452
Int. J. Communication Networks and Distributed Systems, Vol. 2, No. 4, 2009
Cognitive networking of large scale wireless systems Liang Song* and Dimitrios Hatzinakos Department of Electrical and Computer Engineering, University of Toronto, 10 King’s College Road, Toronto, Ontario, M5S 3G4, Canada E-mail:
[email protected] E-mail:
[email protected] *Corresponding author Abstract: We propose the concept of cognitive networking for large-scale wireless systems, which opportunistically utilises network resources including both spectrum bandwidth and radio availability. Both types of resources cannot be predetermined in large-scale wireless systems, due to various reasons such as interferences and dynamic traffic load. The proposed technology not only establishes dynamic wireless networks, but also provides for reliable network quality of services (QoS). The supporting network architecture, embedded wireless interconnect (EWI), is proposed to implement the cognitive networking concept and supply an effective application-programming interface for large-scale data management systems. Two example applications are presented, including wireless mesh networks for broadband wireless internet access and wireless sensor networks for target tracking. Major advantages of the technology are further discussed. We suggest that the performance of the proposed system would improve with larger network scale and the implementation complexity could be independent of the network scale. Keywords: cognitive networks; large-scale wireless systems; network architecture; data management; embedded wireless interconnect; EWI. Reference to this paper should be made as follows: Song, L. and Hatzinakos, D. (2009) ‘Cognitive networking of large scale wireless systems’, Int. J. Communication Networks and Distributed Systems, Vol. 2, No. 4, pp.452–475. Biographical notes: Liang Song received his Bachelor in Electrical Engineering from Shanghai Jiaotong University, China, in 1999, MS in Electronic Engineering from Fudan University, China, in 2002 and PhD from the Department of Electrical and Computer Engineering, University of Toronto, Canada, in 2005. His research and development experience includes the areas of wireless communications and networking and signal processing. His publications include the authoring of about 30 research papers in technical journals and conference proceedings, while he is also the leading inventor of nine patent pending innovations in the areas of wireless communications and networking. He has been the Founder and President of OMESH Networks Inc. (Toronto, Canada) since 2008. Dimitrios Hatzinakos received his Diploma from the University of Thessaloniki, Greece, in 1983, MASc from the University of Ottawa, Canada, in 1986 and PhD from Northeastern University, Boston, MA, in 1990, all in Electrical Engineering. In September 1990, he joined the Department of Electrical and Computer Engineering, University of Toronto, where now he
Copyright © 2009 Inderscience Enterprises Ltd.
Cognitive networking of large scale wireless systems
453
holds the rank of Professor with tenure. His research interests are in the areas of multimedia signal processing, wireless communications and networking. He is author/co-author of more than 200 papers in technical journals and conference proceedings and he has contributed to eight books and three patents in his areas of interest. His experience includes consulting through Electrical Engineering Consociates Ltd. and contracts with United Signals and Systems Inc., Burns and Fry Ltd., Pipetronix Ltd., Defense Research Establishment Ottawa (DREO), Nortel Networks, Vivosonic Inc. and CANAMET Inc.
1
Introduction
Large-scale wireless systems have been proposed for many attractive applications, such as wireless mesh infrastructure for broadband access [e.g., municipal WiFi (802.11) mesh] (Song, 2008), wireless sensor networks (Akyildiz et al., 2002) for industrial or scientific monitoring/controlling, as well as emergency ad hoc wireless communications. Results from theoretical computation (Gupta and Kumar, 2003) show that the transport capacity of wireless networks is determined by the order of O( B ⋅ N ), in terms of bit·metre/sec, where B is the spectrum bandwidth and N is the number of wireless nodes/radios. In large-scale wireless networks, both B and N can be large, due to the possible utilisation of unlicensed bands and the great number of wireless nodes/radios. However, practical solutions that can achieve the above theoretical prediction remain unavailable. In traditional wireless networking, it is often assumed that wireless network resources can be predetermined. More specifically, in the protocol stack, the media access control (MAC) layer is utilised to establish point-to-point wireless linkage with predetermined spectrum allocation. A predetermined network topology can then be established, over which the network-layer routing protocols can relay data packets from source to destination over multiple wireless hops. However, difficulties can arise in large-scale wireless networks, since both types of network resources including spectrum bandwidth and radio availability cannot be usually predetermined. The random spectrum availability is determined by uncoordinated interferences, especially those prevailing in unlicensed bands, e.g., industrial, scientific and medical (ISM) bands. The random radio availability is determined by dynamic traffic load (network congestions) and other factors such as radio failure. As a result, realising the applications in large-scale wireless systems can be very challenging. For example, the dataflow quality of service (QoS), such as throughput, end-to-end delay and delay variance, could degrade fast with the number of wireless hops (Draves et al., 2004), due to link failures and/or network congestions. Moreover, the performance and complexity of network protocols usually does not scale with network size. In this paper, we propose the concept of large-scale cognitive wireless networking to meet these challenges. As illustrated in Figure 1, the network opportunistically utilises network resources including both spectrum bandwidth and radio availability, so as to realise reliable communications in large-scale wireless networks. Different from traditional networking methods, it can establish a dynamic wireless network without predetermined topology and spectrum allocation. Since local random factors in the network are opportunistically utilised, it also produces reliable overall network performance. The proposed cognitive networking concept is implemented into the
454
L. Song and D. Hatzinakos
network architecture of embedded wireless interconnect (EWI) (Song and Hatzinakos, 2007a), which can provide flexible network abstractions with maximised modularity and supply an effective application programming interface for large-scale data management systems. We further present two application examples: the wireless mesh infrastructure for broadband access and the target-tracking wireless sensor networks. By the proposed cognitive networking, we suggest that the system/network performance would improve with larger network scale, whereas implementation complexity could be independent of the network scale. Figure 1
2
Cognitive wireless networking concept
Related works
