Measurement-based Modulation Classification in Unlicensed Millimeter-Wave Bands

Abstract

Automatic modulation classification (AMC) facilitates adaptive modulation schemes, leading to the minimization of pilot signals, thus affecting spectral efficiency and reducing the power consumption in wireless communications systems. Since high-frequency heterogeneous and adaptive networks are established as future projections, AMC will also play a critical role in the millimeter-wave (mmWave) band communications. This study proposes multi-channel convolutional long short-term deep neural network (MCLDNN) model for AMC in mmWave bands. The performance of the proposed method is evaluated under real conditions based on a measurement campaign. 802.11ad signals are utilized for the measurements in 57.24 GHz to 59.40 GHz band. The classification performance of the proposed model is compared with that of well-known deep-learning methods, i.e., convolutional neural network and convolutional long short-term deep neural network. The measurement results imply the robustness of the proposed method to real-life conditions and its superiority against contemporary networks, especially in low signal-to-noise ratio (SNR) region.