An Opportunistic Service Differentiation Routing ... - Semantic Scholar

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.

An Opportunistic Service Differentiation Routing Protocol for Cognitive Radio Networks Kiam Cheng How1, Maode Ma1, and Yang Qin2 1

School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 2

HIT ShenZhen Graduate School, China

Abstract— Cognitive Radio (CR) is a new paradigm that enable nodes to exploit unoccupied frequency spectrum for transmissions. Cognitive Radio Networks (CRNs) have been proposed to enable wireless mesh networks to communicate via dynamic channels. Many existing research consider routing in static CRNs with relatively stable communication channel where the duration of the availability of the communication channel is much longer than the communication time. However, there is limited routing related research in dynamic CRNs where the average available duration of the communication channel can be much shorter than the communication time. To address this, we propose a cross-layer cognitive routing protocol, the Opportunistic Service Differentiation Routing Protocol (OSDRP) for the dynamic CRNs. OSDRP discovers the minimum delay – maximum stability route in CRNs by considering the availability of spectrum opportunity in addition to switching delay and queuing delay across primary user networks. In addition, service differentiation is achieved through a combination of transmit power control and opportunistic routing. Simulation results demonstrate that OSDRP can achieve much better performance in terms of lower delay compared to other existing routing protocols in various scenarios. Keywords- Cognitive Radio, Wireless Mesh Networks, Routing

I. INTRODUCTION

W

IRELESS Mesh Networks (WMNs) are considered an alternative to traditional wired broadband access. Thus existing broadband applications like VoIP and video conferencing are expected to work well on the WMNs with few or no modifications. However, QoS provisioning in WMNs is very challenging due to various factors such as dynamical changing of topology, capacity limitations, link variability and multi-hop communications, etc. The chief factor to limit the performance of the WMNs is the interfering nature of wireless transmissions which reduces the network capacity [1]. In recent years, designs utilizing multiplechannels/interfaces have gained popularity for improving the capacity of WMNs. Particularly, cognitive radio based WMNs or cognitive radio networks (CRNs) [2, 3] have gained much attention due to its capability to exploit frequency bands currently unoccupied and available for use. The set of frequency bands are also known as Spectrum Opportunity (SOP). This has presented new opportunities to improve the performance of WMNs as the SOP can be used to carry existing traffic, thus alleviating the interference problem on existing communication channels. Existing research work on CRNs has mainly focused on media access control (MAC) and physical layer issues [4-7]. These approaches can provide optimal solutions in a single

hop set-up but may become ineffective in a multi-hop scenario. Thus, it is imperative to design appropriate cognitive multi-hop protocols capable of optimizing solutions over endto-end paths [8]. Most legacy wireless routing protocols such as AODV and DSR, etc. select paths based on only one criterion (shortest path, minimum delay etc). The naive routing decision policies based on a simple criterion cannot adequately capture the essence of the network dynamics and thus cannot discover the optimal route in the WMNs. There have been some research proposals to improve routing performance by considering alternative routing criteria like interference, network topology and traffic information, etc. They can generally provide much better performance compared to legacy routing protocols. More recently, Opportunistic Routing (OR) [9] has been proposed to improve the performance of WMNs. When a node transmits, a number of nearby nodes can hear the transmission. A node can then be chosen out of the several candidate forwarding nodes (CFNs) to transmit the received packet. This helps to cut down on unnecessary forwarding by intermediary nodes and improve the performance over the traditional unicast routing [10, 11]. Various routing protocols have been proposed for CRNs. In [12], the authors propose a joint interaction between ondemand routing and spectrum scheduling. A delay-based metric is defined to evaluate the effectiveness of candidate routes. The path selection algorithm aims to minimize the sum of the switching, queuing and back-off delay for the considered route. A capacity-based routing strategy for CRN is proposed in [13]. The scheme can improve performance by shifting traffic to the edge of the network away from the higher density regions. In [14], a spectrum-aware routing solution is developed to opportunistically routes traffic across paths with higher spectrum availability and quality via a new routing metric. The routing algorithm is based on link state routing which requires time to converge and to build a network topology map. This approach may not be suitable due to the dynamic properties of the considered CRN. In [15], the authors aim to find the optimum path in CRN using a novel metric that considers the maintenance cost of a route as channels or links that must be switched due to primary user activity. They model the problem as an integer programming optimization model and shown it to be of polynomial time complexity in case of full knowledge of primary user activity. Most of these solutions [12, 13] assumed a static CRN which considers an available frequency band as a permanent resource indefinitely available during its activity. Clearly, such

