Mesh_Fundamentals_14dec01

Fundamental Characteristics and Benefits
Of Wireless Routing (“Mesh”) Networks
Dave Beyer
Head of Nokia Wireless Routing
Mountain View, California
dave.beyer@nokia.com
Wireless Communications Association
International, Technical Symposium
San Jose, California
January 14-16, 2002
Abstract
Wireless will bring the next significant wave of broadband network services, particularly for
areas difficult or costly to reach with wires. However, before a wireless networking product can
be profitably and widely deployed, it must be proven capable of meeting three key business
challenges:
1. Low total cost of purchase, installation, and maintenance for complete and uninterrupted
wireless coverage despite difficult and changing RF environments;
2. Easy and spectrally efficient network scaling as the customer footprint and density grows,
despite the relatively demanding signal-quality needs of broadband RF links; and
3. Assured and reliable service for broadband subscriber data flows with an end-to-end
communication quality that matches or exceeds that available over wires.
Commercial products are now emerging based on wireless routing, which solves these
challenges using an entirely different approach to the traditional cellular, point-to-multipoint
network architectures. With wireless routing, every subscriber device serves as both the
broadband access device for that subscriber as well as part of the network infrastructure,
automatically forwarding traffic for other subscribers as needed to ensure full and continuous
network coverage and facilitate network growth while providing the service assurances needed
for high-speed, multimedia traffic.
This paper will review the fundamental characteristics and benefits of wireless routing (“mesh”)
network architectures, how they compare with traditional point-to-multipoint networks, and how
they solve the above challenges to enable the coming wave of broadband wireless networks.

Fundamental Characteristics and Benefits of Wireless Routing (Mesh) Networks D.Beyer
Contents
1. Wireless Routing Background ......................................................................................................4
2. Full, Robust Coverage .................................................................................................................6
2.1 Basic RF Propagation Models................................................................................................6
2.2 A Simple Model Based on Link Probability.............................................................................7
2.3 Log-Normal Path Loss Model...............................................................................................10
2.4 Robustness to Link Changes ...............................................................................................12
2.5 Effect of Radio or Other Link Improvements ........................................................................13
2.6 Providing Initial Coverage....................................................................................................15
3. Natural Scaling...........................................................................................................................16
3.1 Signal Versus Interference Propagation...............................................................................16
3.2 Network Capacity in Large, Dense Networks.......................................................................21
3.3 Planning AirHead Sites........................................................................................................24
3.4 Growing with Increasing Customer Density..........................................................................24
3.5 Using Unlicensed Spectrum.................................................................................................25
4. Multimedia, Broadband Service..................................................................................................26
4.1 Throughput over Multiple Wireless Hops .............................................................................26
4.2 Efficient, Fair and Reliable Channel Access.........................................................................29
4.3 Quality-of-Service Support for Multimedia Traffic.................................................................32
5. Conclusion .................................................................................................................................37
6. Acknowledgments......................................................................................................................37
A. Appendix A – Signal-to-Interference Capacity Scaling Derivation...............................................38
A.1 Local Neighborhood.............................................................................................................38
A.2 Receive Power of Desired Signal.........................................................................................38
A.3 Average Transmit Power and Associated Link Distance ......................................................39
A.4 Receive Interference Power.................................................................................................40
A.5 Receive Signal-to-Interference Level ...................................................................................41
Wireless Communications Association, Technical Symposium - 14 January 2001 (rev: 14Dec01) Page 2 of 42

