(CARD) using Cognitive Radio in multi-radio wireless mesh networks

16 downloads 1104 Views 105KB Size Report
existing networking technologies namely, adhoc networks,. Wireless Local Area ... is to dynamically switch to different channels to initiate new connections with ...
Channel Assignment with Route Discovery (CARD) using Cognitive Radio in Multi-channel Multi-radio Wireless Mesh Networks Chittabrata Ghosh and Dharma P. Agrawal OBR Center for Distributed and Mobile Computing Department of ECECS, University of Cincinnati Cincinnati, Ohio - 45219. Email: ghoshc,[email protected]

Abstract: For better spectrum utilization, efficient channel allocation in multi-radio wireless mesh networks has become an active research area. Our proposed CARD algorithm deals with the application of cognitive mesh routers for fixed channel assignments to mesh clients under each router’s domain. The farthest channel assignment by the cognitive radio in mesh routers ensures minimum inter-router and intra-router interference. Initial fixed assignment of channels to clients supporting k-connectivity (k = 3 and 5 sub-channels) shows substantial increase with 15 concurrent transmissions when compared to 4 in case of the CCA scheme with k=3 and 3 channels to choose, i.e., a factor of almost 4 . The improvement in communication delay is about a factor of 80 when compared to SC and a factor of 35 compared to CCA. Keywords- Alpha-beta pruning, wireless mesh networks, cognitive radio

I.

INTRODUCTION

Wireless Mesh Network (WMN) [1], a recent networking technology, is envisioned to improve the performance of all the existing networking technologies namely, adhoc networks, Wireless Local Area Networks, Wireless Private Area Networks and Wireless Metropolitan Area Networks. WMN is similar to that of the adhoc networks in case of selfconfiguration and self-organization, which in turn demands minimum network maintenance. The nodes have the capability of forming a mesh network on an adhoc basis and forwarding packets over multiple hops to the destination nodes not within the communication region of the transmitted nodes. WMN is expected to play a dominant role for future broadband home networks. Cognitive radio is a revolutionary technology that primarily focuses on substantial spectrum efficiency with the aid of advanced spectrum sensing and dynamic channel assignment in licensed bands without actually obtaining a license [2, 3]. The radio is built on a software defined radio and is capable of taking decisions based on its surrounding environment. Opportunistic spectrum sharing of the licensed spectrum [2] by licensed users, hereafter referred to as secondary users (SUs) is one of the key concepts discussed in our research work. For

improved spectrum utilization, a set of sub-channels in an extended C-band (5.925-7.250 GHz) can be assigned to SUs when the sub-channels are not utilized by the primiary users (PUs). Since PUs have higher priority over the secondary users, the former can use the entire C-band whenever necessary. It is the job of the SU to detect the arrival of the primary user also called spectrum sensing and to leave the corresponding sub-channel in the C-band of the recently arrived primary user. Otherwise, the power level of the primary user’s signal, being much higher than that of the secondary user, will entirely corrupt the signal of the latter resulting in interference, called the Primary User Interference (PUI). Again, when an SU uses a particular sub-channel, signals from the adjacent SUs can interfere with this desired user. This interference is called adjacent channel interference (ACI). In case of 802.11 Wireless Local Area Network, the access points generally use a single half-duplex radio with the MAC protocol to support transmissions and receptions over a common channel. Basically, the wireless users compete for the channel using Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA) protocol. But the same strategy can never be applied for wireless mesh networks because of its topology and connectivity constraints. Communication over a single channel can lead to all the mesh routers using the same channel for restoring connectivity. The worst case scenario may occur when mesh routers from neighboring hops of a multi-hop path also use the same channel to ensure connectivity. Therefore, multiple channels are always envisaged for wireless mesh networks. Now we are convinced that multiple channels are essential to ensure better connectivity in wireless mesh networks. But let us walk through the situation of multiple channels being used by wireless mesh routers over a single radio. The obvious problem is to dynamically switch to different channels to initiate new connections with other nodes while restoring existing communication with an ongoing mesh node. Again, switching between channels demands strong synchronization among nodes. Even, slow switching for channels to reduce synchronization requirements will in turn, lead to end-to-end delays. Therefore, to improve the performance of wireless mesh networks in terms of spectrum utilization, multiple radios may be implemented in each mesh router or a mesh client.

