Discussion

2. Channel switching phase: If the packet loss rate for a link reaches a certain threshold, the channel switching phase is executed. In this phase, the sending node searches a dierent node-disjoint path using the connectivity matrix. If a new path is found, the sender initiates a channel switch with the next hop on the path. The least used channel for the link is chosen.

After the successful channel switch, multiple routes for the same destination are available.

Therefore, a route selection process has to be executed. A source routing is used with NNCQ and either the best route is chosen for all transmission or the routes are used round-robin like. The latter method may result in frequent channel switches.

The advantage of this approach is the consideration of the packet loss rate which also incorporates external interference and allows nodes to quickly react to link quality changes.

As a weakness, each node relies on a global connectivity matrix that is never updated and thus renders the approach unable to cope with network topology changes.

2.6 Discussion

2.6.1 Classication Keys

The following classication keys are used to characterize the presented channel assignment algorithms.

 Distributed algorithm (DA): Denes if the algorithm is of distributed nature.

 Channel Switching Frequency (CSF): Denes the frequency of channel switching. In dynamic or fast channel switching approaches, channel switches may occur frequently, up to for every subsequent packet. In static or slow channel switching approaches interfaces are switched to a particular channel for a longer period. Hybrid approaches combine both methods.

 External Interference (EI): Denes if external interference is considered by the algo-rithm. This may be addressed directly by detecting sources of external interference or indirectly by measuring the packet loss rate of a particular link.

 Link Connectivity Preserved (LCP): Denes if all virtual links of the network are preserved after channel assignment.

 Conict Graph Minimization (CGM): Denes if the problem is formulated such that the number of edges in the conict graph shall be minimized.

 Interference Model (IM): Denes which interference model is used for the interference estimation.

 Failure / Mobility (FM): Denes the degree of adaptivity of the algorithm. Node failures and node mobility lead to network topology changes, which result in nodes joining and leaving the neighborhood of other nodes.

20 Chapter 2. Channel assignment in wireless mesh networks

 Fairness (FA): Denes if the approach considers fairness in regard to network re-sources.

 Testbed Evaluation (TE): Denes if the approach was implemented and evaluated in a testbed environment.

 Trac Load (TL): Denes if the expected trac load is considered in the algorithm.

 Channel Oscillation (CO): Denes if the channel oscillation problem is addressed by the algorithm

 Routing Metric (RM): Denes which routing metric is used for the evaluation to exploit channel diverse paths.

The characteristics of each algorithm in regard to the classication keys are summa-rized in Table 2.6.1. The following discussion compares the dierent channel assignment algorithms and investigates trends for future algorithms.

2.6.2 Summary

All of the presented algorithms are of distributed nature apart from BFS-CA [53]. In this centralized algorithm, a dedicated channel assignment server (CAS) calculates the network-wide channel assignment. The algorithm is listed, since it was the rst one to consider external interference by using spectrum sensing to assess the channel conditions.

Of the surveyed approaches, all but NNCQ [70], NET-X [66], and URBAN-X [67] are static, meaning that channels are assigned to interfaces for a longer period of time. One reason for this is the relative long channel switching time, which is in the order of millisec-onds for current IEEE 802.11 hardware [60]. Also the implementation of dynamic channel assignment is more complex. For instance, NET-X [66] and URBAN-X [67] require changes to the Linux kernel and the drivers of the wireless interface in order to reduce the channel switching time [65]. New hardware is likely to reduce the channel switching time, thus making dynamic schemes more attractive for future approaches.

Most of the discussed algorithms try to reduce inter- and intra-ow interference and leave external interference aside, since it can not be controlled. However, with the focus shifting to urban scenarios with a dense distribution of co-located wireless devices and networks, external interference has gained more attention recently. Of the discussed algorithms, BFS-CA [53], Urban-X [67], and NNCQ [70] consider external interference in the channel assignment. While BFS-CA [53] and Urban-X [67] use spectrum sensing to assess the channel conditions, NNCQ [70] indirectly adapts to external interference by periodically monitoring the packet loss rate for a given link.