Acronyms & Terms
802.11 An IEEE working group developing standards for wireless local area networks.
802.16 An IEEE working group developing standards for wireless broadband access networks.
Ad Hoc Not used in this paper due to the confusion between the 802.11 definition referring to a
collection of fully connected nodes, and the definition in the MANET forum referring to multihop
mesh networks.
AirHead The wireless router that attaches a neighborhood wireless routing network (i.e., AirHood) to the
backhaul links that lead to the operator network or Internet.
AirHood The neighborhood of wireless routers currently associated with a particular AirHead (or set of
AirHeads). Somewhat analogous to a PMP sector.
DiffServ Differentiated Services. A standard being developed in the IETF to provide application-specific
service to IP packets.
DSCP DiffServ Code Point. The field in IP packet headers used to determine the service to be applied
to a particular packet.
EIRP Effective Isotropic Radiated Power. Measures the level of power emitted from an RF antenna.
FDD Frequency Division Duplexing. A method often used in PMP networks to avoid interference
between up- and down-stream traffic in neighboring cells.
IEEE Institute of Electrical and Electronics Engineers.
IETF Internet Engineering Task Force.
ISM Industrial, Scientific, and Medical unlicensed frequency bands: 902 to 928 MHz, 2.4 to 2.485
GHz and 5.725 to 5.875.
MAC Medium Access Control. The protocol layer in a wireless network that coordinates transmission
scheduling and related issues.
MANET Mobile Ad-hoc Networks. An IETF working group developing routing protocols for mobile,
wireless, multihop networks.
OFDM Orthogonal Frequency Division Multiplexing. A modulation method which efficiently encodes
and modulates data on multiple orthogonal carrier frequencies, and which has recently gained
favor in standards groups such as 802.11a and 802.16.
PTP Point-To-Point wireless link.
PMP or PTMP Point-To-Multipoint wireless network. PMP refers to wireless “star” networks where the
transmission scheduling and other network behavior of the slave (subscriber) devices are
controlled by a master (base station, or access point) device.
RF Radio Frequency.
TDD Time Division Duplexing. Another method often used in PMP networks to avoid interference
between up- and down-stream traffic in neighboring cells.
TDMA Time Division Multiple Access. A method for synchronizing and scheduling a wireless channel
among multiple devices.
TOS Type of Service. The original field in IP packet headers used to indicate service handling. The
more recent DSCP label was defined within the original TOS field.
U-NII Unlicensed National Information Infrastructure bands: 5.15 to 5.25 GHz (indoor only), 5.25 to
5.35 GHz, and 5.725 to 5.825.
WLAN Wireless Local Area Network.
WR Wireless Routing (as in a WR network), or Wireless Router (as in a subscriber’s WR).

1. WIRELESS ROUTING BACKGROUND
Wireless routing was introduced and developed during over two decades of research work on
“packet radio” technology supported by the U.S. Defense Advanced Research Projects Agency
(DARPA) at a variety of commercial, defense, and university institutions (e.g., see [Kahn78],
[Jubin87], [Beyer90], [Garcia-Luna97], and [Beyer99]). Figure 1 shows a prototype packet
radio system developed in these programs during the early 1980’s.
Figure 1 – Wireless router (“packet radio”) system developed during early 1980’s
in a DARPA-supported research program.
Recently, low-cost commercial products have been introduced that take advantage of the unique
characteristics of wireless routing for broadband access applications using network architectures like the one
presented in
Figure 2. In this architecture, wired or wireless “backhaul” links are used to pipe high-speed
bandwidth into neighborhood access points called “AirHeads.” From there, wireless routers
are used to distribute this bandwidth amongst the subscribers in the neighborhood using self-
configuring routing and channel access protocols which automatically create wireless
neighborhood “mesh” networks (or “AirHoods”) in which each wireless router automatically
forwards traffic for others as needed.
This wireless routing approach has unique advantages regarding the degree of network
coverage that can be provided and robustly maintained at a given cost, the ease and efficiency
of scaling the network in size and density, and the level of service that can be provided to
network subscribers. This paper will review the fundamental characteristics and benefits of
wireless routing networks, how they compare with traditional point-to-multipoint networks, and
how they solve the cost, coverage, scaling, and service challenges to enable the coming wave
of broadband wireless networks.

Figure 2 – Current commercial wireless routing networks for residential broadband Internet access.

