Performance and complexity analysis using a sparse deep learning method for indoor terahertz transmission

Abstract

In this Letter, we propose and experimentally validate a sparse deep learning method (SDLM) for terahertz indoor wireless-over-fiber by transmitting a 16-quadrature amplitude modulation (QAM) orthogonal frequency-division multiplexing (OFDM) signal over a 15-km single-mode fiber (SMF) and a wireless link distance of 60 cm at 135 GHz through a cost-effective intensity-modulated direct detection (IM-DD) communications system. The proposed SDLM imposes the L1-regularized mechanism on the cost function, which not only improves performance but also reduces complexity when compared with traditional Volterra nonlinear equalizer (VNLE), sparse VNLE, and conventional DLM. Our experimental findings show that the proposed SDLM provides viable options for successfully mitigating nonlinear distortions and outperforms conventional VNLE, conventional DLM, and SVNLE with a 76%, 72%, and 61% complexity reduction, respectively, for 8-QAM without losing signal integrity.