In system engineering, a great deal of investigation has been concentrated on the application programming interface for distributed computations, i.e., the middleware design (Hadim and Mohamed, 2006). It has been suggested by many important studies (Madden et al., 2005; Bonnet et al., 2001) that application-specific network design is necessary, in dealing with the requirements of application QoS, network scalability and node complexity constraints. However, the dependence on holistic design might deter the potential mature of middleware technologies, since the design modularity could not be well maintained. In network engineering, most investigations have originated from general-purpose mobile ad-hoc networks (MANET) (Conti and Giordano, 2007). Although more than a decade’s research has been dedicated to MANET, very limited implementations are available in engineering practices. A few pioneering studies have been focused on architectures deviating from MANET (Lilien et al., 2007; Scott et al., 2006), which are usually trimmed to special applications, rather than aiming to provide system designers with re-usable network abstractions. In radio engineering, research so far has been concentrated on the concept of cognitive radio. The concept was firstly proposed by Mitola (1999), in which the radio is envisioned with the intelligence to exploit ambient environment for user-centric communications. Since regulation authorities, e.g., Federal Communications Commission
Cognitive networking of large scale wireless systems
455
(FCC) in the USA, have recognised the inefficiency of legacy static spectrum allocation, dynamic spectrum access (Akyildiz et al., 2006; Haykin, 2006) has captured the most research attention. Existing knowledge in cognitive-radio networking still remains at a preliminary stage. Therefore, our cognitive-network proposal can contribute to a further step, where the network intelligently (dynamically) access both spectrum bandwidth and radio availability. Related works with the EWI architecture have appeared in Song (2008), Song et al. (2008) and Song and Hatzinakos (2007a, 2007b, 2008). The concept of EWI was firstly proposed in some application-specific studies of wireless embedded systems (Song and Hatzinakos (2007a, 2007b). For instance, a protocol for target tracking in wireless sensor networks (Song and Hatzinakos, 2007a) has been the first example to demonstrate the EWI architecture. EWI was further utilised to develop a wireless mesh infrastructure for broadband internet access (Song, 2008; Song et al., 2008). It has also been shown that reliable multi-hop QoS can be supported for real-time services (e.g., voice/video) in large-scale wireless networks (Song and Hatzinakos, 2008).
3
Network architecture
We propose to implement the cognitive networking concept for large-scale wireless systems, with the network architecture of EWI. The key architectural differentiation is based on the abstract wireless linkage, where wireless links are now redefined as arbitrary mutual cooperation among a set of neighbouring (proximity) wireless nodes. In comparison, traditional wireless networking relies on point-to-point ‘virtual wired-links’ with a predetermined pair of wireless nodes and allotted spectrum. The architectural diagram is illustrated in Figure 2 and specific details are presented below. Figure 2
EWI architecture
456
L. Song and D. Hatzinakos
3.1 Primary principles Three primary principles can be summarised from the network architecture. 1
Functional linkage abstraction: Based on the definition of abstract wireless linkage, ‘wireless link modules’ are the building blocks for individual wireless nodes, which can enable different types of abstract wireless links. According to the functional abstractions, categories of wireless link modules include: broadcast, unicast, multicast and data aggregation, etc. This principle results in two hierarchical layers as the architectural basics, including the system layer and the wireless link layer, respectively. The bottom wireless link layer supplies a library of wireless link modules to the upper system layer; the system layer organises the wireless link modules to achieve effective application programming.
2
Opportunistic wireless links: In realising the cognitive wireless networking concept, both the occupied spectrum and the participating nodes of an abstract wireless link are opportunistically decided by their instantaneous availabilities. This principle decides the design of wireless link modules in the wireless link layer. Therefore, the system performance should improve with larger network scale, since higher network density introduces the diversity in the opportunistic formation of abstract wireless linkage. This is also provided by the radio implementation to be elaborated later in Section 4.
3
Global QoS decoupling: Global application or network QoS is statistically decoupled into local requirements of cooperations in proximity wireless nodes, i.e., wireless link QoS. More specifically, by decoupling global application-level QoS, it allows the system layer to better organise the wireless link modules that are provided by the wireless link layer. By decoupling global network-level QoS (e.g., throughput, end-to-end delay and delay jitter), the wireless link module design (e.g., of unicast or multicast) can achieve the global QoS requirements. Therefore, based on the provided wireless link modules, the implementation complexity at individual nodes should be independent of the network scale.
3.2 Architectural interfaces Wireless link modules provide system designers with reusable open network abstractions, where the modules can be individually updated, or new modules may be added into the wireless link layer. The high modularity and flexibility could be essential for middleware or application developments. EWI is also an organising-style architecture, where the system layer organises the wireless link modules (at the wireless link layer); and peer wireless link modules can exchange module management information by padding packet headers to the system-layer information units. Five types of wireless link modules are illustrated in Figure 2, including broadcast, peer-to-peer unicast, multicast, to-sink unicast and data aggregation, respectively. Other arbitrary types of modules may be added, establishing other types of abstract wireless links, without limitation. For example, the broadcast module simply disseminates data packets to surrounding nodes. The peer-to-peer unicast module (Song and Hatzinakos, 2008) can deliver data packets from source to destination over long distance. The
Cognitive networking of large scale wireless systems
457
multicast module sends data packets to multiple destinations, as compared to peer-to-peer unicast. The to-sink unicast module (Song and Hatzinakos, 2007b) can be especially useful in wireless sensor networks, which utilises higher capabilities of data collectors (or sinks), so as to achieve better data delivery. The data-aggregation module (Song and Hatzinakos, 2007a) opportunistically collects and aggregates the context related data from a set of proximity wireless nodes. Shown in Figure 2, two service access points (SAP) are defined on the interface between the system layer and the wireless link layer, which are wireless link SAP (WL SAP) and wireless link management entity SAP (WLME SAP), respectively. WL SAP is used for the data plane, whereas WLME SAP is used for the management plane. The SAPs are utilised by the system layer in controlling the QoS of wireless link modules. Further details about the wireless link modules and the SAPs are elaborated in Section 5.
