Driven by the rapid development of the low-altitude economy, Uncrewed Aerial Vehicle (UAV) networks have become increasingly important for executing tasks such as search, rescue, and monitoring. However, under bandwidth-constrained scenarios, UAV networks still face significant challenges in task execution efficiency and intelligent processing capabilities. To address these issues, we propose Multi-UAV Cooperative Deep Semantic Autoencoders (MC-DSA) for communication networks. The architecture comprises one Task UAV (T-UAV) and multiple Support UAVs (S-UAVs), which jointly enhance communication efficiency and data-processing capability via semantic communication and cooperation. Specifically, we introduce a semantic matching and source selection mechanism, enabling the proposed MC-DSA to handle low-quality images while ensuring effective utilization of high-quality images. The T-UAV broadcasts a task query, and each S-UAV computes the semantic similarity between its local image and the query. Communication links are established only when the similarity exceeds a predefined threshold. The training algorithm optimizes network parameters through gradient descent, improving semantic feature extraction accuracy. The testing algorithm evaluates UAV collaboration efficiency and communication quality. Experimental results demonstrate that, compared with benchmarks, the proposed MC-DSA improves image semantic segmentation accuracy while maintaining task execution efficiency. As a result, it reduces redundant transmissions and offers an effective solution for multi-UAV communication networks.

Speech communication plays a crucial role in the Internet of Things and autonomous networks. However, traditional communication systems suffer from insufficient semantic extraction and limited noise suppression capability. To address these issues, we propose a Speech enhancement-based Deep Semantic Autoencoder (S-DSA) for wireless communications. It consists of a dual-branch semantic encoder, a channel encoder, a channel decoder, and two semantic decoders, enabling joint source-channel coding for robust transmission. A key novelty lies in an adaptive semantic noise decoupling mechanism that disentangles target speech semantics from noise representations through an interaction module, thus maintaining semantic consistency under low SNR conditions. Furthermore, a joint training strategy optimizes speech reconstruction and noise estimation, while a lightweight testing strategy reduces complexity by disabling the noise-decoding path. the proposed S-DSA enables more effective separation between speech and noise, and yields higher robustness as well as better reconstruction quality. These findings highlight the potential of the proposed S-DSA as a practical solution for next-generation intelligent communication systems.

