Best Practices: Voice over IP (VoIP) · Best Practices: Voice over IP (VoIP) Ruckus Wireless, Inc. 880 West Maude Avenue, Suite #101 Sunnyvale, CA

Best Practices: Voice over IP (VoIP) Ruckus Wireless, Inc. 880 West Maude Avenue, Suite #101 Sunnyvale, CA 94085 v2011-‐08-‐31

Table of Contents Overview ................................................................................................................................... 3

Design Considerations ............................................................................................................... 4

VoIP Challenges on a Packet-‐based Network ........................................................................ 4

Delay ................................................................................................................................. 4

Jitter .................................................................................................................................. 5

Quality of Service (QoS) .................................................................................................... 5

Packet Loss ........................................................................................................................ 8

Handsets ................................................................................................................................. 10

5 GHz Support ..................................................................................................................... 10

Roaming .............................................................................................................................. 10

Handoffs .......................................................................................................................... 10

IP Roaming ...................................................................................................................... 11

Authentication ................................................................................................................ 12

Power-‐Save Mode ............................................................................................................... 14

Capacity ............................................................................................................................... 14

802.11n ............................................................................................................................... 15

Deployment environment ....................................................................................................... 16

RF Interference .................................................................................................................... 16

Site survey ........................................................................................................................... 16

Signal Strength ................................................................................................................ 16

Construction Materials .................................................................................................... 18

Advanced Wi-‐Fi Optimization ................................................................................................. 20

Power Symmetry ................................................................................................................. 20

Mitigation ........................................................................................................................ 21

DTIM Configuration ............................................................................................................. 21

Backhaul .................................................................................................................................. 22

VoIP and Wi-‐Fi Mesh ........................................................................................................... 22

VoIP and Wireless Bridging ................................................................................................. 22

Summary ................................................................................................................................. 23

Best Practices: VoIP |3

Copyright © 2011 Ruckus Wireless, Inc.

OVERVIEW

Voice over IP (VoIP) is a method of transmitting voice calls over packet-‐based networks. As it grows in popularity, the demand for reliable networks capable of transmitting voice without

losing voice quality has increased. In particular, VoIP over Wi-‐Fi is a growing trend as VoIP handsets and smartphones/laptops with VoIP applications grow.

The popularity of mobile phones, interestingly enough, has reduced most people’s expectations for voice quality. Mobile phones can have a much lower quality than a traditional landline.

Therefore people can be more tolerant of occasional blips in a call.

A good VoIP design however should be able to deliver high quality voice transmission that is at least as good as a mobile phone and ideally much better.

Wi-‐Fi in particular can pose challenges for voice. Not only is it a packet-‐based network (rather than a traditional circuit-‐based network), it also has the potential for high latency/jitter or

dropped transmissions. All of these contribute to poor voice quality and can impede a VoIP deployment.

Fortunately, with the right Wi-‐Fi technology and careful planning, most if not all of these problems can be avoided. This document discusses best practices for planning and deploying

VoIP over Wi-‐Fi.

4 | Best Practices: VoIP


DESIGN CONSIDERATIONS

Before the actual design of a VoIP network can start, there are some issues that should be considered. These can dramatically impact the final design, deployment and performance. These

include the following:

VoIP handset capabilities

Deployment environment

Site survey

Wi-‐Fi configuration to support voice

Backhaul

VOIP CHALLENGES ON A PACKET-‐BASED NETWORK The first thing to consider is what VoIP requires from a network to function correctly. Voice is considered the most time-‐sensitive application on a network. It does not tolerate any kind of delay.

When planning a voice network, the following guidelines should be used:

Network transmission delay (latency)

Inter-‐packet delay/jitter

Quality of Service (QoS)

Packet loss

DELAY When a person on a VoIP call begins to speak, the handset immediately packetizes the data and transmits it to the network. The network is then responsible for getting this data to the other handset as quickly as possible. Delay, or network latency, introduces a lag in the time one

person speaks and the other person hears them. Minor delays are unavoidable and VoIP can be tolerant of some delay, but not much.

