Cognitive Radio Sensor Networks

Ozgur B. Akan, Osman B. Karli, and Ozgur Ergul, Middle East Technical University Abstract
Dynamic spectrum access stands as a promising and spectrum-efficient communication approach for resource-constrained multihop wireless sensor networks due to their event-driven communication nature, which generally yields bursty traffic depending on the event characteristics. In addition, opportunistic spectrum access may also help realize the deployment of multiple overlaid sensor networks, and eliminate collision and excessive contention delay incurred by dense node deployment. Incorporating cognitive radio capability in sensor networks yields a new sensor networking paradigm (i.e., cognitive radio sensor networks). In this article the main design principles, potential advantages, application areas, and network architectures of CRSNs are introduced. The existing communication protocols and algorithms devised for cognitive radio networks and WSNs are discussed along with the open research avenues for the realization of CRSNs.

y dynamically changing its operating parameters, cognitive radio senses the spectrum, determines the vacant bands, and makes use of these available bands in an opportunistic manner, improving the overall spectrum utilization. With these capabilities, cognitive radio can operate in licensed as well as unlicensed bands. In licensed bands wireless users with a specific license to communicate over the allocated band (the primary users, PUs), have the priority to access the channel. Cognitive radio user, called secondary users (SUs), can access the channel as long as they does not cause interference to the PUs. When a PU starts communication, the cognitive radio user must detect the potentially vacant bands (spectrum sensing), decide onto which channel to move (spectrum decision), and then finally adapt its transceiver so that the active communication is continued over the new channel (spectrum handoff). This sequence of operation outlines a typical cognitive cycle [1], which can also be applied over an unlicensed band by all cognitive radio users. Cognitive radio capabilities may also be exploited by wireless sensor networks (WSNs), which are traditionally assumed to employ fixed spectrum allocation, and characterized by the communication and processing resource constraints of lowend sensor nodes. Depending on the application, a WSN composed of sensor nodes equipped with cognitive radio may benefit from its potential advantages: Dynamic spectrum access: The existing WSN deployments assume fixed spectrum allocation over very crowded unlicensed bands [2] also used by other devices, as in Table 1. Nevertheless, a spectrum lease for a licensed band amplifies the overall deployment cost. Hence, to be able to cooperate efficiently with other types of users, opportunistic spectrum access may be utilized in WSNs.
This work was supported in part by the Turkish Scientific and Technical Research Council (TUBITAK) Career Award under grant #104E043 and by the Turkish National Academy of Sciences Distinguished Young Scientist Award Program (TUBA-GEBIP).

Opportunistic channel usage for bursty traffic: A large number of sensor nodes detecting an event generate bursty traffic and try to acquire the channel to send their readings. This increases the probability of collisions and packet losses, which decreases the overall communication reliability with excessive power consumption. Opportunistic access to multiple alternative channels may alleviate these potential challenges. Adaptability for reducing power consumption: The dynamic nature of the wireless channel causes energy consumption due to packet losses and retransmissions. Cognitive radio capable sensor nodes may be able to adapt to varying channel conditions, which would increase transmission efficiency, and hence help reduce power used for transmission and reception. Overlaid deployment of multiple concurrent WSN: Dynamic spectrum management may significantly contribute to the efficient coexistence of spatially overlapping sensor networks in terms of communication performance and resource utilization. Communication under different spectrum regulations: A certain band available in one specific region or country may not be available in another due to varying spectrum regulations. Sensor nodes equipped with cognitive radio capability may overcome this potential problem. In general, a cognitive radio sensor network (CRSN) can be defined as a distributed network of wireless cognitive radio sensor nodes, which sense event signals and collaboratively communicate their readings dynamically over available spectrum bands in a multihop manner to ultimately satisfy the application-specific requirements. However, the realization of CRSNs and their potential advantages outlined above depends on addressing challenges including inherent resource constraints of sensor nodes, additional communication and processing demands imposed by cognitive radio capability, design of low-cost and power-efficient cognitive radio sensor nodes, and multihop opportunistic communication over licensed and unlicensed spectrum bands in densely deployed sensor networks. Despite the extensive volume of research results on WSN

