Deploy wireless IP voice transmission using Wi-Fi mesh network

Voice is the "killer app" for wireless data networks. And high-performance Wi-Fi mesh systems are the killer IP wireless networks. However, not all mesh networks are created equal. As wireless mesh networks become more popular—new public and private deployments are announced almost every day—the business need to add voice applications requires that the network's overall performance improve in order to handle real-time applications.

Once multiple hops are added to a mesh network, problems such as bandwidth degradation, network latency, and application priority contention can easily occur. These problems are exacerbated if the network covers a large area. Lack of network performance can seriously affect some of the real-time services that Wi-Fi service providers most want to provide, such as VoIP.

Mesh networks have four key performance requirements for real-time applications; three mesh architectures can be used to deploy wireless voice over IP (wVoIP) on Wi-Fi mesh networks; do multi-radio, multi-radio architectures have a positive impact on the capital costs of deployment, and what are the costs of operating and maintaining these networks?

Wi-Fi mesh networks offer many of the benefits of Wi-Fi with the added savings of planning, deploying, and operating such networks, especially in environments where networking is difficult, impossible, and/or metropolitan areas where connecting hundreds of nodes over dozens of square miles is impractical. Converged Wi-Fi networks supporting data and voice can even benefit from mesh architectures because multiple radios are deployed to extend coverage distances and increase bandwidth density. High-capacity mesh nodes typically offer the lowest bandwidth costs for wireless infrastructure equipment and installations, and can also help reduce network operating costs.

Four requirements for mesh networks

Mesh infrastructure performance must be able to provide high throughput, low latency, and end-to-end quality of service, not only between wireless handsets and access points, but also across the mesh links to the wired endpoints (usually IP switches). Because of this, the mesh backbone must provide:

High throughput across multiple hops. The mesh backbone must be able to support the traffic load regardless of the number of hops (typically 3 to 10). The ability to provide high throughput is directly related to the number of voice and data users supported. Insufficient bandwidth across multiple hops will result in unsatisfactory user density; additional equipment and more wired endpoints will be required on the network.

Low latency across multiple hops. High throughput is not enough. To avoid jitter, packet latency must be minimized at each hop. In a mesh network, the waiting time for packets in any node must be minimized (ideally to a negligible 5 milliseconds per hop). Because of this, a packet should be sent before all packets in a stream have been received from the previous node. Data transmission on a mesh network must be asynchronous, not synchronous. In the synchronous case, some kind of highly synchronized packet routing protocol between nodes is required.

End-to-end quality of service—prioritize voice packets. High throughput and low latency alone are not enough if the network is heavily loaded. To handle contention and the spontaneity of load demands, voice traffic must be prioritized across the entire mesh backbone and terminated with end-to-end traffic prioritization. It is no longer enough to simply provide a level of service between a wireless handset and the AP radio that serves that device (just like a wired AP). Mesh networks require quality of service across the entire backbone to avoid contention that can occur at each hop in the mesh. This level of service needs to be automated (depending on the infrastructure) and is best handled through a different virtual LAN/service set identifier (VLAN/SSID) specifically for voice. 802.11e is still a long way from being deployed, so don't expect it to be ubiquitous in the wireless infrastructure and all client devices in the near future.

Layer 2 switched networks. Layer 2 networks minimize the roaming issues that occur in Layer 3 networks. Layer 3 networks also require careful planning for different types of higher-level protocols. These two factors lead to performance issues and protocol configuration issues. Layer 2 wireless networks can act as a high-level "wire."

All four of the above factors directly affect scalability (in terms of the number of users and network coverage) and voice quality. If a specific multi-hop topology does not take these requirements into account, it will inevitably have limited functionality and lack voice quality features.

Three deployment methods

Wireless mesh network solutions vary, but most technologies are derived from the original concept of wireless distribution system (WDS). WDS is a wireless AP mode that uses wireless bridging and wireless relay. The former means that APs only communicate with each other and do not allow wireless clients to access themselves. The latter means that APs can communicate not only with each other but also with wireless clients. It is an inherent feature of all types of mesh networks that user traffic must pass through several nodes (such as through a wired LAN) before leaving the network. The number of relay segments that user traffic must pass through to reach its destination will depend on the network design, link length, technology used, and other uncertain factors.

Single Radio Solution - Multiple Signals on the Same Channel The single radio solution is the weakest wireless mesh solution. The access point uses only one radio channel, which is shared by both the wireless clients and the backbone traffic (which is forwarded between the two APs). As more APs are added to the network, a greater percentage of the wireless bandwidth is dedicated to forwarding backbone traffic, leaving less capacity for the wireless clients since the wireless network is a shared medium. Additionally, the AP cannot transmit and receive at the same time; it cannot transmit when another AP within range is transmitting. This results in unacceptable latency after just three hops.

