(ACROPOLIS) Document Number D8.1. Context Representation for Cognitive Radios and Networks

Transcription

1 Advanced coexistence technologies for radio optimisation in licensed and unlicensed spectrum (ACROPOLIS) Document Number D8.1 Context Representation for Cognitive Radios and Networks Contractual date of delivery to the CEC: 30/09/2011 Actual date of delivery to the CEC: 30/09/2011 Project Number and Acronym: Editor: Authors: Participants: Workpackage: Security: Nature: ACROPOLIS Ragnar Thobaben (KTH) Nicola Baldo (CTTC), Ricardo Blasco- Serrano (KTH), Stefano Boldrini (Uniroma1), Hanna Bogucka (PUT), Sergio Bovelli (EADS), Luca De Nardis (Uniroma1), Maria- Gabriella Di Benedetto (Uniroma1), Eduard Jorswieck (TUD), Adrian Kliks (PUT), Jing Lv (TUD), Jad Nasreddine (RWTH), Marcin Parzy (PUT), Janne Riihijärvi (RWTH), Zhongwei Si (KTH), Dimitri Tassetto (EADS), Ragnar Thobaben (KTH) CTTC, EADS, KTH, PUT, RWTH, TUD, Uniroma1 WP8 Version: 1.0 Total Number of Pages: 61 Public (PU) Report ICT-ACROPOLIS Deliverable D8.1 1/61

2 Executive Summary: This report analyses the context based on which cognitive radios and networks make decisions. Features that define the context are identified and their representation and relevance for the decision- making process are discussed. Typical examples and cognitive radio scenarios are considered. Cognitive radios and networks are expected to act and interact autonomously in their surrounding environment. Obviously, not all aspects of the overall environment are relevant for a cognitive radio to make decisions. But all aspects that are relevant for decision- making define the context. A universal characterization of the context of cognitive radio systems is hard to find; however, in general, it can be assumed that the context of cognitive radios and networks is specified by characteristics of surrounding wireless systems, characteristics of the radio environment, characteristics of the network environment, constraints that are specific for different spectrum- sharing paradigms, demands of specific applications, as well as own capabilities and own limitations. On an abstract level, the context is represented by a set of features, constraints, and parameters, which carry the information which is required for decision making. In practice, there are mainly five representations of context: policies, specifications, and standardization; information obtained from detection, sensing, and estimation; information obtained through direct communication; information obtained from wiretapping, overhearing, and observations; databases and ontologies. Another important representation of a particular aspect of context, self- awareness, is through performance and reliability measures for employed algorithms and components. In real systems, one can expect that all representations of context will be used at the same time to complement each other. Static context components are provided through policies, specifications, and standards. Slowly changing context components can be conveniently provided through databases, and rapidly changing features can be distributed through direct communications or obtained by wiretapping, detection, sensing, and estimation if direct communication is infeasible. Our discussion of context and its representation starts with context components that are characteristic for wireless communication systems, the radio environment, and communication networks. Fundamental parameters of the primary or other coexisting systems like channel access, signalling, and transmission strategies are typically available to cognitive systems through specifications and standardization. In special cases, context components may be as well explicitly communicated to enable and support cognitive transmissions. We discuss the importance of propagation models, which if combined with location information, enable efficient interference control. To illustrate the potential of this concept, benefits of radio environment maps are summarized. Location information provided either through models, databases, or estimation can be further utilized to improve the performance of routing. In this context, if nodes in the network are mobile, further improvements in performance can be obtained if mobility is taken into account for ICT-ACROPOLIS Deliverable D8.1 2/61

3 example by incorporating mobility models. A case study is finally presented which demonstrates simple mechanisms for network identification. Knowing which types of networks are present in the same frequency band can be useful to adjust communication protocols and to avoid interference. For the three dominant spectrum sharing paradigms (interweave, underlay, and overlay), we discuss characteristic context components that are required in order to identify and utilize transmission opportunities. For the interweave spectrum sharing paradigm, we focus on features that are required for adapting sensing algorithms and for judging their reliability, and we summarize two database approaches to interweave cognitive radio. For underlay transmissions we discuss context components which are required to control the interference caused by secondary transmitters if multi- antenna techniques or ultra wideband approaches are considered. For multi- antenna transmissions, the channel from the secondary transmitter to the primary receiver is the most crucial context component. It can be provided for example through channel estimation using a dedicated common pilot channel. For UWB transmission on the other hand, spectral masks are the most relevant context component. They can be provided by databases, through specifications, or explicit communication. Context components that are required to set up cooperative strategies for overlay transmissions are again channel state information for cross links but as well the transmission strategy and parameters of the primary and capabilities of the primary receiver. They can be obtained through specifications or, since coordination between primary and secondary is required in this case, through explicit communication over dedicated cognitive control channels. The last part of this report discusses demands of application, which specify another important context component. They are commonly represented through quality of service, quality of experience, and other quality constraints and requirements. Keywords: Context, context representation, context components, databases, network identification, sensing, spectrum sharing. ICT-ACROPOLIS Deliverable D8.1 3/61

4 Document Revision History Version Date Author Summary of main changes KTH Initial structure of the document KTH Contributions: KTH, RWTH, TUD, EADS, Uniroma1, CTTC KTH Rearranged, additional contributions from PUT and Uniroma KTH Draft for internal review in WP8; only Abstract is missing KTH Revised version, submitted for internal review to the ExeCom KTH Revised version, ready for submission. ICT-ACROPOLIS Deliverable D8.1 4/61

