Addressing the challenge of traditional mathematical model-based quality of transmission (QoT) prediction methods struggling to simultaneously meet the demands of high precision and low computational complexity, this paper introduces three intelligent QoT prediction techniques for single optical paths, multiple optical paths, and cross-topology optical paths. These techniques rely on machine learning models to achieve accurate end-to-end optical path QoT predictions and effectively tackle the following challenges: firstly, how to select appropriate machine learning models and input features amidst the diversity of physical layer parameters. Secondly, how to effectively capture the intricate relationships among optical paths. Thirdly, how to train and continuously optimize network models with limited samples. Finally, the article offers a glimpse into the future development directions of optical path QoT prediction technologies.

To effectively mitigate the negative impact of rerouting on the network in traditional traffic engineering mechanisms,this paper proposes a specific flow routing selection algorithm based on Self-Attention deep reinforcement learning, leveraging the global network perspective and management capabilities of software-defined networking, to reroute a small amount of traffic and achieve near-optimal performance. A neural network model with multi-scale fusion attention mechanism is used to extract features of traffic, and a centralized training distributed execution architecture is adopted to make real-time decisions based on the observed network state. The theoretical research and experimental results show that compared with traditional deep reinforcement learning algorithms and heuristic algorithms, the proposed algorithm has significant improvements in average load and end-to-end delay performance.

A modulation format recognition(MFI) method based on joint residual network(ResNet) and Bottleneck Transformer (BT) is proposed to meet the transmission requirements in future optical network links. This method combines ResNet and BT to identify signals with six different modulation formats, and applies OptiSystem and TensorFlow to simulate them. The simulation results show that within a wide range of optical signal-to-noise ratio (OSNR), the proposed method achieves an accuracy of 99.72% and can effectively cope with the impact of transmission damage. Compared with other deep learning methods, this method significantly improves its performance.

Aiming at the problem that modulation format recognition accuracy is susceptible to factors such as channel environment and background noise interference in visible light communication signal transmission, this paper proposes an improved YOLOv5s(You Only Look Once) algorithm for modulation format recognition of visible light communication signals. Firstly, the Mixup data augmentation method is introduced at the input end of the YOLOv5s algorithm network, and it is combined with the Mosaic data augmentation method in the original network to enhance the robustness of the network and improve the generalization ability of the algorithm among different modulation format signals. Secondly, adaptively spatial feature fusion (ASFF) is introduced into the Neck network to fully extract features from different levels and improve detection accuracy. The experimental results indicate that under mixed signal-to-noise ratio conditions, the mean average precision(mAP) of the proposed improved algorithm reaches 0.903, representing a 0.7% improvement compared to the original YOLOv5s algorithm. Furthermore, the mAP reaches a high of 0.993 when the signal-to-noise ratio is 20 dB.

Aiming at the problem of intensified physical layer damage caused by stimulated Raman scattering (SRS) effect in C+L band elastic optical networks, a spectrum allocation algorithm based on deep reinforcement leaming (DRL) adaptive modulation format is proposed. In the routing stage, the K-shortest routing algorithm is used to pre calculate K shortest candidate paths for business requests. In the stages of band, modulation format, and spectrum allocation, DRL is used for intelligent decision-making, and two reward functions are combined to reduce network blocking rate and improve spectrum utilization efficiency. The simulation results show that the algorithm can effectively reduce blocking rate and improve spectrum utilization.

In order to further reduce the traffic blocking rate of satellite optical networks, this paper proposes a wavelength routing algorithm based on deep Q-networks (DQN) and matching perception. This algorithm first designs a path wavelength matching factor to reflect the common free wavelengths between the current path and the link of the next hop node. Subsequently, the algorithm takes into account multiple factors such as path wavelength matching, adjacent link delay, and the shortest number of hops from the next hop node to the destination node, and constructs a reward function based on these factors. When allocating wave-lengths, in order to have more free wavelengths on adjacent links for subsequent service requests, this paper designs an adjacent link wavelength matching index to describe the alignment of free wavelengths between the path and its adjacent links. The simulation results show that this algorithm can reduce network blocking rate and delay, and improve wavelength utilization.

