This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
Software Radio-Based Decentralized Dynamic Spectrum Access Networks: A Prototype Design and Experimental Results Feng Ge∗ , Aravind Radhakrishnan† , Mustafa Y. ElNainay‡ , Qinqin Chen† , Charles W. Bostian† , Allen B. MacKenzie† ∗ Telcordia
Technologies, Inc., Piscataway, NJ, USA @ Virginia Tech, Virginia Tech, Blacksburg, VA, USA ‡ CSE Department, Faculty of Engineering, Alexandria University, Egypt
[email protected],
[email protected],
[email protected], {chenq, bostian, mackenab}@vt.edu † Wireless
Abstract—Significant progress has been made in the past few years on Dynamic Spectrum Access (DSA) wireless networks which seek to use RF spectrum more efficiently and dynamically. For example, many measurements of current spectrum utilization are available and theoretical analyses and computational simulations of DSA networks abound. In sharp contrast, few network systems, particularly those with a decentralized structure, have been built even at a small scale to investigate the performance, behavior, and dynamics of DSA networks. Our contribution is designing a decentralized and asynchronous DSA network and building a prototype based on software radio technologies, signal detection and classification methods, distributed cooperative spectrum sensing systems, and mobile ad-hoc network (MANET) protocols. This paper details the network’s design and implementation as well as its enabling technologies. Through systematic experiments, we identify several factors influencing performance for decentralized DSA networks.
I. I NTRODUCTION Supporting Dynamic Spectrum Access (DSA) in decentralized wireless networks drives complexity both within individual nodes and over the whole network to an unprecedented level [1]. Both innovative protocols and novel architecture are needed. For example, the most basic requirement in DSA networks is to guarantee non-interference to primary users while using RF spectrum more efficiently and dynamically. This requires a new set of functions related to spectrum sensing at the Physical (PHY) Layer. Node cooperation protocols and policy modules are also needed [1]. The coupling of channel allocation, power control, and topology control is necessary in DSA networks because of the dynamic radio environment [2], [3]. Further, cross-layer architecture is fundamentally required in DSA networks because the operating frequency bands depend on the channel occupancy measured at the PHY layer and directly impact the above layers both within and across nodes [4]. Moreover, new protocols and methods like Disruption Tolerant Networking (DTN) protocols and Content Based Access (CBA) techniques are also being included in DSA network design to cope with network connection disruptions under different environments [5], [6]. The first and third authors were previously with Virginia Tech.
Because of the system complexity and the unavailability of a suitable open platform [7], few physical networks have been built to investigate the performance and dynamics of decentralized DSA networks. Nonetheless, good theoretical analyses and simulations have been carried out. For example, coupled power control and channel allocation optimization are heavily investigated [2], [3]. Game theory is also used to propose new protocols in spectrum sharing and to investigate individual nodes’ behavior in cognitive radio networks [8], [9]. Widely known in this research community are two built networks: DARPA’s XG [1] and Wireless Network after Next (WNaN) projects [6]. The first generated several publications which present significant results and insights in developing spectrum sensing methods, building individual DSA nodes, and testing overall systems in different network scenarios and RF environments [1]. Overall, this project reflects the complexity of DSA node architecture and some unique dynamics in DSA networks under different RF environments. However, the XG project was a centralized network [1]. An important goal of WNaN is to build a decentralized DSA network with about a thousand nodes [6]. Unfortunately, few details about the project are yet publicly available. Considering the complexity of both individual nodes and the network in DSA networks, experiments are necessary to gain insights into network design and to investigate network performance under different scenarios [1]. For example, a protocol for power control must consider newly needed functions, their computation and communication cost, and the associated increase of system/network complexity. Otherwise, achievable power performance derived from theoretical analysis or simulation will not be approached [3]. Further, in research of complex systems like cognitive radio networks, experimental methods can assess system/network conditions that may be neglected by analysis or simulation. Our view of decentralized DSA networks is shown in Figure 1. In this network, nodes opportunistically use vacant channels when primary users are not transmitting. The whole network consists of decentralized mobile nodes and works in a dynamic radio environment where fading and interference
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
Learning Module
DSA Module
Application Controller
Selected Data Fusion Nodes A
Network Controller (OLSR)
Selected Relay Nodes Selected Sensors Ordinary CR Nodes B
Sensor Data Update Route
MAC TX RX SDR System Fig. 2.