Figure 1 Network Delay



Most Wi-‐Fi handset manufacturers recommend network latency that is less than 150ms (end-‐to-‐end)1. Higher latency can impact the quality of the call and can be perceived as an echo, lag

time, clicks, voice cutting in and out or a dropped call. Keep in mind; the latency time includes the entire network path from one handset to the other. This might involve transmission over Wi-‐Fi, a wired network, a public network or VPN.

JITTER Jitter is related to network delay and represents the inter-‐packet delay variation. A good voice call should have a consistent delay (low) between all of the voice packets. This helps create a

smoother call that is more consistent with delivering toll-‐quality voice.

Figure 2 Variable Network Delay (Jitter)

QUALITY OF SERVICE (QOS) Whenever a device transmits data, a prioritization should occur. QoS is a standardized method

to do this. QoS divides data into four types: voice, video, data and background. The highest priority is typically voice. This means whenever a device transmits data it should always send the voice traffic before any other type of data. Wi-‐Fi was not originally designed to carry voice so

there have had to be modifications to increase reliability and add QoS to the wireless networking standard.

How a device classifies traffic becomes extremely important. There are several ways it can occur. WMM (Wi-‐Fi Multi-‐Media) is a Wi-‐Fi Alliance interoperability certification based on the IEEE

802.11e standard. This method is based on the transmitting device tagging the packet, indicating its prioritization. This allows a single device to transmit different types of data (voice, video, etc.) in an order of priority that protects the more delay sensitive applications.

The 802.11e amendment specifies eight user priorities (UP) that align with previous standards

such as 802.1d and 802.1p. These are used to determine Layer 2 link layer frame prioritization. The eight UPs are further grouped into four (4) access categories (AC). Each AC contains two

1 Recommended by the International Telecommunication Union Telecommunication Standardization Sector (ITU-T) G.114.



user priorities. Note that User priority zero (0) as shown below is placed into the Best Effort AC instead of the Background AC for backwards compatibility with non-‐QoS stations.

Priority User Priority (UP) 802.1d Tag Access Category Designation

Lowest 1 BK AC_BK Background

2 -‐ AC_BK Background (Spare)

0 BE AC_BE Best Effort

3 EE AC_BE Excellent Best Effort

4 CL AC_VI Video (Controlled Load)

5 VI AC_VI Video < 100ms latency &

jitter

6 VO AC_VO Voice < 10ms latency & jitter

Highest 7 NC AC_VO Voice/Network Control

Table 1 UP to AC Mappings for Layer 2 Quality of Service

WMM requires a minimum of one set of four (4) queues per WLAN. Each queue maps back to one of the four Access Categories (ACs) mentioned above. Ideally, a Wi-‐Fi solution will support

four queues per client rather than per WLAN. This allows for much finer granularity and network access control.

Why Queue Traffic? There are several ways traffic enters a network. The default, if no QoS mechanisms are used, is First In, First Out (FIFO). In this case, traffic is sent as soon as it is received.



Figure 2 First In, First Out (FIFO)

In the diagram shown above, FIFO is used, i.e. no traffic prioritization. When this happens, traffic transmission can become unpredictable. The above example shows how a laptop might “grab” more airtime because it transmits first and/or faster than one of the phones. Therefore, the

second phone’s traffic is delayed.

When traffic is queued, the network (AP) can choose how and when to transmit each frame of traffic.

Figure 3 QoS Classification and Queuing



In the figure shown above, the traffic from all three devices is sorted into the appropriate queue. The phones are put into the high priority (voice) queue and are transmitted before the lower

priority data traffic from the laptop.

End-‐to-‐End QoS To fully support WMM, the entire network must support it. The VoIP client must identify the traffic correctly, then the Wi-‐Fi AP and finally whatever device sits between the Wi-‐Fi network and the wired network. Wired networks have a similar mechanism to WMM called TOS (Type of

Service) and COS (Class of Service) are bits that can be set within a frame to indicate its priority. Most network switches and routers recognize and honor this setting.

A more recent method is DSCP (Differentiated Services Code Point) also known as DiffServ. This is a more sophisticated replacement for TOS and COS and should be preferred. Most enterprise-‐

class switches and routers support DiffServ.

Another way to implement QoS is for the network itself to recognize and automatically classify traffic based on the application. This is something Ruckus APs already do. The AP will classify traffic into voice, video, data and background queues and transmit the data accordingly.