2. FULL, ROBUST COVERAGE
A well-known key challenge for wireless broadband solutions, particularly for the residential
market, is to ensure sufficient and uninterrupted coverage to the target market despite
cluttered and changing RF environments, and at an overall cost that’s affordable for the
residential consumer business case. Point-to-multipoint (PMP) network approaches to this
challenge include developing and installing better, higher, or more powerful base stations and
subscriber devices, or deploying more base stations in microcellular architectures, to attempt
to overcome even the worst-case “non-line-of-sight” propagation conditions in the target
market.
Although the wireless routing (WR) approach can also benefit from advances in radio
technology, its unique advantage derives from the ability of every device in the network to
forward traffic for other devices as needed. This characteristic results in two main effects
related to coverage:
1. Wireless routing networks self-configure into a microcellular architecture, with a virtual
microcell around each subscriber device, thereby dramatically decreasing the link
distances needed for connectivity (and increasing scalability as discussed in Section 3);
and
2. Wireless routing networks automatically select the best RF links between devices, and
the best multihop routes between each subscriber device and the AirHead, according to
the current local propagation conditions, thus enabling the network to take advantage
of the best RF propagation paths available rather than requiring the operator to plan for
and purchase equipment capable of overcoming the worst-case paths expected for the
lifetime of the network.
The following subsections will analyze the dramatic positive effect that the wireless routing
approach brings to solve the challenge of robust coverage.
2.1 Basic RF Propagation Models
In an otherwise quiet environment, two RF devices will be able to communicate if the RF
propagation characteristics between the devices result in a sufficiently low effective signal loss,
including multipath effects, to permit the receiver to successfully recover each transmission
with high probability (e.g., with a 10-5
bit error rate). However, predicting local RF propagation
in the 2-GHz and higher frequency bands typically used for wireless broadband is complicated
by the fact that the path loss is driven, to a very large degree, by the absence or presence of
obstructions in the links between devices. Approaches to modeling the path loss for wireless
broadband networks include the following (e.g., see [Rappaport96] and [Rappaport98]):
1. Log-Distance Model, which uses a simple, deterministic path loss formula based on
distance (d) of the form dB, where C(d( 0100 /log10)( ddndC + ) 0) is the constant path
loss at a small reference distance (d0), and where the path loss exponent (n) is typically
set to an estimate of the near worst-case exponent between the base station and
subscribers in that region (e.g., 3.5 to 5.5).
2. Log-Normal Model, which extends the Log-Distance Model by including a random
variable Cs resulting in the path loss formula C . The path
loss exponent n is set to the average propagation conditions in the area (e.g., 2.5 to
4.5), and C
( ) sC++ 0100 /log10)( ddnd
s is a zero-mean Gaussian random variable with standard deviation s (in

dB), which accounts for the random likelihood of having greater or fewer obstructions in
the path than the average.
3. Partition-Based Model, which extends the Log-Normal model by separately accounting
for the average path loss through known obstructions over each link, thereby
significantly reducing the standard deviation of the estimate.
Of these, the Partition-Based model is generally the most accurate in environments where the
precise locations and types of primary obstructions are known. However, this model will not be
used in this paper due to the assumption that this type of information is not, in general, known
for target wireless broadband markets. The simple Log-Distance model, using a worst-case
exponent, can be useful for planning point-to-multipoint networks, which must be designed
largely to overcome the worst-case propagation path in any case. However, as discussed in
section 2.3 below, the Log-Normal model is more appropriate for planning wireless routing
networks, due to their ability to take advantage of this random variance in the path loss by
automatically selecting the best links available, while avoiding (routing around) the more
difficult links.
2.2 A Simple Model Based on Link Probability
Before examining the coverage characteristics using the full Log-Normal propagation model in
section 2.3, let’s first examine the characteristics using a simplified version of this model,
where the term Cs completely dominates the formula. In other words, the path loss between
two devices in an AirHood, or between a subscriber device and the base station in a cell, is
completely driven by the presence or absence of obstructions in the path and is independent of
the link distance within that cell or AirHood.1
Thus, the probability of reliable communication
between any two devices within the cell or AirHood (including the base station or AirHead)
simplifies to a certain “link probability” (z), dependent on the RF environment for that area.
With this model, consider the case of adding a new subscriber to an existing, connected
network. For a point-to-multipoint network architecture, the probability that this new subscriber
will also be connected is simply equal to the link probability (z) between itself and the base
station. However, for the wireless routing network, the probability is equal to the chance of
being able to connect to any one or more of the existing devices in the network (including the
AirHead, existing subscribers, and any “seed” nodes). For a wireless routing network with m
devices, this “coverage” probability (Pc(m)) for a new subscriber can be expressed as:
m
c zmP )1(1)( --=
1
Although this model is simplistic, it also resembles reality for areas where the distances between devices varies by
only some low multiple. For an example of the high variability in path loss, refer to the measurements reported in
[Seidel91] performed on the 900-MHz band which resulted in an average path loss exponent of 2.7 and standard
deviation of 11.8. For this type of environment, the longer of two arbitrary links would need to be 7½ times the distance
of the shorter link in order for the first standard deviation areas of the two links’ path loss probability curves to be non-
overlapping.

