Research On Key Technologies Of High-speed Optical Camera Communication Based On Deep Learning

Recently,with the popularity of mobile terminals,wireless communication has entered the era of big data.The decreasing spectrum resources can hardly meet the users' increasing communication bandwidth requirements.In order to solve the prob-lem of spectrum resources exhaustion,visible light communication(VLC)based on the Petahertz(PHz)communication spectrum has received great attention.In visible light communication,optical camera communication(OCC)is a new type of communica-tion.OCC can easily build a multiple-input multiple-output(MIMO)communication link based on widely-used arrayed light screens and image sensors integrated in elec-tronic equipment.However,a series of technical difficulties are also introduced in OCC.Combined with deep learning,this work mainly studies the key MIMO-OCC technolo-gies using either a sparse light source array or a dense light source array and designs corresponding communication systems to verify the communication performance.First,we analyze and introduce the characteristics of the receiver and transmitter of OCC in detail,and clarifies the advantages and technical challenges.We analyze the factors that affect the communication performance of OCC,which mainly include sampling frequency offset and frame mixing,perspective distortion,spatial crosstalk,color crosstalk and nonlinearity.Meanwhile,we introduce the relevant deep learning theory,which lays a foundation for the subsequent discussion of key communication technologies.In sparse light source MIMO-OCC,the light sources at the transmitter are arranged into a sparse array.In order to increase the information transmission capability,we ap-ply color intensity modulation(CIM)to take full advantage of color,intensity and spa-tial dimensions.In the presence of sampling frequency offset and frame mixing caused by the unstable frame rate of the image sensor,we design a synchronization algorithm based on twice oversampling,embedding a few auxiliary light sources in the array.Aware of perspective distortion,we design an ROI detection algorithm based on Hough circle detection and perspective distortion correction.To alleviate color crosstalk and nonlinearity,we design a demodulation algorithm based on deep learning to transform the demodulation problem into a classification problem.Our designed offline experi-mental testbed shows that the sparse light MIMO-OCC system can achieve a transmis-sion rate of 70.56Kb/s at a distance of 0.6 m when LED array transmitter has a refresh rate of 60fps and resolution of 16 × 16,and the receiver's smartphone camera has a frame rate of 120fps.The system BER is 4.79 × 10-5 which is below 7%FEC.In dense light source MIMO-OCC,the light sources at the transmitter are densely distributed.The most effective method to increase the communication rate is to in-crease the number of modulation blocks.However,the difficulty of signal detection and demodulation increases,as the size of the modulation block decreases.Compared to signal intensity,the image sensor is more sensitive to pattern mode.Therefore,we propose color pattern modulation(CPM).In signal detection,a high-precision modu-lation block detection algorithm is designed in conjunction with a fully convolutional network(FCN).Besides,the image super-resolution algorithm based on ESRGAN net-work is used for.signal enhancement.During signal demodulation,the demodulation performance of the convolutional neural network is improved by combining the channel attention mechanism.Finally,we design and implement a corresponding system using a smartphone LCD screen of resolution 1920×1080,which achieves 124.4Kb/frame transmission rate at 17.5 cm and BER is 1.17 × 10-4.Such performance represents the state-of-the-art.