4
Radio implementations
By the proposed concept of large-scale cognitive networking, both the operating spectrum and the participating nodes are opportunistically decided by their availability in the formation of an abstract wireless link. This raises two questions: 1
How to locate the spectrum resource for an abstract wireless link?
2
How to detect the initiation of an abstract link from neighbouring nodes on certain spectrum?
The following radio implementations can provide the solution.
4.1 Basic propositions Previous works in cognitive radio have been focused on opportunistic spectrum access (Zhao and Sadler, 2007), which locates the ‘white-space’ in spectrum for communications. We present two basic propositions for the radio in large-scale cognitive wireless networks, which are the two procedures of ‘sensing’ and ‘polling’, respectively: 1
The radio can opportunistically sense available spectrum resource, so that the selected spectrum usage will not be interfering with other co-existing wireless communications.
2
The radio can opportunistically poll one or more other neighbouring radios on the selected spectrum, so as to realise certain types of local cooperations.
Within the EWI architecture, a wireless node with the proposed radio can initiate an abstract wireless link, i.e., certain types of local cooperations among proximity wireless nodes. Both the wireless nodes and the occupied spectrum are therefore opportunistically decided by their availability. The initiated abstract wireless link would not be interfering with other wireless communications, as protected by the first cognitive proposition.
458
L. Song and D. Hatzinakos
4.2 An implementation Based on the above two propositions, we further describe a prototype implementation of the radio for large-scale cognitive wireless networks. In the example implementation, the radio can access a group of predetermined data channels, where every data channel is also associated with two distinctive frequency tones, i.e., one sensing tone and one polling tone. The radio hardware is therefore comprised of two transceivers, which are the tone transceiver and the data transceiver, respectively. When the radio initiates an abstract wireless link, it senses for an available channel, with the vacant data channel and the vacant sensing/polling tones. The radio then broadcasts the polling tone associated to the selected channel, to poll surrounding nodes. The surrounding nodes can decide autonomously whether to join in the initiated abstract wireless link on the data channel, when they detect the rising edge of the associated polling tone. On joining in an abstract wireless link, the radio of the surrounding nodes also broadcasts the associated sensing tone. As such, both sensing and polling tones protect abstract wireless links from spectrum interferences. Although the use of frequency tones appeared in numerous MANET MAC protocols (Hass and Deng, 2002; Tobagi and Kleinrock, 1975), we demonstrate here the first implementation over multiple channels, in the context of large-scale cognitive wireless networks. The radio is implemented by prototyping, based on a stack of two radio boards that are used respectively as the tone transceiver and the data transceiver. The two radio boards are separated by a half-wavelength distance. The demonstration and experiment results with this prototype appeared in Song and Hatzinakos (2008), Song et al. (2008) and Song (2008). Now assume that there are totally N data channels, as differentiated by frequency, time or spreading codes. Let Sn and Pn denote respectively the sensing and polling tones associated to one data channel n (1 ≤ n ≤ N ). A state diagram of the described radio implementation is shown in Figure 3. Figure 3
Radio state diagram
Cognitive networking of large scale wireless systems
5
459
Wireless link modules
A few embodiments of wireless link modules are presented, which include the broadcast module, the peer-to-peer unicast module and the data-aggregation module. These embodiments are further used in the application examples of Section 6. Although other wireless link modules, e.g., the to-sink unicast and multicast modules, are omitted here, they could be important for other applications. We first introduce the context-based network address and the cost of delivery criterion. A state diagram of the wireless link layer is then shown to provide for a library of wireless link modules, after which the embodiments of the individual modules are described.
5.1 Network address In traditional computer networks, network addresses are based on symbols, e.g., internet protocol (IP) or MAC address. There have also been one-to-one mapping between network addresses and networking devices, where an address can be globally unique as assigned to one networking device. In the proposed large-scale cognitive wireless networks, the network addresses are based on context, e.g., of location coordinates, application-specific address, or logic address, etc. In respect of different contexts, a wireless node (networking device) can acquire more than one addresses. Meanwhile, more than one nodes, e.g., in proximity, can also share the same context address, where these nodes can be considered as identical in the network for the corresponding context. In the design of wireless link modules, a cost of delivery criterion is defined, being applied to context-based network addresses. Given two arbitrary nodes n and m , Cn and Cm denote their addresses related to certain context, respectively. According to the criterion, an estimated cost of delivery cn ,m can be directly calculated from Cn and Cm , which indicates the expected cost of sending one data packet (information unit) from n to m or vice versa. The cost of delivery criterion is necessary for enforcing the principle of decoupling network-wide (global) QoS into proximity-node cooperations. In location-centric networks [e.g., Song and Hatzinakos (2007a)], where wireless nodes are aware of their own locations, e.g., by global position systems (GPS) or by triangulation estimations, the network address of one node n (or m ) is solely decided by the node n (or m ) location coordinates. Therefore, the cost of delivery between n and m , cn ,m can be readily calculated by simple geometry, being defined as the distance between n and m , i.e., cn ,m =| Cn − Cm | .
In data-centric networks [e.g., Intanagonwiwat et al. (2002)], the network address is decided by application-specific context. The cost of delivery cn ,m can be the application data gradient and can be assumed as a monotonically increasing function of the distance between n and m , since the correlation of application data decreases with spatial separation.