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Probability of a reliable link between any two points (z)
Coverageprobabilityfornewnode
PMP
AirHead + 5
AirHead + 10
AirHead + 25
AirHead + 50
AirHead + 100
Figure 3 -- Coverage probability as a function of the link probability between any two points.
Figure 3 shows the powerful effect of the exponential factor in this coverage formula. As more
wireless router devices are added to the network, the probability that the next subscriber will
have coverage increases exponentially.
Using this link probability model, one can also compute the probability of a given subscriber
being at a certain number of hops from the AirHead. Specifically, the chance of being at some
number (y) hops from the AirHead is equal to the chance of not being at less than y hops,
times the chance of being able to connect to one or more of the devices that is at less than y
hops. So, given a network of m devices, the probability of being at y hops from the AirHead
(Ph(m,y)) is:
( )1
)1(1
)(),(
1
1
1
-
--×÷÷
ø
ö
çç
è
æ -
=
×÷÷
ø
ö
çç
è
æ -
=
-
-
-
yMy
yc
y
h
z
m
Mm
MP
m
Mm
ymP
where Mi equals the number of devices that are at i or fewer hops from the AirHead, including
the AirHead. Mi can be computed using the following recursive formula:
( )
1
)1(1)(
)()(
0
1
1
1
11
1
=
--×-=
×-=
å
å
=
-
=
--
-
M
zMm
MPMmM
i
j
M
j
i
j
jcji
j
Figure 4 presents example curves for Ph(m,y) for an AirHood with 25 wireless routers and for
different values for the link probability, z. For environments which exhibit a link probability
between arbitrary devices of 35% or greater, this model predicts that the great majority of
subscriber devices in an AirHood with 25 wireless routers will be either one or two hops from
the AirHead. However, even with the link probability set to 15%, the total coverage within 4
hops of the AirHead is 99.9%. Thus, even in very difficult environments, virtually 100%
coverage is still ensured using this link probability model.

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3
Number of hops from AirHead (y)
ProbabilityofbeingyhopsfromAirHead
15% link probability
4
Figure 4 -- Distribution of hops from AirHead as a function of link probability.
For a comparison with fielded systems, Figure 5 presents the distribution of approximately 100
2.4-GHz wireless routers in two networks, one in downtown Mountain View, California, and
another in Santa Rosa, California, a moderately-wooded residential area. The curves present
the percentage of nodes located at each hop count from the AirHead. Also included is the
curve for the simple link probability model using a link probability (z) of 60% (since this was the
probability of being 1 hop from the AirHead in these networks). Although both networks were
deployed into cluttered RF environments (one with trees, the other with trees and buildings),
the overall area of the networks was fairly contained (within ½ mile radius in each case),
leading to the relatively high direct (1-hop) link probability. In these networks, all of the
subscribers were able to connect to the network at no more than two hops from the AirHead.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3
Number of hops (y) from AirHead
Percentageofsubscribersatyhops
fromAirHead
Downtown, Mountain View, CA
Residential, Santa Rosa, CA
Link probability model -- z=60%
4
Figure 5 – Number of hops from AirHead for subscribers in two fielded mesh networks

