Integrated Network Solutions

Simply Building a
VoIP Service
Over
Wireless Networks
White Paper

By: Yaron Azruel

June 2004
VoIP Telephony over Wireless Networks

1 Introduction

VoIP technology enables packet based IP networks to carry voice, a mission critical
application. With VoIP, operators and service providers can offer voice telephony
service as well as traditional data service over the same data infrastructure, thus,
increasing revenue stream and improving business models.

Constructing a VoIP telephony service over a wireless network requires basic

understanding of the technology, in order to achieve toll quality telephony and to
maximize capacity.

This paper introduces basic elements of VoIP technology and explains how these
elements influence voice capacity and quality over wireless networks. Once the
technology is introduced, this paper provides guidelines on selecting proper VoIP
gateways to work with wireless networks, as well as configuring BreezeACCESS VL
based wireless network for maximizing performances and capacity.

2 VoIP characteristics you should know

VoIP technology, SIP based or H.323 based, uses coder-decoder (CODEC) for
compressing/decompressing the sampled voice signal and RTP over UDP connections
for carrying the packetized voice data over the channel.

Different CODECs exist to allow different optimizations:

1. G.711, in both its a-law and µ-law versions, is the most common and basic
CODEC. It samples the 4Khz voice band at a rate of 8000 samples per
second, each with 8 bit resolution = 64kbps.
This CODEC, referred to as non-compressing CODEC, requires low
computation complexity and provides good voice quality. However, it
consumes 64Kbps, which is relatively high compared to other CODECs
2. G.729, also samples the voice band 8000 times per second, with 16 bit
resolution, but it then performs a compressing algorithm, resulting in a
stream of 8Kbps. This CODEC, referred to as a compressing CODEC,
optimizes the bandwidth used per connection
3. G723 has various versions, some supporting a 6.3kbps sampling rate and
others a 5kbps sampling rate. This CODEC is also referred to as a
compressing CODEC

Some of the principal CODEC parameters are:

1. Frame length
The payload a CODEC generates and the overhead of IP and Ethernet (or any
other layer 2 protocol) headers

INS Group, Alvarion Ltd. -2-

 VoIP Telephony over Wireless Networks

2. Frame rate
The amount of time that passes between the start bit of two consecutive
packets, generated by a CODEC. This also translates to packets per second
(PPS)

The following table summarizes common VoIP CODEC characteristics:

CODEC name Basic rate – Basic Payload – Total Packet Length

G.711 a-law / µ-law 200 pps (5ms) – 40 Bytes (payload) – 98 Bytes (total
packet size)
Abbreviation: basic rate generates 200 packets per
second (a packet every 5ms) with total length of 98 bytes
G.729 100 pps (10ms) – 10 Bytes – 68Bytes
G.723r63 33 pps (30ms) – 24 Bytes – 82 Bytes
G723r53 33 pps (30ms) – 20 Bytes – 78 Bytes
Table 1: common VoIP CODECs’ characteristics
It is important to note that most VoIP Gateways concatenate several basic rate VoIP
packets, thus actual packet size is larger and ppp is smaller (further info in the
following section).

3 When VoIP goes wireless

Wireless networks have their own advantages and limitations, some of which affect
VoIP service performance. The following are important parameters that affect VoIP
capacity and quality over wireless networks:
1. Packet loss
Occurs when some of the data sent does not reach the other end. Packet loss
events degrade voice quality. Packet loss is more common in wireless than in
wire-line networks. Reasons for packet loss include:
a. Link capacity that is smaller than the actual required capacity
b. Degraded link quality that may cause error in reception

2. Delay and jitter

Large delay and/or jitter in packet reception causes voice quality degradation.
This may occur when:
a. Link quality is degraded and requires many retransmissions to
overcome collisions and radio interferences.
b. The same quality of service (QoS) priority is assigned to voice and
data traffic
c. Overloaded /nearly overloaded link capacity may increase the delay
and jitter

3. Sector load
When implementing VoIP service over 802.11 based networks, collisions may
reduce capacity, and increase latency and jitter, to a level that limits the
number of concurrent calls, or alternatively degrade call quality. This should

Alvarion Ltd. -3-

VoIP Telephony over Wireless Networks

be considered when planning the number of subscribers per sector. (Specific

recommended values concerning BreezeACCESS VL can be found in the last
chapter).