978-1-4244-5637-6/10/$26.00 ©2010 IEEE


assumptions do not differ from those considered in multichannel mesh networking. Other proposed works [14, 15] also do not consider the issue of QoS provisioning in CRNs. In this paper, we propose the design and implementation of a cross-layer Opportunistic Service Differentiation Routing protocol (OSDRP) for the dynamic CRNs where the primary band can be exploited by a cognitive user while its intermittent availability can seriously affect the service offered for a CR. The key idea of our scheme is that the availability of SOP across the different PRNs which made up the CRN must be considered in the route selection to achieve an optimal end-toend performance. To achieve this, we make two significant new innovations, which are critical to ensure that OSDRP performs well. They include (1) A multi-metric route selection algorithm, which considers the availability of the frequency band in addition to traditional metric like switching delay and queuing delay. (2) A Combined Opportunistic Routing with Transmit Power Control scheme, which simplifies the selection of CFNs, improving delivery ratios of CFNs and achieving service differentiation to traffic flows with different priorities. The rest of the paper is organized as follows. The system model is described in Section II. The detailed operations of the OSDRP are presented in Section III. We evaluate the performance of OSDRP through simulations in Section IV. Finally, we conclude the paper in Section V.

by cognitive users is at best intermittent. We model the availability of the intermittent channel using an On/Off model by defining the probability p of the primary user transmitting in the next time period (channel unavailable for secondary transmission). The probability of the primary user not transmitting in the next time period (channel available for secondary transmission) is thus 1-p. We also define the mean duration of the channel spent in either status by an appropriate probability distribution with mean τ. Thus, the channel state will alternate between the periods of availability and unavailability according to p and τ (Fig. 1).We assume that the parameters p and τ are available for all nodes in their decision making. It is reasonable given that there are already schemes that allow nodes to monitor and utilize past channel histories to make predictions on future spectrum availability. For each available channel, we also keep track of the elapsed time τavail that the channel has remained available. Each time when the channel state changes from unavailable to available, we set τavail to zero. Thus the expected available transmission time is defined as EATT = (1 − p)τ p − τ avail .

II. SYSTEM MODEL

SOPi ⊆ F and SOPj ⊆ F , where F is the set of all channels in

A. Physical and MAC layer Modeling We consider a dynamic multi-hop CRN consisting of N nodes. Each node has one CR transceiver. The CR transceiver allows each node to communicate with its neighbouring nodes by selecting an overlapping data channel in their respective SOPs. To ensure that routing messages can be received by nodes despite of the inconsistency between their frequency bands, each node is also equipped with a traditional wireless interface to form a common control channel among the nodes. This is a common approach adopted by many MAC and routing protocols [4, 12, 16] proposed for CRNs. Our scheme requires a power-control rate adaptive scheme [17] to be implemented on the MAC layer. For ease of simulation and explanation, we assume that there are 12 nonoverlapping wireless channels specified by the physical layer of the IEEE 802.11a standard. Each of the channels has a bandwidth of 20 MHz with a maximum capacity of 54 Mbps. We also assume that channel 1 is utilized as the common control channel while the remaining 11 channels are available for data communication. The channel switching time is assumed to be td ms. B. SOP Modeling We assume an interweave model [18], i.e., the nodes in the CRN can only transmit when the primary users are not active, and that the set of SOP is same for every node in the same PRN. The availability of the primary band for transmissions