2.3 Log-Normal Path Loss Model
As discussed above, a more realistic model for predicting the path loss (Lab) between a pair of
devices (a, b) is the Log-Normal model.
( ) sC++= 0100 /log10)( ddndCLab
The Log-Normal model is particularly appropriate for estimating the coverage of wireless
routing networks, due to their ability to take advantage of the best links present, while routing
around the more difficult links. Therefore, simulations were used to analyze the coverage
behavior of WR and PMP networks using the Log-Normal model and the following scenario
parameters:
Cell or AirHood radius 1 mile (1609.3 meters)
Mean path loss exponent n 3
Path loss standard deviation s 10 dB
Center frequency f 5.8 GHz
Reference distance d0 1 meter
Reference (free space) path loss C(d0) 20 log10(4p c/f) dB [~47.8 dB]
Path loss allowance for “baseline” system2
A 130 dB
For each simulation trial, the AirHead (or base station) is placed at the center and then
subscriber devices are located randomly within the 1-mile radius AirHood (or cell). A random
path loss is computed between each pair of devices using the Log-Normal formula, and the
resulting connectivity is determined for both the WR mesh and PMP case with links between
each pair of devices (a,b) for which the path loss Lab £ A. 1000 trials were run for each data
point. Figure 6 presents an example trial with 25 subscriber devices where the dark lines
indicate links to subscribers connected in a PMP network (32 %), and the light lines indicate
links to the additional subscribers covered in a WR network (total of 100%).
-1 6 0 9
0
1 6 0 9
-1 6 0 9 0 1 6 0 9
Figure 6 – Example AirHood using Log-Normal path loss model.
2
The path loss allowance accounts for the transmit power plus antenna gains minus receiver sensitivity along with any
cabling and connector loss. Any margin for shadow fading is ignored. For example, 24-dBm transmit power, 8-dBi
transmit antenna gain, 8-dBi receive antenna gain, and - 90-dBm receiver sensitivity gives a 130 dB path loss
allowance.

Figure 7 presents the coverage probability resulting from three scenarios, all using the
“baseline” radio systems: a PMP network, a mesh wireless routing network, and a mesh
wireless routing network where 4 “seed” nodes were first placed ½ mile to the North, South,
East, and West of the AirHead, each installed with a direct link to the AirHead. In each
scenario, the number of randomly located subscriber devices was then varied from 1 to 50. Of
course, the PMP network is unaffected by the number of subscriber devices, the probability for
network connectivity remaining at about 25.1% for all scenarios (using the same baseline radio
used for the mesh simulations, no directional antennas, no high towers, etc.). For the WR
network, the probability of the first subscriber is, of course, also about 25.1%. Then, as more
subscribers are added to the network, the probability increases rapidly (as predicted by the
simple Link Probability Model), reaching about 90% coverage with 15 subscribers, and about
99% with 25 subscribers. With the WR network primed with the 4 seed nodes, the coverage
probability of the first subscriber is around 74%, and increases to 90% after just 8 subscribers.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 10 20 30 40
Number of devices in network
Probabilityofcoverageforanew
subscriber
PMP
Mesh
Mesh + 4 seed nodes
Figure 7 – Coverage probability versus number of devices in example scenario
with Log-Normal path loss model.
Figure 8 shows the distribution for the number of hops from the AirHood for a 50-subscriber
wireless routing network, with and without the 4 seed nodes, and for comparison, also shown
is the distribution for the simple Link Probability model for a 50-subscriber network using a link
probability of 25.1% (set to be the same as the PMP coverage probability).

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3
Number of hops from AirHead (y)
ProbabilityofbeingyhopsfromAirHead
Log-Normal model - Mesh
Log-Normal model - Mesh + 4 seeds
Liink Probability Model - z=25.1%
4
Figure 8 – Distribution of hops from AirHead in 50-subscriber scenario
with Log-Normal path loss model.
2.4 Robustness to Link Changes
When designing and installing PMP wireless networks, extra link fading margin must be
included to overcome some degree of potential changes in the propagation environment (e.g.,
new tree growth, changes in multipath fading due to new reflecting surfaces, etc.). However,
when the changes are so severe that the additional fading margin is insufficient, then a
subscriber outage results, typically requiring an operator (or installer) “repair visit” to reorient or
reposition the subscriber antenna in an attempt to restore service.
The capability of WR networks to automatically switch to alternate paths when necessary
highlights two important, related advantages of WR networks:
1. Automatic recovery from link failures substantially reducing the need for operator repair
visits, and
2. A reduced link fade margin requirement thus easing installation and reducing
equipment costs.
Automatic Recovery from Link Failures
When, despite any fade margin allowance, a link between two devices in a WR network
becomes obstructed or otherwise fails, there is a very high probability that alternate links and
paths exist that can be automatically put to use, thus avoiding subscriber outages and operator
repair visits.
For example, the scenario where all of the links to and from one of the devices are lost entirely
can be modeled by adding the curve “AirHead + 24” to Figure 3. This curve will very closely
resemble the “AirHead + 25” curve. For instance, using a link probability of 25%, the
probability of continued coverage is still 99.92% with the 24-device AirHood (versus the
99.94% probability of coverage for the 25-device AirHood). A similar result can be derived
from the example scenario and Log-Normal model used for the curves presented in Figure 7.

