Smart Distributed Sensing in Adaptive Wireless Networks
In the recent past, great progress has been made in three technological areas of computer science: sensing, softwarization of networks, and machine learning. Currently, a large variety of sensors is available in many devices, and sensors are getting smaller and more energy-efficient. Software-define...
|PDF Full Text
No Tags, Be the first to tag this record!
|In the recent past, great progress has been made in three technological areas of computer science: sensing, softwarization of networks, and machine learning. Currently, a large variety of sensors is available in many devices, and sensors are getting smaller and more energy-efficient. Software-defined networks are becoming more widespread, achieving low latency and high throughput for emerging applications. Machine learning is very successful in creating and improving services in numerous applications at the edge and in the cloud. There is a great potential in the overlap of these areas: (a) smart processing of sensor data using machine learning methods makes potentially huge amounts of data manageable; (b) adaptive networks support the immediate availability of sensor data in several application areas, and (c) sensor data and machine learning methods are already used in the field of adaptive networks to improve the quality of service.
In this thesis, approaches are presented to improve the quality of service, the quality of experience, and the quality of results of algorithms, protocols, and applications using different sensors and sensor sources. The information analysis cost and the achievable quality of different approaches within the same domain are compared, and a novel classification of smart systems is presented. The main challenge is to balance the information analysis cost generated by additional communication, computation, and storage with the quality improvement achievable by the novel methods. This challenge is addressed by presenting different approaches, algorithms, and systems in the areas of environmental monitoring, adaptive disruption-tolerant networking, and transitional wireless networking.
In the area of smart environmental monitoring, flexible single-board computers are used to realize improvements of various sensing tasks, especially spatial movement and visual / acoustic observation of bats, as well as automated recognition of bird species in audio recordings.
In the area of smart adaptive disruption-tolerant networking, different implementations of disruption-tolerant networks, systems for opportunistic execution of functions and workflows, and novel sensor-based routing algorithms are presented.
Insights from the two areas will be used to develop novel approaches in the area of smart transitional wireless networks for classifying network traffic flow using machine learning, for dynamic announcement intervals in service discovery, and for Wi-Fi connection loss prediction to perform seamless Wi-Fi/cellular handovers.