For any two neighboring nodes i, j to communicate, there must be at least one overlapping channel f among their respective SOPs. If there is more than one overlapping channel, the channel with the greatest EATT will be chosen. That is, choose Max (EATT(i, j, f)), where f ∈ ( SOPi ∩ SOPj ) , the primary band and SOPi and SOPj are the set of Spectrum Opportunities at node i and j respectively. III. PROPOSED SCHEME In this section, we first present an overview of the OSDRP scheme for CRNs. Then, we describe the protocol in details. A. Overview The twin goals of the OSDRP are to select the minimum delay - maximum stability route for an end-to-end traffic flow and to provide differentiated service to different traffic priorities. We define an end-to-end traffic flow by the following parameters, (1) flow duration τflow and (2) flow priority Cf. Fig. 2 shows the flow diagram of the OSDRP. OSDRP performs route discovery by implementing the complete set of route discovery mechanisms similar to DSR. By sending out “Route Request” or RREQ packets, nodes can discover the minimum delay routes to arbitrary destinations in the CRN. Once the minimum delay routes have been discovered, OSDRP selects the minimum delay - maximum stability route to fulfil the required flow duration τflow of the traffic flow. To evaluate the stability of a route, we define the minimum expected available transmission time (MEATT) of a route as

978-1-4244-5637-6/10/$26.00 ©2010 IEEE


Step 1 . .

Unavailable channel Fig. 1.

the minimum of the expected available transmission time for all nodes on the selected route. Thus where i and j are nodes on MEATT = Min ( EATT (i , j , f )) 1≤ ( i , j ) ≤ h

the selected route and h is the number of hops. Only the route with MEATT ≥ τ flow will be considered. This ensures that the SOP along the selected route is available during the transmission of the traffic flow in order to avoid costly route re-discovery. Finally, Opportunistic Routing with Transmit Power Control (ORTPC) is performed during actual transmission to provide a differentiated service according to traffic priority Cf. B. Detailed Algorithm Description Step 1: Route Discovery Route discovery is initiated by the transmitting node through the broadcasting of RREQ packets. The routes are stored in route caches on the nodes and they are checked for a valid route before initiating route discovery. Each RREQ packet contains a record listing the address and frequency channel of each intermediate node through which this particular copy of the route request message has been forwarded. The format of the route discovery packet was modified from the DSR version to include two additional fields, “delay” and “MEATT”. Each node, upon receiving a RREQ packet, appends its own address and frequency channel to the route record in the RREQ packet and updates the delay and MEATT field and rebroadcasts the packet to its neighbors if it has not forwarded already or if the node is not the destination node. The delay at each node is made up of two components, switching delay and queuing delay. Switching delay dswitch The switching delay is set to zero if a node ni’s receiving frequency is fixed to the same channel as the previous hop node ni-1’s transmitting frequency. Otherwise, the switching delay is set to dswitch = td ms. Average queuing delay dqueue The MAC layer of the node ni keeps track of the moment when a packet is received and the moment when it is sent. Let the moment that a packet i is received be tireceive, and the moment that the packet is sent be tisend. Therefore, the average queuing delay d = t −t n , where n is the

∑

nsend

Sort route in ascending order of delay

Step 2

Route Decision

Step 3

Differentiated Transmission

Available channel

On/Off Modeling of Spectrum Opportunity in a PRN

queue

Route Discovery

Route Maintenance

Channel 2 Channel 3 . . Channel 11 Channel 12

nreceive

n

window size of samples to keep track of. When the destination node receives the RREQ packet, a Route Reply or RREP

Fig. 2.