2.6. Discussion 21

AlgorithmDACSFEILCPIMCGMFMFATETLCORM DGA•static•3-hopNH••WCETT MICA•static•m-hopNH••• SRI09•static•Fixedrange••• BFS-CAstaticSpectrumsensing•2-hopNH••••WCETT Net-X•hybrid•2-hopNH•••MCR Urban-X•hybridSpectrumSensing•2-hopNH••Custom SAFE•static2-hopNH•• NNCQ•hybridPacketLossRate•Packetloss• Table2.6.1.:Overviewofchannelassignmentalgorithmswiththecorrespondingclassicationkeys:DA:Distributedalgorithm,CFS:Channel SwitchingFrequency,EI:ExternalInterference,LCP:LinkConnectivityPreserved,IM:InterferenceModel,CGM:ConictGraph Minimization,FM:Failure/Mobility,FA:Fairness,TE:TestbedEvaluation,TL:TracLoad,CO:ChannelOscillation,RM: RoutingMetric.

22 Chapter 2. Channel assignment in wireless mesh networks

All surveyed approaches preserve the network connectivity in order to avoid network partitions caused by channel assignment. All approaches apart from SAFE [69] also pre-serve the link-based connectivity. This means that all virtual network links between nodes are preserved. Operating one interface per node on a common global channel to preserve the link-based connectivity is done in DGA [48] and BFS-CA [53].

Most of the surveyed approaches use simple distance-based interference models and dene the interference set as the m-hop neighborhood, with 2 ≤ m ≤ 3. BFS-CA [53]

states that the used 2-hop model is exchangeable with a more accurate, measurement-based model but the eort for measurements is not in the scope of the algorithm study.

With MICA [36], SRI09 [40], BFS-CA [53], several approaches formulate the channel assignment problem as minimizing the number of the edges in the conict graph or multi conict graph. Of these algorithms, only MICA [36] has been evaluated with the fractional network interference (FNI) metric as described in Section 4.2.

Adaptivity to topology changes caused by node failures and mobility is not of a high concern of the presented approaches. MICA [36] and SRI09 [40] are not adaptive because they prevent the re-assignment of a channel-interface combination in order to ensure the convergence of the algorithm. Adaptivity is addressed by running the algorithm peri-odically in DGA [48] and SAFE [69]. One reason why adaptivity is not considered an important feature is the assumed network architecture which is a stationary mesh back-bone. Still, adaptivity becomes more important with the trend to more volatile, large-scale deployments with an increasing number of mobile nodes and possible node failures.

The algorithms BFS-CA [53], Net-X [66], Urban-X [67], and SAFE [69] consider fairness among the nodes for the channel assignment decisions. The fair distribution of network resources is of higher concern in gateway-oriented network architectures such as [56, 71], in which trac patterns are more predictable and are usually limited to ows from mesh router to gateway and vice versa. In the presented approaches the expected trac load is only considered in SRI09 [40] and NNCQ [70]. The reason for this is that most approaches are targeted at peer-to-peer scenarios.

An implementation candidate for a testbed environment has been evaluated for DGA [48], MICA [36], BFS-CA [53], and Net-X [66]. In the eld of channel assignment, a testbed evaluation is of high value due to the complex modeling of signal propagation and the resulting interference eects.

Most approaches prevent channel oscillation by using a three-way handshake to an-nounce channel changes as in DGA [48], MICA [36], SRI09 [40], and NNCQ [70]. Channel oscillation can not occur with centralized algorithms, such as BFS-CA [53].

Several algorithms have been evaluated with interference-aware routing metrics. DGA [48]

and BFS-CA [53] use WCETT [72], while Net-X uses MCR [66]. In the scope of Urban-X [67], a custom routing metric has been designed, incorporating the packet transmission delay, packet error rate, and the channel switching time.

Based on the discussion of the algorithms and classication keys, we try to predict the trends for future research in channel assignment. With the exploding number of wireless