460 Table 1
L. Song and D. Hatzinakos Primitive functions at the service access points
Cognitive networking of large scale wireless systems
461
In data-collecting or fusion networks, e.g., wireless sensor networks, the sink (or data collector) can broadcast a number of identity advertisement packets, which is thereon flooded in the network, by broadcasting. Every node can count the average smallest number of hops to the sink, on receiving the advertisement packets. The count number can be used as the cost of delivery for one node. A similar approach also appeared in Cao and Abdelzaher (2006), where the logic address, i.e., a vector of the estimated distance or hop number to a group of anchors, is maintained for every wireless node. New nodes joining in the network can estimate its own logic address by acquiring such addresses of neighbouring nodes.
5.2 A library of wireless link modules Figure 4 illustrates a state diagram of the wireless link layer, where a library of wireless link modules is provided. Other types of modules may also be added in the library. Table 1 gives a list of primitive functions related to these modules, at the service access points (i.e., WL SAP and WLME SAP in Figure 2). System layer can therefore control wireless link modules by calling their defined primitive functions. Figure 4
Wireless link layer state diagram
The wireless link layer remains in the IDLE state, when no wireless link module is invoked. The pair of primitives, Module.start( ) and Module.sleep( ), are utilised to switch between the IDLE state and the SLEEP state, i.e., a further power-saving mode. While in the IDLE state, the wireless link layer monitors if there are new abstract wireless links being initiated. On detecting an initiation from any one of the surrounding nodes, the wireless link layer transfers from the IDLE state to the Module Request state, where the primitive Module.request( ) is invoked. It provides the system layer with the control of wireless link activities. When the primitive Module.response( ) is received, the
462
L. Song and D. Hatzinakos
wireless link layer either transfers back to the IDLE state or joins in the abstract wireless link, as decided by the response result. On the other hand, on receiving a command to initiate an abstract wireless link: e.g., by the primitives Broadcast/UnicastP. send(...) or Aggregate. start(...), the wireless link layer transfers from the IDLE state to the spectrum sensing state. A corresponding wireless link module can be started, after available spectrum resource has been located. The implementations of ‘spectrum sensing’ and ‘module detecting’ are specified previously in Section 4. By the ‘spectrum sensing’, available spectrum resource is located opportunistically for the abstract wireless link operation, without interfering with other wireless communications. By the ‘module detecting’, an abstract wireless link is set up among a group of proximity wireless nodes, where the wireless nodes are opportunistically determined by their autonomous availability.
5.3 Broadcast module A state diagram of the broadcast module is illustrated in Figure 5, where the ports A, B and C match those in Figure 4. Here, the broadcast module implements the broadcasting and receiving of data packets, where the source initiates a broadcast wireless link by the primitive Broadcast.send(...) and receivers use the primitive Broadcast.indicate(...) to forward data to their system layer. Figure 5
Broadcast module state diagram
The state diagram in Figure 5 should be self-explainable. We also note that some broadcast control information is sent before the data, including the source address and the QoS control. The QoS class, as denoted by the Priority parameter in the primitive function Broadcast.send(...), can be decided by broadcasting range and/or latency requirements. More specifically, broadcasting range indicates the maximal required cost of delivery from the source node by using the broadcast module and broadcasting latency
Cognitive networking of large scale wireless systems
463
indicates the data urgency. The combination of the two gives the optimal power and rate control (resource allocation) for minimising energy consumption cost. Further developments of the module may also include multihop broadcasting, such as the methods presented in Scaglione and Hong (2003).
5.4 Peer-to-peer unicast module A state diagram of the peer-to-peer unicast module is illustrated in Figure 6, whereas the analysis, simulations and experiments are reported in Song and Hatzinakos (2008). The peer-to-peer unicast module can send unicast data packets from source to destination. By the cognitive wireless networking, data packets travel along opportunistically available paths with opportunistically available spectrum on every hop. Figure 6
Peer-to-peer module state diagram
464
L. Song and D. Hatzinakos
The source and relays can use the primitive UnicastP.send(...) to initiate an unicast wireless link. And the destination uses UnicastP.indicate(...) to forward received data to the system layer. Shown in Figure 6, unicast control information includes the source address, the destination address, the data-sender (current relay) address and the QoS control. On receiving those control information, the wireless nodes joining in the unicast wireless link, i.e., relay candidates, locally calculate a time-delay parameter Ta, which can be decided by the wireless channel status and the cost of delivery to the destination address (by the cost of delivery criterion). In principle, the relay candidate with satisfying wireless channel status and the smallest cost should obtain the smallest Ta. As such, this preferred node broadcasts a relay announcement packet and serves as the next-hop relay by receiving the data. Other relay candidates can exit the unicast module on receiving the relay announcement. After having received the data packet, the next-hop relay can invoke the primitive UnicastP.send(...) for further relaying. Otherwise, if the destination has been reached, the primitive function UnicastP.indicate(...) is invoked, which forwards the received data to the system layer. Moreover, if the next-hop relay is not found, the current relay (data-sender) can invoke the primitive UnicastP.send(...) again to re-send the data. A resolution process may be further employed, if more than one relay candidates obtain the same smallest Ta. The QoS class of the peer-to-peer unicast module, as denoted by the parameter Priority in the primitive UnicastP.send(...), can be decided by the data flow QoS levels of throughput, end-to-end delay and delay variance (jitter). These global network QoS can be supported in the unicast module design, by transmitting power control (or joint power and rate control) (Song and Hatzinakos, 2008). More specifically, throughput can be independent of the number of wireless hops; end-to-end delay and delay jitter increase linearly with the number of wireless hops; delay jitter also decreases to zero with higher network density, by which the network performance can be engineered as arbitrarily stable.
