2019/05/08 by Ahmet M. Elbir, Elbir, Ahmet M., Kumar Vijay Mishra +1
Engineering · #Antenna Design and Optimization #Microwave Engineering and Waveguides #Millimeter-Wave Propagation and Modeling
paper · pdf · doi:10.48550/arxiv.1905.03107
In millimeter-wave communications, multiple-input-multiple-output (MIMO)\nsystems use large antenna arrays to achieve high gain and spectral efficiency.\nThese massive MIMO systems employ hybrid beamformers to reduce power\nconsumption associated with fully digital beamforming in large arrays. Further\nsavings in cost and power are possible through the use of subarrays. Unlike\nprior works that resort to large latency methods such as optimization and\ngreedy search for subarray selection, we propose a deep-learning-based approach\nin order to overcome the complexity issue without causing significant\nperformance loss. We formulate antenna selection and hybrid beamformer design\nas a classification/prediction problem for convolutional neural networks\n(CNNs). For antenna selection, the CNN accepts the channel matrix as input and\noutputs a subarray with optimal spectral efficiency. The resultant subarray\nchannel matrix is then again fed to a CNN to obtain analog and baseband\nbeamformers. We train the CNNs with several noisy channel matrices that have\ndifferent channel statistics in order to achieve a robust performance at the\nnetwork output. Numerical experiments show that our CNN framework provides an\norder better spectral efficiency and is 10 times faster than the conventional\ntechniques. Further investigations with quantized-CNNs show that the proposed\nnetwork, saved in no more than 5 bits, is also suited for digital mobile\ndevices.\n