34

0890-8044/09/$25.00 2009 IEEE

IEEE Network July/August 2009

Sensor node platforms Bean, BTnode, Mica2, MANTIS Nymph IMote, MicaZ, SenseNode, XYZ, Sentilla Mini, TelosB Mica, weC ANT EyesIFX v. 1 and v. 2 Iris

Radio chip Chipcon (TI Norway) CC1000 Chipcon (TI Norway) CC2420 RF Monolithics TR1000 Nordic nRF24AP1 Infineon TDA5250 Atmel AT86RF230

Operating bands 315, 433, 868, 915 MHz

Overlapping wireless systems Fixed, mobile, amateur, satellite, radiolocation, broadcasting, telemetry, ZigBee Fixed, mobile, amateur radio as secondary, 802.11b/g/n, telemetry, Bluetooth, ZigBee Fixed, mobile, broadcasting, telemetry, ZigBee Fixed, mobile, amateur radio as secondary, telemetry, 802.11b/g/n, Bluetooth, ZigBee Fixed, mobile, broadcasting, telemetry, ZigBee Fixed, mobile, amateur radio as secondary, telemetry, 802.11b/g/n, Bluetooth, ZigBee

2.4 GHz 916.3916.7 MHz 2.4 GHz 868870 MHz 2.4 GHz

Table 1. Operating spectrum bands of commercial WSN transceivers and overlapping wireless systems.

and ad hoc cognitive radio networks [3], the CRSN is a vastly unexplored field with only a handful of studies. In [4] an energy-efficient and adaptive modulation technique is introduced for CRSNs to achieve high power efficiency toward maximizing the lifetime of resource-constrained sensor networks. In [5] CRSNs are discussed for applications such as health care and telemedicine, which require timely delivery of critical information. The potential of dynamic spectrum access in sensor networks is shown in [6] to achieve high power efficiency in sensing applications. In this article we introduce the main design challenges and principles, potential advantages and application areas, and network architectures of CRSNs. The existing communication protocols and algorithms devised for cognitive radio networks as well as WSNs are explored from the perspective of CRSNs, and the open research avenues are highlighted. Our objective is to provide a clear picture of the potentials of CRSNs, the current state of the art, and the research issues on this timely and exciting topic.

Ad Hoc CRSN In an ad hoc CRSN such as that in Fig. 1a, nodes send their readings to the sink in multiple hops in an ad hoc manner. This topology imposes less communication overhead in terms of control data. However, due to the hidden terminal problem, spectrum sensing results may be inaccurate. Clustered CRSN It is essential to designate a common channel to exchange various control data, such as spectrum sensing results, spectrum allocation data, neighbor discovery, and maintenance information. While it may not be possible to find a network-wide common channel, it is highly possible in a certain restricted locality [7]. Therefore, a cluster-based network architecture such as the one in Fig. 1b is an appropriate choice for effective dynamic spectrum management with a local common control channel approach. Heterogeneous and Hierarchical CRSN CRSN architecture may incorporate special nodes equipped with more or renewable power sources such as actor nodes, which act on the sensed event in wireless sensor and actor networks (WSANs). Here, actor nodes may perform additional tasks like local spectrum bargaining. These nodes may have longer transmission ranges and hence be used as relay nodes. This forms a heterogeneous and hierarchical topology with ordinary CRSN nodes, high power relay nodes (e.g., cognitive radio actor nodes), and the sink (Fig. 1c). Mobile CRSNs Mobility of the architectural elements in a CRSN yields a more dynamic topology, amplifying the existing challenges. This requires mobility-aware dynamic spectrum management solutions over resource-constrained CRSN nodes. Moreover, cognitive radio communication protocols for CRSNs must consider mobility as well. In general, the physical characteristics of a CRSN node and the diverse set of CRSN network topologies discussed above yield many open research issues: CRSN node development: Hardware and software design of efficient, practical, and cost-effective cognitive radio sensor nodes must be investigated. Node deployment strategies: Dynamic spectrum-based optimal node deployment strategies in terms of network coverage, spectrum utilization, and power consumption in CRSN topologies must be analyzed toward design of practical yet efficient deployment mechanisms. Spectrum-aware clustering: For applications requiring cluster-based and hierarchical CRSN topologies, dynamic spectrum-aware group formation and maintenance techniques are required.

