The forthcoming sixth-generation (6G) communication networks are expected to significantly surpass the fifth-generation (5G) across several critical aspects, including latency, connection density, data rate, and coverage area. Furthermore, 6G will integrate artificial intelligence (AI) and sensing capabilities to optimize overall performance. Developing such advanced 6G networks requires the design of multiple access techniques that exceed current standards to ensure reliable, low-latency uplink connectivity for a large number of intelligent devices. Additionally, AI-driven services, such as distributed training and inference, necessitate a complete rethinking of communication, computation, and sensing frameworks, with communication for specific tasks advancing beyond simple bit transmission.

This thesis introduces several multiple access schemes designed to meet the stringent requirements of 6G networks. Initially, a high-data-rate non-orthogonal grant-free random access (GFRA) scheme is presented, utilizing millimeter-wave technology and extremely large antenna arrays to achieve both high data rates and precise centimeter-level localization using the same random access signals. Following this, a low-latency non-coherent GFRA scheme, optimized for aerial base stations to provide wide-range coverage, is proposed. In this scheme, transmit bits are encoded during the preamble selection process and decoded via index estimation from the received signals. The impact of rapidly varying channels is analyzed. Simulation results indicate access latency within milliseconds and an active detection error rate lower than 10^-3, even with radial velocities up to 180 km/h.

Furthermore, the thesis explores a computation-oriented multiple access scheme for federated edge learning (FEEL) in Artificial Intelligence of Things (AIoT) networks. Specifically, a massive digital over-the-air computation (MD-AirComp) approach is introduced, which tailors unsourced massive access for summation computation to improve the efficiency of FEEL among a large number of AIoT devices. This method notably accelerates the convergence of FEEL and requires fewer communication resources compared to the state-of-the-art one-bit quantized AirComp technique.

In summary, this thesis proposes several low-latency massive access techniques for next-generation IoT networks, facilitating sensing, AI-driven services, and wide-area coverage. The findings contribute to the understanding of 6G network design and offer a foundation for further investigation in this area.