5.5 Data aggregation module A state diagram of the data-aggregation module is illustrated in Figure 7. The data-aggregation module collects context-related data packets from proximity nodes, which may be used in the applications of context-aware search, mobile computing and wireless sensor networks. The aggregation node can use the primitive Aggregate. start(...) to initiate a data-aggregation wireless link and collect context-related data packets. It can also terminate the data-aggregation wireless link with the primitive Aggregate.stop(...), when enough data has been collected. The primitive Aggregate.indicate(...) is further used by the aggregation node to forward the received data packets to the system layer. The wireless nodes joining in a data-aggregation wireless link, other than the aggregation node, can use the primitive Aggregate.send(...) to transmit their context-related data to the aggregation node and use the primitive Aggregate.request(...) to supply the system layer with the requested context. Shown in Figure 7, the aggregation control information can include the aggregation address (i.e., of the aggregation node) and the context request, e.g., temperature, humidity or other measurement metrics related to certain events. The QoS class parameter Priority, being used in the primitive function Aggregate.send(...), indicates the specific context-related data quality. Intuitively, data with higher quality should obtain higher Priority for being sent to the aggregation node. Therefore, based on Priority, a time-delay
Cognitive networking of large scale wireless systems
465
parameter Tp can be calculated at the nodes joining in the aggregation, which is a backoff period inversely proportional to the data quality. If the radio channel is busy after the backoff period Tp, i.e., another context-related data packet with higher Priority is being transmitted by another node, the current data at the local node backoffs for another fixed time-period corresponding to the transmission of one data packet. As such, the context-related data packets will be broadcasted to the aggregation node sequentially, according to the data quality. Once enough data has been collected, the aggregation node broadcasts an ‘aggregation stop’ control packet to terminate the data-aggregation wireless link. Therefore, by data-aggregation wireless links, the best-quality data is collected, where the quality is defined by the requested context. The optimisation between application QoS and network energy consumption can be resolved at the system layer (Song and Hatzinakos, 2007a) by a control knob. Figure 7
Data-aggregation module state diagram
466
6
L. Song and D. Hatzinakos
Application examples
The proposed cognitive wireless networking and the EWI architecture can provide for an effective application-programming interface for large-scale wireless systems. We present here the primitive programming with the primitive functions of wireless link modules in Table 1. Other details of application examples appear in the related works (Song and Hatzinakos, 2007a; Song, 2008; Song et al., 2008). The objective is also to inspire future middleware developments for large-scale cognitive wireless systems. Due to the modularity and flexibility of the defined abstractions, multiple wireless link modules can be individually added, removed or updated. The pseudo-code programs are simplified for giving better illustrations, which use the ‘switch’ statement of C in dealing with different messages.
6.1 Wireless mesh networks for broadband access Wireless mesh networking provides for attractive solutions to telecommunication services, where broadband wireless access could be achieved, by exploiting the advantages of frequency re-use and lower engineering costs. For example, municipal WiFi (ANSI/IEEE 802.11 Standard, 2003) mesh is rolling out in a number of cities across North America, providing wireless internet access services, e.g., web-surfing and emails. Illustrated in Figure 8, typical wireless mesh infrastructure is comprised of a number of gateways and a much larger number of mesh stations. More specifically, gateways are the networking devices on the border of wired internet backbone and wireless networks, which are connected to internet backbone by network cables or fibre; mesh stations are the networking devices without any network cable/fibre connections, but provide mobile users with standard wireless access (e.g., by WiFi) and relay user traffics to the gateways and internet thereafter. Figure 8
Wireless mesh infrastructure
Cognitive networking of large scale wireless systems
467
However, state-of-the-art wireless mesh networking technologies can suffer from fast performance (throughput and end-to-end delay) degradation, when the number of wireless hops increases. It is also attractive to reduce the power consumption of mesh stations, so that they may be powered by cost-effective solar cells, which removes any cable attachments and minimises the engineering cost of mesh station deployments. The aforementioned limitations on network scalability and power consumption, can be resolved (Song and Hatzinakos, 2008; Song, 2008) by the cognitive wireless networking. More specifically, it was suggested in Song and Hatzinakos (2008) that reliable end-to-end communications can be established over any number of wireless hops, supporting real-time QoS (e.g., for VoIP), by the unicast wireless link module design. The performance can also improve with larger network scale, but with constant complexity. Figure 9
Protocol stack of mesh stations
We elaborate the programming of mesh stations at the system layer, by using the primitive functions of wireless link modules and some defined standard functions of 802.11 MAC (ANSI/IEEE 802.11 Standard, 2003) in Table 2. The protocol stack of mesh stations is illustrated in Figure 9. The system layer bridges the access network and the backhaul mesh, so that mesh stations act as L2 switches from the standing of IP. Besides forwarding user traffics to a gateway, mesh stations also deal with mobile-user handoff. For the illustration purpose, we assume here that there is only one gateway in the network. Multiple gateways can involve L3 (network layer) roaming for mobile users over different IP sub-nets, which are discussed in other dedicated works (Song, 2008). We also assume here that the network addresses of the mesh stations and the gateway are based on location coordinates, which are either pre-configured or acquired by GPS. We further define a few information variables in Table 3. The primitive programming pseudo-code of the mesh stations, by using the primitive functions provided in Tables 1 and 2, is then shown in Table 4. In principle, every mesh station manages two lists, which are CurrentUserList and CurrentForwardList, respectively. CurrentUserList maintains the list of current mobile users associated to the mesh station, whereas CurrentForwardList maintains a list of mobile users previously associated to the mesh station. On receiving from the gateway a unicast data packet, that appears to be one 802.11 MAC frame, the mesh station searches for the destination-user MAC address in its CurrentUserList. If the destination-user MAC address is found in CurrentUserList, the mesh station uses the 802.11 MAC to send out the MAC payload. Otherwise, if the destination-user MAC address is found in CurrentForwardList, the mesh station helps to forward the unicast data to the user’s current associating mesh station. The lists of