CRSN Architecture
A typical communication architecture of a CRSN is illustrated in Fig. 1a. Depending on spectrum availability, sensor nodes transmit their readings in an opportunistic manner to the next hops and ultimately to the sink. The sink may also be equipped with cognitive radio capability (i.e., a cognitive radio sink). In addition to the event readings, sensors may exchange additional information with the sink including control data for group formation, spectrum allocation, and spectrum-handoff-aware route determination depending on the specific topology.

CRSN Node Structure

The main difference between the hardware structure of a classical sensor and a CRSN node is the cognitive radio transceiver of a CRSN node (Fig. 2). A cognitive radio unit enables the sensor nodes to dynamically adapt their communication parameters such as carrier frequency, transmission power, and modulation. CRSN nodes also inherit the limitations of conventional sensor nodes in terms of power, communication, processing, and memory resources, which also restricts the features of cognitive radio.

CRSN Topology
According to the application requirements, CRSNs may exhibit different network topologies, as explored below.

IEEE Network July/August 2009

35

(a)

(b)

(c)

Conventional CRSN node Spectrum broker

Cluster head

PU connection on licensed band

Opportunistic spectrum access Mobile CRSN node Actor node

Sink

Primary user

Base station

Figure 1. a) Cognitive radio sensor network (CRSN) architecture; b) possible network topologies for clustered CRSN; c) a heterogeneous hierarchical CRSN.

Potential Application Areas of CRSNs

Due to its potential advantages introduced above, CRSN might be a promising solution for some specific sensor network applications, explored below.

terms of bandwidth, bit error rates, and access delay. For example, as the packet travels through multiple hops, each relaying node may use higher frequencies and the highest possible data rate to provide required bandwidth.

Indoor Sensing Applications

Indoor applications (e.g., telemedicine [5], home monitoring, emergency networks, factory automation) generally require the deployment of many sensor nodes within a small area. These nodes use unlicensed bands, such as industrial, scientific, and medical (ISM) bands, which are extremely crowded [2]. This yields significant challenges for achieving reliable communication due to packet losses, collisions, and contention delay. Here, opportunistic spectrum access of CRSNs may help mitigate these challenges such that critical information requiring real-time reliable communication may exploit the potential advantages of dynamic spectrum management even in crowded environments.

Multiclass Heterogeneous Sensing Applications

Some applications may require multiple sensor networks with distinct sensing objectives to coexist over a common area [8]. Data gathered by these networks may be fused to feed a single decision support. Similarly, in a single sensor network different sensor nodes may be deployed over the same area to sample the event signal over multiple dimensions including scalar measurements (e.g., heat, humidity, location, motion) as well as audio-visual readings of the target being monitored. With the help of dynamic spectrum management, multiclass heterogeneous sensor networks may overlap with minimum interference with each other. Furthermore, through coordination and cooperative spectrum management among these multiple CRSNs, their individual performance and the overall spectrum utilization may be improved.

Multimedia Applications
Reliable and timely delivery of event features in the form of multimedia by resource-constrained sensor nodes under varying channel conditions is an extremely challenging objective due to the inherent high bandwidth demands of multimedia. CRSNs may provide sensor nodes the freedom to dynamically change communication channels according to the environmental conditions and application-specific QoS requirements in

Real-Time Surveillance Applications

Real-time surveillance applications like target detection and tracking require minimum channel access and communication delay. In traditional WSNs this objective may not always be achieved due to crowded operating spectrum band. Furthermore, additional latency may occur in case of rerouting due to a link failure caused by degrading channel conditions.

36

IEEE Network July/August 2009

Processing unit

CR transceiver Sensing data, spectrum policy

In CRSNs sensor nodes may opportunistically access the available channel to maintain minimum access and end-to-end delay for effective real-time surveillance applications. With the development of new delay-constrained joint spectrum allocation and routing algorithms for CRSNs, performance of real-time sensing applications may be further improved. In addition, with the spectrum handoff capability, tactical surveillance CRSNs may be less susceptible to interception and jamming threats.