Although the WMM standard does not require it, Ruckus APs also implement per-‐client traffic queues to further differentiate and priority service amongst multiple devices contending for network access2.

PACKET LOSS Like network delay and jitter, packet loss is to be avoided at all costs. Even a 1% packet loss can significantly degrade a call. Packet loss in a Wi-‐Fi network is a particular concern since it is a

contention-‐based, half-‐duplex medium in which only one device may transmit at a time. Multiple transmissions will result in collisions, retransmissions, data corruption and packet loss.

2 The WMM certification only requires a single set of queues for a WLAN.



Another source of packet loss in a Wi-‐Fi network is RF interference. The largest source of RF interference in any Wi-‐Fi network is Wi-‐Fi itself. This is usually the most significant source of RF

by far. However there are cases where other devices may transmit on the same frequency. This is particularly true of the 2.4 GHz spectrum and is a major reason why it should be avoided if possible.



HANDSETS

The most critical piece in a deployment is the handset. It must not only interact with the backend PBX, it needs to seamlessly integrate with the Wi-‐Fi network as well. Not all handsets are the same. Some support more Wi-‐Fi integration options than others. This can dramatically

affect voice quality and reliability. Therefore selection of the handset is the single largest factor in the probable success of a deployment. Ideal features for a handset include:

Dual-‐band (5 GHz support)

WMM

Roaming

Power save/battery life

802.11n

5 GHZ SUPPORT The 2.4 GHz spectrum (802.11bg) is the mostly heavily used in Wi-‐Fi. This means it is potentially subject to more interference than other spectrums. If a handset is able to operate on the 5 GHz

spectrum (802.11a), the chances of interference are much lower. The 5 GHz band offers far more non-‐overlapping channels, lower propagation (smaller cell size), and much less RF interference.

ROAMING Most VoIP deployments occur in a multi-‐AP environment. Therefore the handset must be able to roam quickly and seamlessly to avoid call disruption. This is where many handset manufacturers differ in their implementations.

HANDOFFS In general, a Wi-‐Fi device of any type is the one that makes the decision to roam. Contrary to popular belief, the Wi-‐Fi infrastructure (with one rare exception) has no way to tell a client

device to move to another AP.

Standards-‐based Roaming 802.11r is an IEEE standard that was ratified in 2008. It is designed to address this exact issue with fast handoffs and roaming within a WLAN. Unfortunately, very few clients support it and no VoIP handsets at the time of this document. If an 802.11r handset is available, however, it is

definitely preferable.



When a Handset Roams The main way a handset decides to roam is signal strength. When the signal strength drops below a certain threshold, the device will move to another AP with a stronger signal if one is

available. To assure smooth handoffs and minimal service interruption, it is important to make sure there is always at least one AP with a strong signal ( ~ -‐65 dBm) within range of a handset.

Some handsets expose the roaming settings to an administrator. These may vary by manufacturer. If supported, some parameters that may need to be configured include the

following:

Scan Start (dBm) – When the signal strength reaches this level, the handset will look for a better AP with a stronger signal.

Handoff Start (dBm) – Threshold at which the handset will roam if a better AP is available.

Handoff Delta (dBm) – Difference between signal strength to current AP vs. another available AP.

Each deployment may require different settings, however a good place to start is:

Action Signal Strength (dBm)

Scan Start -‐60 dBm

Handoff Start -‐65 dBm

Handoff Delta 10 dBm

IP ROAMING One significant delay that can occur during roaming is when a handset moves from an AP on one subnet to an AP on a different subnet. Moreover, as soon as the handset’s IP address changes,

any active sessions (RTP, etc.) will be broken.

The simplest way to avoid this is a flat network for the voice SSID. This way the handset can keep its address as it moves between APs in the WLAN. Sometimes different subnets are

unavoidable however. In that case, voice traffic should be tunneled back to the controller.

The controller acts as a proxy for the handset and forwards its traffic to the second subnet while allowing the client to keep its original address. This is completely transparent to the handset and



will not affect voice calls. This type of solution, although not typically a good choice for data, works very well for voice due to the low bandwidth requirements.

Figure 4 Layer 3 Roaming

The figure above shows IP roaming in action. The handset originally connects to AP1 and receives an address from VLAN 100. The phone then roams to AP2, which is on VLAN 200.