Inter−Cluster Sensor Data Synchronization Route A Cluster A Data Routing Path (From A to B)
Fig. 1.
Our view of a decentralized DSA network.
exist and primary users may appear at any time. Because of possible hostile radio environments, distributed cooperative spectrum sensing is necessary [10]. Each node can perform functions like signal detection and classification, node reconfiguration, and data relay. However, it is power inefficient to enable all those functions on each node simultaneously [11]. For example, only a small portion of all nodes are needed for cooperative spectrum sensing. But one node must work on data fusion in each cluster, and there must be some network level sensing data exchange and synchronization. To increase network battery life and reduce network communication overhead, the network dynamically selects nodes to assume functions like sensing and data fusion. This paper presents a network prototype corresponding to Figure 1 (though with fewer nodes) and the results of experiments run with the network prototype. II. C OGNITIVE R ADIO N ODE A RCHITECTURE AND C OMMUNICATION P ROTOCOLS Our network enables DSA by using reconfigurable radio nodes, signal detection and classification methods [12], distributed cooperative spectrum sensing schemes [13], decentralized wireless protocols, and intelligent algorithms at both node and network levels. The current prototype uses a WiFi control channel to support cooperative sensing and network management. The following sections introduce the enabling functions of our network prototype. A. Principles of Operation In this decentralized network prototype, cognitive radio (CR) nodes opportunistically use vacant channels when primary users are not present. All the nodes are mobile under a dynamic radio environment where fading and interference are common and primary users may appear anytime. The network partitions itself into clusters as shown in Figure 1. Each cluster then selects a few nodes and enables their signal detection and classification functions. The selected sensors continuously sense a frequency range and report the results to a selected data fusion node (which we denote the
WiFi Control Channel
Our prototype node architecture.
DSA broker, with functions supported by the DSA module, see Section II-E). Upon any variation in the radio environment, DSA brokers in neighboring clusters exchange their data. All the communications are through a common control channel. Meanwhile, all nodes register with the DSA broker in their cluster. Based on the registration information, DSA brokers determine the local network topology and compute a channel allocation table for each node using the Island Genetic Algorithm (IGA) introduced in Section II-D. Upon receiving such tables CR nodes begin to communicate with each other on assigned channels. When a primary user (PU) returns, CR nodes sense it and pause their transmissions. They then query their closest DSA brokers. At the same time, those DSA brokers realize the presence of a PU through their associated sensors. They select new vacant channels and compute a new channel allocation table. Each DSA broker broadcasts the new tables to their associated CR nodes. Finally, the CR network resumes communications. To reduce power consumption, a DSA broker chooses dedicated signal classification sensors to focus on detecting PUs on channels currently in use by CR nodes, while choosing energy detection-based sensors to cover a wide frequency range looking for vacant channels. B. Overall Node Architecture As shown in Figure 2, our current CR node consists of six components: a learning module, a DSA module, an application controller, a network controller, a SDR system, and a WiFi control channel module. The learning module hosts learning functions and algorithms used at both node and network levels. Currently it runs a heuristic optimization algorithm (IGA) for multi-channel allocation (notice that this algorithm is enabled only when a CR node works as a DSA broker). The DSA module enables cooperative spectrum sensing and distributed spectrum management between a CR and its neighboring nodes. Both modules have database components storing information like the network topology and radio environment map. The application controller manages the execution of different applications under different radio and network dynamics. The network controller makes routing protocols, e.g., OLSR here, work with our multi-channel allocation algorithm for routing data through different links on different channels. The SDR system supports reconfigurable Media Access Control (MAC) and PHY layers. Our network prototype relies on a control channel – currently a WiFi system.
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
RX Path f0
USRP
f1
f2
Data Channel 1 Data Channel 2
Routing Table
Data Frames
Data Channel 3
TX
NO
Set Channels
Carri_sense? Yes Accumulative Backoff
Random Exp Back Off
Multi−chan Allocation
Packet Channel Table
Reset NO
TX Path USRP f0 f1
Data Channel
NO
Data Frame
Reset_Freq
t > Ts
Physical layer structure.
