Wi-Fi-based access networks are proliferating through deployments by service providers as well as through municipal buildouts. Wi-Fi's attractiveness is rooted in its ability to deliver high-speed access connectivity along with the proliferation of Wi-Fi-empowered laptop computing devices.

With the availability of IEEE 802.11a/g technology, Wi-Fi access rates available from single network access points (APs) are now commonly 54 Mb/s. Although this is the maximum, aggregate over-air bit rate, it is not uncommon for these access points to deliver 10 to 15 Mb/s of full duplex user traffic.

When considering the use of Wi-Fi technology in the construction of access hotspots, one often-overlooked facet of the network implementation is the backhaul segment. Poor backhaul network design can contribute to:

1. Reduced access bandwidth

• Sharing access spectrum with in-band backhaul
• Poor backhaul throughput from T1/E1/DSL

2. Poor application/service performance

• High network latency and/or delay variability
• Poor sustainable data throughput

3. Reduced network reliability and availability

• Unlicensed wireless backhaul outages due to interference
• Inability to construct resilient, self-healing backhaul topologies

4. Increased susceptibility to interference

• Unlicensed backhaul outages

5. Poor network financial performance

• High lease cost of fiber/TDM circuits (when used)

6. Delayed time-to-market

• Delays in deployment of fiber (when used)

Wi-Fi Backhaul Technology Options

DSL and T-1/E-1 wireline

This solution is often applied to small hotspots where a low-cost, low-capacity backhaul requirement exists. DSL is better able to address the backhaul needs of bursty traffic applications whilst T-1 is likely a better choice when time-sensitive applications are being considered (i.e. VoIP, VIDoIP).

Optics & T3/E3

When a Wi-Fi hotpsot grows to have many APs and has large a large user density, backhaul capacity may grow enough to demand a connection data rate which can only be met with fiber optics or T-3/E-3. The drawbacks with these solutions are:

• Upfront non-recurring costs
• Leasing costs
• Line availability (or time to deploy if unavailable)

In-band Wireless

In-band wireless backhaul may use Multi-Point or Point-to-Point techniques to provide low-speed backhaul links, on the order of 10 Mb/s or less. The central drawback with this solution is that the same spectrum that is used for Wi-Fi access bandwidth has to be shared with the backhaul segment. This "sharing" inherently reduces access capacity. As much as 50% of the available access capacity can be lost. This can be significant when there may likely also be further reductions in "useable" bandwidth associated with the likely reality that there will be other interfering spectrum users present. This will also cause the access and backbones to be highly susceptible to interference, making them unreliable.

Out-of-band Wireless

Out-of-band wireless backhaul usually employs licensed Multi-Point or Point-to-Point techniques. The reason for this is that the unlicensed spectrum is typically employed for the access layer (2.4 and 5.8 GHz). An exception to this is the 24 GHz ISM unlicensed spectrum in which high capacity Point-to-Point backhaul links can be operated without stealing capacity from the access layer.

An additional attraction of licensed wireless backhaul is the fact that it delivers interference-free operation, which means the backhaul links will perform properly and with the expected throughput. In addition, out-of-band solutions typical provide high backhaul bandwidths enabling multiple hotspots to be connected in a mesh, without sacrificing capacity.

Wireless Backhaul Network Topology Comparison

Multi-Point

Multi-point solutions for Wi-Fi backhaul may include LMDS, WiMax-like¹ or other similar system. The solutions share a common technical kernel, which is shared RF carrier operation in which a number of end-sites are multiplexed onto shared RF carrier(s)². The result is that a limited amount of bandwidth is generally available to any one subtended end-site. The result is that this technology is generally targeted at lower speed applications, i.e. < 10 Mb/s per end-site. These solutions also typically have high latency and jitter and lack a redundancy solution. This prevents high availability services from being delivered the latency and jitter will prevent voice and video services from being offered.

Point-to-point

Point-to-Point technology can be used to implement high speed backhaul links and is generally the choice where high bandwidth, bandwidth scalability and low latency are concerns.

Point-to-Point links can be implemented in single links, daisy-chained or deployed in "virtual" multi-point arrangements referred to as "hub & spoke" multipoint. In the latter, the advantage over conventional multi-point technology is the tremendous potential for bandwidth scalability of any of the end-site links without affecting the attributes of the links to the remaining end-sites. In addition, each link can be hardened with redundancy for higher availability, and can provide enhanced services through ultra-low latency.

Mesh

There are two basic implementations of mesh technology:

• Unconstrained mesh
• Constrained mesh

These network topologies generally rely on Ethernet technology since they both require connectionless networking to achieve their functionality. As such, self-healing, very high resiliency, inherent redundancy, and very high availability are attainable network attributes³.

In the unconstrained case, the mesh nodes are connected to numerous other mesh nodes and re-route traffic based on higher level protocols. When a service disruption occurs, the disrupted flows negotiate alternate paths thru the network without a pre-determined idea of where the traffic should go. The downside of this is that a lot of network bandwidth can be consumed during this process and it can take a "long" time⁴ when considering that the "network" work thinks in terms of SONET-like 50ms network outages. Additionally, when delay control is a necessary network attribute, the "indeterminate" nature of the unconstrained mesh usually results in a lot of network delay variability.

In the constrained mesh case, the network nodes are linked (typically with point-to-point links) to a constrained number of other network nodes. Using high speed protocols such as Rapid-Spanning-Tree (RSTP), the nodes autonomously detect network failures and switch traffic through pre-determined backup routes. Since the routes are pre-defined, the fail-switch-over deadtime/outage can be reduced to 50ms - 100 ms through co-ordination with the wireless layer. The controlled nature of the constrained mesh topology also allows the network designer to control the network path delays, making it a superior solution for delay-sensitive networks (i.e. those handling VoIP, VIDoIP, etc).

