Statistical QoS Provisioning Over Uncertain Shared Spectrums in Cognitive IoT Networks: A Distributionally Robust Data-Driven Approach

With the soaring wireless traffic for Internet of Things (IoT), spectrum shortage becomes an extremely serious problem, leading to the paradigm shift in spectrum usage from an exclusive mode to a sharing mode. However, how to guarantee the quality of service (QoS) when using the shared spectrum is not straightforward due to its uncertain availability. In this paper, from a session-based view, we propose a metric to evaluate how much data can be delivered via a shared band during a session period, named probabilistic link capacity (PLC), which offers us an effective way to guarantee the QoS statistically. Different from most existing works where the distributional information is assumed exactly known, we develop a distributionally robust (DR) data-driven approach to estimate the value of the PLC based on the first and second order statistics. Two cases are considered that the statistics are exact or uncertain with estimation errors. For each case, to calculate the DR-PLC, we formulate it into a semidefinite programming problem based on the worst-case of conditional-value-at-risk. With the proposed metric, we further design a service-based spectrum-aware data transmission scheme, which allows us to efficiently use different kinds of spectrums to satisfy the diverse IoT service requirements.