4. The 802.11 standard has a built-in algorithm that reduces collisions in a

wireless medium, and thus reduces packet loss. Along with the obvious
advantages of this algorithm, some overhead per packet is created. Therefore,
proper planning should include traffic planning that reduces the number of
packets per second (pps) generated, thus reducing the overhead affect. (For
further information, see next section).

Question: can we get good VoIP quality over wireless network?

Answer: yes! By choosing the right VoIP equipment for working
over wireless networks

4 Choosing the right VoIP equipment

In addition to the trivial considerations for choosing VoIP equipment, such as VoIP
preferred standard (H.323, SIP, MGCP etc.) network architecture and others, the
wireless network characteristics should also be considered. This mainly refers to:
1. Link capacity both in BW terms and packets per second terms
2. Latency and jitter (may be higher than in wire-line networks)

4.1 Link Capacity

There are three important parameters that affect the number of calls that can be
carried by a single sector of a wireless access network:

4.1.1 CODECs (Voice compression)

As access networks provide, by definition, limited BW per subscriber (e.g., compared
to a LAN), compressing CODECs is strongly recommended for saving BW and
increasing capacity. The basic non-compressing CODEC which most equipment
supports is G.711 (64kbps). The next common CODEC, which is a compressing
CODEC, is G.729 (8kbps). Another common CODEC is G723r63 – a compressing
CODEC, which utilizes 6.3kbps. As described above, using a compressing CODEC
can save more than 8 times the capacity of a non-compressing CODED.

4.1.2 Concatenation
Concatenation is a very critical consideration when choosing VoIP equipment for a
wireless network, due to the limitation that a wireless medium imposes on the link in
terms of packets per second. (Note that this may happen in wired as well due to the
access router capability).
Table 1 above specifies, among other things, how many packets per second (pps) are
needed to transmit voice in each direction. This is specified for the basic rate of each
CODEC.

INS Group, Alvarion Ltd. -4-

VoIP Telephony over Wireless Networks

According to table 1, G.711 generates 200 pps per call direction, which aggregates to
400pps, at G.711 basic rate. Assuming a VoIP gateway concatenates several CODEC
frames together and transmits them within a single Ethernet packet, this would reduce
the pps per call, minimizing overhead and processing. For example: if 12 basic G.711
CODEC frames are concatenated and sent within a single Ethernet packet, a single
voice call will generate only 16.5 pps, instead of 200 - (per direction).

With Ethernet packets transmitted every 5ms (packing factor of 1 CODEC frame per
packet):
98Bytes/packet x200pps=19,600Bytes/second (156.8kbps)

With Ethernet packets transmitted every 60ms (packing factor of 12 CODEC frame
per packet):
538(40 bit payload x 12+ 58 overhead)Bytes/packet x16.67pps=8966Bytes/second (71.73kbps)

This means over double the capacity of calls per wireless link by simply using a
higher packing factor of 12 CODEC frames per packet (60ms) instead of 1 (5ms).

4.2 Latency and jitter

For good voice quality, even under extreme conditions, where jitter may be higher
than “normal”, the VoIP equipment needs a large dynamic jitter buffer. The size of
the buffer determines how much jitter/latency variation can be compensated without
degrading the voice quality. The dynamic sizing of that buffer allows latency to be
minimized according to the link quality and its jitter (which may also vary according
to network load).

By selecting VoIP gateways with a large and dynamic jitter buffer, one can greatly
improve voice quality over a wireless network, even without any specific
configuration change in the wireless network itself.