Processor

RF front-end

PU statistics spectrum management rules

Operating frequency, modulation, transmit power, coding scheme

Demodulator (QAM, QPSK, etc.)

Dynamic Spectrum Management in CRSN

The major challenges and open research issues regarding the efficient dynamic spectrum management framework primarily required for the realization of CRSNs are explored in this section.

Memory Decoder (FEC, block codes, LDPC, etc.)

Spectrum Sensing

Battery

Power Spectrum sensing is one of the major functionaliunit ties distinguishing CRSNs from traditional WSNs. Recharging unit Since nodes can operate on spectrum bands of the licensed PUs in an opportunistic manner, they must gather spectrum usage information via Figure 2. Hardware structure of a cognitive radio sensor node. spectrum sensing prior to transmission. However, the advantages of opportunistic spectrum access approaches and addressing the trade-off between sensing (e.g., more bandwidth, lower error rate due to the ability to accuracy and complexity, must be investigated. switch to the best channel, smaller contention delay) come Cooperative sensing: Distributed cooperative spectrum with the additional power consumption imposed by spectrum sensing mechanisms that strike a balance between sensing sensing and distribution of sensing results. Therefore, a thoraccuracy and overhead need to be developed. ough analysis of cost vs. benefits for a specific CRSN applica Sensing based on collaborative PU statistics: Intelligent and tion must be performed. In the literature there are various distributed methods, which collectively estimate and then spectrum sensing methods, which are examined below in make use of PU channel usage statistics, must be explored terms of how they can be applied to CRSNs: to improve sensing accuracy. Matched filter: This is the optimal spectrum sensing method in Gaussian noise; however, it requires a priori knowledge Spectrum Decision about PU transmission and additional hardware for CRSN nodes to synchronize with the PU. CRSN nodes must analyze the sensing data and make a deci Energy detection: If the measured energy on a channel is sion about the channel and transmission parameters (e.g., below a threshold value, the channel is considered availtransmission power and modulation). Spectrum decision able. Its simplicity and low signal processing requirement methods proposed for cognitive radio networks consider make this method very attractive for CRSNs. However, power consumption as a secondary issue. Processing capabilienergy detection requires longer measurement duration ties and the amount of extra control packets to transmit are leading to higher power consumption. almost always ignored. Clearly, these settings do not match Feature detection: This method can be used when certain CRSN characteristics due to inherent constraints of sensor features of the PU transmission (e.g., carrier frequency and nodes. cyclic prefixes) are known. It exploits the spectrum correlaOn the other hand, the spectrum sensing results will be tion of the PU signal, and hence is very robust against varisimilar in any given locality [7]. If nodes try to access the ations of noise. However, it has very high complexity; channel depending only on their individual spectrum decision therefore, it may not be suitable for CRSNs. results, collision probability increases. Furthermore, since Interference temperature: Nodes calculate how much interfernodes run the same algorithm, when a collision occurs, they ence they would cause at the PU receiver and adjust their all try to switch to another channel, leaving the previous chantransmission power such that their interference plus the nel empty and colliding again on the new channel. Therefore, noise floor does not exceed a certain interference temperaspectrum decisions in CRSNs must be coordinated to increase ture level. This method requires a priori PU location inforoverall utilization and maximize power efficiency. Furthermation, and is computationally too intensive for a low-end more, a spectrum decision mechanism for CRSNs must have CRSN node. low complexity as well. Along these main approaches as outlined in Table 2, many Coordination and spectrum decision can be handled by spectrum sensing methods exist for cognitive radio that are centralized or distributed approaches. A sink might be designot directly tailored for constrained sensor nodes. Therefore, nated as the centralized entity to reach a network-wide optithere are several open research issues on spectrum sensing in mal spectrum decision, which also imposes additional traffic, CRSNs: resulting in excessive power consumption. However, in dis Hybrid sensing techniques: New hybrid sensing techniques, tributed coordination, nodes share their spectrum sensing and exploiting the advantages of main spectrum sensing decision results only with their immediate neighbors or within

IEEE Network July/August 2009

37

Spectrum sensing method Matched filter

Disadvantages Requires a priori info on PU transmissions, and extra hardware on nodes for synchronization with PUs. Requires longer sensing duration (high power consumption). Accuracy highly depends on noise level variations. Requires a priori knowledge about PU transmissions. Requires high computational capability on nodes. Requires knowledge of location PU and imposes polynomial calculations based on these locations.

