Existing wireless systems are designed to support human-generated traffic, which is predominantly a monolith class with a few dominating applications such as voice telephony, Internet browsing, and video streaming. The success of current cellular systems is in large part due to the design of an efficient connection-based downlink for transferring data from the base station to the user equipment. This has been facilitated by the fact that we have not placed stringent demands on the quality of service. Indeed, design metrics have centered around peak/average downlink data rates and aggregate data throughput rather than on strong statistical guarantees for individual users or resiliency. Two factors are likely to upend status quo in the design philosophy of wireless systems: the changing profile of traffic and the focus on resiliency rather than on average performance measures. Growth in wireless traffic will come from distributed content creators, massive number of unattended machines enabling distributed learning, mobile robots, and video monitoring devices. These devices will demand a substantially robust and efficient uplink from the user equipment to the base station. In addition, these devices tend to generate a rich diversity of traffic profiles and demand profiles, thereby creating the need for adaptable and resilient infrastructures. With this in mind, this project addresses the need to optimize and robustify uplink wireless access for heterogenous devices. From a broad perspective, the project seeks to strengthen wireless network infrastructures. It provides opportunities for collaboration with the industry for technology transfer to wireless standards. Concurrently, the project contributes to the training of a globally competitive science, technology, engineering, and mathematics (STEM) workforce and offers upskilling opportunities to practitioners in the wireless industry.
From a technical viewpoint, the project considers the paradigm of cell-free multiple input multiple output systems with distributed signal processing as a solution to improve resiliency of cellular systems. It explores unsourced multiple access as a means to reduce coordination at the physical and medium access control layers, thereby enabling the operation of an efficient uplink within the cell-free paradigm. It creates innovative solutions to several key problems that need to be addressed in order to design a robust uplink based on cell-free unsourced random access MIMO, especially to enable distributed learning. The problems tackled in this project can be grouped into roughly three interrelated categories: (i) the design of codes and receiver signal processing algorithms for uncoordinated, non-orthogonal multiple access in cell-free MIMO systems, (ii) the design of novel joint compression, coding and multiple access algorithms for federated learning and edge computing, and (iii) the joint design of RF front ends and baseband signal processing for mitigating non-linearities that result from multiband receivers that are emblematic of future devices. The proposed solution methodologies are based on promising recent results leveraging connections between multiple access and sparse recovery, over the air federated learning, new RF front-end designs based on machine learning, and combining model-based signal processing and machine learning. Some aspects of our solution methodologies are also based on classical and highly-effective results in multi-terminal information theory and rateless codes but used in the context of distributed learning.