of radio spectrum. Spectrum bands are categorized as licensed and unlicensed bands. Licensed bands are used by licensed users, called Primary Users (PUs). They have the priority to use the spectrum. Unlicensed users, called Secondary Users (SUs) can use the licensed bands as long as the PUs are not temporally using it; or as long as the PUs’ can properly be protected. However, SUs should vacate the licensed bands immediately when a PU is detected to be active. This will significantly improve spectrum utilization. SUs detect the conceivable vacant bands, determine operational channel, and eventually adjust their parameters. Thus, efficient spectrum sensing techniques are key to the successful operation of CR networks.
In CR systems, SUs should be able to [14]:
Sense the spectrum bands and determine the possible channels as well as activity of PUs.
Decide on the quality of available channels that satisfy users’ requirements.
Share the available channels with other SUs.
Avoid harmful interference to PU who is starting to use the any channel by vacating the channels PU just start operating on.
The detection of PU’s presence is a major challenge in CR. This process needs complex sensing technologies. This complexity stems from the nature of the electromagnetic signals, the multipath fading, and the changing interference. Spectrum sensing for CR systems is a very interesting research area. Researchers try to develop quick and accurate methods to detect the PU’s activity.
The transmission cycle of a SU can be represented as seen in Figure 2.8. The sensing process is performed periodically using one of the sensing methods. During the sensing time, the SU does not transmit data. After sensing, SU can decide to transmit on the same channel or switch to another channel, depending on PU presence. The transmission continues until the next sensing period. Clearly, the sensing time should be as short as possible, but at the same time enough for accurate sensing. Therefore, there is a trade-off between protecting the PU’s QoS and improving the QoS of SUs.
Figure 2.8 The transmission cycle of SU.
2.6.1 Sensing Methods
The fundamental objective of sensing methods is to increase the positive detection probability and decrease the false detection probability of PUs. This leads to more protection of PUs and at the same time more utilization of the available spectrum. In this section, we present the main sensing methods. However, several other methods can be found in the literature.
1. Energy Detection
The most commonly adopted method because of its simplicity and the associated computation overhead. It requires short sensing interval. An energy detector is used to detect a narrowband channel. The detected energy is compared to a predefined threshold. If the measured energy is found to be larger than the threshold, then PU is judged to be active. The selection of the threshold is a challenge because the noise level is normally unknown.
2. Cyclostationary Feature Detection
This method aims at distinguishing between PUs’ signals, interference, and noise. This is achieved by identifying the cyclostationary features of signals, including modulation type, carrier frequency, and data rate. The implementation of this method needs sufficient prior information about these features of the PUs’ signals so that the method can use this knowledge as a base during the matching of measured features with those belong to PUs’ signals. Hence, sufficient number of samples is needed for accurate performance, leading to long sensing intervals.
3. Matched Filtering
This method is considered to be the most accurate method that achieves higher detection probability in short sensing intervals. The basic idea of this method is that the sensed signal is passed thought a filter that is matched to the PUs’ signals. Despite its accuracy, the method is considered to be impractical in cases wherein PUs transmit signals of different features.
2.7 ML for Wireless Networks: Challenges and Solution Approaches
In previous sections, we discussed various topics about recent trends in network configuration, control, and management. We indicated recent trends in network architecture design in order to respond to the requirements of large-scale networks. Large-scale networks feature a significant increase in the number of users and smart technology-based systems that have become an integral part of human life, especially with the vast amount of software applications that require high speeds and fast real time response. In addition to that, different applications emerged in the last years to allow easy and high quality communication between people. Such applications have become an important tool used even by the largest media stations.
In the following sections, we review and discuss important challenges in wireless networking that can be better tackled, exploiting ML approaches. The application of ML in wireless networking aims at reducing human interaction and creating a self-driven network, that are able to optimize and configure themselves. We focus on recent published research and try to shed light on important research aspects of the present and future. Our goal is to assist readers to identify the scientific areas and specific issues that need further research and exploration. We divide the discussion into three parts. In the first part, we focus on Cellular networks, while the second part focuses on wireless local area networks (WLANs). The third part is devoted to cognitive radio networks.
2.7.1 Cellular Networks
2.7.1.1 Energy Saving
With the steady increase in the number of users of wireless networks and the need to deploy large number of base stations, and since base stations consume large energy; operating the network with minimum energy is a challenge. One way to reduce energy consumption is the idea of turning off some base stations if users can be served from others, while maintaining a reasonable QoS level. Learning the operation of the network over time helps in improving decisions about which base stations might be switched off.
An SDN-based ML system for energy saving is proposed in [15]. Performance of neural networks and SVM algorithms is compared. The network trains itself using data collected from base stations and recommends the operator time periods during which some base stations are predicted to handle very low traffic and therefore can be switched off.
The authors of [16] propose a Q learning method for base station on-off switching. The switching of base stations is defined as the actions, while the traffic load is defined as the state. The overall objective is to minimize energy consumption. Policy values are used by the controller to decide on switching. After performing a switch operation, the system state is changed and the energy cost of the former state is computed. If the energy cost of the newly executed action is smaller than energy costs with other actions, then the controller updates the policy value in order to increase the probability of selecting this action. With time, the optimal switching mechanism is obtained.
2.7.1.2 Channel Access and Assignment
The effective use of wireless channels has become an urgent necessity, as many heterogeneous systems operate in the same frequency band. Thus, coexistence and organized access of the shared frequency chunks by systems are necessary. Consequently, any design of the wireless channel sharing mechanism should be based on a prediction of the behavior of networks users.
In [17], the authors propose deep reinforcement ML-based MAC protocol for the coexistence of multiple heterogeneous networks. The method allows time-sharing access of the spectrum, by a series of observations and actions. The MAC protocol does not have to know the MAC mechanism of other networks and tries to maximize the throughput of all coexisting networks. The authors of [18] employ reinforcement learning for managing cell outage and compensation. The system state is constituted by the