Advantages Best in Gaussian noise. Needs shorter sensing duration (less power consumption). Requires the least amount of computational power on nodes. Most resilient to variation in noise levels. Recommended by FCC. Guarantees a predetermined interference to PU is not exceeded.

Energy detection

Feature detection Interference temperature

Table 2. Overview of spectrum sensing methods. small clusters. This approach leads to suboptimal utilization, which can still be close to global optimal [7], and yet is considerably simpler to implement and incurs less communication and power overhead than the centralized approach. Clearly, there are many open research issues for the development of new spectrum decision techniques for CRSNs: Spectrum decision parameters: Parameters to be used in spectrum decision (e.g., signal-to-noise ratio, channel capacity, delays and holding times of PUs) must be explored, and new algorithms that yield optimal spectrum decisions must be designed. Distribution of control data: Using a network-wide common control channel for coordinated spectrum decision is generally not feasible; however, finding such a channel within a given locality has high probability. Therefore, energy-efficient centralized and distributed methods of sharing spectrum decision in CRSN data must be thoroughly investigated.

Physical Layer
The physical layer of a CRSN node should provide the capability of reconfiguring its operating frequency, modulation, channel coding, and output power without hardware replacement. Software-defined radio (SDR)-based RF front-end transmitters and receivers are required for reconfigurability of CRSN nodes. However, this is a significant challenge due to the low-cost and resource-constrained nature of sensor nodes. On the other hand, limited capabilities of analog-to-digital (A/D) converters and heavyweight signal processing algorithms make spectrum sensing a challenging issue as well. Detecting weak signals, and hence the presence of PUs, while there are SUs is a significant sensing challenge in CRSNs. Wideband spectrum sensing, advanced modulation schemes, and cognitive learning capabilities cannot be fully realized in a CRSN node due to its limited computational power. Clearly, there are many fundamental open research issues on the physical layer design for CRSNs: SDR-based transceiver: For energy-efficient and practical dynamic spectrum access, low-cost SDR-based transceivers must be designed for CRSN nodes. Optimal waveform design: Design of an optimal waveform to be used over multiple channels with different transmission parameters needs to be studied. Adaptive power control: Methods addressing the trade-off between transmission power and interference must be designed for densely deployed CRSN.

Spectrum Handoff
When a PU starts using a previously available channel, CRSN nodes must detect this activity within a certain time through spectrum sensing methods. Then they immediately move to another available channel decided on by an effective spectrum decision mechanism, even if they have ongoing transmissions. Nodes may also want to switch channels if channel conditions get worse, reducing communication performance. This fundamental functionality of cognitive radio is called spectrum handoff. When spectrum handoff is needed, first an alternate channel must be determined. Then a receiver-transmitter handshake must be performed on the new channel. Only then may nodes continue their transmissions. All of these additional operations incur long delays and hence buffer overflows, which lead to packet losses, degradation in reliability, and ultimately resource waste in CRSNs. In [5] a central spectrum allocation scheme that tries to minimize spectrum handoff has been proposed for CRSNs. However, none of the previous studies on spectrum handoff consider the challenges posed by the inherent limitations of CRSNs. As an open research issue, minimizing the effect of spectrum handoff on various communication layers must be analyzed for CRSNs. At the same time, the development of central and distributed spectrum handoff solutions for CRSNs must be investigated.