Instead of forcing the phone to drop its IP address and acquire a new one, the AP creates a tunnel to the ZoneDirector containing the phone’s traffic on VLAN 100. The ZoneDirector is connected to the router via a trunked port and forwards the phone’s traffic to the network. In

this way, the phone’s original IP address is preserved and it does not need to drop its connection after a Layer 3 roam.

AUTHENTICATION There are several supported authentication mechanisms over Wi-‐Fi. The most popular are:

Open (no authentication/encryption)

WPA2 with a pre-‐shared key (PSK)

WPA2 with 802.1X.

When a Wi-‐Fi device moves from one AP to another, it must authenticate as per the standards. Authentication can take time and introduce delay, so it is important to understand the impact

authentication has on roaming.



Roaming in an open network is simple as there is no delay due to authentication negotiation. The downside of an open network is data is not encrypted.

WPA2 with a pre-‐shared key provides encryption via a shared secret (PSK). Since the

authentication happens with the WLAN, it is also fairly quick. When a device roams to a new AP, it presents the PSK, which the AP can verify very quickly. Since data is encrypted and roam times are low, this is a popular choice for VoIP implementations.

WPA2 with 802.1X is the most secure option that authenticates the device with the user’s

credentials rather than a single, shared key. Because of this, outside devices are introduced into the authentication process such as a RADIUS server, routers, etc. The delay introduced can be significant and negatively impact voice quality. Remember the standards require the handset

authenticate every time it roams. So not only does this authentication delay happen when the handset first connects to the WLAN, it occurs every time the handset moves through the network.

To get around this delay, there are two techniques in use today: PMK caching and Opportunistic

Key Caching (OKC).

The PMK (Pair-‐wise Master Key) is created after the first successful authentication occurs. If both the handset and the AP support PMK caching, the client device can save the master session key on an AP. When the client roams to another AP, it must do a full authentication again,

however if it roams back to the first AP it can use the PMK to skip the full authentication process (RADIUS server, etc.). This allows the client to roam much faster. But it does have the disadvantage of only helping a client roam back to a recently visited AP. It does not help when

the handset moves to a new AP it has not visited.

Opportunistic Key Caching attempts to solve this problem. In this case the Wi-‐Fi network distributes the client device’s PMK to APs near the client that are likely to be roaming candidates.

When the handset moves, it can offer its PMK to the AP, which already has it cached and skip full authentication and handshaking.

If 802.1X authentication will be used, the handsets and APs must support OKC. This should be a requirement for any handset selection. Otherwise it is highly likely the latency introducing from

full authentication during roaming will degrade voice quality.

Ruckus offers a compromise that can be useful for those who dislike the weak security of a single shared PSK and yet want to avoid the potential roaming issues inherent with 802.1X or have devices that do not support OKC. The solution is Dynamic PSK. With Dynamic PSK, each Wi-‐

Fi device is given its own PSK, which is bound to that device’s specific MAC address.



This is a superior solution to typical PSK networks; each device has a unique key and yet there is no 802.1X service, which side steps many support issues.

POWER-‐SAVE MODE Handsets have a very limited battery life. Constant radio use for Wi-‐Fi can quickly drain the battery. To get around this problem, there are standardized ways a device can inform the Wi-‐Fi network that it needs to go into power save mode. When a handset is not off hook (active voice

call) it can use this to reduce the amount of time it uses the radio and increase battery life.

There are several ways to implement power save in a Wi-‐Fi network. Any handset selected for a VoIP implementation should support at least one.

CAPACITY A VoIP connection transmits relatively little traffic, ranging in the hundreds of kilobits per second. It seems strange to discuss network capacity. But capacity is even more important for voice than data. The reason for this requirement comes back to latency and jitter. If multiple

phones are on the same AP, the AP will rotate receive or transmit that data for that phone before moving to the next phone. If there are enough phones on an AP the time between each phone getting access to the network can be long and add to network delay and jitter.

Figure 5 Capacity vs. Delay for VoIP

For this reason it is recommended no more than about 20 phones should be on a single AP.



802.11N As 802.11n gains popularity it is also starting to see widespread adoption in the VoIP handset

market. It might seem strange for a phone to support Wi-‐Fi connections in the hundred(s) of megabits per second when a voice call typically consumes less than a hundred or so kilobits per second.