5 Table of Contents 1. Introduction Context and Context Representation Context Context Representation Radio and Network Environment Characteristics of Wireless Communication Systems Channel Access Strategies Signalling and Transmission Strategies Radio Environment Characteristics of the Wireless Channel Propagation Models Radio Environment Maps Summary of Section Characteristics of Wireless Networks Geographical Network Layout Network Topology Impact of Mobility and Mobility Modelling Network Awareness and Identification Ontologies and Policy Representation Summary of Section Context and Context Representation for Specific Spectrum Sharing Paradigms Interweave Cognitive Radio Systems Transmission Opportunities Spectrum Sensing Approaches using databases Underlay Cognitive Radio Systems Transmission Opportunities Multiple-Antenna Techniques for Underlay Transmission Ultra Wide Band Techniques for Underlay Transmission Overlay Cognitive Radio Systems Transmission Opportunities Transmitter Cooperation for Spectrum Overlay Cognitive Primary Systems and Receiver Cooperation for Spectrum Overlay Application Requirements Applications Control applications Data Applications Cognitive Learning Applications Quality of Experience (QoE) Quality of Service (QoS) Quality of Context (QoC) Open Issues Summary and Conclusion Appendix - Spectrum Sharing Paradigms and Concepts Interweave Cognitive Radio Systems Underlay Cognitive Radio Systems Overlay Cognitive Radio Systems References ICT-ACROPOLIS Deliverable D8.1 5/61

6 1. Introduction This deliverable analyses the context based on which cognitive radios and networks make decisions. Features that define the context are identified. We discuss how they can be made available to cognitive radios and networks, and we will explain their relevance for the decision making process. The key problem for cognitive radios and networks to solve is the recognition of the context they are being used in in terms of both user requirements and their surroundings and to adapt their operation accordingly. The term context in relation to cognitive radios should be understood in a very general sense, including both the context of the user and applications, as well as the radio environment. The user and application context determines the objectives for the communications, whereas the radio environmental context provides valuable inputs and boundary conditions on the learning and optimization process a cognitive radio or network has to carry out in order to reach those objectives. In this deliverable, we start with a general discussion of context and context representation in Section 2. The context of cognitive radio systems and networks is defined, and the most important context representations are discussed. A discussion of fundamental features of the radio and network environment and ways to represent them is provided in Section 3. Fundamental parameters of primary secondary systems like channel access, signalling, and transmission strategies are considered. Typical representations are specifications and standardization, and in special cases, context components may be as well explicitly communicated. The importance of propagation models is explained, which if combined with location information, enable efficient interference control. To illustrate the potential of this concept, benefits of radio environment maps are summarized. Location information provided either through models, databases, or estimation and mobility models can be further utilized to improve the performance of cognitive systems. A case study is finally presented which demonstrates simple mechanisms for network identification. In Section 4, specific context components and their representation that are required in order to identify and utilize transmission opportunities under the three dominant spectrum sharing paradigms (interweave, underlay, and overlay) are discussed. For the interweave spectrum sharing paradigm, we focus on features that are required for adapting sensing algorithms and for judging their reliability. We summarize as well two database approaches to interweave cognitive radio. For underlay transmissions we focus our discussion on context components that enable interference control at the secondary transmitters considering that multi- antenna techniques or ultra wideband approaches are used. For multi- antenna transmissions, the most important context component is the channel state information, and for UWB transmission, spectral masks are typically used to define interference constraints. Context components that are required to set up cooperative strategies for overlay transmissions are again channel state information for cross links but as well the ICT-ACROPOLIS Deliverable D8.1 6/61

7 transmission strategy and parameters of the primary and capabilities of the primary receiver. Section 5 discusses demands of application, which specify another important context component. They are commonly represented through quality of service, quality of experience, and other quality constraints and requirements. Open issues are discussed in Section 6, and our discussions are summarized in Section 7. ICT-ACROPOLIS Deliverable D8.1 7/61

8 2. Context and Context Representation 2.1 Context In an idealized setup, cognitive radios and networks are expected to act autonomously in their surrounding environment. Depending on their degree of sophistication, they form networks, identify transmission opportunities, make decisions, act, and learn from past events. In all decisions and actions that cognitive radios and networks make, they take into account their interaction with the environment. Obviously, not all aspects of the overall environment are relevant for a cognitive radio to make decisions. But all aspects that are relevant for decision- making define the context. Since the context depends on the chosen cognitive radio and network architecture, it is difficult to give a universal characterization of the context of cognitive radio systems. However, in general, it can be assumed that the context of cognitive radios and networks is defined by Characteristics of surrounding wireless systems, Characteristics of the radio environment, Characteristics of the network environment, Constraints that are specific for different spectrum- sharing paradigms, Demands of specific applications, as well as by Own capabilities and own limitations. Especially the last point is worth mentioning since it indicates that context is heavily defined by the employed algorithms, methods, and solutions. Clearly, they define which actions are feasible in different situations, which parameter sets need to be known, but as well how likely it is that a chosen action satisfies or violates the given constraints. 2.2 Context Representation The relation between environment, context, context representation, decision- making and actuation is summarized in Figure 1. On an abstract level, if the underlying cognitive radio architecture and its implementation is not or only partly specified, the context is represented by a set of features, constraints, and parameters, which characterize the context and which carry the information which is required for decision making and for operation. Sets of features and constraints, as an abstract form of context representation, are mainly used in research and development if only components, algorithms, etc. of cognitive radio systems are considered. For example, communication strategies or sensing algorithms are often developed assuming that certain model parameters are available without addressing how they can be made available. ICT-ACROPOLIS Deliverable D8.1 8/61