Data Link Layer

Error Control The main error control schemes assumed by WSNs are forward error correction (FEC) and automatic repeat request (ARQ). Despite the simplicity of ARQ approaches, their retransmission-based mechanism causes extra energy consumption and reduces bandwidth utilization. Therefore, similar to traditional WSNs, FEC schemes are promising for resource-constrained cognitive radio sensor nodes. However, a fixed FEC scheme may not yield optimal results for every channel under spectrum mobility in CRSNs. Hence, dynamic spectrum-aware and low-complexity FEC mechanisms for CRSNs need to be developed. Accordingly, optimal packet size distribution in CRSNs must be analyzed. Medium Access Control In general, a medium access control (MAC) protocol aims to provide the sensor nodes with means to access the medium in a fair and efficient manner. In CRSNs, depending on the specific topology, sensor nodes may handshake to negotiate on the channel before transmitting packets. Both topology forming and channel negotiations require control packet exchange. Therefore, compared to conventional WSNs, the MAC layer of a CRSN node must handle additional challenges such as silent spectrum sensing

Communication in CRSNs
Here, we investigate the specific design considerations, existing approaches, and open research issues of each communication layer in close relation with dynamic spectrum management, as illustrated in Fig. 3

38

IEEE Network July/August 2009

Application layer Application data Transport layer Delay packet loss End-to-end transmission Network layer New route and spectrum decision Spectrum handoff Joint route and spectrum decision Routing Routing information Rerouting delay Data link layer Medium access Channel Bad channel characteristics PU activity Spectrum sensing Physical layer Reconfiguration Transmission Sensing information Frequency, modulation Reconfiguration Link state information Phy. layer reconfiguration Application req. Bandwidth, delay, reliability

Reliability, congestion Lower layer reconfiguration Upper layer requirements

Figure 3. Interaction between the communication and dynamic spectrum management functionalities. periods temporarily inhibiting transmission, the impracticality of broadcast over a network-wide common channel, and the need for high-priority PU PU F3 F3 F3 access mechanism for the distribution of spectrum sensing and decision results. F1, F2 Node 4 There is a substantial amount of work in the , F2 Node 2 F1 Event F1 literature on MAC solutions for ad hoc cognitive radio networks [9]. However, these solutions have Node 1 F1, F3 various disadvantages that render them impractiF2, F2, F3 F1, cal for CRSNs, such as a requirement for multiSink Node 3 ple transceivers, dependence on network-wide Sensing synchronization, and poor performance under F3~F2>F1 area bursty traffic in densely deployed networks. Hence, new approaches are needed for CRSNs. Figure 4. Three frequencies, F1, F2, F3, with different transmission ranges, R1, The open research areas can be listed as: R2, and R3, such that R1 > R2, R3, are available. F3 is used by the PU. When Novel MAC solutions: A new spectrum-aware node 1 selects F1 for transmission, it connects to node 2 directly, or over nodes and energy-efficient MAC technique that can 3 and 4. However, if it selects F2, it can only connect to node 2 over nodes 3 or make full use of the multiple alternative chan4. If it selects F3, it connects to node 2 over node 3 as a path over node 4 cannel availability must be developed. not be used due to PU activity. Minimum communication overhead: Solutions for CRSNs must operate with minimum exchange of control data with no additional hardware jointly considered with route determination in CRSNs, as requirements, such as an extra transceiver or GPS for synillustrated in Fig. 4. chronization. Furthermore, with spectrum handoff, routes may be obso Opportunistic duty cycle: New duty cycle mechanisms must lete during new spectrum sensing and assignment phases. be designed that jointly consider neighbor discovery, and After spectrum assignment, variation in channel characterisspectrum sensing and allocation. tics mandates recalculation of routes according to the metrics in use. Unlike cognitive radio networks, rerouting algorithms Network Layer should also be highly energy-efficient in CRSN. Moreover, the absence of a global common channel makes neighbor discovExisting ad hoc cognitive radio routing schemes aim to proery and messaging for route establishment quite challenging vide joint spectrum and routing decisions, but do not consider for CRSNs. Therefore, heavyweight and high-maintenance the inherent resource constraints of CRSNs. Similarly, routing routing algorithms may not be practical for CRSNs. schemes developed for WSNs mainly aim to minimize energy The open research issues for effective network layer soluconsumption and do not handle dynamic spectrum access. tions in CRSNs are: Due to spectrum mobility in CRSNs, hop-based channel Opportunistic spectrum-aware routing: New energy-efficharacteristics like channel access delay, interference, operatcient cognitive radio routing metrics and mechanisms that ing frequency, and bandwidth are new metrics to consider in consider spectrum mobility, resource constraints, and dense the design of new routing techniques. As spectrum handoff deployment must be developed. incurs delay, the number of channel switches along a path Analytical framework for routing optimization: The effibetween source and sink also affects the route decision and ciency and complexity of routing in conjunction with opporhence must be considered by routing algorithms. In addition, tunistic spectrum access must be analyzed to reach optimal a spectrum decision can change the neighboring status of a networking solutions for CRSNs. CRSN node [10], and hence is directly related and must be