This paper deals with multi-radio multi-channel wireless mesh networks and study the throughput and delay performance. The central idea is to increase the capacity and additionally, the throughput of the wireless mesh networks by using multiple radios per node for simultaneous communications over interfaces tuned to multiple free sub-channels, unlike the base single channel assignment scheme. Channel assignment is a major problem when discussing multi-radio multi-channel WMN. One of the existing schemes to solve this problem is the common channel assignment scheme (CCA) [4] where the different radio interfaces are tuned to the same set of channels. But this scheme is inefficient for large number of channels being assigned to a small number of radio interfaces. The inevitable reason behind this is the pronounced intra-channel interference between the radio interfaces. This leads us to a different channel assignment strategy utilizing the relation between channel assignment and channel interference. While multiple radios operate on different channels, two nodes can communicate with each other if they have a radio interface tuned to a common channel. Again, if the nodes have different radio interfaces assigned to same set of frequencies, then it may provide better connectivity but can lead to severe interference. Therefore, strategic channel assignment is vital in overall performance evaluation of multi-radio wireless mesh networks. The other way of looking into the problem is to design a trade-off between connectivity and channel interference. Our proposed CARD scheme has taken care of these two parameters using opportunistic spectrum sharing and updating the set of sub-channels to be used by mesh routers at a particular time instant. The rest of this paper is organized as follows: Section II describes the underlying alpha-beta pruning scheme used in the CARD algorithm. Section III describes about our proposed CARD algorithm. Section IV evaluates the performance measures using our proposed CARD algorithm. Finally, section V draws the conclusions and future research directions. II.

ALPHA-BETA PRUNING

Alpha–Beta technique is used on the game tree to compute the best move by the player and in turn, ignore branches that do not contribute further to the outcome. The advantage of this scheme is that not all the outcomes need to be checked for the best possible move. In other words, if the current outcome results in a worse move when compared to our best possible choice, then the first move that the opposition could make would be our last possible choice. The few assumptions needed for the Alpha-Beta pruning as follows: (i) all paths end up to fixed depth limit, d and (ii) the opponent will always choose the best move. To get into the details of the game, each node has memory enough to store four parameters namely, Alpha, Beta value for itself, current score and, if not terminal nodes, must also store the node address of its recently selected child node. The fourth parameter will allow backtracking down to the terminal nodes. To start with the game, Alpha and Beta values are assigned initial values of –Infinity and +Infinity respectively. Then the

Figure 1. Parent node and its children with d=3.

following steps are followed for the best move to be taken by the player/opponent. (1) Track down to the depth of the game tree (2) Calculate the utility of each terminal node based on certain parameters that better describes its characteristics. The point to be noted over here is that the same parameters should be used to compute the utility function for all the nodes. in its parent and also replace the score at the parent node with this new score. (3) Propagate the Alpha and Beta values along with the path traced based on the following options: •

If the opponent makes the move to be backtracked:

(i) If the current score obtained from its child node is less than the stored score at its parent, store the path from the bottom and the Beta value in its parent node and replace the stored score with this new score. (ii) If this newly stored score is less than the Alpha value stored in its parent node, prune all the branches under this node. If greater than the Alpha value in its parent node, replace the Alpha value with this score and proceed with the next child node and sending the Alpha and Beta values down. If no child exists, these values are propagated up the tree and the Alpha value becomes the MIN score as shown in Fig. 1. •

If the player makes the move to be backtracked:

(i) If the current score of the child node is greater than the score at its parent, replace the stored parent’s score with this new score and store the path from the bottom and the Alpha value in its parent. (ii) If the newly stored score is greater than the Beta value in the parent node, prune all the existing child nodes and backtrack the parent’s Alpha and Beta values up the tree. If less than the Beta value at the parent node, replace the Beta value with this new value and proceed to the next child and sending the Alpha and Beta values down. If no child exists, these values are propagated up the tree and the Beta value becomes the MAX score as shown in Fig. 1. When the search is complete, the Alpha value at the top node gives the minimum score which is guaranteed to attain if the path stored in the parent node is backtracked. The entire algorithm for the implementation of the Alpha-Beta pruning is

Figure 2. Algorithm to implement Alpha-Beta pruning

explained in Fig. 2. We have used this algorithm for our CARD algorithm explained in the next section. III.

