Deep and robust resource allocation for random access network based with imperfect CSI

A deep and robust resource allocation framework was proposed for the random access based wireless networks,where both the communication channel state information (C-CSI) and the interference channel state information (I-CSI) were uncertain.The proposed resource allocation framework considered the optimization objective of wireless networks as a learning problem and employs deep neural network (DNN) to approximate optimal resource allocation policy through unsupervised manner.By modeling the uncertainties of CSI as ellipsoid sets,two concatenated DNN units were proposed,where the first was uncertain CSI processing unit and the second was the power control unit.Then,an alternating iterative training algorithm was developed to jointly train the two concatenated DNN units.Finally,the simulations verify the effectiveness of the proposed robust leaning approach over the nonrobust one.