OSDRP Flow Diagram

packet would be generated with a copy of the accumulated route record from the RREQ packet. The reply will be send to the source node through the reverse route the RREQ packet had traversed. When the RREP packet is received by the source node, the route information, delay and MEATT information within it will be used to update the route cache. The delay information within the RREP packet reflects the sum of the delays caused by switching and queuing for each node along the route and is defined as Delay(n0,nh) = (d +d ) where h is the number of hops between

∑

switch

queue

h

nodes n0 and nh. Finally, the route information will be sorted in ascending order of delay. Route maintenance is performed for every valid route in the cache. The MEATT field of each route is reduced by 1 every second. Once the value of the MEATT field drops to zero, the route entry will be deleted from the cache. Step 2: Route Decision The minimum delay – maximum stability route is chosen among the sorted route information identified in Step 1. A route is stable when its MEATT is greater than the required flow time τflow of the considered traffic flow. From the list of routes with MEATT > τflow, the one with the minimum delay will be selected. Step 3: Differentiated Transmission Differentiated service is achieved through ORTPC. We consider that traffic flows have 3 different traffic priorities c1, c2 and c3 with corresponding minimum required data rate of 24, 12 and 6 Mbps, respectively. With a given BER (≤ 10-5), a node ni forwarding traffic to another node nj with priority ct can calculate the required transmitter power Pi based on the minimum required data rate rt, corresponding SINR threshold γ t using SINRij = Pi Gij /( N 0 + Noise) ≥ γ t where Gij

(

)

denotes the link gain from node ni to node nj, N0 denotes thermal noise of node nj and Noise is the measured noise due to other sources at nj. Higher priority traffic is likely to require larger transmission power and a corresponding increase in the transmission range. OSDRP utilizes this fact to vary the range of the CFNs selection according to the traffic priority. For example, to choose the CFNs for traffic with priority ct, node ni examines the remaining traversing path to identify nodes within a certain hop count (ni+1, ni+2, .., ni+range). An example is depicted in

978-1-4244-5637-6/10/$26.00 ©2010 IEEE


2

3

4

3

4

5

2

Candidate forwarding set at node 1 Fig. 3.

TABLE I AVAILABILITY OF SOPS IN PRNS Primary Networks SOP Availability, U (τ- a, τ+ a) PRN1 a = 5, τ = 15 PRN2 a = 5, τ = 45 PRN3 a = 5, τ = 10 PRN4 a = 5, τ = 30

CFNs Selection

Fig. 3, where node 1 is forwarding traffic to node 5. The candidate set at node 1 is 4, 3 and 2 for a selection range of 3 hops. Our CFN selection strategy has three benefits compared to existing OR protocols. Firstly, the selection of CFNs is simpler as only nodes further along the selected path are considered. Secondly, candidate nodes are range limited according to traffic priority and this can provide service differentiation. Finally, duplicate transmissions are minimized as nodes will not be selected on possibly diverging paths. To ensure that only the node closest to the destination perform the forwarding, we implement the prioritized acknowledgement scheme [9] where candidate nodes acknowledge the successful reception of a packet in a prioritized manner according to the sequence indicated in the packet, i.e. a candidate node with higher priority will send its acknowledgement before any lower priority candidates. For the example in Fig. 3, node 4 will send its acknowledgement before node 3, and node 3 will send its acknowledgement before node 2. The complexity analysis for OSDRP is similar to DSR. For a network consisting of N nodes and diameter d, the time complexity for route initialization is O(2d). Since OSDRP maintains multiple routes to the same destination, the time complexity after a link failure is 0 for a cache hit and O(2d) otherwise. The communication complexities for both initialization and after a link failure are both O(2N). IV. PERFORMANCE EVALUATION Simulation experiments were performed using OPNET Modeler 14.5A to evaluate the performance of the proposed OSDRP scheme. A. Simulation Setup We model a scenario with 24 nodes spanning across 4 PRNs. Each PRNs is consists of 6 nodes which are randomly scattered within a circular region of radius 500 metres. The centres of each PRN are 800 metres apart, forming a square. For ease of explanation and without loss of generality, the probability p of the primary user coming online is assumed to be the same for all channels in each PRN. TABLE I lists the parameters defining the availability of SOPs in each PRN. In our experiments, we vary the probability p between 0.1 – 0.9. Nine sets of traffic flows with different priorities were deployed. The traffic generation is a Poisson process and the mean packet inter-arrival time is exponentially distributed between 10ms to 40ms. The packet size is fixed at 512 bytes.