Examples of the Application of Wireless Mesh Technology to Metro Wi-Fi Backhaul

Constrained mesh networks used for metro backhaul offer a number of attributes, which are key to a variety of types of access networks:

• High bandwidth
• High degree of bandwidth scalability
• Ease of adds/changes
• Rapid self healing with "hitless" fail-switch-over⁵
• High availability
• Inherent N+1 equipment redundancy
• Low latency
• Economical

Among the key network applications that employ Wi-Fi access technology and require some or all of the above backhaul attributes provided by the constrained mesh are;

• HotSpot Backhaul
• Wi-Fi-based Municipal & Emergency Preparedness Backhaul
• Nano-Cellular Mobile Networks

When considering the use of constrained meshes, these can be constructed using the following frequency/spectrum arrangements;

• Unlicensed point-to-point: i.e. 24 GHz ISM
• Site-licensed: i.e. 18 or 23 GHz
• Area licensed: i.e. 24 GHz DEMS, 28 GHz LMDS

When considering the licensing options, it can be seen that there is virtually no limit to spectrum availability...and thus to network capacity. Choosing the appropriate backhaul wireless technology enables mesh sub-circuit bandwidths to be scaled from low initial bandwidths (i.e. 10 Mb/s) up to 200 Mb/s and 400 Mb/s (full duplex, sustained).

Coupling high capacity and capacity scalability with low latency (< 200 us per mesh hop) is mandatory in insuring that all service applications operating in the user access layer will operate properly.

Figure 1 employs distributed Wi-Fi access points located to create full metro coverage using "nano-cell" implementations with each Access Point site delivering access over a few square city blocks. This would be reflective of a municipal Wi-Fi network, an emergency preparedness Wi-Fi network, or a nano-cellular mobile network. The mesh is constructed using 2 layers, the first layer creates mesh sub-circuits which connect the Wi-Fi Access Points, the mesh sub-circuit root nodes are then meshed on a wider area basis and deliver the traffic to metro PoP locations.

The Wi-Fi access points in a network like this may require 1Mb/s - 20 Mb/s of backhaul depending upon a number of factors (applications, locations, on-line user count, etc). Clusters of 5 - 10 Access Points are grouped onto individual mesh subcircuits. The backhaul demand on these sub-circuits may easily reach 50 to 100 Mb/s full duplex or more (200 Mb/s aggregate). In turn, meshing the sub-circuit root nodes together in clusters of ~ 5 may demand backhaul bandwidths of 100Mb/s - 400 Mb/s full duplex. Additionally, consideration must be given to bandwidth scaling in the future. Recall the days when wireless Access Points with 2 or 3 Mb/s were "fast"... it was only a couple of years ago!

If the backhaul bandwidth is not able to deliver the required sustained data rates, buffering⁶ and packet discard will result. The goal of these functions is to smooth peaky data and/or discard low priority data when insufficient bandwidth is available. Although some "best-effort" services can cope with this, delay and packet-loss sensitive applications do not do well, nor to time-sensitive network routing functions such as cell-to-cell handoffs used in mobility-capable networks.

On a wider area basis, mobile voice networks are often considered as being candidates in doing "double-duty" as Wi-Fi or WiMax enabled data networks. As such, the traditional mobile backhaul model, being leased T1/E1 lines, becomes ill-equipped to deliver high value services. The reasons for this are related to:

• The high degree of statistical multiplexing required to funnel high bandwidth applications thru the narrowband T1/E1 backhaul
• Inadequate bandwidth available for the data service "user experience" to be positive. Users are effectively acclimatized to "high-speed broadband access" and so are the applications that they want to use.

Figure 2 illustrates a backhaul network in which a number of mesh sub-circuits are used to backhaul cellular mobile BTS locations distributed throughout a large metro area. The mesh sub-circuits employ root nodes, which are PoPs on a metro fiber ring (shown bold purple). In this case the traffic on the backhaul network is a combination of high-speed data and cellular voice traffic (carried as TDM-over-Ethernet). This allows the operator to provide high bandwidth data services, mesh resiliency & scalability, forward-looking IP/Ethernet centricity, whilst not abandoning any voice revenue or stranding G1/G2 cellular BTS equipment, which requires T1-based backhaul.

Summary

Ethernet constrained mesh backhaul offers an optimal backhaul topology for Wi-Fi(and WiMax) access networks. Mesh technology cost-effectively provides;

• High bandwidth
• High degree of bandwidth scalability
• Ease of adds/changes
• Rapid self healing with "hitless" fail-switch-over⁷
• High availability
• Inherent N+1 equipment redundancy
• Low latency
• Economical

These attributes make the mesh topology ideally suited for a variety of different types of metro access networks, including; municipal Wi-Fi networks, emergency preparedness networks, cellular mobile voice+data networks or WiMax multi-point metro access networks.

¹ At the time of this writing, WiMax certified systems were not yet available
² This sharing process is referred to as TDMA.
³ It should be noted that these are also the hallmarks of a good "carrier class" network, or a network designed for disaster-recovery or defense preparedness.
⁴ Tens of seconds or even minutes of outage time are typical.
⁵ "Hitless" refers to the deadtime being "SONET-like", 50 - 100 ms, which enables applications to operate without connection loss or service disruption during a user session set-up or whilst the session is in process.
⁶ Buffering translates into delay and delay variability.
⁷ "hitless" refers to the deadtime being "SONET-like", 50 - 100 ms, which enables applications to operate without connection loss or service disruption during a user session set-up or whilst the session is in process.

Erik Boch is CTO and VP of engineering for DragonWave Inc.

Visit DragonWave online.