PROPOSED CARD ALGORITHM

Our proposed CARD algorithm is based on the hierarchical mesh networks with k-connectivity (node with k mutually independent connected radio links). Each mesh router is equipped with a cognitive radio which periodically scans and detects the free channels [5] in five different sub-bands (0.265 GHz each) of the entire C band). These free channels are stored in each mesh router’s free channel pools (FCP), each pool having free channels from a single sub-band. In our scheme, each channel selection from the FCP by a mesh router is broadcasted to its one hop mesh routers and in turn to its two hop routers. This assures that the same set of channels is not assigned to mesh clients of adjacent mesh routers, till two hops. On the other hand, this channel assignment scheme can lead to primary user interference and substantial adjacent channel interference. Therefore, our proposed CARD scheme incorporates cognitive routers with its two-fold strategy that helps in reducing intra-hop interference to a great extent and as well as the inter-hop interference: i) each channel request from the same hop mesh clients is assigned from a different sub-band and (ii) additional channel requests within the same hop

should be based on farthest channel assignment. The reason for this assignment strategy is that usually all mesh nodes request for one free sub-channel which, if assigned from different sub-bands will not interfere with the transmissions from the neighboring nodes. On the other hand for subsequent channel requests, the cognitive radio must assign the farthest channel with respect to the already assigned channel in the same sub-band as that of the requesting node. The control messages are sent over a common control channel but data transfer takes place over assigned free sub-channels. First, the one-hop mesh clients have their parent node (PN) discovered as the mesh router with one radio interface connected to the PN and channels assigned as per the above mentioned strategy. The possible detection of the PN is based on the received broadcast packet by the mesh node with hop count =1. Now if two one hop members are within each other’s communication range and have the same PN, then they request for and share a common channel for their second radio interface as shown by the dashed lines in Fig. 3. This common channel is decided by the router and selected from the subband different from the ones used by the sharing nodes’ radio interfaces. This helps in constructing the mesh with kconnectivity with k=2. Now, the two-hop clients under each of these one-hop clients can request for channels. On request, each one-hop members are discovered as PN for their corresponding one-hop neighbors and channels assigned similarly as explained above.

function is multiplied by its access rate, if MIN nodes are the terminal nodes and divided by the access rate if MAX nodes are the terminal nodes. This manipulation is performed to avoid giving access to a particular parent node having its terminal node with maximum number of packets. The division minimizes the chances of the same MIN node being selected by a MAX node and multiplication increases the chances of a MAX node when MIN node is the one to choose its move.

Figure 3. Initial channel assignment to mesh clients by the parent node

Thus, all data packets have to be routed through the discovered PN. For future communication, one mesh client at a time, the third radio interface of the PN requests for the same channel as that of the desired member node interface. This way, the k-connectivity (k=3) mesh network with fixed channel assignment is created through our proposed scheme as shown in Fig. 4. Additionally, the CARD algorithm also discovers the route through intermediate PNs from each and every mesh router till its two-hop neighbors. The Alpha-Beta pruning algorithm has been modified for our research in the following manner: the utility function is calculated by the parent node for all the terminal nodes under it. This function is a ratio of the total number of packets under this parent node by the number of forwarding packets for the concerned terminal node. The intention for this calculation is to minimize the utility function for a node having maximum number of packets to be forwarded. Again, when the utility function is finally distributed to the terminal nodes, the utility

Another modification made in the Alpha-Beta pruning algorithm is done to utilize the multi-radio facility. The parent node checks for the destination address from the incoming packets. If this address is related to its own hierarchical game tree, it stores the packets and waits for its chances to be selected as a MAX or a MIN player. If this address is not within its own hierarchy, the parent node utilizes its different radio interface tuning its radio interface to a different channel and communicates to the different adjacent hierarchy. In cases for different destination addresses other than its own hierarchy, then the parent node does not forward the packets to its own parent node within its own hierarchy. After storing the packets within its buffer, it tunes to a different channel to a different mesh node in a different hierarchy and forwards packets to that node. The improvements achieved have been shown in our simulation results. IV.

SIMULATION RESULTS

In this section, we have studied our proposed CARD scheme using network simulations. The single channel assignment scheme has served as a baseline for comparison purposes. We have also included CCA scheme [5] as the comparative algorithm for channel assignment. The entire simulation of the hierarchical wireless mesh networks has been performed using GloMoSim [7]. As stated earlier, we have used throughput and average delay parameters for evaluation and comparison of CARD scheme with other mentioned schemes. The entire simulation has been carried out with 50 randomly placed nodes in a 650m × 650m area. The transmission range is assumed to be 200 m which in turn leads us to an interference range of 525m. Our simulation results are based on two distinct categories: (i) Evaluation based on connectivity and topology and (ii) Single hop performance for multi-radio mesh networks. The topological performance in wireless mesh networks has been studied in our research using the maximum concurrent transmission as the deciding parameter. The reason for this choice is to verify efficiency of the radio interfaces to switch between different free sub-channels from the free channel pool. The more the number of successful transmissions, the better is the utilization of the spectrum and hence, throughput and also the capacity of wireless mesh networks increases. Fig. 5 shows the comparison between the CARD and the CCA algorithm using three channels. The linear increase in the curve for CARD algorithm shows the efficiency of our algorithm over CCA with gradual increase in the number of radio interfaces per node. As shown in Fig. 3, CARD algorithm has been

Figure 4 Assignment of sub-channels to ensure k-connectivity, k=3.