Reduced Link Fade Margin Requirement
Because WR networks are able to automatically switch to the best links available at the time,
and because the probability that such links exist is high, the additional link margin used for
network planning can be reduced substantially when compared with PMP networks. For
example, whereas a 10-15 dB fade margin may be employed in PMP wireless systems to
overcome most (e.g., 99.9% of) fading situations, a much smaller fade margin can be sufficient
for planning WR networks because there is typically no reliance on specific links.
To quantify this fade margin difference, simulations were run to determine the difference in link
performance required to go from a 99.9% coverage probability to a 99.9% probability of not
only having coverage, but of also having at least one alternate path for each device in the
network. Specifically, 25- and 50-device AirHood simulations were run using the above
scenario assumptions with the Log-Normal path loss model, but varying the path loss
allowance by 1 dB for each series of trials. The path loss allowance required for a 99.9%
coverage probability was compared with the path loss allowance required for a 99.9%
probability of having coverage plus at least one alternate path for each device. With the 25-
device AirHood tests, the resulting path loss allowance difference (or the required “mesh fade
margin”) was 5 dB for both the scenarios with and without the 4 seed nodes. With the 50-
device AirHoods, the resulting difference was a 6-dB mesh fade margin requirement for both
the scenarios with and without seed nodes.
Therefore, assuming a 12-dB fade margin requirement for PMP networks and 6 dB for mesh
WR networks, then WR networks enjoy a 6-dB fade margin advantage.
2.5 Effect of Radio or Other Link Improvements
Of course, both PMP and WR networks can take advantage of advancements in digital radio
technology to increase their link ranges, data rates, and robustness to interference. These
advancements can be roughly summarized as an improvement (in dB) of the signal-to-noise or
signal-to-interference level that can be overcome, and corresponding receiver sensitivity, at
each modulation rate. To determine how this improvement would effect coverage, the Log-
Normal simulations were repeated on 25- and 50-device networks but with a “link
improvement” factor (I) varying from –30 dB (degraded radios or link propagation paths) to +40
dB. The link connectivity check was modified accordingly to Lab £ A+I.
Figure 9 presents the results for four scenarios: PMP, PMP + 6 dB fade margin, 50-device
mesh, 50-device mesh + 4 seeds, 25-device mesh, and 25-device mesh + 4 seeds. The new
scenario, “PMP + 6 dB fade margin” adds 6 dB of additional fade margin for the PMP network
due to its reliance on single links between the subscribers and the base station (refer to
Section 2.4 for more details).

0
10
20
30
40
50
60
70
80
90
100
-30 -20 -10 0 10 20 30 40
Link performance improvement in dB
Probabilityofcoverage
PMP
PMP + 6dB fade margin
50-device mesh
50-device mesh + 4 seeds
25-device mesh
25-device mesh + 4 seeds
Figure 9 – Coverage improvement versus link performance (such as improved radio systems)
For this scenario (average path loss exponent 3, AirHood or cell radius of 1 mile, etc.), Figure
9 leads to the following conclusions:
Ø For an 85% coverage target, the “cost” in link performance required for the “PMP + 6dB
fade” margin solution is approximately
o 35 dB greater than that required for the 50-device mesh solution, and
o 30 dB greater than that required for the 25-device mesh solution.
Ø For a 95% coverage target, the “cost” in link performance required for the “PMP + 6dB
Ø For a 99% coverage target, the “cost” in link performance required for the “PMP + 6dB
Ø Between 15% and 85% coverage, the rate of increasing coverage probability is:
o 2.9% more coverage per dB of improved link performance for the PMP solution,
and
o 8.4% to 11% more coverage per dB of improved link performance for the 25-
and 50-device mesh solutions, respectively.
Note that the “link performance improvement “ figures can also be interpreted as the result of
combining multiple mechanisms to improve the link including the use of directional antennas at
PMP subscriber sites, and mounting the PMP base station antenna on a high tower. However,
this huge advantage in the required link performance for mesh wireless routing versus point-to-
multipoint approach translates into mesh radio systems with much lower cost, AirHead