The advantages of 802.11n for voice lie in the reduced airtime, i.e. the handset gets on and off

the air much more quickly. This results in less delay and improves overall performance. 802.11n can also take advantage of multipath to improve signal quality3.

Although 802.11n is not required for a good voice implementation, it is very desirable.

3 Legacy (non-‐802.11n) Wi-‐Fi networks do not deal well with signal reflections (multipath) and prefer line of sight (LoS) connections.



DEPLOYMENT ENVIRONMENT

The physical location of a Wi-‐Fi deployment can have a significant impact on the overall design and eventual performance. Key factors that can affect this include:

RF interference

Site surveys

Construction materials

RF INTERFERENCE As mentioned previously, RF interference can affect overall Wi-‐Fi network performance and

introduce significant delay/jitter. VoIP requires a reliable, predictable connection to function well. Therefore, anything the Wi-‐Fi network can do to improve its connection to the handset or reduce overall RF noise/interference will greatly improve voice quality.

Environmental factors that can introduce noise:

The Wi-‐Fi network itself

Other Wi-‐Fi networks

Non-‐802.11 RF (microwaves, non-‐DECT phones, RFID, etc.)

Some noise is unavoidable. Every Wi-‐Fi network will produce some level of noise that can

interfere with its own operations. There are also other sources of noise that may or may not be under control for mitigation: neighboring companies’ Wi-‐Fi, non-‐802.11 security systems, etc.

SITE SURVEY Whenever a Wi-‐Fi network must support voice, a site survey is crucial. As noted above, every location has its own unique combination of RF noise, construction materials, layout, etc. that can affect Wi-‐Fi performance.

SIGNAL STRENGTH Most VoIP handset manufacturers recommend minimum signal strength of -‐65 dBm. This is contrasted with laptops, which can often function with signal strength of -‐75 dBm. A good site

survey will take this into account and survey to -‐65 dBm rather than a lower rate.

SNR (Signal to Noise Ratio) must be part of any signal strength analysis since it takes into consideration the proportion of interference to the signal. For example, a survey might yield



signal strength of -‐65 dBm in the surveyed area. This seems like good coverage, however if the noise floor (background noise) is very high, it will reduce the effectiveness of that signal. If the

noise factor is large enough, e.g. -‐70 dBm, that yields an effective SNR of only 5 dBm, which is very poor. Effectively, the noise is so loud it is drowning out the real signal. To maintain excellent signal strength and quality, an SNR between 20-‐25 dBm is usually recommended.

It is also important to perform the survey to ensure the correct signal strength is being recorded.

This includes:

Passive vs. active survey tools – in particular, Ruckus APs with BeamFlex should always

be surveyed in active mode (passing traffic to the client device)4

Use the right client device for the survey

A common mistake when surveying for voice is to measure signal strength with a laptop. This

can result in insufficient signal strength for phones, which typically have much weaker radios. Instead, always make sure the survey is performed with the exact handset that will be deployed. If this is not feasible, use a device with similar radio/power characteristics. Using the right client

device will help ensure the survey is useful and yields accurate results for the design.

4 For more information on site surveys, please see the Ruckus document Best Practices: Site Surveys.



Figure 6 Representative Link Budgets for Clients

The diagram shown above gives some examples of different clients and their power and range. In general, devices with less Tx power will have less range to maintain a signal. This example uses a -‐70 dBm floor, which would not be appropriate for VoIP.

CONSTRUCTION MATERIALS RF will degrade as an inverse square over distance in free space. Within a building it typically degrades much faster due to walls, doors, elevators, etc. The amount a signal is reduced as it

goes through a wall is the attenuation rate. This is a primary reason why physical site surveys are recommended for VoIP deployments. Only an on-‐site analysis can fully compensate for building construction and layout.

Not all construction materials are conducive to good Wi-‐Fi operations. In particular, if non-‐

802.11n handsets are used, multipath can cause problems. Multipath occurs when there is reflection, diffraction or scattering of the signal. Highly reflective materials such as mirrors and glass can cause this.

