Beyond rate maximization : wireless communication techniques for edge information systems

Abstract

The traffic in mobile Internet is growing at a tremendous rate due to the proliferation of mobile devices including smartphones and Internet of Things (IoT) sensors. The availability of enormous mobile data and recent breakthroughs in artificial intelligence (AI) motivate the industry and academia to develop edge information systems (EIS) for supporting intelligent applications. The mission of EIS is to achieve efficient content retrieval and fast intelligence acquisition, involving two major techniques: edge caching and edge learning, with big data exchanging across the air-interface. Accomplishing this mission calls for the con- vergence of caching, learning, and wireless communications, leading to innovations on communication techniques beyond rate maximization. This dissertation con- tributes to the emerging area of EIS by designing data-content aware wireless communication techniques for maximizing the data utility. For content delivery in edge caching systems, one characteristic is that many signals coexisting in the air carry identical popular contents, which could cause mutual interference if their modulation and coding (MAC) schemes follows the classic approach. To retrieve useful content from interference, a novel algorithm called content adaptive MAC (CAMAC) is proposed to ensure that all signals carry identical content are encoded using an identical MAC scheme. In the process of network-performance analysis using a stochastic-geometry model, an open mathematical problem concerning the distribution of the ratio of two shot noise processes is solved. Applying this result yields the content-delivery probability in closed form, which shows that spatial signal correlation in a network can be exploited in a simple way to substantially enhance the efficiency of content delivery. For wireless data acquisition in edge learning systems, the radio resource management is revamped by exploiting the non-uniform distribution of data importance to improve the communication efficiency, thereby improving the learning performance. Specifically, a new retransmission protocol, called data-importance aware automatic-repeat-request (importance ARQ), is proposed by adapting the retransmission decision to both data importance and reliability. Underpinning the protocol is a derived elegant communication-learning relation between two corresponding metrics, i.e., signal-to-noise ratio (SNR) and data uncertainty. This relation facilitates the design of a simple threshold based policy for importance ARQ. As a result, the allocation of channel uses is biased towards protecting important data samples against channel noise while ensuring the quantity of acquired data. On the other hand, an importance-aware scheduling scheme is proposed for selecting an active user to transmit the most important data sample. To this end, a novel selection metric is designed to be the expected received data uncertainty, which integrates the SNR and data uncertainty in a summation form. Therefore, the importance-aware scheduling can exploit the multi-user diversity from two aspects: one is channel diversity due to the independent fading states, and the other is data diversity due to data samples at different devices. The deployment of the proposed scheme is shown to lead to a remarkable performance improvement compare with the conventional schemes exploiting only a single type of diversity.