antennas mounted lower and less exposed to outside interference than their base station
counterparts resulting in higher link data rates, and subscriber devices which are easier to
install and more robust to link changes.
Although wireless routing methods can be used in combination with virtually any radio
technology, it essentially shifts the system economics and planning from sophisticated but
expensive infrastructures and radio technologies, which attempt to overcome obstructions and
subsequent environmental changes by “brute force” radio and signal processing methods, to a
mesh of smart devices running protocol software on components that have enjoyed the
benefits of a couple decades on the “Moore’s Law" cost-performance curve.
2.6 Providing Initial Coverage
Of course, in order for the coverage and robustness benefits of mesh wireless routing
networks to take effect, some number of WR devices must be operating in the area. Various
approaches that can help ensure this initial and continued robust coverage include:
Ø Using a “targeted marketing” approach to roll out pockets of coverage quickly, one area
at a time, rather than in a scattered manner.
Ø Installing some number of “seed nodes” into an area, either at first-adopter home or
office sites, or on utility poles or other locations to which the operator has access. As
illustrated in Figure 7, installation of only a handful of seed nodes can have a dramatic
effect on the coverage probability for the first subscribers. These “seed nodes” need
only be used for initial network coverage, and can optionally be removed after a
sufficient number of subscribers are connected to the wireless router network.

3. NATURAL SCALING
Growing the footprint and density of wireless routing networks benefits from the unique
characteristic that interference is attenuated at a faster rate (per distance traveled) than the
rate for the desired signals in typical networks. This section will review the cause and nature
of this powerful trait, which allows wireless routing networks to grow and re-use channels with
less planning and increased spectral efficiency than alternative methods.
3.1 Signal Versus Interference Propagation
For a given RF channel, system scalability can be considered as the modulation rate that can
be sustained over an active link despite the sum of interference power being received by all of
the other active transmitters out to the radio horizon. Critical parameters for this computation
are the average path loss exponents for the propagation paths followed by the interfering
signals versus that for the intended signals.
Due to the usual strategy of mounting PMP base stations on towers to achieve sufficient
coverage, each base station must be designed and deployed to be capable of receiving
transmissions from a range of subscriber devices within its cell (with propagation paths around
a mean 1/r3
path loss factor, for example), despite interference from other base stations (e.g.,
close to 1/r2
path loss factor - “I1” on the left side of Figure 10), or from subscriber devices in
other cells that happen to have a relatively clear path to the local base station (e.g., also close
to 1/r2
path loss factor - “I2” on the right side of Figure 10). Even when base stations use
sector antennas oriented to avoid direct co-channel interference in the main lobes of the
antennas, the front-to-back ratios of these antennas are still often insufficient to counter the
near free-space path conditions between the base stations. Therefore, at the base station
receiver, the propagation paths of the major sources of interference often have a more
favorable path loss exponent than that for the paths to many of the subscribers within its cell.
Raising the base station antenna further improves the coverage within that cell, but at the cost
of stronger interference.
Figure 10 – Interference scenarios at point-to-multipoint base station receivers.
Also, a similar problem is due to the interference received at a subscriber device from
transmissions by other base stations that happen to have a clear path to this subscriber.
These problems are exacerbated when subscriber devices are moved indoors, due to the
increased variations in the path loss between the subscriber devices and base stations.