RF absorption should also be mentioned here. When planning coverage, the materials must be

taken into consideration. A large number of glass walls, metal, etc. can reduce the Wi-‐Fi signal to



the point that it is not useable by a handset. Given their weaker radios, handsets typically need a much stronger connection than other devices such as laptops.

Voice typically requires 100% coverage to ensure seamless service for users. This means it is

particularly important to ensure the minimum signal strength is available everywhere.



ADVANCED WI-‐FI OPTIMIZATION

There are several things that have not been discussed yet that a WLAN can do to improve voice performance. These include:

Transmit power symmetry

DTIM setting

POWER SYMMETRY A commonly overlooked optimization is matching AP and handset transmit (Tx) power. Depending on local regulatory restrictions, APs can transmit at 100 mW or higher. This is in

contrast to a typical handset, which transmits at around 14 mW5.

When Tx power is not balanced (symmetric) a situation can occur in which the handset can hear the AP quite well and thinks it has a solid connection. The weaker handset radio means the signal received at the AP is much lower. If the delta gets too great the handset will not roam and

yet packets are dropped between it and the AP due to the lower strength signal.

Figure 7 Tx Power Asymmetry

Symptoms of this condition include client stickiness (does not roam), dropped calls and constant connects/disconnects from the WLAN.

5 Transmit power can vary by manufacturer.



MITIGATION The simplest way to reduce this problem is to reduce the Tx power of the AP. The power should match the handset power. Reducing Tx power on an AP will reduce its coverage area, so implementing this change will change coverage. However, the need to have consistently high

signal strength everywhere would mandate a similar configuration anyway. So this is rarely a problem.

DTIM CONFIGURATION A Delivery Traffic Indication Message (DTIM) is included in the AP beacon. An AP uses this to

notify a client that there is data waiting. This is particularly important for devices, such as handsets, that use power save mode to conserve battery life.

DTIM settings can vary by handset manufacturer depending on how they implement power save and wake up. A good number to start with is 1 or 2. This variable is set on a per-‐WLAN and per

AP basis via the CLI.



BACKHAUL

Since most VoIP installations will traverse some type of backhaul (wired, wireless, etc.) to reach the PBX and other handsets, the backhaul can also be optimized for voice deployments. In

particular: latency and preservation of QoS.

VOIP AND WI-‐FI MESH A common question in Wi-‐Fi design is whether voice may be deployed over a mesh network. In general, mesh should be avoided since it will introduce some delay. How much depends on the

connection speed of the mesh backhaul, reliability, etc.

Latency is a mesh network increases with the number of mesh hops, i.e. how many times a client’s data must hop from one AP to another before it reaches the wired network. In a multi-‐hop mesh network this can introduce considerable delay. Delay can be even worst if there are

two Wi-‐Fi handsets involved – traffic must traverse the Wi-‐Fi from one handset and then downlink to a handset somewhere else. This effectively doubles the latency introduced by the mesh network.

Ideally, any Wi-‐Fi network that supports voice should avoid mesh. Wired APs will give the fastest

performance and introduce the least delay. If mesh is absolutely required the network designer could consider a single hop mesh network.

To ensure the reliability and performance of a mesh backbone, the following guidelines should be used:

All mesh nodes must be connected with a very high signal strength/transmit rates/SNR

The mesh nodes should be dual-‐radio with the backbone traffic on the 5 GHz radios only.

Following these simple steps will greatly increase the performance of VoIP over a mesh network.

VOIP AND WIRELESS BRIDGING If VoIP traffic must traverse two buildings over a wireless bridge, much of the previous section is also true. A bridge with a poor connection between the bridge endpoints will not be able to

maintain voice calls well.



SUMMARY

This document was written to provide an overview of the important design issues involved in deploying VoIP over a Wi-‐Fi network. Voice network requirements are very stringent requiring

minimal latency and jitter.

Some critical design issues that must be considered as part of any deployment include:

VoIP handset selection

Handset radio (spectrum) capability

Quality of Service/WMM support

Fast roaming/authentication

100% coverage with high signal strength

Minimal RF interference

Site surveys

WLAN optimization for voice

Reliable, predictable backhaul

Best Practices: Voice over IP (VoIP) · Best Practices: Voice over IP (VoIP) Ruckus Wireless, Inc. 880 West Maude Avenue, Suite #101 Sunnyvale, CA

Documents