All traffic flows last for 30 seconds and the OR candidate selection range is 3, 2 and 1 for priorities c1, c2 and c3 respectively. Channel switching delay is assumed to be 1ms. B. Simulation Results To show the effectiveness of our scheme, we compared it to the Delay motivated On-demand Routing Protocol (DORP) [12] and a modified version of DSR that exchanges control information through a common control channel so that it can work in a CRN. The common control channel DSR (ccDSR) also randomly chooses an available channel for communication if the existing channel becomes unavailable. Fig. 4 shows the average number of hops for the different traffic priorities for p between 0.1 – 0.9. The average number of hops taken by the priority c1 is lower than those of other flows. This is because the higher transmission power coupled with larger OR candidate selection range for priority c1 traffic enables it to skip more nodes along the route. The result also shows that the average number of hops is relatively constant under different values of p. Fig. 5 shows the average end-to-end delay for the different traffic flows for p between 0.1 – 0.9. As expected, the higher priority traffic encounters less delay as compared to the lower priority traffic. Furthermore, as p increases, the end-to-end delay also increases due to the increased channel switching and route rediscovery overheads. However, the delay for all cases remains below 120ms which is acceptable for most multimedia broadband applications. To verify the effectiveness of the OSDRP notwithstanding the dynamic nature of the SOP, we customise a scenario where the route taken by all schemes initially goes through PRN2. We then set the average τ for PRN1, PRN3 and PRN4 to 60s and progressively reduced the τ for PRN2 from 45s to 10s in steps of 5s. Fig. 6 shows the average end-to-end delay for different traffic flows when p is set to 0.5. As τ decreases, the 8.00

c1

c2

c3

ccDSR

DORP

7.00 Average number of hops

1

6.00

5.00

4.00

3.00 0.00

0.20

0.40

0.60

0.80

p

Fig. 4.

978-1-4244-5637-6/10/$26.00 ©2010 IEEE

Average Number of Hops vs. p

1.00


140.00

c1

c2

ccDSR

DORP

performance differences get larger as the values of τ decreases.

c3

V. CONCLUSION

End-to-end delay (ms)

120.00

100.00

80.00

60.00

40.00

20.00 0.00

0.20

0.40

0.60

0.80

1.00

p

Fig. 5.

Average End-to-End Delay vs. p

150.00

c1

c2

ccDSR

DORP

In this paper, we have proposed an OSDRP scheme. Our scheme utilizes multi-metric routing together with the consideration of SOP availability to discover the minimum delay – maximum stability route. In addition, TPC and OR are performed to provide differentiated service for different traffic priorities. Simulation results indicated that the proposed OSDRP scheme can provide a lower end-to-end delay for traffic flows at different levels of spectrum availability across different PRNs. The simulation results also show that the OSDRP can provide clear service differentiations for different traffic classes under various scenarios. REFERENCES

c3

[1]


130.00

[2]

110.00

90.00

[3] 70.00

[4]

50.00

30.00 10.00

20.00

30.00

40.00

50.00

[5]

τ

Fig. 6.

[6]

Average End-to-End Delay vs. τ

delay for DORP and ccDSR begin to increase due to the frequent changes in SOP in PRN2. This is due to DORP and ccDSR being unable to react to changes in SOP. The delay encountered by OSDRP is quite constant despite the variation in τ, due to the selection of alternative routes through PRN3 when τ drops below certain level in PRN2. We also verify the effectiveness of the OSDRP in the event that ORTPC features are unavailable. As expected, Fig. 7 shows that there are minimal differentiations for all traffic priorities under the OSDRP. However, the OSDRP still performs better compared to DORP and ccDSR with its ability to select a minimum delay – maximum stability route. The 160.00


150.00

c1

c2

ccDSR

DORP

[7]

[8] [9] [10] [11] [12] [13]

c3

140.00

[14]

