Optimum Power Allocation for HARQ-Aided NOMA With Proportional Fairness on Fading Channels

Power-Domain NOMA is one of the enabling technologies for future wireless communication networks of the fifth and sixth generations. This work addresses some key features of Power-Domain NOMA, including the impact of block fading on interference cancellation (leading to outage events), the limited channel state information available at the transmitter (consisting in the simple statistic distribution of the channel state), the fairness of the user achievable information rates (according to the Proportional Fairness criterion), and the optimization of the outage probability in the presence of a simple or hybrid ARQ protocol. After recalling some basic results on the achievable outage information rate region, the Proportional Fairness criterion is used to optimize the power allocation rates required to achieve specific outage probabilities. This is achieved by properly choosing the outage probabilities in conjunction with the hybrid ARQ protocol. To this purpose, Maximum Ratio Combining is used to enhance the achievable rate for multiple retransmissions. The system throughput analysis and optimization resort to a Markov chain representation of the hybrid ARQ protocol. This allows to assess the impact of retransmissions on the throughput. The latency involved is assessed by evaluating the average value and the standard deviation of the packet transmission delay. Numerical results are reported for two different system models: 1) symmetric scenario, where all users have the same average SNR, which varies according to Rayleigh fading; 2) asymmetric scenario for a single-cell broadcast channel, where the users are uniformly located over a disk and their average SNR depends on the distance from the transmitting base station at the center. For the latter scenario, the base station is assumed to know the user distances, which corresponds to a partial knowledge of the channel state at the transmitter. Both scenarios are thoroughly analyzed, and the impact of several system factors is discussed in detail. The results show, among other things, that very high outage probabilities may be required to optimize the throughput in low average SNR conditions, and that the optimum power allocation at the transmitter may reach a wide dynamic range when the SNR is large.