Channel Estimation And Energy Efficiency Optimization In IRS-Assisted MmWave Communication

With the development of video service,the demands for mobile traffic grow rapidly.To achieve the 1000x capacity increase in wireless communication system,a variety of advanced technologies have been widely investigated,including massive multiple-input multiple-out(massive MIMO),millimeter wave(mmWave)and so on.However,the high frequency of mm Wave signal leads to severe path loss and makes it vulnerable to blocking obstacles.To compensate the high path loss of mm Wave signal,massive MIMO arranges a large number of antennas and radio frequency(RF)chains,which causes unaffordable hardware cost and energy consumption.Therefore,current wireless systems still suffer from high hardware and energy cost.To solve these problems,lens antenna array and intelligent reflecting surface(IRS)are introduced.Specifically,lens antenna array is composed of an electromagnetic(EM)lens and a matching antenna array located in the focal region of lens.IRS is a man-made surface consisting of massive reconfigurable reflecting elements,which are capable of changing the EM properties of the incident signal.Since IRS does not use a transmitter module but only reflects the received signal as a passive way,it can reshape the wireless environment and improve communication quality with low power consumption.By integrating both lens antenna array and IRS,an efficient system called IRS-Lens-mmWave system is designed,where lens antenna array is adopted at base station(BS)to reduce RF chains cost and IRS is deployed between BS and users to reflect the signal to circumvent the obstacles.To take full advantages of lens antenna array and IRS,related channel state information(CSI)should be acquired and joint optimization algorithm sholud be designed.For the channel estimation(CE)problem,this paper proposes a CE scheme based on super-resolution(SR)neural network.To reduce the time delay brought by the large number of IRS reflecting elements,a least square(LS)estimation with partial on-off is proposed.During the uplink pilot transmission stage,users send pilot symbols and part ofthe IRS reflecting elements are serially switched on.Based on the received pilot symbols and LS estimation,an estimated cascaded channel matrix with low dimension and low accuracy can be otained.This estimated matrix is further expanded by linear interpolation and can be considered as 2-dimension low-precision image.This image is fed into SR network to enhance the CE performance.Due to the correlation among channels at adjacent antennas and elements,the SR network is effective to improve the estimation accuracy.Simulation results demonstrate that the proposed CE scheme achieves better CE performance and subsequent IRS optimization performance compared with benchmarks.To enhance the system performance,this paper formulates an energy efficiency(EE)maximization problem.To cope with this difficult non-convex problem,the beamforming(BF)vectors at BS and IRS phase shifts are jointly optimized based on alternating optimization method.For given BF vectors,the original problem is transformed to a convex problem by mathematical methods,such as fractional programming,semidefinite relaxation and so on.Then the convex problem is solved by available solvers such as CVX to obtain the optimal IRS phase shift.For given IRS phase shift,the BF vectors are optimized by zero-forcing precoding and power based antenna selection.Simulation results show that the proposed algorithm outperforms benchmarks under different parameters setting.