468
L. Song and D. Hatzinakos
CurrentUserList and Current-ForwardList can be updated over time, provide by the association/deassociation procedures of 802.11 MAC. When a user gets newly associated to one mesh station, the mesh station broadcasts one mobility control packet, which notifies its neighbouring mesh stations about the handoff. On receiving the mobility control packet, the previous mesh station updates its local lists CurrentUserList and CurrentForwardList. The list CurrentForwardList can also be subject to defined time-out, where entries of previously associated users are deleted after a predetermined period of time. Since, the gateway also keeps a list of current mobile users in the IP sub-net, the mesh station also unicasts one mobility update packet directly to the gateway, when a mobile user is newly associated. Table 2
Some IEEE MAC primitive functions
Function
Description
Caller
MLME-ASSOCIATE. indication (UserMACAddr)
Indicate the association of a mobile user: UserMACAddr (user MAC address);
802.11 MAC layer
System layer
MLMEDISASSOCIATE. Indication (UserMACAddr, ReasonCode)
Indicate the disassociation of a mobile user: UserMACAddr (user MAC address);
802.11 MAC layer
System layer
MA_UNITDATA. request (SourceMACAddr, DestiMACAddr, MACDATA, others)
Request to send the MACDATA (MAC payload) to the destination: DestiMACAddr (user MAC address);
System layer
802.11 MAC layer
MA_UNITDATA. indication (SourceMACAddr, DestiMACAddr, MACDATA, others)
Indicate the received MACDATA (MAC payload) from the source: SourceMACAddr (user MAC address).
802.11 MAC layer
System layer
Table 3
Responder
Mesh station information variables
Variable
Description
Addr
Network address of the local mesh station
GatewayAddr
Network address of the gateway
GatewayMACAddr
MAC pseudo-address of the gateway
CurrentUserList
List of current users associated to the local mesh station
CurrentForwardList
List of previously associated users and their current mesh station’s addresses
UserPriority
QoS class of user traffic
Therefore, the primitive programming of mesh stations achieves seamless user roaming. The mesh stations also periodically broadcast router advertisements via the 802.11 MAC to their associated mobile users, which contain the MAC pseudo-address of the gateway router GatewayMACAddr.
Cognitive networking of large scale wireless systems Table 4
Primitive programming of mesh stations
Switch (message) { Case UnicastP.indicate (SA, DATA): if ((SA==GatewayAddr)&&(DATA is of MAC frames)) { Obtain the destination MAC address UserMACAddr and the MAC payload MACDATA from DATA; if (UserMACAddr ∈ CurrentUserList) MA-UNITDATA.request (GatewayMACAddr, UserMACAddr, MACDATA, Others); else if (UserMACAddr ∈ CurrentForwardList) { Obtain the associating mesh station address MeshAddr, from CurrentForwardList; UnicastP.send(GatewayAddr, MeshAddr, DATA, UserPriority); } } break; Case Broadcast.indicate(SA,DATA): if (DATA is of mobility control packets) { Obtain the user MAC address UserMACAddr from DATA; if (UserMACAddr ∈ CurrentUserList) { Remove UserMACAddr from CurrentUserList; Add UserMACAddr in CurrentForwardList; } } break; Case MA-UNITDATA.indication (SourceMACAddr, DestiMACAddr, MACDATA, Others): if (DestiMACAddr == GatewayMACAddr) { Obtain DATA (MAC frame) from MACDATA; UnicastP.send(Addr, GatewayAddr, DATA, UserPriority); { break; Case MLME-ASSOCIATE.indication (UserMACAddr): Add UserMACAddr in CurrentUserList; Generate the mobility control packet MocDATA; Broadcast.send(MocDATA, UserPriority); Generate the mobility update packet MopDATA; UnicastP.send(Addr, GatewayAddr, MopDATA, UserPriority); break; Case MLME-DISASSOCIATE.indication (UserMACAddr, ReasonCode): Remove UserMACAddr from CurrentUserList; Add UserMACAddr in CurrentForwardList; break; }
469
470
L. Song and D. Hatzinakos
6.2 Wireless sensor networks for event tracking Wireless sensor networks have been proposed for a lot of scientific and industrial applications, which can acquire real-world data from integrated sensors and intelligently process the data by massive distributed computing. Although wireless sensor networking has been considered as a sub-division of MANET research in earlier days, it appears to be a broader research area. This translates to the need for more wireless link modules in the large-scale cognitive wireless networks. Shown in Figure 10, we now consider that a large number of wireless sensor nodes are deployed in a surveillance area for certain event detection and tracking. More specifically, a set of leader nodes are elected along the moving track of the target event, which is shown as an intruding vehicle in Figure 10. A leader node collects sensing data from surrounding sensor nodes and generates context-related tracking records. The use of the EWI architecture in such applications has been demonstrated in Song and Hatzinakos (2007a). We further elaborate the programming of wireless sensor nodes with the primitive functions of wireless link modules in Table 1. Figure 11 shows the protocol stack of a wireless sensor node. And some additional system primitive functions are defined in Table 5. Figure 10 Sensor networks for event detection and tracking
Figure 11 Protocol stack of wireless sensor nodes
The network address of a sensor node is acquired dynamically from the sensing data, based on the interested context, i.e., the event under surveillance. Derived from the sensor reading, the parameter SenPriority can be obtained, indicating the sensing data quality as related to the context. Nominally, SenPriority can be proportional to the physical proximity between the wireless sensor node and the event under surveillance. Based on the parameter SenPriority, the system layer also decides whether the node can serve as a leader node, i.e., a leader candidate, which is denoted by the Boolean parameter Candidacy. For example, Candidacy may be decided by whether SenPriority is above a
Cognitive networking of large scale wireless systems
471
predetermined threshold and other related metrics such as the node computation capability. The combination of the parameters SenPriority and Candidacy decides the network address, which can be acquired by using the primitive function CalcAddress(...) in Table 5. Table 5
System primitive functions
Function
Description
SensorReading (Context&DATA) CalcAddress (DATA, Context, &SenPriority, &Candidacy) Table 6
Read the DATA (sensing data), related to the Context; Calculate the network address, i.e., composed of SenPriority and Candidacy from the DATA (sensing data), based on the Context.