A little math shows that in this single-radio scenario, each wireless client has only a limited amount of throughput to work with. For example, if you have five APs, each with 20 wireless clients connected to it, because all the APs and clients share the same 802.11b channel (5Mbps), that translates to only 50K to 100Kbps per user—the same throughput as a dial-up connection. And because all wireless clients and APs must operate on the same channel, network contention and RF interference can result in unpredictable latency.

Dual Radio Solution - Shared Backhaul With a dual radio solution, one radio can be dedicated to supporting wireless clients while the other radio is dedicated to supporting wireless backhaul - the backhaul channel is shared by both inbound and outbound traffic. Because the dual radio solution provides a single radio for both client access and backhaul, this alleviates some client-side congestion (low throughput, low latency), but the backhaul mesh channel is shared by both inbound and outbound traffic because the backhaul radio must still switch between backhaul mesh inbound and backhaul mesh outbound. This solution does little to alleviate backhaul bottlenecks and only helps slightly improve latency across the mesh compared to a single radio architecture.

In a multi-radio or "structured mesh" approach, there are several dedicated link interfaces and each network node uses at least three radios, including one radio for wireless client traffic, a second radio for inbound 802.11a wireless backhaul traffic, and a third radio for outbound 802.11a backhaul traffic. This wireless mesh networking approach greatly improves performance over single or dual radio approaches. It allows dedicated mesh backhaul links to transmit and receive simultaneously because each link is on a different channel.

Because the three functions of client inbound, backhaul outbound, and backhaul inbound are handled by dedicated wireless devices, you can:

(1) Ensure high throughput over 10 hops;

(2) The delay of each hop is also limited to 4 to 5 milliseconds, and the total delay of 10 hops is only 50 milliseconds - far less than the 120 milliseconds required for voice.

(3) If each wireless device supports quality of service and supports multiple SSID/VLAN, voice traffic will be properly prioritized throughout the entire process from the wireless phone through the mesh network to the wired endpoint.

How to achieve optimal operation?

The investment required to purchase and install wireless infrastructure depends on the capital cost per megabit and the number of subscribers that can be served in a given area. Ongoing operating expenses include not only the management and maintenance of the network, but also the fees paid to the infrastructure broadband provider, typically for DSL, T1 lines, T3 lines, and/or OC3 broadband termination points.

The capital expense for service providers is primarily a matter of the cost per radio (1 radio provides 54 megabits of bandwidth, equivalent to supporting a certain number of users based on the amount of bandwidth allocated to each user) and the cost of deploying radios at the required density to meet user demand. Multi-radio, multi-radio, and multi-channel wireless mesh architectures enable the highest density of radios to be deployed in the most cost-effective manner. The cost per megabit to purchase and deploy six single-radio nodes should be lower than that of three dual-radio nodes. A partitioned multi-radio node should be able to handle at least three times the number of concurrent voice calls as a typical single-radio or dual-radio node.

Broadband point-of-presence (PoP) providers typically charge a per-pipe, per-PoP bandwidth fee as a significant operating expense. That is, while one T3 line has the same bandwidth as 30 T1 lines, the cost of a T3 line is roughly three times that of a T1 line, so the bandwidth cost of a T3 line is only one-tenth of that of the latter. Ultimately, using three T3 lines instead of 90 T1 lines can significantly reduce costs. From a deployment architecture perspective, for example, T3 links allow each PoP to use ten times more mesh nodes, so providers can fully utilize all T3 bandwidth. The only way to achieve this is by using a multi-radio, multi-radio, multi-channel wireless mesh architecture to make the most and most efficient use of wired PoPs. For example, using a single radio node would require 10 endpoints per 5 square miles; using multi-radio nodes instead, one wired endpoint is sufficient.

Multi-radio, multi-radio nodes typically cost less per radio, are less expensive to install, and can leverage much more economical broadband endpoints by using lower cost per megabit but higher capacity bandwidth. To meet the demands of real-time communications applications such as voice, Wi-Fi mesh requires a multi-radio, multi-radio, multi-channel architecture. Deploying cost-effective mesh networks with the necessary capacity and coverage for high throughput, low latency, and high priority voice traffic requires high capacity nodes dedicated to supporting client inbound, mesh backhaul inbound, and mesh backhaul outbound.