NOMA With Guaranteed Weak User QoS: Design and Analysis

Non-orthogonal multiple access (NOMA) in the power domain serves multiple users at the same time and frequency slot. The significance of NOMA is its ability to increase the sum rate of the system and serve more users relative to orthogonal techniques. In this paper, we look at the high rate NOMA implementations where a single beamforming vector is shared by multiple users in the same cluster in a NOMA zero-forcing beamforming (NOMA-ZFBF) system. In order to maintain user fairness, it is important to guarantee the quality of service (QoS) of all clustered users. First, considering a target signal to interference plus noise ratio (SINR) for the weak user, i.e., guaranteeing the weak user's QoS (W-QoS), we propose a simple W-QoS-based NOMA (W-QoS NOMA) algorithm for NOMA cluster formation and power allocation that always exceeds the time division multiple access rates. Second, the correlation between the users plays a vital role during NOMA cluster formation in NOMA-ZFBF system. Therefore, we integrate the proposed W-QoS with partially correlated user clustering algorithms in order to analyze the effect of using the W-QoS technique. The numerical results confirm that the integration of W-QoS with partially correlated user clustering algorithms improves the sum capacity of the system. Third, considering an arbitrary correlation between users, we derive the signal to noise ratio (SNR)/SINR expressions of strong and weak users. These expressions show the importance of correlation factor which has a higher impact on the power penalty of the strong user, which in turn affects the clustering rate. Finally, we devise a mathematical technique using a perfectly correlated channel for the strong and weak users in each NOMA cluster to derive the exact SNR/SINR distributions of singletons, strong, and weak users. This gives a closed form expression for the expected sum rate, which gives an upper bound on performance. Furthermore, we investigate how close that realistic partially correlated user clustering can get to this bound.