C. SDR System Our current SDR system relies on GNU Radio, an open source toolkit for building software radios and a USRP I, an openly designed low-price SDR hardware platform. In particular, the SDR system can transmit on one channel and receive 3 channels simultaneously using frequency multiplexing. Further, we designed a MAC layer protocol to control the PHY layer and enable the DSA mechanism. 1) PHY Layer: As shown in Figure 3, the PHY layer receives 3 channels simultaneously and reconfigures its transmitter for different frequencies. Specifically, the USRP downconverts a wideband RF signal into an IF signal (with a programmable bandwidth). The receiver path uses filters to separate this IF signal into three channels. Signal samples in each channel are demodulated and saved in a queue which interfaces with a callback function used by the MAC layer. However, because USRP’s RF front end has a poor nonlinearity performance and lacks in filtering separation [13], the transmitter side only supports one transmitting channel. It is able to reconfigure on-the-fly for switching transmitting frequencies among three channels. 2) MAC Layer: Fundamentally, SDR achieves its development and operation flexibility by moving the PHY/MAC layer functions into the software domain. This, however, comes with a cost, e.g., execution latency and dynamic computing resource contention [7], [13]. Both problems limit the performance of SDR-based MAC functions, details are available in [13]. The goal of this paper is not to tackle such problems, but rather to investigate the performance of SDR-based DSA networks. To support network communications, we designed a carrier sense multiple access (CSMA) based MAC protocol which enables multi-channel communications among CR nodes under DSA environments. This protocol is shown in Figure 4. Since its receiver side is fairly simple, we only show the transmitter side in the figure. It is exclusively for data communication on vacant channels, not for the control channel. At the transmitter side, the MAC layer first checks a sending packet’s destination IP address, then queries the routing table and the channel table, finally uses our channel allocation algorithm to determine the channel for sending this packet. The node sends out this packet’s data frames when it senses that no other nodes are transmitting in the selected channel.
Yes
Wait for DSA Broker
Resume App/PHY
Yes Pause App/PHY
f2
Fig. 3.
Primary?
Fig. 4.
Query Nearby DSA Broker
MAC layer enabling DSA.
If the node senses that the channel is occupied, it waits for a random time interval (determined using the truncated binary exponential backoff algorithm) before trying to send the frame again. After each data frame is transmitted, the MAC enables a much shorter random time backoff so that other nodes do not have to wait for a long time to access the channel. Under a DSA environment, however, primary users may appear at any time. A CR node may not be able to immediately realize that a busy channel is actually occupied by primary users, instead of by its peer nodes. To solve this problem, the MAC layer also enables an accumulative time counter of all the backoff time between the interval bounded by transmitting two consecutive data frames. If the whole backoff time exceeds a time threshold ts , the MAC signals the application layer and the SDR system to pause. Then this CR node queries its DSA module to confirm whether a primary user is present in the current channel. If a primary user is not present, the CR node assumes that its peer nodes are transmitting over the channel. It resets its accumulative time counter to zero and continues its random time backoff before trying its next transmission. If a primary user is present, the CR node queries its DSA module for a new channel. Meanwhile, its DSA module negotiates with its neighboring CR nodes and they together determine new channels for their associated CR nodes. Once a CR gets a new channel, it reconfigures itself to transmit over that channel. It also resumes the data communication at its application layer. At the same time, the DSA module, once sensing the presence of in-band primary users, also makes its CR node, together with its communicating neighbors, switch to a new channel. D. Network Controller As shown in Figure 2, the network controller works with the learning module and the DSA module to enable multichannel allocation. The learning module currently applies a distributed learning and optimization algorithm – the Island Genetic Algorithm (IGA) developed in [3]. The IGA is a distributed implementation of classic GAs. Briefly, it divides candidate solutions into subpopulations that evolve separately at each different node. Nodes communicate to share candidate solutions to increase local diversity using a migration policy that defines the migration rate and topology. Finally the distributed computation converges and finds a solution, either
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
Cluster 1
t2
t t Signal 3 ... 4 Detector N hop
Signal Recognizer 1
CR Node 1
DSA Broker
t 4’
...
t 3’
t2 Signal Recognizer
t 5 t 5’
DSA Broker 1 CR Node 2
Signal Detector 1
Incumbent User 1
Incumbent Signal
DSA Broker 2 CR Node 4 Signal Detector 2
Cluster 2
Fig. 5.