Objective In recent years, visible light communication (VLC) has emerged as a promising next-generation wireless communication technology, particularly suitable for high-speed indoor access. However, its performance is heavily dependent on the line-of-sight (LOS) link. In indoor environments, obstacles such as pedestrians or furniture may obstruct the LOS path, resulting in shadowed areas where communication performance deteriorates significantly, thereby degrading user experience and undermining system reliability. Although some studies have attempted to address this issue by increasing the number of light sources or integrating complementary technologies, these approaches often lead to increased cost and system complexity. Furthermore, most existing research on incorporating intelligent reflective surfaces (IRS) into VLC systems relies on idealized models and fails to effectively guarantee communication quality in complex and dynamic environments involving multiple light sources and obstacles. To address these limitations, this study proposes an innovative and robust system optimization framework aimed at mitigating the shadowing effect in indoor VLC. The primary objective is to develop a collaborative strategy that compensates for transmission link degradation and enhances both communication performance and illumination quality in environments with severe occlusion. This work seeks to validate the feasibility and advantages of the proposed approach in realistic deployment scenarios, offering a novel perspective for the development of high-quality, highly reliable indoor wireless networks. Methods A shadow modeling and system optimization method for environments with multiple light sources and multiple obstacles is proposed in this paper. The system model comprises multiple LED light sources installed on the ceiling, an array of photodetectors (PDs) on the receiving plane, and controllable IRS arrays evenly distributed on the four walls, designed to compensate for blocked LOS signals by reconstructing non-line-of-sight (NLOS) links. A dynamic shadow area analytical model is developed based on spatial geometric projection and Boolean operations. For a single light source, the shadow effect of a cylindrical obstacle on the receiving plane is accurately derived using three-dimensional geometric relationships. For multiple light sources, the concepts of "full shadow" and "half shadow" are introduced, and a feature matrix is innovatively applied for quantitative analysis. Logical operations are then employed to precisely calculate the channel gain under complex occlusion conditions. At the system optimization level, the whale optimization algorithm (WOA) is adopted to obtain the global optimal configuration of the IRS azimuth, and further integrated with a light source layout optimization strategy to form a collaborative solution for mitigating the shadow effect. The optimization objective function covers three key performance indicators: signal-to-noise ratio mean square error, illumination uniformity, and communication outage probability, to meet the performance-focused requirements in different application scenarios. Results and Discussions The effectiveness of the proposed method is systematically validated through simulation experiments. The results indicate that, with fixed light source and obstacle numbers, the IRS-VLC system optimized via WOA achieves significant performance gains across all metrics compared with a traditional VLC system. As the number of obstacles increases, the compensation effect of IRS becomes increasingly prominent, effectively mitigating the degradation of communication and illumination quality caused by severe occlusion. For example, in a complex scenario with two light sources and 16 obstacles, the introduction of IRS reduces the signal-to-noise ratio mean square error by 1.7714 dB and improves illumination uniformity by 0.6314 dBm. More importantly, the joint optimization of light source layout and IRS azimuth significantly outperforms the scheme that optimizes only the IRS. In a representative case with four light sources and eight randomly distributed obstacles, the joint optimization scheme improves the SNR mean square error by 41.2 degrees o and illumination uniformity by 37.5 degrees o over the traditional VLC system when outage probability is used as the optimization objective, and can even completely eliminate communication outages. Conclusions This paper successfully constructs and verifies an innovative channel compensation scheme that combines the assistance of IRS and the optimization of light source layout for the problem of obstacle shadows in indoor VLC systems. Research has confirmed that the proposed shadow model based on geometric optics and Boolean operations can accurately quantify the channel attenuation characteristics in multi-light source and multi-obstacle environments. On this basis, the simulation results clearly show that the introduction of IRS can significantly improve the signal-to-noise ratio distribution and illumination uniformity of the system, and its performance gain is also more obvious when the occlusion environment becomes increasingly complex. Most importantly, jointly optimizing the light source layout with the IRS Angle is the most effective strategy for solving the shadow problem. This strategy can maximize the overall performance of the system without increasing additional hardware costs, and even completely eliminate communication interruptions in specific scenarios, demonstrating strong robustness. The research results of this paper not only demonstrate the necessity and great potential of deploying IRS in VLC systems, but also provide an important theoretical basis and technical solution for the future deployment of high-performance VLC networks in dynamic and complex indoor environments.