9 Context Representation Features, Constraints, and Parameters Context Policies, Specifications, Standardization Direct Communication Data- bases Detection, Sensing, Estimation Wiretapping, Overhearing, Observations Decision- Making Environment Action Figure 1: Relation between environment, context, context representation, decisionmaking, and action. The problem how features, parameters, and constraints are made available to cognitive radio systems becomes important if complete system architectures are considered and cognitive radio concepts are implemented. As already indicated by Figure 1, there are mainly five different representations considered in on- going discussions of practical systems: Policies, specifications, and standardization Detection, sensing, and estimation Direct communication Wiretapping, overhearing, and observation Databases and ontologies Another important representation of a particular aspect of context, which is however often neglected, is through ICT-ACROPOLIS Deliverable D8.1 9/61

10 Performance and reliability measures for employed algorithms and components (self- awareness). In real systems, one can expect that all five representations of context will be used at the same time to complement each other. One possible assignment of features to practical representations is discussed in the following: 1. Static features and features that change only slowly over time (for example within months or years), are typically captured by policies, specifications, and standardizations. They can provide a cognitive radio system for example with a set of permitted actions, required actions, models, assumptions, and constraints. These features constitute typically the foundation of the cognitive radio system and network. 2. Features and constraints that change on a time scale of minutes or hours and features that are specific for some geographic regions can conveniently be made available through databases given that they are widely accessible. These databases can carry for example time variant model parameters for specific geographic locations, local constraints, or parameter sets for algorithms. 3. Rapidly changing properties of the context are typically made available through detection, sensing, and estimation, through direct communication, or can be overheard (observed) in some cases. Here, detection and sensing are applied if parameters cannot be obtained in other ways or if direct interaction between different systems is not permitted. A typical example in cognitive radio is spectrum sensing or network identification. If direct communication is feasible, for example through common control channels, it is a convenient way to share important context parameters. In this way, one system can inform the other about actions which it plans to take or to provide and share important model parameters such as for example signal attenuation and path loss. It is worth mentioning that combined approaches are as well considered in the current literature. Learning mechanisms can for example be used to maintain databases based on inputs from detection and estimation and other sources like direct communication. 4. Whenever models are used to describe context and whenever detection, estimation or learning techniques are used to infer context parameters and features, uncertainty is induced into the system, which limits the performance and the system s capabilities. For example, if the estimation error of the signal- to- noise ratio in energy detection is larger than the energy of the signal that we wish to detect, energy detection breaks down. In a similar way, communication schemes break down or show a degraded performance if they are adapted based on inaccurate channel parameters. Knowing these limitations (self- awareness) is important for decision- making. It can be made available to the decision- making engine through reliability information in terms of probabilities, likelihoods, etc. (for example detection and false- alarm probabilities, error rates, etc.). ICT-ACROPOLIS Deliverable D8.1 10/61

11 3. Radio and Network Environment The context of cognitive radios and networks is characterized by a large number of features, which describe for example the radio and network environment. The relevance of different features relies clearly on the considered scenario. In this section we identify fundamental context components, describe how they are typically made available to cognitive radios, and discuss their importance for different scenarios. Here, we focus on features of wireless communication systems, the radio environment, and wireless networks and protocols. 3.1 Characteristics of Wireless Communication Systems Characteristics of wireless systems (see e.g. [3], [4], [5]) define one important set of features that describe the context of cognitive radios. For example, in order to be able to decide for a particular spectrum sharing and channel access strategy, a cognitive radio system needs to be aware of the channel access strategies which are used by the licensed primary user or other cognitive radios. Features of transmitted signals can be furthermore utilized for detection and classification or are required by cooperative strategies. In the following, we list related context components and explain their relevance and how they are typically represented. For a detailed description and fundamental definition of wireless systems and networks, we refer to [3], [4], [5] Channel Access Strategies The channel access technique of the licensed primary network specifies and characterizes the properties of transmission opportunities (e.g., spectral holes). The traditional approach to provide multiple users in (coordinated) wireless systems access to the channel is to orthogonalize the resources of the channel (time, frequency, code, and space) and to assign them to the users. Relevant context components are the following. Frequency bands in which primary users are active or in which cognitive communication is permitted. Representation (See Section 2.2) Relevance Regulation; Databases. Knowledge of relevant frequency bands is required to adapt communication and sensing protocols. This knowledge can for example be used to reduce the complexity of spectrum sensing. Frequency bands determine furthermore properties of transmission opportunities (e.g., propagation models, etc.). Example TV bands which are opened for cognitive communications. Time- frequency resolution of primary and coexisting communication systems Representation (See Section 2.2) Relevance Standardization, specifications. Adaptation of communication and sensing protocols (as above). ICT-ACROPOLIS Deliverable D8.1 11/61