Interlayer interaction considering energy efficiency 39

Link state and routing info. QoS req.

IEEE Network July/August 2009

 Adaptive and QoS-aware routing: New techniques considering application-specific QoS requirements for varying channel conditions need to be designed for real-time communication in CRSN.

Transport Layer
In sensor networks the transport layer is mainly responsible for end-to-end reliable delivery of event readings and congestion control to preserve scarce network resources while considering application-based QoS requirements. To achieve successful detection and tracking of an event signal, event features must be reliably communicated to the sink without leading to congestion, which wastes power and communication resources. While this delicate balance between reliability and energy efficiency is also inherited by CRSNs, dynamic spectrum management brings additional challenges such as varying channel characteristics (e.g., link delay, channel bit error rate, capacity, additional delay, and packet losses due to spectrum handoff) and temporal inability to transmit due to silent sensing periods. Furthermore, some applications such as target tracking and surveillance may also impose additional real-time delay bounds on the reliable communication requirements. Moreover, effective transport layer solutions are also required, with even tighter reliability requirements, for delivery of queries, commands, and code updates in the reverse path (i.e., from the sink to the sensors) in CRSNs. Although there exist several transport layer solutions for reliable delivery with minimum energy consumption and congestion avoidance in WSNs, none of them considers dynamic spectrum access. In [3] a transport layer mechanism addressing dynamic spectrum access in cognitive radio ad hoc networks is presented. However, it does not consider the inherent constraints and requirements of sensor networks. Hence, there are several open research issues for transport layer in CRSNs: Spectrum-aware reliability: New reliability definitions, objectives, metrics, and analytical modeling must be studied in order to incorporate the fundamental variables of dynamic spectrum access. Opportunistic energy-efficient transport: Energy-efficient cross-layer spectrum-aware transport protocols must be developed for both event-to-sink and sink-to-sensor communication in CRSNs. Real-time cognitive reliable transport: Adaptive real-time transport solutions must be developed to address application-specific real-time reliability and QoS requirements under varying spectrum characteristics.

plexity have been widely studied. However, in CRSNs transceivers of the nodes may be tuned to different channels; thus, a node cannot hear all transmitted data around it. New aggregation and fusion techniques that address CRSNs limitations and take advantage of its additional capabilities must be investigated. Another open research issue is developing schemes to perform sampling of the event signal and gather sensing data based on spectrum availability. If sensing results are sent to the sink periodically, mechanisms to schedule sampling and sensing based on spectrum availability must be investigated. If sensing data is sent based on queries from the sink, new query methods that take spectrum availability into account must be developed.

Conclusions
In this article we investigate CRSNs formed by incorporating cognitive radio capabilities in WSNs. We discuss advantages and limitations of CRSNs, and explore the applicability of the existing techniques for cognitive radio and WSNs in CRSNs along with their shortcomings. Clearly, there are many significant challenges in the realization of CRSNs. We anticipate that this article will provide better understanding of the potentials for CRSNs and motivate the research community to further explore this promising direction.