Objective In recent years, the search for efficient, low-cost, and interference-resistant indoor positioning technologies has become a key research area. Visible light communication (VLC), based on light-emitting diodes (LEDs), is an emerging technology that offers advantages such as low hardware cost, high bandwidth, strong independence, and no electromagnetic interference. Using visible light positioning (VLP), full coverage of indoor environments and information transmission can be achieved, providing support for device positioning and monitoring. However, the application of deep neural network (DNN) in VLP is limited by high computational complexity and overfitting, which affect its practical use. Deep forest, an ensemble learning method, is more robust in data analysis compared to DNN. Its advantages include lower complexity, fewer hyperparameters, and better adaptability to data. Deep forest automatically generates cascading forest layers, enabling the model to adjust its complexity and perform well even with limited training data. In recent years, deep forest has shown great potential in solving wireless signal positioning problems. This paper proposes an indoor VLP method based on reinforced deep forest. By introducing a self-attention mechanism, the method improves the model's ability to select and weigh input fingerprint features, enhancing its adaptability and discriminative ability in complex signal environments. This method not only improves positioning accuracy but also overcomes the limitations of traditional DNN models in practical applications. Methods The experimental environment for this study is a closed space of 5 m & times; 5 m & times; 3 m. Signal collection is performed using a photodetector array to capture light intensity and incident angle data. The positioning system uses multiple LED emitters to ensure overlapping light coverage, allowing any location in the space to receive optical signals from multiple LEDs. In the offline phase, the receiver area is scanned at 0.2 m intervals, dividing it into 676 grids, and corresponding received signal strength (RSS) feature vectors are collected. These RSS vectors serve as features, and physical position coordinates are used as labels to build a fingerprint database for training the reinforced deep forest model. The input data is processed through the self-attention mechanism, which helps the model identify and emphasize the most important features for prediction while suppressing irrelevant or redundant features. The processed features then pass through a multi-scale scanning phase, which generates feature sub-samples at different scales using a sliding window method. These sub-samples are then input into the random forest for processing, improving feature diversity and expressiveness. The features obtained from the multi-scale scanning are fed into the cascaded forest part of the deep forest. In this part, feature representations are optimized in layers, and each layer's output is used as the input for the next, with the final prediction result obtained from the ensemble of multiple cascaded forests. With the introduction of the self-attention mechanism, deep forest can more accurately weigh and filter features, significantly improving its ability to identify fingerprint features in complex environments and making the model robust in noisy environments. Results and Discussions Self-attention reinforced deep forest (SA-Deep Forest) achieves an overall average error of 0.065 m, with a maximum error of 0. 306 m, outperforming the other three traditional methods in both average and maximum positioning errors. The cumulative distribution function (CDF) curve for SA-Deep Forest shifts to the left. To evaluate the stability of the algorithms, box plots of the error distributions for the four algorithms are generated, and the median, interquartile range (IQR), and variance are assessed. SA-Deep Forest performs the best, with the lowest median and smallest IQR, indicating high stability. Random forest performs second, while DNN and decision tree have wider error distributions, with large differences between the upper and lower quartiles, indicating poor performance. To validate deep forest's advantage with small datasets, the fingerprint database size is reduced by increasing the sampling interval. SA-Deep Forest's curve is closest to the center, indicating that the algorithm has the smallest average root mean square error (RMSE) across various dataset sizes, demonstrating superior and stable performance, mainly due to the self-attention mechanism and the cascaded structure. To assess the robustness of the algorithms under noise, noise is added to the training dataset at various signal-to-noise ratio (SNR) levels. SA-Deep Forest outperforms the other algorithms, maintaining relatively low positioning errors even under low-SNR conditions, which indicates stronger robustness and stability. Conclusions To improve the performance of deep forest in indoor VLP, this paper proposes an reinforced deep forest-based indoor visible light fingerprint dataset positioning method. By introducing a self-attention mechanism, the method dynamically adjusts and optimizes the weights of input fingerprint features, effectively enhancing the expression of key features and suppressing noise and redundant information. This significantly improves deep forest's ability to capture and distinguish fingerprint features in complex environments. The method constructs a fingerprint database using RSS data and inputs it into the reinforced deep forest model for training. In the online stage, real-time collected fingerprint features are input into the model, enabling quick and accurate location estimation. Experimental results show that in a 5 m & times;5 m & times;3 m indoor positioning space, the algorithm achieves an average positioning error of 0.065 m and a maximum error of 0.306 m. Approximately 85% of the test points have positioning errors under 0.1 m, outperforming random forest, DNN, and decision tree in both average and maximum error metrics. This demonstrates superior positioning accuracy and reliability. Additionally, the algorithm shows good stability and robustness, especially when trained with small-scale datasets, where the model still exhibits strong adaptability. Overall, the reinforced deep forest-based indoor visible light fingerprint positioning method not only improves positioning accuracy but also significantly enhances the model's generalization ability and robustness in noisy environments and small sample conditions. This method achieves centimeter-level positioning accuracy, meeting the needs of most complex indoor scenarios, with high practical value and potential for further application.

As the maritime industry continues to thrive and maritime services diversify, the demand for highly reliable maritime communication systems has become increasingly prominent. However, harsh marine conditions pose significant challenges to communication systems. In this work, we propose a Maritime AutoEncoder (MAE) communication system based on Attention Mechanisms (AMs) and DenseBlock (namely AM-Dense-MAE). AM-Dense-MAE utilizes DenseBlock and long short-term memory to extract deep features and capture spatio-temporal relationships, addressing the issue of “long-term dependency”. Furthermore, the decoder incorporates spatial attention modules and convolutional block attention module to enhance the preservation of crucial information and suppress irrelevant data. We employ the Rician fading channel model to simulate maritime communication channels. A substantial volume of data is utilized for model training and parameter optimization. Simulation results demonstrate that, in comparison to the benchmarks, the proposed AM-Dense-MAE exhibits better block error rate performance under various signal-to-noise ratio conditions and showcases generalization capabilities across diverse parameter settings.

Maritime monitoring has become a critical issue for both industry and academia, with the dual goals of ensuring safety and mitigating environmental impacts. This paper proposes a semantic communication system tailored for maritime monitoring, which is based on the Swin transformer-based Deep Semantic Autoencoder (Swin-DSA). The proposed Swin-DSA consists of the camera, encoder, decoder, and classifier. The camera, mounted on an unmanned aerial vehicle, captures images of the maritime monitoring area, whereas the encoder employs swin transformer to extract and encode semantic information from the images, which are then transmitted through maritime wireless communication channels to the shore-based command center. The decoder decodes the signals and restores the semantic information utilizing the swin transformer. Furthermore, the classifier employs the GoogleNet V2 to recognize and classify the decoded images. The proposed Swin-DSA optimizes its network parameters by minimizing both the reconstruction loss and the classification loss. Experimental results demonstrate that the proposed Swin-DSA not only achieves the extraction, transmission, restoration, and reconstrution of image semantics, but also outperforms the benchmark in terms of peak signal-to-noise ratio and classification accuracy. Therefore, the proposed Swin-DSA provides a new technical reference for maritime monitoring.