12 In addition, the time- frequency resolution and resource allocation of the primary system define further properties of transmission opportunities (e.g., duration in time and bandwidth). Especially the time resolution constrains the timeframe for a cognitive cycle (perceive, decide, act). Example Standardization of wireless communication systems like Wifi, UMTS, LTE, WiMAX, etc. Medium access: scheduled access and random access strategy Representation (See Section 2.2) Standardization, specifications; Direct communication. Relevance Adaptation of communication and sensing protocols. Scheduled access: In this case, channel access is granted by a control unit based on a specific resource alignment protocol, and control information is typically communicated over control channels. If control channel information is not encrypted, it can be overheard (observed) by cognitive radios, and context parameters can be extracted. In this case, the primary system shares information (direct communication) in order to enable cognitive transmissions without significantly changing its own architecture. Random access: The performance obtained by Carrier Sensing Multiple Access (CSMA) is heavily affected by two phenomena, the well known hidden terminal and exposed terminal problems. In order to solve the hidden and exposed terminal problems, practical implementations of MAC protocols combine carrier sensing with a handshake between transmitter and receiver; these protocols are commonly referred to as CSMA with Collision Avoidance (CSMA- CA). Knowledge on handshake mechanisms, etc. can be utilized for sensing and network detection. Example The Distributed Foundation Wireless MAC (DFWMAC) is an example of CSMA- CA, adopted for the MAC layer of the IEEE standard [6] adopts a Clear Channel Assessment (CCA) function which performs channel sensing in two different ways, either by measuring the received power and comparing it with a threshold, or by performing a true carrier sensing by detecting another signal on the same channel. Features of code division multiple access (CDMA) Representation (See Section 2.2) Standardization, specifications; Databases; Direct communication. ICT-ACROPOLIS Deliverable D8.1 12/61

13 Relevance Adaptation of communication and sensing protocols; Interference control. If a CDMA based primary system is not fully loaded, it can be expected to be able to cope with the interference from cognitive transmitters assuming that interference constraints are respected by the cognitive radio. If spreading codes used by the primary are furthermore made public (for example through a common/public control channels or databases), the cognitive radio can utilize this knowledge for signal detection and spectrum sensing or consider this knowledge in the design of its transmit waveforms in order to minimize the interference to on- going primary transmissions. The noise- like character of CDMA signals with their low and flat power spectral densities makes them generally attractive for underlay transmissions. For a more detailed discussion we refer to Section Features of spatial multiplexing and multiple- antenna techniques Representation (See Section 2.2) Relevance Standardization, specifications; Direct communication. Adaptation of communication and sensing protocols; Interference control. Multiple antennas at the transmitter and/or the receiver can be used to shape the overall antenna beam (transmit beam and receive beam) in a certain way, for example, to maximize the overall antenna gain in the direction of the target receiver/transmitter or to suppress specific dominant interfering signals. That is, having knowledge that the primary system uses directed antennas (for example through standardization, by overhearing control channels, or through direct communication), may in some cases allow the cognitive radio to relax interference constraints. Furthermore, as the discussion in Section will show, multiple- antenna techniques can be used for shaping cognitive transmissions and avoiding undesired interference to the primary system. The required channel state information can be obtained for example through common pilot channels. Example Direct communication: common pilot channel, brought up in the standardization of reconfigurable radio system of ETSI [70] Signalling and Transmission Strategies As mentioned above, the features of the transmitted signals are important for cognitive radio systems since they can be utilized for spectrum sensing. More sophisticated cognitive ICT-ACROPOLIS Deliverable D8.1 13/61

14 radio concepts require even decoding of primary messages. This requires that all features like coding parameters, signal constellation, and transmission pulses are known by the secondary user. A summary of important context components and potential ways of making them available to cognitive radios are provided in the following: Transmit waveforms and single- /multicarrier modulation Representation (See Section 2.2) Specifications and standardization; Detection, sensing, and estimation; Direct communication; Databases and ontologies. Relevance Adaptation of communication and sensing protocols; Network identification and classification. It can be expected that the set of possible transmission parameters of the primary or other coexisting systems is defined through standards and specifications. This information may be refined through direct communications (a system may indicate which parameters it uses), as a result of spectrum sensing and network identification, or through databases, which keep track of learned parameters or record their statistics. In general, knowledge of transmission parameters can be used to improve the performance of sensing and detection algorithms. Knowledge of transmit waveforms and transmit parameters is furthermore essential in overlay cognitive radio architectures when the cognitive radio serves as a cognitive relay. In this case, it can be expected that the cognitive radio is instructed through direct communication. Signal- space constellations, channel coding, rate allocation Representation (See Section 2.2) Specifications and standardization; Detection, sensing, and estimation; Direct communication; Databases and ontologies. Relevance Adaptation of communication and sensing protocols; Network identification and classification; Interference control. As explained above, features of transmitted signals can be utilized for sensing, detection, and learning, and knowing them is essential for cooperative cognitive radio strategies. In addition, knowledge on rate allocation and the employed coding techniques provides the cognitive radio systems with information on the primary system s robustness. This knowledge can be utilized for interference control. ICT-ACROPOLIS Deliverable D8.1 14/61

