Modulation Classification for OTFS-NOMA in Heterogeneous User Mobility Profile

An Orthogonal Time Frequency Space (OTFS) is particularly proposed for challenging communication scenarios where the users are moving at high speeds. Traditionally, the bandwidth resources are entirely utilized by the high mobility users. To overcome this issue, a new non-orthogonal multiple access (NOMA) protocol that utilizes OTFS called OTFS-NOMA is proposed for users with heterogeneous mobility profiles. In OTFS-NOMA, the signal for high and low mobility users are combined in power domain that utilize the same spectrum. In order to effectively utilize the available bandwidth, an adaptive modulation is used by the transmitter based on their channel conditions. At the low mobility user, the signal of high mobility user needs to be demodulated in order to perform successive interference cancellation (SIC). We generate the dataset for downlink OTFS-NOMA system considering four possible modulations for high and low mobility users and two power ratios for Signal to Noise Ratio (SNR) ranging from −10 dB to 20 dB. In this paper, we extract multiple features from the data and explore multiple machine learning based algorithms for AMC in OTFS-NOMA system. The classification performance of different machine learning models are evaluated for different scenarios. Among the various ML models, the Random Forest classifier performs the best and achieves more than 90% classification accuracy at 8 dB or higher SNRs.