Signal Detection And Channel Estimation In Ambient Backscatter Communication Systems

With the explosive growth of internet of things(Io T)devices,energy constraint and spectrum resource shortages are gradually becoming two major factors limiting the development of Io T.Ambient backscatter communication(AmBC)is a new low-energy communication technology,which achieves passive information transmission by modulating the signal to be sent to the RF signal in the surrounding environment,without the need for expensive RF components and without occupying an additional spectrum.With the wide application of AmBC,how to achieve high speed and highly reliable reflection communication has become a hot spot for domestic and international research.The reflection link is affected by the double fading effect and the traditional single antenna backscatter device(BD),resulting in the reflection link signal is usually very weak,which limits the transmission rate and communication coverage of backscatter communication.intelligent reflecting surface(IRS)is an auxiliary communication device consisting of a large number of reconfigurable reflecting units,which achieves beam assignment with controllable reflected signal strength and direction by controlling the reflection coefficient of the reflecting units.Applying IRS to AmBC,it can be used as BD for passive information transmission and enhanced reflective link gain.Due to the unknown and uncontrollable nature of the ambient RF signal,the demodulation of the reflected link signal in AmBC is strongly interfered with by the direct link RF signal.For this reason,a new symbiotic radio(SR)communication technique has been proposed.In the SR system,BD modulates its signal to the RF signal of the arriving autonomous transmitter to achieve passive information transmission,and the transmission of BD signal provides additional multi-path gain for direct link communication.the receiver of SR uses joint decoding to achieve highly reliable backscatter communication.In recent years,deep learning has become a favorite tool for solving complex problems in future wireless communication networks by virtue of its natural advantages in processing data and its powerful learning capability.With the help of deep learning methods,this paper conducts an in-depth study on the passive beam design and signal detection problems in IRS-based AmBC systems and the guide frequency design and channel estimation problems in multi-BD SR systems.For IRS-based AmBC systems,this paper proposes a deep learning-based algorithm for IRS passive beam assignment and signal detection.We propose an optimization problem for IRS beam assignment design with the objective of minimizing the detection of bit error rate(BER)for backscattered symbols in the AmBC communication model where IRS is used as BD for passive information transmission.The problem is a nonconvex optimization problem,which cannot be solved directly by conventional optimization problems.In this paper,an IRS passive beam design and signal detection network structure combining data-driven and model-driven deep learning methods is designed based on the communication system model and expectation maximization(EM)detection algorithm to realize the nonlinear mapping relationship between the IRS reflection coefficient matrix and BER.The online learning-based EM neural network(OEMNN)consists of two sub-networks.The first sub-network is constructed based on the signal transmission model,which effectively handles the unit-mode constraint of the IRS reflection unit in the optimization problem by transforming the IRS phase shift to be optimized into trainable neural network parameters.The second sub-network is obtained by unfolding the iterative process of the EM detection algorithm and is used to obtain estimates of the backscattered symbols.The parameters of the optimized neural network are obtained by training OEMNN,and these parameters are the phase shifts of the IRS reflection coefficient matrix.The simulation experimental results show that the symbol detection performance of OEMNN significantly optimizes the IRS random passive reflection beam design scheme and the IRS-free AmBC scheme.For the SR system with multiple BDs,a deep learning-based joint pilot design and channel estimation(JPDCE)scheme is proposed in this paper.The system considers two modulation modes,on-off keying(OOK)and higher-order backscattering of BDs.The JPDCE scheme minimizes the mean square error of channel estimation by constructing a guide frequency designer and a channel estimator.At the transmitter side,this paper constructs a guide frequency designer using a self-encoder network structure and a signal flow graph,maps the guide frequency signal model into the corresponding guide frequency design network,and the neural network parameters after training are the desired guide frequency signal.At the receiver side,the channel estimator first uses deep residual network(DRN)to denoise the received signal,and then estimates each channel in turn based on the SIC algorithm,where the channel estimation is done using the constructed deep neural network.In addition,their corresponding gradient update methods under the two modulation constraints of BD are given in the back-propagation process of network training,respectively.The simulation results show that the channel estimation performance of the JPDCE scheme outperforms the conventional linear minimum mean square error(LMMSE)and the deep learning scheme without SIC,and has better estimation performance with the introduction of DRN.