Design of distributed signal processing algorithms and scalable hardware platforms for energy-vs-performance adaptive wireless acoustic sensor networks.

Wireless acoustic sensor networks (WASNs) consist of a set of battery-powered nodes, distributed over a wide area and equipped with acoustic sensors (microphones) as well as processing and wireless communication facilities. The nodes cooperate by exchanging pre-processed microphone signals to perform signal processing (SP) tasks, such as for instance a speech enhancement task. State-of-the-art instantiations however suffer from fast battery depletion, since their operating parameters (e.g. transmission range, sampling rate, number of fused signals) are typically fixed at design time and kept identical across nodes. As a result, the system is incapable to dynamically adapt to a varying context, such as changing performance requirements or operating conditions, in order to avoid energy wastage. To increase lifetime, the energy-vs-performance (EvP) trade-off should be fully exploited across system levels. To achieve this, a close collaboration between the SP algorithm and the hardware platform is required. Therefore, this project targets the development of a framework for collaborative optimization of EvP aware distributed SP algorithms, in  articular speech enhancement algorithms, and the supporting EvP scalable hardware platforms, with wide-range linear EvP scalability. The two domains are linked through EvP models, autonomously learned by the platforms at run-time. The ultimate goal is to demonstrate efficient, selforganized, operation in challenging WASN scenarios.