Research On Channel Estimation Method Of Millimeter Wave Massive Mimo Based On Deep Learning

As a distinct technology in next generation communication,the millimeter wave and large-scale multiple-input multiple-output(MIMO)could figure out the difficulty,which does not possess enough spectrum resources,nevertheless,which brings new problems for millimeter wave channel estimation.Deep learning is incorporated to study the channel estimation solution of millimeter-wave massive MIMO system..In this paper,the fading characteristics of millimeter-wave communication channel are overall expounded,and the channel estimation algorithm of traditional millimeter-wave large-scale MIMO system is researched in depth.Aiming at the problem that the extant channel estimation algorithm model is not applicable and the sparseness of the beam domain channel can lead to lower estimation accuracy,two channel estimation algorithms are put forward to settle these problems,one is TransNet and the other is LUAMP..In allusion to the problem that the model is not applicable to the training of low-resolution ADC millimeter-wave large-scale MIMO system in different locations,a channel estimation algorithm based on TransNet network is proposed.At first,in order to establish a pre-training model,CNN is the basis of this solution,and the model apply the data of source,so as to extract the channel general features.Secondly,the transfer learning idea is adopted to transfer the channel general feature capability,train the target domain data and fine-tune the network parameters to form the channel estimation model based on TransNet network,so that the channel specific features of the target domain can be get.Compared with LS and LMMSE,the estimation performance of the raised one is more excellent,what’s more,it can also bring down amount of calculation while traning than other two distinct traning manners.What come up with is that the algorithm can deal with channel estimation called LUAMP.Firstly,by introducing the sparse structure of updated sparse signal,the shrinkage threshold adaptively changes and adding the re-weighting of constraint conditions,which improves the defect that the threshold control parameters of traditional AMP algorithm can only be fixed according to the experience value.Secondly,the improved AMP algorithm is combined with deep learning to build a LUAMP network for estimating beamspace channels by using DNN’s ability to solve nonlinear difficulty.What shows in the simulation is that LUAMP’s performance promote rapidly.