Figure 5 Comparison in number of concurrent transmissions for CARD and CCA algorithms for 3 channels per node.

successful in 13 concurrent transmissions when compared to 5 in CCA. Similarly, Fig. 6 indicates better spectrum utilization by CARD algorithm over CCA when considering 12 channels to be switched by progressively increasing number of radio interfaces per node. The distinct feature to be observed in this figure is that the linear increase in the number of concurrent transmissions saturates for CARD algorithm after 4 radio interfaces per node. This indicates that the adjacent channel interference and intra-channel interference affects in concurrent transmissions after adding more number of radio interfaces. Therefore, as seen from Fig. 6, the trade-off for designing multi-radio wireless mesh networks can be limited to 5 radio interfaces per node when 12 free sub-channels can be utilized for concurrent transmissions.

Figure 6. Comparison in number of concurrent transmissions for CARD and CCA algorithms for 12 channels per node.

Figure 7 Throughput comparison of CARD algorithm with CCA and Single channel base cases using 3 channels.

Fig. 7 gives a detailed comparison of our scheme when compared to CCA and the single channel case. CARD algorithm provide a substantial improvement in throughput calculations – up to a factor of 4 with 3 free sub-channels and 3 radio interfaces per node. Similar throughput improvements can also be achieved using increasing number of channels with 3 radio interfaces as shown in Fig. 8. The point to be noted over here is that the increase in throughput is minimal with increase in the number of radios per node when compared to that in Fig. 7 for 3 channels and 3 radio interfaces per node. The reason for this is accounted for the same adjacent and intra-channel interference. The average delay performance comparison has been shown in Fig. 9. The curves show a distinct improvement in average delay- about a factor of 2 when compared to single channel and around a factor of 1.75 when compared to CCA with 2 radios. The reason for this minimal delay is due to minimal number of transmission needed and with increasing switching technique used in CARD algorithm. The same can be observed in Fig. 10 with 10 channels and 6 radios but better results are obtained for average delay because of increasing number of radios switching between 10 sub channels from the free channel pool.

Figure 8 Throughput comparison of CARD algorithm with CCA and Single channel base cases using 10 channels.

V.

CONLUSION

In this paper we have dealt with cognitive radio based fixed channel assignment in multi-radio wireless mesh networks. Extensive set of simulations emphasize the efficacy of the CARD scheme in multi channel interference reduction with multiple number of radio interfaces per node.

Our future work will concentrate on theoretical performance characterization and evaluation of the CARD scheme in a multi-radio mesh networks. Dynamic channel assignment must also be considered in our research work.

The CARD algorithm has proved to be efficient when compared to CCA or the single channel case when considering the topological characteristics like maximum concurrent transmissions. Again while considering the single hop multiradio wireless mesh networks, our algorithm proves to be efficient when considering the average delay and throughput parameters.

REFERENCES [1] [2]

[3]

[4]

[5]

[6]

[7]

Figure 9 Average delay performance comparison for CARD and CCA algorithms using 3 channels.

Figure 10 Average delay performance comparison for CARD and CCA algorithms using 3 channels.

I. F. Akyldiz and X. Wang, “A survey of wireless mesh networks,” IEEE Comm. Mag., Vol. 43, no. 9, Sept.”05, pp. S23-S30. S. Haykin, “Cognitive radio: brain-empowered wireless communications," IEEE Journal Selected Areas in Communications, vol. 23, no. 2, pp. 201 - 220, Feb. 2005. S. Nandagopal, C. Cordeiro, and K. Challapali, “Spectrum Agile Radios: Utilization and Sensing Architectures,” Proceedings of IEEE DySPAN2005, Baltimore, USA, Nov 8-11, 2005. A. Adya, P. Bahl, J. Padhye, A. Wolman and L. Zhou, “A multi-radio unification protocol for IEEE 802.11 wireless networks,” Proc. BroadNets, 2004. T. Banerjee, C. Ghosh and D. P. Agrawal, “Wireless sensor network based dynamic channel selection in cellular communication”, CROWNCOMM ’06, Greece M. K. Marina and S. R. Das, “A topology control approach for utilizing multiple channels in multi-radio wireless mesh networks”, IEEE 2005, pp. 412-421. X. Zeng, R. Bagrodia and M. Gerla, “GloMoSim: A library for parallel simulation of large-scale wireless networks,” Proc. 12th PADS, Banff, Canada, May’98, pp.154-161.