Novel QoS-Aware Proactive Spectrum Access Techniques for Cognitive Radio Using Machine Learning

Abstract

Traditional cognitive radio (CR) spectrum access techniques have been primitive and inefficient due to being blind to the occupancy conditions of the spectrum bands to be sensed. In addition, current spectrum access techniques are also unable to detect network changes or even consider the requirements of unlicensed users, leading to a poorer quality of service (QoS) and excessive latency. As user-specific approaches will play a key role in future wireless communication networks, the conventional CR spectrum access should also be updated in order to be more effective and agile. In this paper, a comprehensive and novel solution is proposed to decrease the sensing latency and to make the CR networks (CRNs) aware of unlicensed user requirements. As such, a proactive process with a novel QoS-based optimization phase is proposed, consisting of two different decision strategies. Initially, future traffic loads of the different radio access technologies (RATs), occupying different bands of the spectrum, are predicted using the artificial neural networks (ANNs). Based on these predictions, two strategies are proposed. In the first one, which solely focuses on latency, a virtual wideband (WB) sensing approach is developed, where predicted relative traffic loads in WB are exploited to enable narrowband (NB) sensing. The second one, based on Q-learning, focuses not only on minimizing the sensing latency but also on satisfying other user requirements. The results reveal that the first strategy manages to significantly reduce the sensing latency of the random selection process by 59.6%, while the Q-learning assisted second strategy enhanced the full-satisfaction by up to 95.7%. INDEX TERMS Cognitive radio networks, dynamic spectrum access, machine learning, self-organizing networks, Q-learning.

Q-Learning Assisted Energy-Aware Traffic Offloading and Cell Switching in Heterogeneous Networks

“…Reinforcement learning is a risk and reward kind of learning whereby the agent (or macro cell) gets information from the environment and then tries to take action and is rewarded or penalized depending on whether the action taken is right or wrong. Reinforcement learning is applied in this work due to its suitability to handle this kind of tasks that involve making decisions out of a wide-range of options [13]. As a illustration in our study, the macro cell interacts with the network environment, obtains information about the traffic loads levels of the small cells through its backhaul connection with them and then decide which combination of small cells to switch off per time.…”

Spectrum Cost Optimization for Cognitive Radio Transmission over TV White Spaces using Artificial Neural Networks

“…ANN has already being used in wireless communication network optimization [9]- [12] due to its abilities in dealing with large dimensions and ease of implementation. Moreover, it outclasses the statistical methods in regression problems, as it does not require any prior knowledge about the underlying distribution in the data [2]. On the other hand, the current case is treated as a classification problem, where the elements in A constitute the classes, and ANN provides a suitable solution due to its proven performance in solving the classification problems [15].…”

“…The rapid increase in the number of machine type devices and internet of things (IoT) applications would result in an enormous demand for wireless spectrum [1]. Cognitive radio (CR) provides a means of opportunistically exploiting unused frequency channels for communication by an unlicensed or secondary user when the licensed or primary user is absent [2]. However, exploiting the licensed frequency spectrum might not be sufficient to handle the enormous demands that would be placed on the network due to anticipated massive deployment of IoT devices in 5G networks [3].…”

AMC With a BP-ANN Scheme for 5G Enhanced Mobile Broadband

“…Through ML, it is possible to solve the needs of adaptive and 5G services and application scenarios. Machine learning technology has the characteristics of flexibility and self-adaptation, and many studies have applied machine learning to cellular networks [11], which is often used in quality of service (QoS) [12], HARQ [13] [34][35][36] and network slicing [14]. In [15], this study proposed the generalized minimum-sum decoding algorithm with machine learning to decode LDPC codes in 5G network.…”