15 3.2 Radio Environment The context of cognitive radios and networks is heavily defined by the radio environment, which is mainly characterized by two aspects, the properties of the wireless medium and the geographical network layout. While the relevance of the geographical network layout is discussed in Section 3.3.1, we list in the following important characteristics of the wireless channel and explain how they can be made available to cognitive radios through propagation models and radio environment maps Characteristics of the Wireless Channel The most important context components that describe characteristics of the propagation environment are the following: Path loss: it describes the distance dependent attenuation of the signal power; Shadowing: it describes the attenuation of the signal power due to obstacles; Fading: it describes the signal attenuation due to the superposition of multi- path components; Delay spread and frequency selective fading; Coherence time and Doppler spread; Transmit powers and spatial power density distribution; Environmental features, like terrain, buildings, etc. These features are typically described by e.g., propagation models, radio environment maps, power density distributions, or models based on terrain data. These models typically include and are based on Frequency dependent path- loss models for different propagation environments (e.g., urban, sub- urban, rural); Statistical models describing shadowing and fading distributions; Power- delay profiles describing multi- path propagation; Statistical models based on spatial power density distribution. A detailed discussion is given in the following Propagation Models The success of wireless communication networks has lead to an increase in the complexity of these networks and made them more sensitive to the uncertainty of the used information. In particular the precision of the models representing the propagation losses and radio environment is becoming a critical issue for radio resource management techniques and planning tools. For instance when determining the allowed power with which the secondary user can transmit using a licensed channel to a primary user, the path loss model with all its components (i.e., distance- dependent, shadowing, and fast fading component) should be modelled with high precision [7], [8]. In the worst case, uncertainty models for these models should be developed [7]. Furthermore, the precision of these models is very important in femtocell scenarios due to the aggressive interference patterns, in addition to the fast changing and unpredictable indoor environment. A reliable ICT-ACROPOLIS Deliverable D8.1 15/61

16 propagation model will create a trust liaison between primary and secondary networks, which will motivate the former to share its spectrum with the latter. In all systems using Dynamic Spectrum Access (DSA) the propagation loss at least between the primary receiver and secondary transmitter in addition to the one between primary transceivers should be modelled with precision in order to estimate the impact of secondary activity on primary signal. In addition, when sensing is used as a detection mechanism, the propagation between the primary and secondary transmitters should be modelled with precision also. In [7] the authors stress the importance of the propagation models for DSA techniques in order to motivate primary networks to share their spectrum by establishing a trust liaison between primary and secondary networks. To do so, they propose to use a quantile model for the fading and shadowing distributions; the fading and shadowing are modelled as a class of distributions instead of only one distribution. In [8] the authors show the impact of correlation between the different links on the performance of DSA methods. It was shown in the paper that the allowed transmit power of a specific secondary user has a range of about 50 db, only by changing the correlation factors. In addition the authors showed that the probability that primary users' QoS is degraded significantly can reach 0.16 in some scenarios if the secondary assumes that there is no shadowing between the different links while the shadowing factors range is between - 1 and +1. Such a probability is unacceptably high for primary networks. This result shows the importance of defining reliable models of shadow correlations for DSA application Radio Environment Maps FARAMIR project [9] has been developing architecture and the corresponding tools to build Radio Environment Maps (REMs) in a dynamic way. REMs include, among others, information about the radio environment, power density distributions, and propagation models. These models and information can be used by any type of RRM techniques to enhance their performance. Recently, the authors in [10] showed that most of the statistical propagation models give good results in terms of protecting primary TV channels. However it was also shown that complex and computationally- intensive models based on terrain data, such as Longley- Rice model with terrain, lead to a low percentage of lost spectrum opportunities compared to other propagation models that do not enable efficient detection of the spectrum holes. In addition, the authors in [11] show quantitatively that basic path loss models perform poorly at predicting path loss in even relatively simple outdoor environments. As a solution, the authors advocate a renewed focus on measurement- based, adaptive path loss models built on appropriate statistical methods. This problem becomes even more difficult in indoor environments where the understanding of propagation and radio characteristics is less than in outdoor due to the hostile environment. ICT-ACROPOLIS Deliverable D8.1 16/61

17 3.2.4 Summary of Section 3.2 The discussions from Section and are summarized in the following table: Radio environment Representation (See Section 2.2) Relevance Specifications and standardization (propagation models); Databases (radio environment maps); Reliability information. Adaptation of communication and sensing protocols; Interference control; The discussion above has provided examples on how propagation models and databases in form of radio environment maps improve the reliability of cognitive radio systems. It has been emphasized that the reliability and robustness of the models have large impact. Examples Propagation models have been provided by various research projects on European level. Especially, the models developed in the WINNER projects [12] have attracted significant attention. 3.3 Characteristics of Wireless Networks The discussion of the relevance of propagation models and radio environment maps above illustrates the importance of other network related features like the geographical network layout and the network topology. The geographical network layout is typically represented in form of databases, and its relevance for the decision process is discussed in the following. After this, we will discuss the relevance of topology, mobility and mobility modelling, network identification, and ontologies Geographical Network Layout Having knowledge about the locations of primary transmitters provides important a- priori information to the system assuming that a propagation model is available. Such a model can be further extended by taking into account the typical behaviour of the network (e.g., described by traffic patterns). This information allows the system to improve the reliability of its decisions. It should be noted that the location of base stations can be normally obtained by the regulator as in the case of the TV towers in the USA [13], [14] or cellular base stations as in USA and some European countries [15]. The interest in using location information in DSA was initiated by the FCC which mandated the use of database for TV white spaces, in addition to some research works as in [16], [17]. The use of location information in TV white space is motivated by the fact that TV towers positions are fixed and can be identified easily. In addition the number of these towers is relatively low. If the positions of the active towers in a given channel are stored in a database, the protection zone for each channel can be determined and stored in the ICT-ACROPOLIS Deliverable D8.1 17/61