Alvarion Ltd. -5-

VoIP Telephony over Wireless Networks

5 BreezeACCESS VL Specifics

5.1 BreezeACCESS VL Introduction:

BreezeACCESS VL is a high capacity, IP services oriented broadband wireless access
(BWA) system. The system employs wireless packet switched data technology to
support high-speed IP services including fast Internet, virtual private networks
(VPNs) and telephony with voice over IP (VoIP) technology. BreezeACCESS users
are provided with a network connection that is always on, supporting immediate
access to the Internet and other IP services at high data rates. The system is designed
for cellular-like deployment, enabling the system architecture to vary in size and
structure. A system can include any number of cells, each containing several Access
Units (AUs) for better coverage of densely populated areas.
To better support data services as well as real time protocol (RTP) applications as
VoIP telephony, the system supports layer-2 traffic prioritization based on IEEE
802.1p and layer-3 traffic prioritization based on IP ToS (RFC791), as well as
committed information rate (CIR) and maximum information rate (MIR) per
subscriber unit (SU). BreezeACCESS VL products operate in unlicensed frequency
bands in time division duplex (TDD) mode, using orthogonal frequency division
multiplexing (OFDM) modulation with forward error correction (FEC) coding. Using
the enhanced multi-path resistance capabilities of OFDM modem technology,
BreezeACCESS VL enables operation in near and non-line-of-sight (NLOS)
environments. These qualities enable service providers to reach a previously
inaccessible and broader segment of the subscriber population.

5.2 Optimization of BreezeACCESS- VL configuration for VoIP

The goal in this optimization is to:
• Reduce jitter and delay of voice packets
• Reduce the loss of voice packets in case of over-subscription of the
sector
The above will directly affect the voice quality of the telephony in the sector.

5.2.1 Retransmission
Since voice packets are only “relevant” for short periods of time (as opposed to data),
retransmissions of voice packets should be reduced. Also, since retransmission of
long data packets can affect the jitter and delay of voice packets in other SUs, it is
important to lower the retransmission attempts in a mixed voice/data network to an
acceptable level for both the telephony application and for the data application.

The BA-VL parameters that affect the retransmissions are:

Number of HW Retries – the number of times a packet will be retransmitted, if it has
not received and ACK from the destination. This parameter is by default 10.
It is recommended that in a mixed environment, this parameter be reduced to 3-5.

Software Retry Option – This option enables retransmission of a packet in a lower

modulation level, in case the HW retransmissions has failed to deliver the packet.
This option influences even more the delay and jitter that voice packets may suffer in

INS Group, Alvarion Ltd. -6-

VoIP Telephony over Wireless Networks

the network. This option is by default ON – which is good for a data only
environment.
However, in a mixed Data/Voice environment, it s recommended to set this parameter
to OFF.

5.2.2 QoS through Priority Queuing

The BreezeACCESS VL supports low, mid and high priority queuing packets QoS
markings. (High is typically reserved for control.)
There are two mechanisms available in the BreezeACCESS VL, which enable
mapping VoIP packets to the mid queue of the VL (the high priority queue is reserved
for management traffic):
• CoS – according to the priority marked in the VLAN Tag of an Ethernet
packet (if it is tagged)
• ToS – according to the ToS field of an Ethernet packet (assuming the VLAN
tag does not contradict the ToS marking).

To ensure that data packets are mapped to the low priority queue, and Voice packets
are mapped to the mid priority queue, the VoIP packets marking need to be aligned
with the VLs:
• ToS Precedence Threshold (set to 3 by default)
• VLAN Priority Threshold (Set to 3 by default)
This ensures that even if there are many long data packets waiting their turn to be
transmitted over the air, the newly arrived high priority voice packet will be
transmitted first.
In addition, this can reduce packet loss of telephony packets, even in conditions where
the network is congested with low priority data packets.

