Research On The Energy-Efficient Constellation Design For Photon-counting Optical Wireless Communication

As one of important technique for underwater wireless communication,wireless optical communication(OWC)is regarded as an ideal method for realizing underwater high-speed communication due to its lower time latency and higher bandwidth.However,because of the high path-loss existing in the transmission of the light signal,those low-sensitivity signal detectors adopted in traditional OWC will seriously impede the performance of underwater OWC and thus cannot meet various needs of future underwater applications.In order to respond to this challenge,a series of high-sensitivity detectors have drawn much attention.Owing to the unique photon-counting-based signal detecting method adopted by these devices,there exists a high difference between the model of photon-counting channel and traditional Gaussian channel.The design theory of reliable transmission for Photon-counting channel is not mature and cannot directly follow those for the Gaussian channel.Thus,towards the needs for reliable transmission of OWC system,this work aims at the problem of how to design the energy-efficient constellations for the photon-counting channel.In this thesis,our main research contents and results are listed as follows:1.Distance-optimal multilayer modulation constellation design for Poisson photon counting channel.For the problem that there dose not exist a currently-available general constellation scheme for the superposition coding under ideal photon counting boardcast channel.We propose a multilayer decomposable constellation,which can realizes the parallel transmission of the multiple users' information in power domain in a nonorthogonal way,and provide the corresponding jointly detection and decoding algorithm.Under the double constraint of Hellinger distance criterion and the uniquely decomposable property,the proposed scheme provides the designs of multiuser joint one-dimensional Hellinger distance optimal constellation and the corresponding uniquely decomposable mapping,which realizes the energy-efficiency of the constellation and the unique identification of multiuser information.Additionally,this scheme can realize the layer-allocation according to the users priority by hierarchically modulating the multiusers information.The corresponding joint detection can use the special algebraic structure of the layer coefficient for the proposed mapping to let the user directly decode the massage in target layer.By simulations,the proposed multilayer modulation has a better performance than the currently available time division multiple access,which is a scheme focusing on the fairness of multiple users.2.Multidimensional multilayer collaborative coded modulation constellation design for Poisson photon-counting channel.Catering to the need of flexible rate and reliable transmission demands and aiming at the problem of how to further improve the rate flexibility and energy-efficiency of the proposed multilayer modulation,we propose a time-collaborative multilayer coded modulation constellation and the corresponding fast square-root receiver scheme.With the aid of the non-integer rate of linear block code,the proposed scheme can realize the flexible rate adjustment by a cascade structure between the multipath block encoder and the multilayer modulator,which can be proved that the proposed constellation can get a larger minimum distance by coding the low layers.Then,considering that the complexity problem of the maximum likelihood receiver is more serious for high dimensional constellation,lowcomplexity receiver based on square root transform is devised.Finally,we analyze the achievable rate with infinite-alphabet-input for the currently available superposition coding and the proposed multilayer modulation.By simulation,the proposed time-collaborative constellation is more energy-efficient than the time-independent one.Correspondingly,the proposed lowcomplexity receiver can realize an exponentially reduction in the algorithm complexity compared with the maximum likelihood receiver at the cost of a minor performance loss.Compared with the currently-available superposition coding,the proposed multilayer modulation and nonlinear superposition coding can exceed the achievable rate region of the former.3.Data-driven energy-efficient constellation design for Poisson photon-counting channel.To deal with the problem that the shortage of the constellation design tools for ideal-photoncounting channel,and the currently-available constellation design tool suffers performance reduction because the model approximation condition is weaken by the channel parameters,we propose two autoencoder-based data-driven constellation design tools and provide a novel active function for the sparse constellation learning task under peak-limited constraint.When constructing the design tools,we adopt two different ways to surmount the implementation problem of the currently-available autoencoder caused by the discreteness of channel output and the unavailability of reparameterization for channel distribution: With the aid of square root transform,we propose a square root autoencoder(SR-AE)based on the currently available training algorithm at the cost of the performance loss.By designing a piece-wisely training algorithm,we propose a double neural network autoencoder(DNN-AE)at the cost of relatively high algorithm complexity to obtain a better learning performance.Additionally,by using the regularization effect,we propose a novel active function to relieve the gradient vanishing problem in the task of sparse constellation learning with peak-limited constraint.According to the simulations,the proposed autoencoder can learn the constellation with better performance when the channel condition gets worse.The DNN-AE always obtains the best constellation and the square root transform can cause the error rise at the point near the zero point.Additionally,the proposed active function can improve the learning performance when the autoencoder learns a sparse constellation.4.Data-driven model-free energy-efficient constellation design for non-Poisson photoncounting channel.Towards the barrier of channel modeling caused by the complex physical process existing in the actual device and the problem that artificial analysis method cannot fulfill the constellation design task,we propose a policy-gradient-based model-free autoencoder constellation design technique,finish the design of policy function and the convergence proof of the corresponding training algorithm in general photon-counting channels,and propose two-way symmetrical autoencoder for training with noisy feedback.First,based on the proposed double-neural-network autoencoder model,introducing the policy-gradient method to deal with the gradient-dependent problem of the original training algorithm,we obtain a modelfree autoencoder scheme and the corresponding double-network synchronous update fast training method;Then,considering the nonnegative and peak-limited signal property,we design a policy function and finish the convergence proof of the corresponding training algorithm in general photon-counting channels.Finally,considering the fact that practical implementation of the PGA MF scheme needs a preexisting feedback link,we propose a two-way symmetric autoencoder which can learn the desired transceiver without a preexisting feedback link.As shown by simulations,the proposed model-free constellation design tool can learn the constellations which can perform the same as or better than the currently-available constellation in the considered two photon-counting channels and has the advantage in the algorithm convergence rate than the existing model-free tool.Additionally,the proposed two-way symmetry autoencoder can obtain the constellation under the noisy feedback condition which has almost the same performance as that learned in the ideal feedback condition.