Deep Learning at the Edge for Channel Estimation in Beyond-5G Massive MIMO
Massive multiple input and multiple output (mMIMO) is a critical component in upcoming 5G wireless deployment as an enabler for high data rate communications. mMIMO is effective when each corresponding antenna pair of the respective transmitter-receiver arrays experiences an inde- pendent channel. While increasing number of antenna elements increases the achievable data rate, at the same time computing the channel state information (CSI) becomes prohibitive expensive. In this paper, we propose to use deep learning via a Multi- Layer Perceptron architecture that exceeds the performance of traditional CSI processing methods like Least Square (LS) and Linear Minimum Mean Square Error (LMMSE) estimation, thus leading to beyond fifth-generation (B5G) networking paradigm wherein machine learning fully drives networking optimization. By computing the CSI of all pairwise channels simultaneously via our deep learning approach, our method scales with large antenna arrays as opposed to traditional estimation methods. The key insight here is to design the learning architecture such that it is implementable on massively parallel architectures, such as GPU or FPGA. We validate our approach by simulating a 32-element array base station, and a user equipment with a 4- element array operating on millimeter wave (mmWave) frequency band. Results reveal an improvement up to 5 and 2 orders of magnitude in BER with respect to fastest LS estimation and optimal LMMSE respectively, substantially improving the end- to-end system performance, and higher spatial diversity for lower SNR regions, achieving up to 4 dB gain in received power signal compared to performance obtained through LMMSE estimation.
We use Communication Toolbox and Phased Array Toolbox within MATLAB to set up an mMIMO transmitter/receiver scenario. Specifically, we simulate a downlink end-to-end transmission from a BS equipped with NT = 32 URA antennas and a UE with NR = 4 ULA antennas, resulting in a 32x4 MIMO channel. Devices operate on a carrier frequency of 28 GHz, using 100 MHz bandwidth and FFT size of 256, resulting in 234 usable sub-carriers. We use a geometric scattering channel model without a line of sight (LoS) path with 100 scatterers that, for every transmission, are randomly placed on a spherical surface around the UE, which has a radius of 10 percent of the distance between UE and BS, while the position of UE and BS are assumed to be fixed, with a distance of 500 m. We generate Channel Sounding preamble frames, as explained earlier, having length L = NT OFDM symbols, and simulate transmission through a multi-path scattering channel model with Ns = 100 scatterers, as well as adding thermal noise. For training, we simulate 9000 complete transmissions, that is, including both Channel Sounding and Data Transfer phases, which are divided in 85 percent and 15 percent ratios for training and validation. In order to generate enough variation in channel realizations, we uniformly sample random seeds in the range U[1, 107], used to generate unique channel states.