Driven by the continuous growth of maritime activities such as shipping, resource development, and rescue operations, the demand for highly reliable maritime communication is steadily increasing. This paper proposes a maritime end-to-end autoencoder communication system based on compressed channel feedback (CF-Dense-MAE), building on the existing maritime autoencoder communication system, aimed at enhancing the reliability of data transmission in maritime environments. CF-Dense-MAE integrates feedback encoder and decoder to learn richer signal features through channel feedback and quadratic coding. To reduce feedback overhead, we design an efficient compressed channel feedback mechanism by adjusting the output dimension of the feedback encoder and reconstructing the data in the feedback decoder. CF-Dense-MAE has been trained in environments with Rician fading channels and additive white Gaussian noise, and its model parameters are optimized with a large amount of data. Simulation results show that CF-Dense-MAE outperforms the baseline in terms of block error rate performance, validating its stronger generalization capabilities and communication reliability. With lower feedback overhead, CF-Dense-MAE can provide stable and highly reliable communication services, adapting to complex maritime environments.

With the rapid growth of the global marine economy and flourishing maritime activities, the marine Internet of Things (IoT) is gaining unprecedented momentum. However, current marine equipment is deficient in data transmission efficiency and semantic comprehension. To address these issues, this paper proposes a novel End-to-End (E2E) coding scheme, namely the Turbo-based Deep Semantic Autoencoder (Turbo-DSA). The Turbo-DSA achieves joint source-channel coding at the semantic level through the E2E design of transmitter and receiver, while learning to adapt to environment changes. The semantic encoder and decoder are composed of transformer technology, which efficiently converts messages into semantic vectors. These vectors are dynamically adjusted during neural network training according to channel characteristics and background knowledge base. The Turbo structure further enhances the semantic vectors. Specifically, the channel encoder utilizes Turbo structure to separate semantic vectors, ensuring precise transmission of meaning, while the channel decoder employs Turbo iterative decoding to optimize the representation of semantic vectors. This deep integration of the transformer and Turbo structure is ensured by the design of the objective function, semantic extraction, and the entire training process. Compared with traditional Turbo coding techniques, the Turbo-DSA shows a faster convergence speed, thanks to its efficient processing of semantic vectors. Simulation results demonstrate that the Turbo-DSA surpasses existing benchmarks in key performance indicators, such as bilingual evaluation understudy scores and sentence similarity. This is particularly evident under low signal-to-noise ratio conditions, where it shows superior text semantic transmission efficiency and adaptability to variable marine channel environments.

With the prosperous development of the shipping industry, a variety of emerging maritime services are continuously emerging, highlighting the increased importance and necessity for high-reliability maritime communication. Building upon existing research on end-to-end autoencoder communication systems, this paper proposes a densely connected convolutional networks (DenseNet)-based maritime autoencoder communication system (DenseNet-MAE). Specifically, the DenseNet module is employed to take advantage of its cross-layer connectivity and feature reuse for the extraction of deeper features of the transmitted signal. Furthermore, In order to improve the system’s reliability in low signal noise ratio (SNR) environments, the convolutional code (CC) concept is introduced into the DenseNet-MAE using recurrent neural network (RNN). The DenseNet-MAE utilizes the long short term memory (LSTM) to extract the spatio-temporal correlation of data and solve the "long-term dependency" problem. The maritime channel model is simulated utilizing the Rician fading channel. In order to demonstrate the superiority of the DenseNet-MAE, we utilize a large amount of data to train the network and optimize network parameters. Simulation results show that the proposed DenseNet-MAE exhibits excellent error performance and superior generalizability across varying Rician K factors, channel time slots, and coding rates, compared to the benchmark.

