Context-Aware Cognitive Radios Learning from Data Using Machine Learning Techniques

Wired or wireless, connectivity has been a vital commodity of life and more sorecently in the realm of information age. Those who have access to faster, morereliable and more ubiquitous connectivity—put simply, those who are “betterconnected”—will have significant advantages in commerce, research and a hostof other arenas. In regards to wireless communications, due to the explosionin demand for higher capacity networks, availability of free spectrum resourceshave become increasingly scarce. The UHF spectrum band in particular, dueto its excellent electromagnetic properties, has been reported as inefficientlyused and congested by many spectrum regulators of the world. This spectrumresource scarcity issue combined with the ongoing research and developmentfor more intelligent, autonomous and self-aware radio communication led to avast amount of research on the concept of Cognitive Radio.This thesis researches the learning unit of cognitive radios. The learning unitis responsible for processing information and autonomous decision making. Inparticular, the research is focused on the extraction and usage of contextualinformation from the radio environment (e.g. Radio Access Technology type,channel access pattern learning/recognition) and how such information couldbe exploited to improve the performance of the cognitive radio. The key metricsdiscussed will be based on information extraction under noise, channelblocking and interference reduction to primary users.We present a set of novel works involving Machine Learning, which is a branchof Artificial Intelligence. New implementation and use cases of state-of-the-artmachine learning algorithms are presented that learn from real-life data. In atestbed setup we program software defined radios to recognize different RadioAccess Technologies and their channel access patterns. The main techniqueused in the majority of the thesis is Artificial Neural Networks, concretely:Multi-Layer Perceptron Neural Nets, Self-Organizing Neural Nets, and DeepAuto-Encoders. In some of the works these neural network architectures havebeen combined in a novel way with Support Vector Machines, and ReinforcementLearning algorithms for channel classification and access.In this thesis we show that it is possible to achieve 95% correct classificationat -25 dB among three different radio access technologies, namely, DVB-T,WCDMA and IEEE 802.11a, where, consequently, we can reason over theoutcome of this classification to differentiate between primary and secondarytransmissions. We also show that, through the use of the proposed autoencoderapproximate Q-learning technique, such context-aware cognitive radiocan achieve better key performance metrics in dynamic spectrum access ascompared to previously researched Q-learning algorithms