlinear Schrödinger equation remains invariant during fiber-based nonlinear dispersive transmission. However, practical applications involving use of such systems are limited by the occurrence of fiber loss and amplified noise that induce eigenvalue distortion. Thus, several time-domain neural-network-based approaches have been proposed and demonstrated to enhance receiver sensitivity toward eigenvalue-modulated signals. However, despite the substantial improvement in power margin realized using time-domain neural-network-based demodulators compared to their conventional counterparts, these devices require rigorous training for each transmission distance owing to changes in time-domain pulses during transmission. This paper presents a method for demodulation of eigenvalue-modulated signals using an eigenvalue-domain neural network and demonstrates its utility through simulation and experimental results. Simulation results obtained in this study reveal that the proposed demodulator demonstrates superior generalization performance compared to its time-domain counterpart with regard to the transmission distance. Moreover, experimental results demonstrate successful demodulation over distances from zero to 3000 km without training for each distance.

Demodulation scheme based on multilabel eigenvalue-domain neural network for a 4096-ary eigenvalue-modulated signal is demonstrated experimentally. Successful demodulation with a 2.5 dB power margin compared with multiclass single-label classification is achieved at 10.7 Gb/s.

Demodulation scheme based on multilabel eigenvalue-domain neural network for a 4096-ary eigenvalue-modulated signal is demonstrated experimentally. Successful demodulation with a 2.5 dB power margin compared with multiclass single-label classification is achieved at 10.7 Gb/s.

We experimentally demonstrate the transmission of a 4096-ary eigenvalue-modulated signal using a neural network-based demodulator and SD-FEC. The experimental results indicate a successful operation with an error-free transmission through a 1200-km optical fiber line.

Large-scale disaster occurs all over the world frequently, disconnects telecommunications, and destroys communication equipment. In recent years, unmanned aerial vehicles (UAVs) network systems have been studied to work on the reconstruction activities safely and flexibly. The more means of telecommunication, the better because the UAV networks are used for emergency communication. Therefore, this paper studies optical camera communication (OCC) systems using RGB-LED-mounted drones and a high-speed camera for disaster recovery and proposes the RGB-LED-mounted drone's detection scheme and the signal equalization technique to suppress the RGB interference. We detect the drone using the algorithm of a deep neural network (DNN) based object detection called YOLOv3. This paper adds a new function to reduce the frame rate in object detection. Consequently, the proposed scheme reduces the frame rate to a rate that can conduct real-time operations less than 20 fps from 600 fps. Moreover, the experimental results indicate the feasibility of the proposed scheme that can communicate in error-free operation at a 300-m distance.

Deep Joint Source-Channel Coding (DeepJSCC) has demonstrated significant advantages in wireless image transmission by integrating source and channel coding through deep learning. However, most existing DeepJSCC methods use a fixed compression rate, which can lead to either excessive compression or inefficient use of transmission resources. To address this limitation, we propose a PSNR-guaranteed variable compression scheme that dynamically adjusts the compression rate to maintain a target Peak Signal-to-Noise Ratio (PSNR). Our method utilizes an exhaustive search algorithm to optimize compression parameters based on the characteristics of the input image. We evaluate the proposed scheme through both simulations and real-world 5G outdoor field experiments. The results demonstrate that our approach achieves superior compression efficiency while ensuring the desired image quality. Compared to conventional DeepJSCC, the proposed method achieves a better trade-off between compression rate and quality, particularly in low Signal-to-Noise Ratio (SNR) environments. Furthermore, since the proposed method functions as an external module to DeepJSCC, it can be seamlessly integrated with existing DeepJSCC frameworks. We also evaluate the computational complexity of the proposed method and show that it supports real-time processing despite employing an exhaustive search strategy. These findings suggest that our approach is highly suitable for practical deployments in wireless image transmission over 5G and beyond.

This paper presents the first outdoor field trials of deep joint source-channel coding (DeepJSCC) for image transmission over a 5G system. DeepJSCC is a deep learning-based end-to-end method that unifies source and channel coding to enable robust and low-latency image transmission, particularly in low signal-to-noise power ratio (SNR) environments. Unlike conventional methods, DeepJSCC inherently avoids the cliff effect and maintains stable image quality under harsh channel conditions. To evaluate its feasibility of practical environments, we modified a commercial 5G base station (gNB) and user equipment (UE) to support DeepJSCC signal transmission and reception. Extensive experiments were conducted under indoor and outdoor settings, including line-of-sight (LoS) and non-line-of-sight (NLoS) conditions. A key contribution of this study is the empirical verification that even a DeepJSCC model trained solely on an additive white Gaussian noise (AWGN) channel can maintain stable reconstruction performance in real 5G environments, demonstrating its generalization capability. Compared to baseline systems using JPEG2000 and LDPC, DeepJSCC achieved higher PSNR stability and was able to restore image content even when conventional schemes completely failed. These findings suggest that DeepJSCC is a promising candidate for next-generation visual communication over 5G infrastructure.

The deep joint source-channel coding and modulation (JSCCM) is a promising technology to realize efficient communication over extreme environments such as underwater area. In previous works, it is shown that deep convolutional neural networks (CNN) can successfully learn JSCCM encoder and decoder, outperforming conventional separation-based coding and modulation schemes in low signal-to-noise ratio settings. This paper proposes a new architecture for deep JSCCM based on the self-attention mechanism. We show that the proposed architecture achieves significant performance improvement compared with the CNN-based schemes while requiring a smaller network size in terms of the number of weight parameters. Furthermore, we present efficient hardware implementation of the proposed JSCCM encoder on a field programmable gate array (FPGA). In particular, we demonstrate that a systolic-array-like structure is effective for FPGA implementation of the proposed JSCCM scheme based on the self-attention mechanism.

Optical Camera Communication (OCC) is an emerging technology for wireless communication between smart devices such as smartphones. A light source such as a LED light and a LED panel is employed as a transmitter in OCC. Among the various modulation schemes, color shift keying (CSK) has attracted considerable attention to improve throughput. CSK exploits the design of three-color LED luminaires. Although CSK is a promising modulation technique, the relationship between symbol colors and external environments has not been considered yet. Some colors become difficult to identify depending on environments including ambient light. To address this problem, this paper proposes an adaptive (N+1)-CSK to define one more color in addition to general CSK. The goal of the proposed scheme is to reduce bit errors by explicitly defining a base color, i.e. NULL color. One symbol is adaptively selected as the base color from the N+1 symbols according to external conditions, and data signals are transmitted using other symbols. We also provide preliminary experimental results to validate the proposed scheme. The bit error rate (BER) is significantly suppressed by appropriately setting the base color depending on the pilot signals of each symbol.

We propose a novel method for all-optical modulation format conversion from DP-PAM4 to DP-QPSK using PM-HNLF. Successful operation of the proposed conversion is experimentally demonstrated, and the conversion performance is numerically investigated.