Caller System layer System layer
Responder Sensor N/A
Sensor node information variables
Variable
Description
Addr
Network address of the local sensor node
EventContext
Context information about the interested event under surveillance
LeaderAddr
Group network address of the leader nodes, i.e., Candidacy = 1
TrackPriority
Traffic priority of the event tracking
Table 7
Primitive programming of wireless sensor nodes
Switch (message) { Case UnicastP. indicate (SA, DATA): if ((SA is of LeaderAddr)&&(DATA is of tracking record packets)) Aggregate.start(EventContext); break; Case Aggregate. indicate (SA,DATA): if (DATA is related to EventContext) if (Enough sensing data has been collected) { Aggregate.stop (EventContext); Generate the tracking record packet TraDATA; UnicastP.send (Addr, LeaderAddr, TraDATA, TrackPriority); } else Save the DATA for further generating the tracking record; break; Case Aggregate. request (AA, Context): EventContext=Context; SensorReading(EventContext, &DATA); CalcAddress(DATA, EventContext, &SenPriority, &Candidacy); Addr=[SenPriority, Candidacy, EventContext]; Aggregate.send(AA, DATA, SenPriority); break; }
472
L. Song and D. Hatzinakos
We further define some information variables in Table 6. With the provided primitive functions in Tables 1 and 5, the primitive-programming pseudo-code of the wireless sensor nodes is shown in Table 7. In principle, the programming utilises data-aggregation modules to opportunistically collect context-related sensing data at leader nodes. The peer-to-peer unicast module is used for forwarding the generated tracking record packet to the next leader node along the event track. Since these two abstract wireless links are used consecutively along the event moving trajectory, the related sensing data along the track is collected by the leader nodes, generating the event profile. The use of peer-topeer unicast modules and the group address of leader node LeaderAddr can pick up one leader candidate, if more than one of such candidates exist. Not shown in the primitive programming, the wireless sensor nodes also periodically read their sensors for the initial event detection.
7
Comparative advantages
The advantages of the proposed cognitive networking for large-scale wireless systems have been envisioned and proved in related implementations: •
System performance would improve with larger network scale, since larger number of wireless nodes can contribute to the diversity in setting up abstract wireless links. For example, by the unicast wireless links (Song and Hatzinakos, 2008), reliable wireless multi-hop QoS can be supported for real-time services such as multimedia streaming; and by the data-aggregation wireless links (Song and Hatzinakos, 2007a), better data can be collected for the related context. In the experiments of Song (2008), the performance improvement over traditional methods in dataflow throughput is above 100% in a nine-node network.
•
System complexity could be independent of any network scale, due to the abstractions of wireless link modules. Different from traditional networking, where routing tables need to be maintained for global network information or at least neighbourhood nodes, the operation of wireless link modules does not depend on acquiring or maintaining neighbourhood node information. The advantage can also contribute to lower power consumption of wireless nodes.
•
Lower wireless network planning/deployment costs can be provided, by creating the dynamic drop-and-play network without predetermined topology. ‘Drop-and-play’ suggests that: 1 instantaneously adding a new wireless node improves the network 2 instantaneously removing a working wireless node do not introduce any breaks.
•
With robustness to interferences, the network can be ideal for operating in unlicensed bands with large free bandwidth (e.g., ISM). This is provided by the opportunistic utilisation of spectrum (wireless channels).
•
It is compatible with all the established standards, being off-the-shelf. The technology works with WiFi (IEEE 802.11 Standard, 2003)/Zigbee (IEEE 802.15.4 Standard, 2003) based, or any other physical radio implementations. The protocol establishes a cognitive wireless link layer, which can be seamlessly integrated with internet protocols (including TCP/IP) by setting up gateways (Song, 2008).
Cognitive networking of large scale wireless systems
473
•
Better network resource utilisation is provided, by the opportunistic utilisation of both spectrum bandwidth and radio availability. It centrally approaches the capacity of large-scale wireless networks.
8
Conclusions
We have proposed the cognitive networking concept for large-scale wireless systems, which differentiates from traditional networking by the opportunistic utilisation of network resources including both spectrum bandwidth and radio availability. It is shown that the proposed methods could resolve the scalability problems in large-scale wireless systems where both types of resources, i.e., spectrum and radio availability, can not be predetermined. As a result, the large-scale cognitive wireless networks are highly dynamic without predetermined network topology and spectrum allocation, where resource utilisations could reach their instantaneous maximum. We have also proposed to use the network architecture of EWI, in the implementation of the proposed cognitive networking concept. The EWI architecture provides the network abstractions with high modularity and flexibility to support the effective application programming in large-scale wireless systems. In conclusion, compared to traditional computer networks, where the industrial success begins with the development of individual systems (e.g., personal computers), the possible proliferation of large-scale wireless systems should start with enabling network technologies, since any application on top cannot be afforded by a single node but needs the mass collaboration among all. In traditional computer networks, the TCP/IP stack has been offering a scalable language to general purpose computers. We suggest that the research in this paper could be contributing to the possible standardisation of large-scale cognitive wireless networks, which might offer similar benefits to wireless embedded systems.