18 database. Then the secondary users can determine in which channel they can transmit if they know their own positions. In [10], the authors propose a database driven white space network where the database is used by the secondary users to determine the white spaces. In this paper the authors also study the importance of knowing the location of the secondary with precision and they concluded that the position of the secondary should be known within a few hundred meters. In addition the authors show that knowing the class of TV transmission increase the performance of secondary networks since the FCC has put different protection contours for different classes of digital or analog TV transmitters. In [17] the authors use the information about the location of primary base stations to determine the allowed transmit powers. In this paper the primary user can be any wireless network including cellular network with uplink and downlink transmission. In [18] the authors show how the location and power information can be used in order to develop DSA system based on a dynamic threshold in the context of primary network with multiple transmitters. The threshold is determined by assuming that the positions of both secondary transmitter and primary base stations are known in addition to the transmit power and the coverage area of the primary network. The authors show that by knowing this information the secondary can transmit with power comparable to IEEE equipment in a relatively large portion of the surface covered by the primary network. The authors also show that the power as well as the region where transmission is allowed increases significantly if information about long- term primary frequency allocation (e.g., in the case of frequency reuse higher than one) is known by the secondary. Developing reliable models for topology and node positions is not only useful in operational DSA networks but also to evaluate the performance of these networks by researchers. In [19] the authors study the influence of the structure and dynamics of the primary network on the quality of the secondary transmission. In contrary to most of the existing works where a uniform distribution of the positions is used, the authors of this paper use a combination of actual node location data sets of primary base stations and carefully fitted and validated statistical node location models for secondary transmitters in order to obtain reliable results and enable greater flexibility and realism, while still allowing general conclusions to be made. For the secondary transmitters the authors use a stochastic node location model developed in [20]. This model specifies a probability density function for a collection of node locations in a region of interest, measuring how much more or less likely a given set of locations is to occur compared to the uniformly random case of unit density. In [20] it was established that the so- called Geyer saturation process [21] yields particularly versatile class of models, yielding good fits to both clustered network deployments (such as user- deployed femtocell networks), as well as regular, planned networks. Depending on the parameter used, the Geyer density function can be reduced to the uniformly random case or becomes a hardcore one with no node pair being closer than distance r apart, for instance. In addition, the paper highlights the importance of determining the coverage area of primary network and adopts a protective approach towards the primary users. This will motivate primary users to share their spectrum. The authors first approximate the coverage area with Voronoi polygons to have a pervasive coverage. However, since with typical radio technologies it is not possible to limit the actual coverage of the transmitter to the irregular shape of the Voronoi polygon, the protected zone of a disc centered at the base station and ICT-ACROPOLIS Deliverable D8.1 18/61

19 completely enclosing its service area represented by the Voronoi polygon. Moreover, the authors study the impact of primary traffic pattern on secondary transmission. The results show that even in dense primary networks significant opportunities for secondary use can arise. For high activity patterns (i.e., higher than 30 50% of the time) there are almost no useful spectrum opportunities even if the primary network is rather sparsely deployed. The results also show that the spatial structure of the deployment of the secondary network has significant influence on the capacity that can be achieved by using the arising spectrum opportunities. This latter result can help the operators that are willing to deploy secondary networks to deploy their systems Network Topology Above we have discussed the role of the geographical layout of the nodes of the wireless communications system on spectrum use and interference. These issues also have, of course, implications on the topology of the network in the sense of connectivity as measured on a particular layer. The simplest mathematical framework that has been used in the literature for describing this dependency between node locations and network topology is that of unit disk graphs or random geometric graphs [22], in which network topology is formed as graph connecting nodes that are closer than a prescribed threshold distance apart. The random geometric graph formalism has been used to study questions such as what should the threshold distance be for the connectivity and for a given node location distribution to ensure certain level of connectivity within the network. While these results do shed some light on problems related to network planning and dimensioning, they rarely enable accurate non- asymptotic results to be derived. The main problems are usually the model of fixed connectivity distance, and limiting the relationships between the nodes to one type only, i.e., connectivity. Both these limitations can be removed by considering more general classes of models. Starting from models based on the use of signal- to- interference- plus- noise ratio (SINR) levels computed using models for activity patterns of nodes enables reasoning of both connectivity as well as capacity. These results can be again discretized to yield limited number of interactions between different nodes. For example, in [23] two types of edges were used to model interactions in CSMA networks: one type for connectivity as above, and a second type for interactions triggering the carrier sense mechanism provided that either of the involved nodes is active. Such a combination of SINR based network models and subsequent discretization in order to obtain labelled graph models provides a flexible framework for modelling and reasoning about the network topology as means of network awareness for wireless systems. The discussion has focused so far on topological information, which can be indeed very useful in several key network functions. For example, typical topology- related information, such as the average node connectivity (that is the average number of edges that a node shows in the graph representing the network) can be used in selecting nodes for specific network functions (e.g. clusterheads or gateways between network clusters). On the other hand, information related to the reciprocal physical position of network terminals could further improve network performance if properly used. ICT-ACROPOLIS Deliverable D8.1 19/61