With the rapid development of maritime transportation, ocean tourism, monitoring, surveillance and aquaculture, the maritime communication industry has encountered new opportunities. Maritime End-to-end (E2E) communication systems based on deep learning (DL) have become diverse, while semantic communication systems have become a critical research point due to their ability to understand and analyze natural language. In this paper, we propose a deep semantic autoencoder (DSA) for maritime wireless communications. The DSA consists of a semantic encoder, channel encoder, channel decoder, and semantic decoder. The DSA employs transformers and convolutional neural networks to extract and restore the meaning of sentences while minimizing semantic errors during transmission. In order to enhance the effectiveness of the DSA in transmitting maritime data, the Rician fading channel is utilized to simulate complex and varied marine environments. Furthermore, the system parameters of the DSA are optimized through training the maritime data and calculating the loss functions. Extensive simulation experiments indicate that compared to the benchmark, the DSA demonstrates superior bilingual evaluation understudy and sentence similarity performance under different signal-to-noise ratio conditions, indicating its ability to effectively extract semantics.

为了充分利用信道反馈编码系统中反馈信息,提出了一种针对带有反馈信道的高可靠性自编码器系统设计方案.该方案通过在发送端引入反馈解码器模块,将编码与反馈信息的提取分成两部分进行,利用端到端的学习方式使发送端和接收端寻找到适合信道的编码和解码算法.针对反馈解码器模块,给出了基于卷积神经网络和双向门控循环单元的两种实现方案.此外,提出了一种节约反馈发送功率的部分反馈方案,通过抽样间隔灵活控制反馈发送功率,同时利用反馈解码器对反馈信息进行恢复和提取.仿真实验表明,在不同码速和不同反馈信道信噪比的情况下,提出的两种信道反馈编码方案的系统误码性能均优于已有文献给出的方案,并且在反馈信道环境较差时,提出的部分反馈方案在节省反馈发送功率的同时仍有0.1 dB的误码性能提升.

针对通信系统中长序列建模存在维度诅咒的问题,提出了一种基于深度卷积生成对抗网络(Deep Convolutional Generative Adversarial Networks,DCGAN)信道建模的端到端通信系统改进方案.该方案将卷积神经网络(Convolutional Neural Network,CNN)和条件生成对抗网络(Conditional Generative Adversarial Network,CGAN)结合,利用CNN与全连接层(Fully Connected Layer,FC)的局部连接特性对传输长序列的信道进行建模.通过对参数重新设计及网络结构调整,获得了适应不同调制方式和信道类型的学习网络,将其应用端到端通信系统中,作为收发机之间梯度反向传播的桥梁.仿真实验表明,改进的DCGAN能够以减小的网络规模以及计算量成功地实现长序列建模,并且表现出良好的泛化能力.此外,将建模结果运用到端到端通信系统设计中,可以获得与传统数字调制系统相近的误比特率性能.

Driven by the rapid growth of maritime business, the research of reliable maritime communications has attracted great attention from both academic and industry. This paper proposes a novel convolutional neural networks (CNN) -based autoencoder with channel feedback (CNN-AE-CF) for intelligent maritime communications with complex and changeable environment. A CNN-AE-CF is comprised of CNN layers, which inherits the breakthrough characteristics of CNN, such as generalization, feature learning, classification, and fast training convergence. In the CNN-AE-CF, we introduce a feedback channel and a feedback decoder at the transmitter to decode the feedback signals. Then, the transmitter combines feedback signals with input signals for secondary encoding. We leverage Rician fading channel to simulate the marine environment, in which CNN-AE-CF is trained. Finally, simulation results illustrate the superiority in terms of reliability and robustness. The proposed autoencoder is promising for intelligent maritime communications.

This paper introduces a hybrid modulation end-to-end communication system based on convolutional neural network (CNN) to optimize the structure and performance of the hybrid quadrature amplitude modulation (QAM) and pulseposition modulation ( PPM) modulation system applied to visible light communication. This scheme used the designed loss function to train the network in multiple stages to realize QAM and PPM. Accordingly, the two modulations were combined to realize hybrid modulation. With regard to demodulation, a method for recognizing the pulse of the received signal by changing the kernel size of CNN is proposed to improve the pulse-recognition accuracy and reduce the calculation complexity. The simulation results show that under the additive white Gaussian noise and Rayleigh fading channels, the proposed technical scheme exhibits fine generalization ability for the hybrid modulation method with different pulse time slots and modulation levels. When the symbol error rate is 10-3, the error performance improvement range is 0. 4 dB. 2. 8 dB compared with the traditional demodulation method.