Network performance prediction is the key to achieving efficient network management of software defined optical networks(SDON), but there is an urgent need for a network performance prediction model that can accurately predict key indicators at limited cost. A graph neural network-based SDON performance prediction model is proposed, which combines BiGRU and Self-Attention mechanisms to learn the complex relationships between network topology, routing, and traffic matrices, accurately estimating the packet delay, jitter, and packet loss rate of the source/destination in the network. This model can be applied to networks that have not been encountered during training. The experimental results show that in different traffic model tests, the proposed model has a significant improvement in average absolute percentage error (MAPE) performance compared to the baseline model.

In order to meet the real-time and efficient computing power scheduling requirements of hot and cold services, a computational load prediction model (abbreviated as C-TCN model) based on adaptive noise complete set empirical mode decomposition(CEEMDAN) and time convolutional network(TCN) is proposed, and a resource cooperative scheduling algorithm(CTQ algorithm) based on C-TCN and Q learning is designed. The C-TCN model is used to sense the load change at the next time in advance, and the optimal wavelength partitioning and edge storage allocation scheme is found through Q learning. The experimental results show that the CTQ algorithm not only has better scheduling performance than the existing scheduling algorithms,but also can meet the requirements of hot and cold service scheduling performance, and improve the wavelength utilization rate.

Due to factors such as equipment failure and environmental interference, lidar often encounters problems of missing data or noise interference during the data collection process, which seriously affect the subsequent analysis and application of the data. To solve this problem, the diffusion Transformer network (DT-Net) is introduced and used as a generator in combination with a Self-Attention unit discriminator. Additionally, a diffusion mechanism is designed for lidar data completion. The experimental results show that compared to the PoinTr method, the proposed approach achieves significant improvements in lidar data completion tasks, with an average Chamfer distance(CD) value reduced by approximately 1.79% and an F-Score value increased by approximately 1.88%.

In order to better solve the routing, modulation format and spectrum allocation (RMSA) problems of elastic optical networks(EON), and further reduce the network blocking rate, an RMSA algorithm based on deep reinforcement learning(DRL) is proposed. This algorithm will consider two indicators, resource occupancy and spectral adjacency, which affect RMSA decision making in reward design, to encourage agents to prioritize selecting paths with low resource occupancy and high spectral adjacency to establish optical paths, and compare the performance of this algorithm with other algorithms in different networks.The simulation results show that compared with several typical DRL algorithms, the proposed algorithm has a lower network blocking rate.

Aiming at the problems of high complexity and computational intensity in optical signal-to-noise ratio (OSNR) estimation, a high-order quadrature amplitude modulation (QAM) signal OSNR estimation method based on lightweight random forest (RF) algorithm is proposed. This method maps high-order QAM signals with different OSNRs into different constellation diagram datasets, and uses these datasets to train the RF model, thereby achieving rapid OSNR estimation The simulation results show that when the real value of system OSNR is between 5~30 dB, the accuracy of OSNR estimation for 64QAM and 128QAM signals based on lightweight RF algorithm is close to 100%, the mean absolute error(MAE) of OSNR estimation for 64QAM signals is 0.08 dB, and the MAE for 128QAM is 0.12 dB, which is more accurate than the signal OSNR estimation results based on long short-term memory(LSTM) algorithm.

In order to address the security problem arising from the numerous devices and limited terminal resources in the internet of things (IoT) environment, a method for radio frequency (RF) fingerprint signal recognition of IoT devices based on lightweight omni-dimensional dynamic convolutional neural network(LR-ODCNN) is proposed. Firstly, the LR-ODCNN model is designed. Then, the baseband signals of the devices are collected using an optical transmission system, and the I and Q signals are extracted from the baseband signals as the input to the network. Finally, the LR-ODCNN model adapts to the signal characteristics of different devices based on a multi-dimensional attention mechanism and performs signal feature extraction and recognition. The experimental results show that the average recognition accuracy of the LR-ODCNN model is 94.35% at transmission distances of 10 m, 400 m, 1.7 km, and 8.6 km., which is an improvement of 5.35% and 10.13% compared to the McAFF model and the Oracle model respectively. Additionally, it boasts strong robustness and lightweight.