20 Routing is a network task that can take advantage of the availability of position information. Several position- based routing protocols have been proposed in the past, using position information either to reduce overhead during route search procedures ([24], [25], [26]) or to select the next hop in data packets forwarding ([27], [28]). Performance evaluation of such protocols traditionally assumed that position information was obtained by means of GPS receivers, thus not introducing any overhead on the radio network. When this assumption is no longer valid, as for example in indoor networks, retrieving position information calls for the introduction of dedicated algorithms and protocols taking advantage of cooperation between network terminals, causing thus an additional overhead that must be taken into account. Such overhead is related to the required exchange of distance information between terminals (which are not aware of their own spatial position and only have measurements of distances to their neighbours) and to the introduction of more complex algorithms in order to build a reliable network topology from distance information. A key issue in the adoption of this class of protocols is the availability of accurate measurements of distance between terminals. Such measurements can be obtained by the terminals in different ways [29]: RSSI (Received Signal Strength Information): with this technique the distance is measured based on the attenuation introduced by the propagation of the signal from transmitter to receiver; a precise propagation model and accurate calibrating procedures are required to obtain reliable measurements [30]; nevertheless terminal mobility and unpredictable variations in the channel behaviour can occasionally lead to large errors in distance evaluation. AOA (Angle of Arrival): in this case the distances between terminals are reconstructed from the angle between them; this technique requires however the adoption of expensive arrays of directional antennas, and the results are highly sensitive to multi- path, NLOS effect and array precision. TOA (Time of Arrival): in this technique the distance is reconstructed from the propagation delay between transmitter and receiver; in a distributed scenario this requires a handshake protocol between the two terminals in order to synchronize them with the requested precision. TDOA (Time- Difference of Arrival): the main difference with respect to TOA techniques is that the difference between times of arrival in several receivers is used to reconstruct the transmitter s position. This requires a highly precise synchronization between the receivers, but not precise synchronization between transmitter and receivers. Solutions based on range estimation (RSSI, TOA, or TDOA) are by far the most common ones. In the specific case of RSSI/TOA solutions, once that the range measurements between n reference terminals and each terminal i with unknown position are available, the position map is obtained by solving the triangulation problem, described by the following set of equations for each terminal: ICT-ACROPOLIS Deliverable D8.1 20/61

21 "( X 1! X Ti ) 2 + Y 1! Y $ Ti $ ( X 2! X Ti ) 2 + Y 2! Y Ti # $! $ ( X n! X Ti ) 2 % $ + Y n! Y Ti ( ) 2 + ( Z 1! Z Ti ) 2 ( ) 2 + ( Z 2! Z Ti ) 2 ( ) 2 + ( Z n! Z Ti ) 2 & $ " $ $ ' = # $ $ $ % $ ( $ To obtain a tri- dimensional position at least n 4 reference points are needed; more reference points can be used for redundancy in order to reduce the effect of errors in range measurements. When positioning is obtained by means of terminal cooperation, typically in a distributed fashion ([31], [32], [33]), the evaluation of the impact of the introduction of position information on routing performance becomes a quite challenging task. The issue was addressed for example in the case of underlay networks based on UWB technology in [34]. The UWB technology is indeed particularly suitable for the introduction of position information based on terminal cooperation, as on the one hand it provides the means for obtaining high accuracy position information thanks to its high ranging accuracy, while on the other hand it can heavily benefit from position- based optimizations due to the coexistence requirements under which it operates. Position information can for example be used in conjunction with primary systems databases and Radio Environment Maps to allow the operation of UWB systems while minimizing harmful interference towards and from other systems. The problem of routing in a cognitive UWB network was further investigated in [35]. R 1Ti R 2Ti! R nti & $ ' $ ( $ Impact of Mobility and Mobility Modelling Cognitive radios are conceived to operate across different spectrum bands and heterogeneous radio access technologies, driven by the exciting vision of an always best connected society that will enable innovative ways of social interaction and personalized services and applications. In this context, the absence of, or even the introduction of significant limitations on, user mobility is unsustainable. Node mobility significantly affects the performance of wireless networks and it is a common cause of route outages, packet loss, and delay. The state- of- the- art research on cognitive radio networks has focused on the time varying character of the available spectrum resources, leaving specific terminal mobility issues mostly unexplored. Most spectrum sensing [36], [37] and dynamic spectrum access [38] approaches, so far, address cognitive radio networks with static topology. The introduction of node mobility imposes new challenges, making it necessary to review current network and protocol design approaches for, among others, spectrum sensing, spectrum sharing, interference management and routing [39]. A reliable model for user mobility is, in this context, a key point in the design and performance evaluation of algorithms and protocols for cognitive wireless networks of mobile nodes. ICT-ACROPOLIS Deliverable D8.1 21/61