Incumbent User 2
The experimental system.
the optimum one or close to it. Details are available in [3]. For the multi-channel allocation problem, our IGA takes inputs of the radio environment table (available channels at each node) and the network topology table, and determines the best channel for each communication link so that the overall network throughput is maximized. Our network prototype uses an open source implementation of OLSR as our routing protocol. It is a proactive link-state routing protocol which uses Hello and Topology Control (TC) messages (currently on both the SDR and control channels) to discover and then distribute link state information throughout the ad-hoc network. CR nodes use this topology information to compute next hop destinations for all nodes in the network using shortest hop forwarding paths. E. DSA Module The DSA module enables two functions: cooperative spectrum sensing and distributed spectrum management. Cooperative spectrum sensing is widely endorsed in DSA networks [10] because it allows detection of primary users even under conditions of shadowing, multipath, and interference. Distributed spectrum management is necessary in a decentralized DSA network because CR nodes must coordinate to access dynamically available spectrum. To support both functions, we developed a distributed sensor network using signal detection and classification methods [12], [13], distributed database technologies [13], and MANET wireless protocols. In particular, we use energy detectors to identify unused channels and use the Universal Classifier and Synchronizer algorithm [12] to recognize and differentiate signals from primary and secondary users. These two sensor types are called signal detectors and signal recognizers respectively. The DSA module can work as a DSA broker, processing data from spectrum sensors and coordinating with neighboring CR nodes to monitor all nodes’ spectrum usage and allocate channels. See details in [13]. III. E XPERIMENTAL S ETUP AND R ESULTS The experimental network setup is shown in Figure 5. It emulates two clusters of the network shown in Figure 1. We
CR
t 6’
TX
Control Channel
Signal Recognizer 2
CR Node 3
t6
Secondary Channel
t1 Fig. 6.
Experiment setting to measure sensor network performance.
manually partitioned the network and selected the sensors and the DSA brokers. We are working on network management protocols which will automate such network partition by methods like [14]. Here, each cluster has five nodes including one DSA broker, one signal detector, one signal recognizer with the signal classification algorithm, and two other CR nodes1 . In Figure 5, the four CR nodes form a chain topology and use spectrum opportunistically. Both a signal recognizer and a signal detector associate with each DSA broker. Spectrum sensing and allocation respectively by the sensors and the DSA brokers is coordinated through the WiFi control channel, which also works as a MANET. We used two SDR transmitters to emulate primary users. All the nodes consist of a 2.0 GHz dual-core laptop, a USRP I, and a WiFi card . Because we had only limited physical space for this experiment, we were not able to separate all the nodes widely enough. All the WiFi and USRP can sense each other. We artificially created the above network topology by blocking MAC addresses of non-neighbor nodes in Figure 5. Therefore, the physical media contention was higher than it in theory. In addition, the experiments were completed in a laboratory environment, exacerbating multi-path effects. A. Experimental Results In measuring the network performance, we followed guidance from [1]: must do no harm; must work; must add value. Specifically, we measured this network’s reaction time to active signals, its data communication performance, and its reforming time upon a primary user’s return. 1) Distributed Cooperative Spectrum Sensing: To measure the timing performance of distributed cooperative spectrum sensing, we had a separate experiment setting shown in Figure 6, where spectrum sensors are put multiple hops away from a DSA broker while the CR node is only one hop away from it. We then inserted timestamps at multiple locations to get the timing performance of this distributed cooperative spectrum sensing network. The time symbols used in Table I are the same as those in Figure 6. Our quantitative results are shown in Table I. The meaning of each item is explained by the time difference shown in Figure 6. As we can see, signal detection takes only about 22 ms while signal classification using our algorithm [12] takes significantly longer, about 780 ms. Although our results 1 Our original design and development intended that a single CR node could support all these functions. Unfortunately, the laptops we used were not powerful enough and we had to split the functions on different nodes.
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
TABLE I P ERFORMANCE OF THE DISTRIBUTED COOPERATIVE SPECTRUM SENSOR NETWORK .
