University of Rostock

The Internet of Things is an often-sought-after vision of a fully connected world in which not only people but also machines communicate with each other. Through this connectivity, a new degree of digitization and automation of industrial processes and services shall emerge often denoted as Industry 4.0. Thereby, the Internet of Things is not confined to smart factories or smart cities. In general, also spatially distributed applications such as smart agriculture, the monitoring of logistics chains (asset tracking), and maritime applications can benefit from the Internet of Things. However, in maritime applications as well as in very rural areas, traditional cellular networks often cannot meet the latency, coverage and data rates requirements. Therefore, the so-called satellite IoT has been discussed which leverages skylink constellations. This idea gained practical relevance with the development of so-called Cube Sats. Cube Sats are very small, not very powerful but very cheap satellites. Therefore, they cannot effort sophisticated signal processing but are restricted to simple operations.

The main goal of this proposal is the optimization of distributed compression under individual rate constraints. Consider a process that is scanned by several sensors forwarding their measurements to a common receiver. A direct communication between the sensors is not allowed. Due to capacity limitations of the forward links, the measured signals have to be locally compressed at the sensors such that

• all rate constraints of the forward links are fulfilled
• and a maximum overall relevant information about the process is preserved.

This scenario is known in information theory as the Chief Execution Officer (CEO) problem. Although it has been investigated for decades, only partial solutions have been found.

In this project, we want to develop algorithms for finding feasible solutions of the CEO problem. Using the logarithmic loss function as a distortion measure, the optimization problem can be formulated as a distributed version of the Information Bottleneck problem. Based on this principle, we want to derive solutions and compare them with simple scalar quantization or the vector quantization. The influence of the number of sensors, the loss due to non-cooperative distributed compression, the influence of the optimization order and other aspects will be investigated. Moreover, the robustness with respect to imperfect system knowledge will be analyzed.