References
[1] S. Haykin, Cognitive Radio: Brain-Empowered Wireless Communications, IEEE JSAC, vol. 23, no. 2, Feb. 2005, pp. 20120. [2] G. Zhou, J. A. Stankovic, and S. H. Son, Crowded Spectrum in Wireless Sensor Networks, Proc. 3rd Wksp. Embedded Net. Sensors, 2006. [3] I. F. Akyildiz, W. Y. Lee, and K. R. Chowdhury, CRAHNs: Cognitive Radio Ad Hoc Networks, Ad Hoc Net., 2009; doi:10.1016/j.adhoc.2009.01.001 [4] S. Gao et al., Energy Efficient Adaptive Modulation in Wireless Cognitive Radio Sensor Networks, Proc. IEEE ICC, June 2007, pp. 398086. [5] S. Byun, I. Balasingham, and X. Liang, Dynamic Spectrum Allocation in Wireless Cognitive Sensor Networks: Improving Fairness and Energy Efficiency, Proc. IEEE VTC, Sept. 2008, pp. 15. [6] S. Gao, L. Qian, and D. R. Vaman, Distributed Energy Efficient Spectrum Access in Wireless Cognitive Radio Sensor Networks, Proc. IEEE WCNC, Apr. 2008, pp. 144247. [7] J. Zhao, H. Zheng, and G. Yang, Distributed Coordination in Dynamic Spectrum Allocation Networks, Proc. IEEE DySPAN, Nov. 2005, pp. 25968. [8] A. K. M. Azad and J. Kamruzzaman, A Framework for Collaborative Multi Class Heterogeneous Wireless Sensor Networks, Proc. IEEE ICC, May 2008, pp. 43964401. [9] C. Cormio and K. R. Chowdhury, A Survey on MAC Protocols for Cognitive Radio Networks, Ad Hoc Net., Feb. 3, 2009; http://www.sciencedirect.com [10] Q. Wang and H. Zheng, Route and Spectrum Selection in Dynamic Spectrum Networks, Proc. CCNC, Jan. 2006, pp. 62529.

Biographies
OZGUR B. AKAN [M00, SM07] (akan@eee.metu.edu.tr) received a Ph.D. degree in electrical and computer engineering from the Broadband and Wireless Networking Laboratory, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, in May 2004. He is currently an associate professor with the Department of Electrical and Electronics Engineering, Middle East Technical University, Ankara, Turkey, and director of the Next-Generation Wireless Communications Laboratory (NWCL). His current research interests are in next-generation wireless networks, cognitive radio networks, sensor networks, and bio-inspired nano-scale and molecular communications. He is an Associate Editor for IEEE Transactions on Vehicular Technology, Editor for ACM/Springer Wireless Networks Journal and Editor for International Journal of Communication Systems (Wiley). He received the IBM Faculty Award 2008 and Turkish Academy of Sciences Distinguished Young Scientist Award 2008 (TUBA-GEBIP). OSMAN B. KARLI (obkarli@aselsan.com.tr) received his B.S. degree in electronics and communication engineering from Istanbul Technical University, Turkey, in 2005. He is currently an M.S. student in the Department of Electrical and Electronics Engineering, Middle East Technical University. His current research interests include ad hoc cognitive radio networks and transport layer protocols for ad hoc wireless networks. OZGUR ERGUL (oergul@eee.metu.edu.tr) received his B.S. degree in electrical and electronics engineering from Hacettepe University, Ankara, Turkey, in 2000. He is currently a research assistant in the NWCL and pursuing his Ph.D. degree at the Department of Electrical and Electronics Engineering, Middle East Technical University. His current research interests include ad hoc cognitive radio networks and cooperative communication schemes for ad hoc wireless networks.

Application Layer
Application layer algorithms in sensor networks mainly deal with the generation of information and extracting the features of event signals being monitored to communicate to the sink. Other services provided by the application layer include methods to query sensors, interest and data dissemination, data aggregation, and fusion. Clearly, each of these services must utilize the capabilities of a CRSN while conforming to its limitations. Therefore, existing application layer protocols must be revisited with these capabilities and limitations in mind. One of the potential areas that needs an application layer protocol is overlay multiclass heterogeneous sensor networks described earlier. For such applications, an application layer protocol that analyzes and organizes user queries in a heterogeneous network for efficient transmission is needed. Similarly, for multimedia sensor applications an open research issue is adaptive coding schemes that can employ various coding methods depending on channel conditions and handoff rate. On the other hand, data aggregation and fusion techniques that aim to decrease communication cost with additional com-

40

IEEE Network July/August 2009