22 Modelling mobility for wireless mobile networks has proved to be difficult to achieve and as such a plethora of models have been proposed in the past that address a wide range of possible network scenarios (see [40] for a more extensive description). Several criteria for classifying mobility models have been proposed in the literature ([40], [41]). The most common criterion for classification is based on the distinction between entity (or standard) models in which each node in the network follows its own mobility pattern, and group models in which node movement patterns are not mutually independent. The rigid distinction between entity and group mobility models is due to the fact that the most popular group mobility models (see the Reference Group Mobility Model [42]) do not track the actual position of each node in a group. Although such models are suitable for scenarios where a rough description of user mobility is sufficient (e.g. in modeling mobility of cellular subscribers over movement spans of hundreds or thousands of meters), this is no longer the case for application scenarios proposed by new, short range wireless technologies where small scale movements can dramatically change network topology [43]. This is the case, in particular, for application scenarios typical of emergency, rescue, and security operations, where groups of people move in a coordinated manner while keeping a significant degree of independence in their individual mobility pattern; such scenarios are indeed some of the most promising for the deployment of underlay cognitive Network Awareness and Identification The capability of automatically detecting, identifying and classifying wireless systems coexisting in the same band without requiring detailed information on the transmitted signals can be extremely useful in improving the performance of a cognitive wireless network. Based on this idea, a method for automatic recognition and classification of wireless technologies is currently under investigation. The considered frequency band is the Industrial Scientific and Medical (ISM) 2.4 GHz band. Many different and widespread networks operate in this band, as it is open for use without any license: these characteristics make this band particularly appealing for coexistence studies. Well- known examples of technologies operating in this band are: Bluetooth (IEEE ); Wi- Fi (IEEE ); ZigBee (IEEE ). ISM 2.4 GHz band is also utilized by many wireless mice and keyboards, cordless Wi- Fi phones and also by cameras for security closed- circuit TVs. Moreover, common interference at 2.4 GHz band comes from microwave ovens and DECT cordless phones (operating at 1.9 GHz); these can compromise the quality of the radio link of the other technologies, and should be also taken into account by the cognitive radio recognition system. The proposed approach consists of exploiting features of the MAC sub- layer of the different wireless technologies. The idea that resides under this approach is that every network has its own particular and peculiar MAC behaviour, as expressed in the standard that defines each technology. Based on the study of each standard, a MAC peculiar behaviour can be identified for each type of network. Furthermore, some features that reflect these MAC behaviours can be found, and through these features, a recognition and classification process can be carried out. ICT-ACROPOLIS Deliverable D8.1 22/61

23 In particular, a time- domain packet diagram must be obtained. This diagram shows the presence vs. absence of a packet in every instant. With packet here it is intended a MAC sub- layer information unit, that in some technologies is actually called packet, while in other ones is referred to as frame or datagram. Note that the effective content of these packets, i.e. which bits they are carrying, is not relevant for the scope of this recognition. What is important is only the packet pattern, that is whether a packet is present or not. An analysis of this packet exchange pattern can be very useful for revealing the technology that is currently in use, leading to network recognition. The standard that defines a wireless technology deeply describes every aspect of its functionalities, and of course its MAC sub- layer behaviour. This means that there can be maximum or minimum durations for certain types of packets, or even fixed durations. The same rules can be determined for the silence gaps that fall between the packets. Other rules that the standard may specify can be a regular and predetermined transmission of a packet (usually these are control packets, needed for specific system functionalities), or the transmission of acknowledgment packets after the reception of data packets. All these rules are specific for every single technology, i.e. each different network may present a MAC behaviour that is proper and peculiar of that technology. This means that an identification of each single behaviour can be useful for the identification of each technology, leading to the final goal of the network recognition. For this reason, based on the study of the standard, some MAC features were identified for the three technologies taken into account: Bluetooth, Wi- Fi, and ZigBee. These features can highlight the MAC specific behaviour and can be therefore exploited for recognition. This approach integrates the cognitive concept at the network layer, having the big advantage, respect to the widely used spectrum sensing approach, of being extremely simple, and thus keeping a high computational efficiency. In fact, in order to obtain the mentioned time- domain packet exchange diagram, only a simple and rudimentary device is needed: an energy detector. Through this, the short- term energy that is present on the air interface can be computed. After defining a threshold value, all the consecutive short- term energy values that are higher than the threshold can be considered as a packet. In this way, the packet diagram can be formed using energy detection. Based on the study of the standards of the three technologies mentioned before (Bluetooth, Wi- Fi and ZigBee), the following features were identified: packet duration; packet inter- arrival interval. duration of silence gaps identified as Short Inter- Frame Space (SIFS). An important aspect that must be observed is that all these features are very simple to extract from the packet exchange diagram, and are also simple to analyze, requiring low computational load and algorithms. This approach can be relevant for decision making because it might be a simple but effective method for automatic network recognition and classification. This step is important in order to take further decisions based on the active networks eventually present in the ICT-ACROPOLIS Deliverable D8.1 23/61

24 environment in that instant, i.e. if the cognitive radio knows whether other technologies are operating in that area in that instant, it can adapt its parameters or take other decisions, but in a conscious way. An example of the validity of this approach is presented in the following. Wi- Fi and Bluetooth networks are considered in this example, and two features are exploited. Four linear classifiers are used; after the training, they are used for a classification test IEEE Figure 2: Features space with Wi- Fi and Bluetooth points and trained classifiers. Table 1: Classification test result. Classifier Input network Classification into Wi- Fi Classification into multi- slot Bluetooth Perceptron Wi- Fi 98.86% [348/352] 1.14% [4/352] Perceptron Bluetooth 0.43% [2/462] 99.57% [460/462] Pocket Wi- Fi 98.86% [348/352] 1.14% [4/352] Pocket Bluetooth 0% [0/462] 100% [462/462] LMS Wi- Fi 99.43% [350/352] 0.57% [2/352] LMS Bluetooth 34.85% [161/462] 65.15% [301/462] SOE Wi- Fi 99.72% [351/352] 0.28% [1/352] SOE Bluetooth 29.87% [138/462] 70.13% [324/462] The obtained classification test results, even for a simple case reported here as an example, show that the adopted approach is valid, because very high correct classification rates can be obtained only through features simple to extract and simple algorithms. In order to recreate a possible real multi- network scenario, multi- network packet traffic was generated, by mixing two test sets. Furthermore, three different scenarios were considered: ICT-ACROPOLIS Deliverable D8.1 24/61