Signal processing techniques for cooperative spectrum sensing in trusted and untrusted networks

Abstract

The rapid growth in the demand for wireless broadband applications in both licensed and unlicensed frequency bands has led to an ever-increasing need for radio spectrum. As confirmed by several measurements about spectrum occupancy, the fixed policy adopted by Governments and regulatory Agencies concerning the assignment of the spectrum results in an under-utilization of its usage below 1GHz. This behaviour severely reduces the number of available (i.e., vacant) frequency bands viable to deploy new communication services or to enhance the existing ones. In addition, the continuous development of new technologies requires a more flexible and efficient management of the spectrum to satisfy the goals of the EU Digital Agenda and the future market demands for mobile and broadband services. A new emerging technology, namely Cognitive Radio (CR), addresses the issue of spectrum scarcity and aims at improving the efficiency of the spectrum. CR-based devices are indeed able to gather information about the surrounding radio environment to dynamically adjust their operational parameters and improve their performance. CR Technology (CRT) promises several benefits such as: the interoperability with new and legacy radio systems; the ability to implement on each radio networking tasks that are transparent to users; the possibility to implement reconfigurable and cost effective architectures for wireless devices. In addition, CRT paves the way for broadband usage in secure communications – which is currently restricted to only a few available technologies – harmonizing the needs of commercial, public, safety and military users. As a matter fact, CRT can also coexist with the current telecommunication technologies and licensed legacy users, namely Primary Users (PU), allowing unlicensed users, namely Secondary Users (SU), to opportunistically access the unused frequency bands (i.e., the spectrum holes or white spaces). SUs can employ both cooperative and non-cooperative techniques to sense the spectrum. As a matter of fact, spectrum sensing is the main task of SUs, given their ability to detect a PU signal in a certain frequency band. However, in a non-cooperative scenario, with each SU performing the spectrum sensing independently, it might be hard to obtain a reliable decision about the spectrum occupancy. For this reason, several approaches suggest sensing the spectrum through the exploitation of a cooperative scheme that combines the local decisions of SUs in order to make a global decision about spectrum occupancy. Cooperative spectrum sensing (CSS) can be classified in three categories: centralized, distributed, and relay assisted. In particular, the centralized CSS is widely considered the conventional solution. In such approach, SUs perform the spectrum sensing independently, sending their decisions to a cognitive base station, namely fusion center (FC), which is responsible for combining them and then reaching the global decision about the spectrum occupancy. This behaviour allows SUs to create cognitive radio networks (CRNs) to communicate without interfering with the primary communications, therefore fulfilling the main constraint of the CRT. Implementing CRNs, however, requires knowledge in different fields of expertise related not only to scientific and technological capabilities for the physical deployment of such networks, but also to marketing and management for the regulation of the coexistence of primary and secondary users. Even though CSS ensures improvements in terms of performance and spectrum utilization, it poses several challenges that are not to be overlooked in the design and implementation of a cooperative environment. Two typical challenges are: the cooperative sensing in presence of correlated SUs’ observations, and the security of CRNs. In particular, when the proximity among SUs results in correlated observations, the performance of the cooperative sensing degrades if the correlated observations are under the conventional threshold of techniques based on the Energy Detection (ED), which leads SUs to interfere with primary communications and to discourage PUs from sharing their licensed spectrum. In addition, the openness of low layers protocol stacks makes CRT vulnerable to different kind of attacks that aim to destroy the typical cognitive operations of legitimate SUs and to allow malicious users to join in the CRN to exploit the available spectrum holes. Then, malicious users can act as the relay nodes of the network accessing the transmitted information, and affect the efficiency of the spectrum, interfering with primary communications. Therefore, the definition of an efficient, secure CRN is a critical challenge that requires not only the definition of techniques to improve the performance of the spectrum sensing (both cooperative and non-cooperative), but also the definition of ad-hoc countermeasures to identify all malicious users, discarding them from the cooperative sensing and the CRN. The proposed doctoral dissertation aims at addressing the problem of spectrum scarcity by proposing novel signal processing techniques for the performance improvement of CSS in both trusted and untrusted networks (i.e. without and with the presence of attackers). In particular, a novel cooperative sensing approach is introduced to improve the detection performance in presence of correlated SUs’ observations and in trusted CRNs. The proposed method exploits two tests at once to recover those correlated observations that are under the conventional threshold of ED-based techniques, and to increase the cooperative performance, in the presence of a communication channel affected by additive white Gaussian noise (AWGN) and different levels of noise uncertainty. Successively, a statistical, fine-grained analysis framework for the identification of the root causes of packet losses to distinguish between packet drop and jamming attacks is defined. In particular, the approach builds a statistical model for determining optimal thresholds and testing variables, and for setting an individual threshold for each link of the network by using the typical packet information provided by both CR-based and traditional wireless sensors networks. Finally, we provided the definition of a new centralized reputation-based CSS in untrusted CRNs affected by Byzantine attackers. In particular, the proposed CSS scheme is based on two features (i.e. a new reputation method, and three dynamic lists) that state the reliability of CR users in a CRN and allow the FC to properly identify Byzantine attackers, without penalizing legitimate users that misbehave due to channel noise. Even though our method is completely blind (i.e. no need for any a priori information), it can also involve trusted nodes to enhance the system’s performance.