In this paper, we investigate the optimal power allocation at the MAC layer in both the asymptotic and the finite blocklength regimes in a two-user NOMA uplink network, under statistical delay constraints captured through the link-layer rate. Using the link-layer effective rate analytic expressions, we provide closed-form expressions of the optimal power coefficients in low and high signal-to-ratio (SNR) regimes. These are validated by an extensive set of simulations, showing that that the proposed power allocation policy optimizes the performance of NOMA compared to OMA at Layer 2.

To enhance the sum secrecy rate of uplink networks, we propose the use of full-duplex transmission and wiretap coding in conjunction with NOMA. In further detail, in this paper we investigate the secrecy rates and provide closed-form expressions for the secrecy outage probability in a full-duplex two-user NOMA uplink network. We derive the optimal power allocation that maximizes the sum secrecy rate while guaranteeing to each user a target secrecy rate. Numerical results show that the proposed power allocation guarantees zero secrecy outage probability to both users, provided that the two users are paired so that the distance between the users exceeds a certain threshold.

We investigate the optimal power allocation and user-pairing in non-orthogonal multiple access (NOMA) uplink networks, in the asymptotic (infinite blocklength) and in the finite blocklength regimes. Firstly, in the asymptotic blocklength, we identify the optimal power allocation that maximizes the Layer-1(L1) sum-rate in a two-user network, while ensuring that the individual NOMA rates are lower-bounded by the corresponding orthogonal multiple access (OMA) rates, and, further provide an analytic expression for the Layer-2 (L2) optimal power allocation in high SNRs. Secondly, in the finite blocklength, we provide analytic expressions for the L1 optimal power coefficients in high SNRs; at L2, we first derive novel analytic expressions for the individual effective rates (ERs) validated through Monte-Carlo simulations. These expressions are used to obtain the L2 optimal power coefficients in high SNRs. What emerges from this analysis is that reducing the delay beyond a certain threshold results in a performance degradation of NOMA compared to OMA. Finally, with respect to user-pairing, our results show that the optimal pairing reduces to near-far pairing, i.e., pairing the user of index k with the user of index 2K-k+1; interestingly, this conclusion is drawn for both L1 and L2, in both the asymptotic and the finite blocklengths.