Acknowledgements The authors would like to acknowledge the support from Natural Sciences and Engineering Research Council of Canada (NSERC), Ontario Centres of Excellence (OCE) and Ontario Research Commercialisation Program (ORCP).
References Akyildiz, I., Su, W., Sankarasubramaniam, Y. and Cayirci, E. (2002) ‘A survey on sensor networks’, IEEE Communications Magazine, Vol. 40, No. 8. Akyildiz, I.F., Lee, W., Vuran, M.C. and Mohanty, S. (2006) ‘Next generation/dynamic spectrum access/cognitive radio wireless networks: a survey’, Computer Networks Journal, Vol. 50, pp.2127–2159, Elsevier. ANSI/IEEE Standard for Local and Metropolitan Area Networks Part 11 (802.11) (2003) ‘Wireless LAN medium access control (MAC) and physical layer (PHY) specifications’.
474
L. Song and D. Hatzinakos
Bonnet, P., Gehrke, J. and Seshadri P. (2001) ‘Towards sensor database systems’, Proc. 2nd International Conference on Mobile Data Management, pp.314–810. Cao, Q. and Abdelzaher, T. (2006) ‘Scalable logic coordinates framework for routing in wireless sensor networks’, ACM Transactions on Sensor Networks (TOSN). Conti, M. and Giordano, S. (2007) ‘Multihop ad hoc networking: the theory’, IEEE Communications Magazine, Vol. 45, No. 4, pp.78–86. Draves, R., Padhye, J. and Zill, B. (2004) ‘Routing in multi-radio, multi-hop wireless mesh networks’, Proc. ACM MobiCom, Philadelphia, PA. Gupta, P. and Kumar P.R. (2003) ‘Towards an information theory of large net-works: an achievable rate region’, IEEE Trans. on Information Theory, Vol. 49, No. 8, pp.1877–1894. Haas, A.J. and Deng, J. (2002) ‘Dual busy tone multiple access (DBTMA) – a multiple access control scheme for ad hoc networks’, IEEE Trans. on Communications, Vol. 50, No. 6. Hadim, S. and Mohamed, N. (2006) ‘Middleware challenges and approaches for wireless sensor networks’, IEEE Distributed Systems Online, Vol. 7, No. 3. Haykin, S. (2006) ‘Cognitive radio: brain-empowered wireless communications’, IEEE Journal on Selected Areas in Communications, Vol. 23, No. 2. IEEE Standard for Local and Metropolitan Area Networks Part 15.4 (802.15.4) (2003) ‘Wireless medium access control (MAC) and physical (PHY) layer specifications for low-rate wireless personal area networks (LR-WPANs)’. Intanagonwiwat, C., Govindan, R., Estrin, D., Heidemann, J. and Silva, F. (2002) ‘Directed diffusion for wireless sensor networking’, IEEE/ACM Trans. on Networking, Vol. 11, No. 1, pp.2–16. Lilien, L., Gupta, A. and Yang, Z. (2007) ‘Opportunistic networks for emergency applications and their standard implementation framework’, Proc. 1st International Workshop on Next Generation Networks for First Responders and Critical Infrastructure (NetCri07), New Orleans, Louisiana. Madden, S., Franklin, M. and Hellerstein, J. (2005) ‘TinyDB: an acquisitional query processing system for sensor networks’, ACM Trans. Database Systems, Vol. 30, No. 1, pp.122–173. Mitola, J. (1999) ‘Cognitive radio: making software radios more personal’, IEEE Personal Communications, Vol. 6, No. 4, pp.13–18. Scaglione, A. and Hong, Y.W. (2003) ‘Opportunistic large arrays: cooperative transmission in wireless multihop ad hoc networks to reach far distances’, IEEE Trans. on Signal Processing, Vol. 51, No. 8, pp.2082–2092. Scott, J., Hui, P., Crowcroft, J. and Diot, C. (2006) ‘Haggle: a networking architecture designed around mobile users,’ Proc. 3rd Annual IFIP Conference on Wireless On-demand Network Systems and Services (WONS), Les Menuires, France. Song, L. (2008), ‘Mesh infrastructure supporting broadband Internet with multimedia services’, Proc. IEEE International Conference on Circuits and Systems for Communications, Shanghai, China. Song, L. and Hatzinakos, D. (2007a) ‘A cross-layer architecture of wireless sensor networks for target tracking’, IEEE/ACM Trans. on Networking, Vol. 15, No. 1, pp.145–158. Song, L. and Hatzinakos, D. (2007b) ‘Architecture of wireless sensor networks with mobile sinks: sparsely deployed sensors’, IEEE Trans. on Vehicular Technology Part 1, Vol. 56, No. 4, pp.1826–1836. Song, L. and Hatzinakos, D. (2008) ‘Real time communications in large scale wireless networks’, International Journal of Digital Multimedia Broadcasting, ID 586067, doi: 10.1155/2008/586067. Song, L., Hatzinakos, D. and Wang, X. (2008) ‘Wireless mesh infrastructure for ubiquitous voice and video’, Proc. IEEE 5th Consumer Communications and Networking Conference (CCNC) – Demonstration, Las Vegas, NV.
Cognitive networking of large scale wireless systems
475
Tobagi, F.A. and Kleinrock, L. (1975) ‘Packet switching in radio channels, II. The hidden terminal problem in carrier sense multiple-access and the busy-tone solution’, IEEE Trans. on Communications, Vol. 23, pp.1417–1433. Zhao, Q. and Sadler, B. (2007) ‘A survey of dynamic spectrum access’, IEEE Signal Processing Magazine, Vol. 24, No. 3.