35 30 Network Throughput Jitter (ms)
Node Level Time Performance Mean Detection Algorithm (t3 − t2 ) 22.2 Classification Algorithm (t!3 − t2 ) 780 DSA broker Processing (t5 − t4 ) 94 Radio Reconfiguration 7.8 Network Level Network Awareness (t5 − t1 ) One Hop 184 Two Hops 228 Three Hops 262 CR Awareness (t6 − t1 ) One Hop 208 Two Hops 246 Three Hops 281 Network Recognization (t!5 − t1 ) One Hop 993 (Classification) (756)
(ms) STD 2.2 154.6 30 1.1
15.7 36.3 46.0
multi−channel network single channel network
25 20 15 10 5 0 0
57.2 54.6 12.7
20
40
Fig. 8.
174.5 (164.6)
Time (s)
60
100
Jitter comparison.
60
18
80
multi−channel network single channel network
50
14
Packet Loss (%)
Network Throughput (kb/s)
16
12 10
40
30
20
8 10
6 4 2 0
multi−channel network single channel network 20
Fig. 7.
40
Time (s)
60
80
100
Throughput comparison.
depend on our selected hardware and algorithms, such a big time difference illustrates that a signal recognizer’s execution time and not only its accuracy will be a critical limiting factor for a DSA network’s performance. Our SDR-based sensors can reconfigure within 10 ms; this is also the performance for our CR node reconfiguration. Also shown in Table I is the network level performance. The network awareness time performance shows that it takes about 40 ms more for each extra hop. Similarly, increasing by one hop in the CR awareness takes about 40 ms more for the network to react to the presence of an active signal. Given this trend, we only measured the one hop case for the time performance of network level recognization. 2) Multi-Channel Allocation: We set the IF band sample rate of 1 MS/s, the frequency separation between two adjacent sub-channels as 200kHz. We used the GMSK modulation and set the data rate as 50 kb/s at each sub-channel . We measured the performance of multi-channel allocation from three aspects: throughput, jitter, and packet loss. The acquired results were based on UDP packets and averaged with an interval of 5 seconds over a period 100 seconds. We set
0 0
20
Fig. 9.
40
Time (s)
60
80
100
Packet loss comparison.
the UDP buffer length as 1 kB and the bandwidth as 10 kB/s. Figure 7 compares the network throughput of the multichannel network to the single-channel network. It indicates that the multi-channel network throughput is higher than the single-channel network while their fluctuations are similar. As shown in Figure 8, the single-channel network almost always experiences higher jitter than the multi-channel one. Further, both networks experience an increase followed by a decrease in jitter. Figure 9 shows that the packet loss percentage is stable and almost zero in the multi-channel network, while the single-channel experiences abrupt jumps in packet loss. A higher collision rate in the single-channel network than the multi-channel network may explain this performance contrast. 3) Channel Switching and Network Reforming: To measure the network prototype’s performance in DSA, we started its operation by letting four CR nodes wait for a vacant channel from their associated DSA brokers. We then emulated an initial radio environment by transmitting two signals in different bands. They were immediately detected by both signal detectors, which then updated the DSA brokers. Next the DSA brokers calculated the radio environment map and
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.
18 16
Network Throughput (kb/s)
14 12 10 8 6 4 2 0 0
20
40
Time (s)
60
80
100
spectrum sensing, multi-channel allocation, and dynamic spectrum access. Experimental results from this network prototype show that (1) Today’s SDRs, particularly those based on general purpose processors, have performance limitations which lead to poor performance in scaled networks built from them. (2) Differentiating different signals–a basic requirement in DSA applications–requires efficient algorithms. One possible solution is designing signal classification algorithms which exploit the distinctive features of signals used in typical communication systems. (3) Achieving DSA in decentralized networks requires further research and development in radio platforms, signal processing methods, and network protocols. ACKNOWLEDGMENT
Fig. 10. Throughput variation in dynamic spectrum access. The primary user began its transmission at 59 seconds.
