Abstract

In an era of Internet of Things (IoT), new services are expected to generate heterogeneous traffic that involves both human-to-human and machine type communications (MTC). There will be MTC services that require massive connectivity, higher throughput, and low latency. 5G networks are under development to meet the needs of MTC. Non-Orthogonal Multiple Access (NOMA) has currently gained a traction in 5G for the realization of MTC. NOMA enables simultaneous utilization of resources by multiple users. This thesis considers Power Domain NOMA (PD-NOMA) among other variations for uplink communications. In PD-NOMA, users transmit signals simultaneously at pre-determined receive power levels. The base station decodes the signals using Successive Interference Cancellation (SIC) technique starting from highest power level. Each decoded signal is subtracted from the received signal to enable decoding of the next higher power signal. The decoding continues until a collision is detected at a power level. The signals at that power level as well as signals transmitted at the lower power levels cannot be decoded. The previous work on PD-NOMA assumed uniform user access to the system power levels. This thesis considers random non-uniform selection of power levels by the users.
The system model under consideration assumes that the new packets to be transmitted arrive according to a Poisson process and the time axis slotted. As a user may have only a single packet to be transmitted during any slot, each new packet is assumed to be generated by a different user. We consider two packet service strategies which are with and without packet loss.
In the packet loss service strategy, a packet can only be transmitted once, and packet is lost if that transmission is not successful. We determine the throughput of the system and then determine the optimal user access probabilities to the power levels that maximizes the throughput. We also determine the receive power levels for SIC as a function of the Signal-to-Interference-plus-Noise Ratio (SINR). Numerical results show that the optimal access probabilities are non-uniform, and they are a function of packet arrival rates. Further, in uniform access, the throughput of the system drops to zero at higher power levels whereas, in optimal non-uniform choice, the system maintains a non-zero throughput. We also consider the case that the users may choose not to transmit their packets to reduce the collisions in the system. The packets that are not transmitted, are lost, however the system benefits from reduced collisions. Numerical results show that the optional transmission of packets achieves the same maximum throughput as compulsory packet transmission, but the throughput does not decrease with increasing packet arrival rate.
In the service strategy without packet loss, a user will keep transmitting a packet until it is successfully received by the base station. We derive the Probability Generating Function (PGF) of the distribution of the number of packets in the system at the steady-state by imbedding a homogenous Markov chain at the end of the slots. We determine enough number of equations to solve for the unknowns in the PGF. Then, we obtain the mean packet delay from the PGF of the number of packets in the system through application of the Little’s result. Mean packet delay is a function of the user access probabilities to the power levels. We show how to determine the optimal access probabilities both for optional and non-optional transmission of a packet during a slot. We plot the mean packet delay as a function of the packet arrival rate for optimal user access probabilities both for optional and non-optional packet transmission, as well as for uniform choice of power levels. The numerical results show that the optimal access with optional transmission results in lowest mean packet delay and therefore in highest throughput.