Moving Toward Intelligence: Detecting Symbols on 5G Systems Through Deep Echo State Network

Abstract

Due to the nonlinear distortion caused by radio-frequency (RF) components in the transceiver, detecting transmitted symbols for multiple-input and multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) systems can be challenging and resource consuming. In this work, we introduce a Deep Echo State Network (DESN) to serve as the symbol detector for 5G communication networks. Our DESN employs memristive synapses as the dynamic reservoir layer to accelerate the learning algorithm and computation. By cascading multiple dynamic reservoir layers in a hierarchical processing structure, our DESN processes received signal in both spatial and temporal domains. The resulting hybrid memristor-CMOS co-design provides the nonlinear computation required by the reservoir layer while significantly reduces the power consumption. From the benchmark on nonlinear system prediction, our DESN exhibits 10.31 X reduction on the prediction error compared to state-of-the-art neural network designs. Moreover, our DESN records a bit error rate (BER) of $5.76 \times 10^{-2}$ on the high-speed transmitted symbol detection task for MIMO-OFDM systems, yielding 47.73% more precise than state-of-the-art techniques in the literate for 5G communication networks.