Machine learning explores the study and development of algorithms that can learn from and make predictions and decisions based on data. Applications of machine learning in wireless communications have been receiving a lot of attention, especially in the era of big data and IoT, where data mining and data analysis technologies are effective approaches to solving wireless system evaluation and design issues. This edited book presents current and future developments and trends in wireless communication technologies based on contributions from machine learning and other fields of artificial intelligence, including channel modelling, signal estimation and detection, energy efficiency, cognitive radios, wireless sensor networks, vehicular communications, and wireless multimedia communications. The book is aimed at a readership of researchers, engineers and students working on wireless communications and machine learning, especially those working in big data and artificial intelligence multi-disciplinary fields related to wireless communication technologies.
Inspec keywords: data mining; radiocommunication; learning (artificial intelligence); telecommunication computing; Big Data; data analysis
Other keywords: data mining; data analysis; wireless communications; wireless system design; machine learning; wireless system evaluation; Big Data
Subjects: Radio links and equipment; Communications computing; Knowledge engineering techniques; Data handling techniques
Machine learning, as a subfield of artificial intelligence, is a category of algorithms that allow computers to learn knowledge from examples and experience (data), without being explicitly programmed. Machine-learning algorithms can find natural patterns hidden in massive complex data, which humans can hardly deal with manually.In wireless communications, when you encounter a complex task or problem involving a large amount of data and lots of variables, but without existing formula or equation, machine learning can be a solution. Traditionally, machine-learning algorithms can be roughly divided into three categories: supervised learning, unsupervised learning and reinforcement learning (RL). In this chapter, we present an overview of machine-learning algorithms and list their applications, with a goal of providing useful advice and references to fundamental concepts accessible to the broad community of wireless communications practitioners.
In this chapter, we present an introduction to the use of machine learning in wireless propagation channel modeling. We also present a survey of some current research topics that have become important issues for 5G communications.
In this chapter, the authors address the wireless channel prediction using state-ofthe-art machine-learning techniques, which is important for wireless communication network planning and operation. Instead of the classic model-based methods, the authors provide a survey of recent advances in learning-based channel prediction algorithms. Some open problems in this field are then proposed.
Wireless communication has been a highly active research field. Channel estimation technology plays a vital role in wireless communication systems. Channel estimates are required by wireless nodes to perform essential tasks such as precoding, beamforming, and data detection. A wireless network would have good performance with well-designed channel estimates. In this chapter, we first review the channel model for wireless communication systems and then describe two traditional channel estimation methods, and finally introduce two newly designed channel estimators based on deep learning and one expectation-maximization-based channel estimator.
As an intelligent radio, cognitive radio (CR) allows the CR users to access and share the licensed spectrum. Being a typical noncooperative system, the applications of signal identification in CRs have emerged. This chapter introduces several signal identification techniques, which are implemented based on the machine-learning theory.
This chapter introduces the fundamental concepts that are important in the study of compressive sensing (CS). We present the mathematical model of CS where the use of sparse signal representation is emphasized. We describe three conditions, i.e., the null space property (NSP), the restricted isometry property (RIP) and mutual coherence, that are used to evaluate the quality of sensing matrices and to demonstrate the feasibility of reconstruction. We briefly review some widely used numerical algorithms for sparse recovery, which are classified into two categories, i.e., convex optimization algorithms and greedy algorithms. Finally, we illustrate various examples where the CS principle has been applied to deal with various problems occurring in wireless sensor networks.
In this chapter, the authors study the enhancement of the proposed IEEE 802.11p medium access control (MAC) layer for vehicular use by applying reinforcement learning (RL). The purpose of this adaptive channel access control technique is enabling more reliable, high-throughput data exchanges among moving vehicles for cooperative awareness purposes. Some technical background for vehicular networks is presented, as well as some relevant existing solutions tackling similar channel sharing problems. Finally, some new findings from combining the IEEE 802.11p MAC with RL-based adaptation and insight of the various challenges appearing when applying such mechanisms in a wireless vehicular network are presented.
We present in this chapter the advantage of applying machine-learning-based perceptual coding strategies in relieving bandwidth limitation for wireless multimedia communications. Typical video-coding standards, especially the state-of-the-art high efficiency video coding (HEVC) standard as well as recent research progress on perceptual video coding, are included in this chapter. We further demonstrate an example that minimizes the overall perceptual distortion by modeling subjective quality with machine-learning-based saliency detection. We also present several promising directions in learning-based perceptual video coding to further enhance wireless multimedia communication experience.
