Binary MIMO Detection via Extreme Learning Machines in 5G Network and Beyond

5G network and beyond uses massive multi-input multi-output (MIMO) to improve spectral efficiency and network capacity. The massive MIMO detection requires significant processing and computational resources along with hardware limitations. Such issues make massive MIMO detection infeasible for edge devices in dense wireless 5G network. Recent studies reveal that coarsely quantized MIMO detection schemes help in reducing the hardware requirement and the power consumption at the communication service provider's side. Classical MIMO detection schemes assume the received signal vectors are unquantized. Many of these detectors are inapplicable to binary/one-bit MIMO detection. This paper demonstrates the application of extreme learning machine (ELM), or random vector functionallink (RVFL) net, to the problem of binary MIMO detection under the effect of amplifier distortion in simple edge devices. A parallel network structure is proposed to leverage the problem settings and the stochastic nature of the detector. A convolution-based parameter initialization technique is presented to lighten the memory burden on the implementation of the detector. Numerical results suggest that our proposed design outperforms conventional schemes in catering one-bit MIMO detection with non-linearly distorted symbols, such as the high-power amplifier at the mobile user end.