130.00

[15]

120.00

[16]

110.00

[17] 100.00 90.00

[18]

80.00 10.00

20.00

30.00

40.00

50.00

P. Gupta and P.R. Kumar, "The Capacity of Wireless Networks," IEEE Transactions on Information Theory, Vol. 46, No.2, Mar. 2000, pp. 388404 S. Haykin, "Cognitive Radio: Brain-Empowered Wireless Communications," IEEE Journal on Selected Areas in Communications, Vol. 23, No.2, 2005, pp. 201-220 K.R. Chowdhury and I.F. Akyildiz, "Cognitive Wireless Mesh Networks with Dynamic Spectrum Access," IEEE Journal on Selected Areas in Communications, Vol. 26, No.1, Jan. 2008, pp. 168-181 J. Jia, et al., "HC-MAC: A Hardware-Constrained Cognitive MAC for Efficient Spectrum Management," IEEE Journal on Selected Areas in Communications, Vol. 26, No.1, 2008, pp. 106-117 X. J. Zhang and Yi Gong, “Adaptive power allocation for multihop regenerative relaying with limited feedback,” IEEE Transactions on Vehicular Technology, vol. 58, pp. 3862-3867, Sept. 2009. X. J. Zhang and Yi Gong, “Adaptive power allocation for regenerative relaying with multiple antennas at the destination,” IEEE Transactions on Wireless Communications, vol. 8, pp. 2789-2794, June 2009. Hang Su and Xi Zhang, "Cross-Layer Based Opportunistic MAC Protocols for QoS Provisionings Over Cognitive Radio Mobile Wireless Networks," IEEE Journal on Selected Areas in Communications, Vol. 26, No. 1, pp. 118--129, January 2008. H. Khalife, et al., "Multihop Cognitive Radio Networks: To Route or Not to Route," IEEE Network, Vol. 23, No.4, 2009, pp. 20-25 S. Biswas and R. Morris, "ExOR: Opportunistic Multi-Hop Routing for Wireless Networks," SIGCOMM Comput. Commun. Rev., Vol. 35, No.4, 2005, pp. 133-144 A. Zubow, et al., "An Opportunistic Cross-Layer Protocol for MultiChannel Wireless Networks," in Proc. IEEE PIMRC 2007, pp. 1-5. E. Rozner, et al., "SOAR: Simple Opportunistic Adaptive Routing Protocol for Wireless Mesh Networks," IEEE Transactions on Mobile Computing, Vol. 8, No.12, 2009, pp. 1622-1635 G. Cheng, et al., "Joint On-Demand Routing and Spectrum Assignment in Cognitive Radio Networks," in Proc. IEEE ICC, pp. 6499-6503. Y. Liu and D. Grace, "Improving Capacity for Wireless Ad Hoc Communications Using Cognitive Routing," in Proc. CrownCom 2008, pp. 1-6. I. Pefkianakis, et al., "SAMER: Spectrum Aware Mesh Routing in Cognitive Radio Networks," in Proc. IEEE DySPAN 2008., pp. 1-5. I. Filippini, et al., "Minimum Maintenance Cost Routing in Cognitive Radio Networks," in Proc. IEEE MASS 2009, pp. 284-293. H. Khalife, et al., "Probabilistic Path Selection in Opportunistic Cognitive Radio Networks," in Proc. IEEE GLOBECOM 2008, pp. 1-5. M. Ma, et al., "A Power-Controlled Rate-Adaptive MAC Protocol to Support Differentiated Service in Wireless Ad Hoc Networks," in Proc. IEEE GLOBECOM 2006, pp. 1-5. A. Goldsmith, et al., "Breaking Spectrum Gridlock With Cognitive Radios: An Information Theoretic Perspective," Proceedings of the IEEE, Vol. 97, No.5, 2009, pp. 894-914

τ

Fig. 7. Average End-to-End Delay vs. τ

978-1-4244-5637-6/10/$26.00 ©2010 IEEE