With the exponential growth of wireless connections and the limited bandwidth of the radio spectrum, extensive research on multiple access is crucial for facilitating efficient multi-user communication in future wireless networks. Traditional orthogonal multiple access (OMA) divides resources independently and assigns them to different users to ensure interference-free transmission. However, this approach limits the number of wireless connections and spectral efficiency. Non-orthogonal multiple access (NOMA) overcomes these limitations by superposing users’ data for transmission and using successive interference cancellation (SIC) to eliminate interference among users. Recently, over-the-air computation (OAC), an analog multiple access scheme, has gained attention for jointly computing a function over a multiple-access channel by superposing their transmission signals, effectively leveraging interference for computation. Both NOMA and OAC demonstrate high spectral efficiency and are suitable for different wireless networks.

The thesis aims to investigate the potential of NOMA and OAC in enhancing existing techniques for various application scenarios, with a specific focus on improving energy efficiency and reducing communication delay, which are crucial performance metrics in wireless networks. Given the inherent tradeoff between energy and delay, this thesis considers both delay-constrained and energy-constrained scenarios.

In delay-constrained scenarios, the objective is to minimize the total transmit energy while adhering to a given delay constraint. The particular scenario under consideration involves the application of NOMA in caching networks, which presents a challenging energy minimization problem due to the existence of cache-aided interference cancellation (CIC) and its combination with other transmission techniques like index coding. To tackle these challenges, a cross-layer unified framework is proposed to integrate NOMA and index coding, and two novel algorithms, index-coded NOMA and dynamic coded-NOMA are designed.

In energy-constrained scenarios, the objective is to minimize the total transmit delay under energy or power constraints. The thesis first considers the uplink multi-subcarrier NOMA system, where the minimization of the maximum completion time, a crucial delay metric, is investigated. The associated problems of power allocation, user pairing, and scheduling are analyzed. The second scenario involves the uplink of distributed machine learning over the network edge. In this case, conventional multiple access techniques like OMA or NOMA result in significant communication delay, thereby hindering learning efficiency. Therefore, OAC, a delay-efficient multiple access technique, is considered in wireless distributed learning networks. Inspired by gradient coding, this work proposes leveraging computing capabilities through data repetition to achieve diversity gain. Data assignment and power allocation are jointly designed to minimize the mean square error (MSE) of the aggregation data.

Through these investigations, the thesis harnesses the potential of NOMA and OAC to enhance energy efficiency and reduce communication delay across various wireless network scenarios. Practical schemes are designed, and their superior performance is demonstrated through computer simulations.