Saliency detection has been widely studied to predict human fixations, with various applications in wireless multimedia communications. For saliency detection, we argue that the state-of-the-art high-efficiency video-coding (HEVC) standard can be used to generate the useful features in compressed domain. Therefore, this chapter proposes to learn the video-saliency model, with regard to HEVC features. First, we establish an eye-tracking database for video-saliency detection. Through the statistical analysis on our eye-tracking database, we find out that human fixations tend to fall into the regions with large-valued HEVC features on splitting depth, bit allocation, and motion vector (MV). In addition, three observations are obtained from the further analysis on our eyetracking database. Accordingly, several features in HEVC domain are proposed on the basis of splitting depth, bit allocation, and MV. Next, a support vector machine (SVM) is learned to integrate those HEVC features together, for video-saliency detection. Since almost all video data are stored in the compressed form, our method is able to avoid both the computational cost on decoding and the storage cost on raw data. More importantly, experimental results show that the proposed method is superior to other state-of-the-art saliency-detection methods, either in compressed or uncompressed domain.
In this chapter, we incorporate deep learning for indoor localization utilizing channel state information (CSI) with commodity 5 GHz Wi-Fi. We first introduce the state-ofthe-art deep-learning techniques including deep autoencoder network, convolutional neural network (CNN), and recurrent neural network (RNN). We then present a deep-learning-based algorithm to leverage bimodal CSI data, i.e., average amplitudes and estimated angle of arrivals (AOA), for indoor fingerprinting. The proposed scheme is validated with extensive experiments. Finally, we discuss several open research problems for indoor localization based on deep-learning techniques.
In this chapter, we shall focus on the formulation of radio resource management via Markov decision process (MDP). Convex optimization has been widely used in the RRM within a short-time duration, where the wireless channel is assumed to be quasi-static. These problems are usually referred to as deterministic optimization problems. On the other hand, MDP is an elegant and powerful tool to handle the resource optimization of wireless systems in a longer timescale, where the random transitions of system and channel status are considered.These problems are usually referred to as stochastic optimization problems. Particularly, MDP is suitable for the joint optimization between physical and media-access control (MAC) layers. Based on MDP, reinforcement learning is a practical method to address the optimization without a priori knowledge of system statistics. In this chapter, we shall first introduce some basics on stochastic approximation, which serves as one basis of reinforcement learning, and then demonstrate the MDP formulations of RRM via some case studies, which require the knowledge of system statistics. Finally, some approaches of reinforcement learning (e.g., Q-learning) are introduced to address the practical issue of unknown system statistics.
Because of the time-varying nature of wireless channels, it is difficult to guarantee the deterministic quality of service (QoS) in wireless networks. In this chapter, by combining information theory with the effective capacity (EC) principle, the energy-efficiency optimization problem with statistical QoS guarantee is formulated in the uplink of a two-tier femtocell network. To solve the problem, we introduce a Q-learning mechanism based on Stackelberg game framework. The macro users act as leaders and know the emission power strategy of all femtocell users (FUS).The femtocell user is the follower and only communicates with the macrocell base station (MBS) without communicating with other femtocell base stations (FBSs). In Stackelberg game studying procedure, the macro user chooses the transmit power level first according to the best response of the femtocell, and the micro users interact directly with the environment, i.e., leader's transmit power strategies, and find their best responses. Then, the optimization problem is modeled as a noncooperative game, and the existence of Nash equilibriums (NEs) is studied. Finally, in order to improve the self-organizing ability of femtocell, we adopt Q-learning framework based on noncooperative game, in which all the FBS are regarded as agents to achieve power allocation. Numerical results show that the algorithm cannot only meet the delay requirements of delay-sensitive traffic but also has good convergence.
Vehicular networks have been recently attracting an increasing attention from both the industry and research communities. One of the challenges in this area is the understanding of vehicular mobility and further propose accurate and realistic mobility models to aid the vehicular communication and networks design and evaluation. In this chapter, different from the current works focusing on designing microscopic level models that are describing the individual mobility behaviors, we are exploring the use of open Jackson queuing network frameworks to model the macroscopic level vehicular mobility. The proposed intuitive model can accurately describe the vehicular mobility, and further predict various measures of network-level performance. These measures include the vehicular distribution and vehicular-level performance, such as average sojourn time in each area and the number of sojourned areas in the vehicular networks. Model validation based on two large-scale urban vehicular motion traces reveals that such a simple model can accurately predict a number of system measure concerned with the vehicular network performance. Moreover, we develop two applications to illustrate the proposed model's effectiveness in the analysis of system-level performance and dimensioning of vehicular networks.