为提升现有端到端通信系统的适应性与信道估计的准确性,提出了一种适用于多种空时编码方案的多输入多输出(Multiple-Input Multiple-Output,MIMO)自编码器通信系统.该系统将基于卷积神经网络的自编码器引入到MIMO系统中,并结合信道估计网络实现信道均衡,通过端到端的学习方式实现各种空时编码方案下信号映射传输,以获得最佳接收性能.仿真实验表明,在不同信道环境、调制方式以及空时编码方案下,所提系统表现出了良好的适应性和泛化能力;并且在垂直空时分层码方案下,所提系统可在低信噪比情况匹配最大似然检测性能,在高信噪比情况获得优于传统非线性方法的信号检测性能.与传统方法相比,所提的信道估计方法在长期演进的扩展典型城市信道下可获得5.5 dB左右的性能提升.

为提高可见光通信链路传输性能,提出一种无载波幅度相位(CAP)和幅度差分脉位调制(ADPPM)混合的调制系统.该系统通过将频谱高效的多维CAP调制与功率高效的ADPPM相结合,利用ADPPM的时隙脉冲搭载CAP信号进行信息传输;在高斯白噪声信道模型下,推导了 CAP-ADPPM混合调制系统的误帧率解析表达式并进行了仿真验证;分析了混合调制系统的阶数、幅度取值等参数对系统误帧性能的影响.实验结果表明:当频谱效率相同时,CAP-ADPPM混合调制系统的可靠性优于传统单一调制系统,并且可以提供多种参数选择方案,为可见光通信提供了更为灵活的设计选择.

In order to improve the performance of Wi-Fi fingerprint indoor positioning technology, a method based on Convolutional Neural Networks (CNN) for Channel State Information (CSI) fingerprint indoor positioning is first proposed. This method combines the CSI amplitude difference and phase difference information to train the CNN model in the offline stage. Positioning experiments are carried out in two different indoor positioning scenarios in the gallery and the laboratory, and the average positioning errors of 25 cm and 48 cm are obtained respectively; then, on this basis, the focus is on improving the timeliness of CNN-based CSI indoor positioning, and the volume is introduced. The Convolutional AutoEncoder (CAE) realizes the dimensionality reduction processing of CSI. Under the premise of ensuring the accuracy of the original positioning method, the positioning time is increased by 40% and the memory consumption is reduced to 1/15 of the original algorithm. The experimental results verify the effectiveness of the proposed algorithm.

针对应用于可见光通信的多脉冲位置调制(MPPM)方式的性能优化问题,提出了一种基于自编码器模型的MPPM传输设计方案.该方案分别利用全连接网络(DNN)和卷积神经网络(CNN)搭建自编码器模型,编码器端通过采用多阶段训练策略和自定义损失函数调控,实现了 MPPM信源符号的生成;解码器端通过全连接层或一维卷积层构建的网络完成信道和MPPM信号特征学习等功能.仿真结果表明:基于自编码器的MPPM传输系统的误码性能可以达到且略优于传统最大似然序列检测性能.

As the increasing throughput of commercial ports and the carrying capacity of large-scale container terminals, the phenomena such as dangerous goods' wrong declaration and false reporting have become increasingly prominent. Intelligent supervision of dangerous goods is essential. Aimed at the problems of poor scalability, high communication overhead, and low efficiency of the existing Practical Byzantine Fault Tolerance (PBFT) algorithm in the consortium blockchain's scenario, we design a hybrid consensus algorithm Link-Identification-Based Raft and PBFT Hybrid Consensus Mechanism (LRPBFT). Compared with the existing traceability platform of dangerous goods based on the blockchain, the LRPBFT algorithm enables the platform to have less delay, greater throughput and higher security while uploads information to the chain.

This paper proposes a visible light fingerprint positioning scheme based on a convolutional neural network (CNN) to improve the performance of indoor visible light positioning systems. In the proposed scheme, optical intensity signals are employed as the features of the reference node LED, and receiver coordinates are employed as training labels to construct fingerprint database. In addition, a positioning model based on light intensity information is constructed, and a one- dimensional CNN learning model is adopted for training. CNN application solves the problems of low-positioning accuracy and poor stability of the fully-connected feedforward neural network method. In an indoor-positioning scene (size: 5 m x 5 m x 3 m), the proposed positioning scheme obtained high positioning accuracy with an average positioning error of 4.44 cm. In addition, the performance of several different indoor visible light positioning methods was compared and analyzed in simulation experiments, and the results verified the technical advantages of the proposed scheme.