chose a vacant channel for four CR nodes. Meanwhile, the DSA brokers made their associated signal recognizers tune to the center frequency of the chosen channel. Upon receiving channel parameters, four CR node configured and started their SDR subsystems and network controllers. The SDR subsystem used GMSK modulation on its three sub-channels. Finally the CR network established its data communication. Before transmitting an in-band primary signal, we started an Iperf client and an Iperf server to monitor data traffic variation over the network. We then emulated a primary user by transmitting a signal with DBPSK modulation2 on one of the three used subchannels. The signal recognizers sensed and classified this new signal. It then sent the signal’s parameters to the DSA brokers. The DSA brokers compared this new signal with those stored in their databases, found it to be a primary signal and notified their associated CR nodes to stop transmitting on this subchannel. The CR nodes then reconfigured their SDR subsystems and continued the data communications on the other two subchannels. The Iperf recorded the network throughput variation during this process. We repeated the above scenarios multiple times and found that the performance variation was similar. One case is shown in Figure 10, where we introduced the primary signal at 59 seconds. Please notice that the throughput fluctuation is due to memory buffering between the USRP and the laptop. Detailed explanations are available in [13]. To reveal more details, we don’t average the results. As shown in Figure 10, the network experienced data loss when switching channels. This transition period is about 2 seconds. Also shown in the figure, there was occasional packet loss at other times. IV. C ONCLUSION We have developed a decentralized DSA network using lowcost SDR platforms. Further, we introduced a signal classification algorithm, instead of using simple energy detection, to differentiate signals between primary users and secondary users. Overall, this network enables distributed cooperative 2 We assume that primary and secondary users use signals at least with different modulations.
This project is supported by National Science Foundation under Grant No. CNS-0519959 and by National Institute of Justice, Office of Justice Programs, US Department of Justice under Award No. 2005-IJ-CX-K017. The opinions, findings, and conclusions or recommendations expressed are those of the authors and do not necessarily reflect the views of the Department of Justice and the National Science Foundation. R EFERENCES [1] M. McHenry et al., “XG DSA radio system,” in IEEE Symposium on New Frontiers in Dynamic Access Networks (DySPAN), 2008. [2] R. Thomas, R. Komali, A. B. MacKenzie, and L. A. DaSilva, “Joint power and channel minimization in topology control: A cognitive network approach,” in IEEE International Conference on Communications, June 2007, pp. 6538–6543. [3] M. Y. ElNainay, D. H. Friend, and A. B. MacKenzie, “Channel allocation and power control for dynamic spectrum cognitive networks using a localized island genetic algorithm,” in IEEE Symposium on New Frontiers in Dynamic Spectrum Access Networks, Oct. 2008, pp. 1–5. [4] V. Srivastava and M. Motani, “Cross-layer design and optimization in wireless networks,” in Cognitive Networks: Towards Self-Aware Networks, Q. Mahmoud, Ed. Wiley, 2007, pp. 121–146. [5] S. Farrell, V. Cahill, D. Geraghty, I. Humphreys, and P. McDonald, “When TCP breaks: Delay- and disruption- tolerant networking,” IEEE Internet Computing, vol. 10, no. 4, pp. 72–78, 2006. [6] [Online]. Available: http://www.bbn.com/technology/networking/wnan [7] L. A. DaSilva, A. B. MacKenzie, C. da Silva, and R. W. Thomas, “Requirements of an open platform for cognitive networks experiments,” in 3rd IEEE Symposium on New Frontiers in Dynamic Spectrum Access Networks, Oct. 2008, pp. 1–8. [8] M. van der Schaar and F. Fu, “Spectrum access games and strategic learning in cognitive radio networks for delay-critical applications,” Proceedings of the IEEE, vol. 97, no. 4, pp. 720–740, April 2009. [9] J. Suris, L. DaSilva, Z. Han, and A. B. MacKenzie, “Cooperative game theory for distributed spectrum sharing,” in IEEE International Conference on Communications, June 2007, pp. 5282–5287. [10] S. M. Mishra, A. Sahai, and R. W. Brodersen, “Cooperative sensing among cognitive radios,” in IEEE International Conference on Communications, 2006, pp. 1658–1663. [11] R. Jurdak, Wireless Ad Hoc and Sensor Networks: A Cross-layer Design Perspective. Springer, 1st Edition, 2007. [12] Q. Chen et al., “Universal classifier synchronizer demodulator,” in The 1st IEEE International Workshop on Dynamic Spectrum Access and Cognitive Radio Networks, 2008. [13] F. Ge, “Software radio-based decentralized dynamic spectrum access networks: A prototype design and enabling technologies,” Ph.D. dissertation, Virginia Polytechnic Institute and State University, 2009. [14] Y. M. Gottlieb et al., “Policy-controlled dynamic spectrum access in multitiered mobile networks,” in IEEE Military Communications Conference, 2010.
978-1-4244-5638-3/10/$26.00 ©2010 IEEE
