Adaptive Communications for Intelligent and Autonomous Systems in the Urban Internet-of-Things (IoT)

Wireless communications for intelligent and autonomous systems in the urban Internet-of-Things (IoT) involve immense challenges, especially for time-sensitive and mission-critical applications and/or disruptive environment scenarios. Moreover, the enormous scale of wireless devices with heterogeneous technologies like LTE, Wi-Fi or other device-to-device (D2D) communications often share common wireless frequency spectrum and causes interference in the dense environment of urban IoT. The aforementioned scenarios result in high latency in the application, data losses in the wireless medium and often disruption in connections among communicating devices, thus failing to meet the quality of service or computation. Modeling the behaviors of the complex interactions of communication, computation and control in the real-world autonomous systems, e.g., unmanned aerial vehicles (UAV) in the dynamic environment is practically impossible. Our proposed approach learns the behavior of the application performance through the semantics of the data and also the cross-layer information from the network protocol stack, and formulates dynamic adaptive policies for improving the performance of the application and/or mission. This dissertation proposes mitigation of these problems by realizing approaches, e.g., data-driven prediction in information constrained systems for dynamically selecting the best mode of communication and computation for a given objective function, Stochastic optimization for cognitive interference control, Software defined networking protocols and network virtualization to facilitate the adaptive framework. For the prediction of quality of wireless networks in a multi-channel or multi-network environment, the notion of probes, resembling the data semantics is introduced to predict the available unused channels or networks. In complex scenarios involving more environmental dynamics, e.g., UAV based applications, prediction based on cross-layer information and physical parameters is used to facilitate dynamic network selection for the UAV mission. In case of interference from coexisting wireless devices sharing frequency spectrum, we propose a cognitive interference control framework where based on the relevance of the data content, i.e., if the data is part of a reference block or differentially encoded block, a Markov Decision Process (MDP) determines a transmission policy that achieves required data quality while optimizing the throughput of the secondary users. Additionally, in order to minimize the latency of decision making for data processing and/or forwarding in a multi-scale computing architecture, we incorporate Network Function Virtualization (NFV) employing inbuilt kernel feature, called extended Berkeley Packet Filter (eBPF) to parse the packets in the earlier phase of the packet reception in the kernel space. The packet processing and filtering functions implemented in the user space can be called by eBPF for fast packet filtering and forwarding on layer2 to minimize the socket buffer allocation; thus, saving both system resources and latency.The proposed solutions are implemented both in simulation and on real-world systems, e.g., UAV equipped with sensors for data processing, and software defined radios (e.g. USRP B210, B200mini) for communication. This research also contributes to the development of an open-source based UAV-network simulator called ``FlyNetSim" and open-source based NFV implementation with eBPF on Linux kernel, for free use in the research community