In order to address the challenge of achieving precise motion control based on real-time feedback signals in soft robot,a motion detection method for soft robot based on the hue of photonic crystal structural color sensors is proposed. The working mechanism of this method lies in the fact that when the soft robot moves, the color of the photonic crystal structural color sensor integrated on it will change. By detecting these hue changes through image processing techniques and utilizing the mapping relationship between hue signals and motion states, the tracking and monitoring of the soft robot's motion state can be achieved. The results of hue detection stability test show that the average hue values under different brightness and saturation are 179.97° and 179.67°, and the standard deviations are 0.51 and 0.36, respectively.

In order to investigate the bit error rate (BER) performance of convolutional codes with different code rates in underwater turbulent channels, this paper employs accept-reject sampling to simulate multiplicative interference in turbulent channels.Binary phase shift keying(BPSK) modulation is chosen, and a simulation model of a Gamma-Gamma turbulent channel communication system is established. The simulation results indicate that convolutional coding can improve the BER performance of the system in turbulent channels with different intensities. The smaller the code rate of convolutional code is, the more significant the improvement of system BER performance will be. As the signal-to-noise ratio (SNR) increases, the faster the system BER decreases when the memory depth is longer. Using soft decoding can improve the gain by at least 2.82 dB compared to using hard decoding. The decoding of convolutional codes is not only affected by the current information, but also related to the previous symbol information.

In order to reduce the bit error rate of visible light communication (VLC) systems, a VLC video transmission system based on binary on-off keying (OOK) modulation is designed. This system improves the system's bit error performance and expands the system's -3 dB bandwidth by using an improved 4B5B encoding and post-equalization amplification circuit, and realizes OOK demodulation by analyzing the correlation characteristics of the data based on adaptive threshold decision. The experimental results show that the system can achieve real-time, lossless transmission of 1 080 P high definition video at a distance of two meters. When the transmission rate is 60 Mb/s, the improved 4B5B encoding can reduce the bit error rate of the uncoded system from 2.2x102 to 2.81x10-6.

To further investigate the influencing factors of Terahertz radiation generated by femtosecond laser-excited water lines,experiments are conducted to study the effects of laser power, the position of the laser focus within the water line, and water line temperature on the Terahertz radiation power. The experimental results indicate that as the laser power increases, the Terahertz radiation power also increases accordingly, but the conversion efficiency decreases. Compared to the center of the water line, the Terahertz wave radiation power is higher at the edges of the water line. Additionally, as the temperature of the water line increases, the Terahertz wave radiation power at the center of the water line also rises. Furthermore, by establishing a two-dimensional plane model of the water line and the adjacent air domain and calculating the distribution of Terahertz electric field intensity, it is found that as the frequency of the Terahertz wave increases, the electric field intensity of the Terahertz wave radiated into the air also increases accordingly.

Aiming at the need for generating complex waveforms, a periodic optical arbitrary waveform signal generation scheme based on polarization multiplexing modulation is proposed. By carefully adjusting the modulation index, radio frequency offset,and bias phase shift of the quadrature phase shift keying (QPSK) modulator, the modulators on the two orthogonal polarization states can output optical intensities in the form of sine and cosine harmonic superpositions in the Fourier series, respectively.When these optical intensities are combined, a functional waveform with precisely controllable amplitude and phase can be obtained, and it also has the characteristic of tunable functional waveform signal output. The simulation results show that when the root mean square error is less than or equal to 0.05, this scheme can generate three typical waveforms: a symmetric factor adjustable (20%~80%) ramp wave, a duty cycle adjustable (30%~100%) triangle wave, and a half-duty cycle adjustable(20%~80%) ramp wave. This verifies the feasibility of the proposed solution.