5.2.3 BW management
Another feature in the BreezeACCESS VL, which can help improve the voice quality
and capacity, is its ability to control the amount of traffic that each station can issue.
The idea is to limit the over-subscription of a BreezeACCESS VL sector, to the
lowest possible value:
MIR: this parameter helps to control the amount of traffic a single SU can send over
the air link, i.e., the amount of bandwidth a single SU can consume.
The recommendation is:
1. The sum of all SUs MIR should be as close as possible to the maximum
BW available in a sector (the actual BW which is the mean of the
modulation level of all SUs in that sector), with as little as possible over
subscription. E.g., an over subscription rate of 5 (sum of MIR = 5 times
the BW available in that sector) is better than 10
2. Allocate a CIR value to each SU that is equal to the bandwidth required
per voice call multiplied by the number of calls expected from this SU.
This will ensure that if the sector is congested, the SU will still get air
allocation for voice.

Alvarion Ltd. -7-

VoIP Telephony over Wireless Networks

5.2.4 Sector Load

As stated above, the number of active SUs in a BreezeACCESS VL sector will
strongly affect its capacity, latency and jitter. Although this is hardly an issue for data
users (the latency and jitter), it does affect the VoIP application that runs in that
sector.
The allowed number of concurrent transmitting / receiving SUs in a sector depends on
various link characteristics:
1. What is the lowest link rate of weakest SU (in term of radio reception)
2. How many interferences exist in that sector (that causes retransmissions)

Therefore, the relevant parameters when planning for over subscription in a VoIP
enabled sector, should include:
1. Capacity over subscription (total MIR/CIR committed to all SUs in that sector
vs. available capacity in that sector)
2. Number of SUs in a sector, vs. the estimated number of concurrently active
SUs allowed in that sector, before affecting the voice quality.
This number can vary between 2 and 45 concurrently active SUs in a sector,
depending on link characteristics, as mentioned above.

Another parameter, which can increase sector load, without degrading voice quality is
voice packet concatenation: The longer the concatenation, the larger the jitter, without
affecting the voice quality (and thus more SUs can concurrently transmit):
• In a network where voice packets are transmitted every 20ms, jitter must be
smaller than 20ms.
• In a network where voice packets are transmitted every 60ms, jitter must be
smaller than 60ms (3 times the jitter allowed in a ‘20ms’ network).

INS Group, Alvarion Ltd. -8-

VoIP Telephony over Wireless Networks

5.3 BreezeACCESS capacity

According to our experience and simulations, a VL sector can support maximum 40
concurrent calls with a call per SU. When the number of concurrent calls is smaller,
the rest of the capacity can be utilized for best effort data.

CODEC Topology Voice capacity Comments

G.711 (Voice packet Single call 6 call No recommended
every 5ms) per subscriber
(worst case)
G.711 with 12 voice Single call 35 calls No more than 40 active
frames per packet per subscriber subscribers in total,
(voice packet every (worst case) sending data and/or
60ms) voice.
Maximum capacity is 35
calls + 8Mbps data.
G.711 with 12 voice Multi calls 70 calls
frames per packet per SUs (10
(voice packet every SUs)
60ms)
G.723/G.729 with 12 Single call 40 calls No more than 40 active
voice frames per per subscriber subscribers in total,
packets (voice packet (worst case) sending data and/or
every 60ms) voice.
Maximum capacity is 40
calls + 8Mbps data.
G.723/G.729 with 12 Multi calls 100 calls No Data
voice frames per per SUs (10
packets (voice packet SUs)
every 60ms)
Table 2: BreezeACCESS VL capacity

6 Summary

Voice over IP telephony is an excellent technology that allows carriers and Wireless
Internet Service Providers (WISPs) to offer telephony services on their existing
wireless access data networks, which are usually built for standard Internet access
application.
However, as shown in this document, the right VoIP equipment that can support all
the right functionality, can make a huge difference in the capacity of the telephony
application over the wireless network. By optimizing both the VoIP equipment and
the wireless infrastructure, voice capacity can be increased by 200-500% - a
difference that can absolutely turn the business model to a great success.
Moreover, by correctly configuring and building the wireless IP network, good
telephony quality can be achieved, and total voice and data capacity can be increased.

Alvarion Ltd. -9-