Joint design of compressed sensing and network coding for wireless meshed networks

点击翻译按钮获取中文摘要

Based on the impressive features that network coding and compressed sensing paradigms have brought separately, the idea of bringing them together seems obvious. By combining them, we can realize low latency communication for large-scale sensing scenarios just by reducing the amount of data significantly. Our first phase proposal aimed to break with the agnostic combination of the two key techniques and replace it with a holistic approach for wireless meshed networks. We identified the relevant scenarios and applications, in order to design a robust joint encoding/recoding/decoding and compression scheme. Currently, we are deploying our joint approach in real-life for industrial IoT-devices for audio and video transmissions. We will deliver a full-fledge demonstrator by the end of the first phase.In the second phase of the project, we would like to continue our research work with novel ideas and concepts in order to achieve low-latency, scalability, and security for future communication systems. The main goals of phase two are i.) finite field compressed sensing, ii.) joint coded computation, iii.) adaptive learning strategies for optimal coding/compression decisions per node, and iv.) group testing with user activity detection. While network coding already operates in finite fields, the challenge is to change compressed sensing from real field to finite field computation. We expect that finite-field compressed sensing will overcome the drawback of costly computational complexity, which has a direct impact on latency. Furthermore, we would like to continue the work in combinatorial group testing and coded computation for user activation and secure wireless distributed storage. In this context, we expect that joint network coding and compressed sensing design with our new ideas will enhance tremendously the built-in security and reliability of any future communication system. By bringing in machine learning, and in particular learning based on deep neural networks, we expect to considerably reduce the latency and complexity for real-time applications.