A number of PMP strategies have evolved to combat this unfavorable interference problem
that include the use of:
Ø Frequency Division Duplexing (FDD) or synchronized Time Division Duplexing (TDD)
among all base stations within each region, to avoid interference from other base
stations, at the cost of essentially fixing the up- & down-stream bandwidths throughout
the region as well as increased equipment costs (for duplexers needed for FDD) or
increased scheduling overhead (due to the long guard times needed for synchronized
TDD throughout a region),
Ø Directional antennas at the subscriber device to focus its transmit power, and receive
gain, on the intended base station, at the cost of reliance on the single link between the
subscriber and the base station, more costly installations which require trained
personnel, and more operator repair visits when this link fails,
Ø More careful sectorization and channel planning, with increased installation costs, and
Ø More sophisticated radio equipment to filter out the interference and better extract the
desired signal, at increased equipment expense.
In wireless routing networks, this relationship between the propagation path loss exponents of
the interference versus the desired signals has the opposite relationship! As discussed in
Section 2.1, local path-loss characteristics are highly variable, dependent largely on the
presence and type of any obstructions in the paths. In the case of wireless routing networks,
the selection of the active links to be used is not random; the links selected are the best
available which ensure overall network connectivity. In addition, because of the lack of high
base station towers (even at the AirHead), and assuming that the nearby devices are
coordinated to avoid interfering transmissions (see Sections 3.2 and 4.2), then the propagation
paths of the interference signals will more closely follow the mean path-loss exponent of the
region. For example, for interfering signals to follow a clear path to receiving devices in a
wireless routing network, it is typically not enough for the interferer to be located in a remotely
visible location (as is the case for the PMP scenarios in Figure 10), but both the interfering and
receiving subscriber or AirHead devices must be located in this way, with a clear path to each
other. More often, even when a distant, potentially visible interfering device is transmitting in a
WR network, the receiving device will still be shielded by neighborhood “clutter” (as in Figure
11).
Therefore, the active links in a mesh network are not only generally shorter than the links
needed in a PMP network, but will also tend to encounter path loss exponents that are lower
(e.g., 1/r2.5
) than the mean path-loss exponent of the region (e.g., 1/r3
) over which the
interfering signals propagate.

Figure 11 – Interference scenario in wireless routing network.
This is a key, unique characteristic of mesh networks which allows it to scale to large networks
while maintaining dynamic TDMA transmission scheduling, adaptive to the up- and down-
stream traffic needs, easing installation and planning costs, and using high-volume radio
components which are affordable for residential business cases. In other words, the basic
physics of scaling to large networks favors wireless routing.
To estimate a typical path loss distribution for the active links in a wireless routing network, the
simulation scenario described in Section 2.5 was run for a 50-device mesh network, with a link
performance degradation of -6 dB, which results in a 99% coverage probability (see Figure 9).
Then, each link in the mesh network was assigned a modulation rate based on the highest
supportable rate for the path loss over that link, in accordance with Table 1.
Table 1 – Modulation rates supportable over a range of path loss and
signal-to-noise values for “baseline” radio.
Link data rate (Mbps) 3
6 9 12 18 24 36 48 54
Max. link path loss (dB) 124 123 121 119 116 112 108 107
Max. signal-to-noise (dB) 16 17 19 21 24 28 32 33
(Note that in this scenario, the maximum path loss allowance for any link is 124 dB rather than
the 130 dB specified in Section 2.3, due to the -6 dB link degradation used.) The active links
were then determined by computing the minimum delay paths between each subscriber device
and the AirHead using the Dijkstra shortest path algorithm with the reciprocal of the data rate
as the delay “cost” of each link. (See Section 4.1 for a related analysis on this choice of
routing paths.)
Figure 12 presents an example trial showing all of the wireless routing links on the left, and
only the active, minimum-delay-path links on the right. Notice that due to the random variable
used in the Log-Normal path loss model, the minimum delay paths often follow trajectories that
3
The path loss figures correspond to an 802.11a-compliant radio (see [802.11a]) improved by 6 dB over the minimum
802.11a receive sensitivity requirements, if using 8 dBi omni-directional antennas and 20 dBm transmit power for
example. The baseline signal-to-noise values were set using a fixed offset (representing reasonable noise figure and
implementation loss values) from these receive sensitivity values.

are not the most geographically direct between each subscriber and the AirHead.4
As a side
comparison, Figure 13 presents another example trial using the Log-Distance model with the
path loss exponent reduced to 2.5, to ensure connectivity. As expected, with all randomness
in the path loss formula removed, the minimum delay paths do follow trajectories that are much
more geographically direct (e.g., notice the absence of any crossing links in Figure 13).
-1 6 0 9
0
1 6 0 9
-1 6 0 9 0 1 6 0 9
-1 6 0 9
0
1 6 0 9
-1 6 0 9 0 1 6 0 9
Figure 12 -- 50-device wireless routing network showing all of the usable links (on left) and
only the active, minimum-delay-path links on the right (path loss exponent 3, standard deviation 10 dB).
-1 6 0 9
0
1 6 0 9
-1 6 0 9 0 1 6 0 9
Figure 13 – 50-device wireless routing network with active, minimum-delay-path links shown
(path loss exponent 3, standard deviation 0 dB).
Returning to the Log-Normal model, and the problem of determining a typical path loss
distribution for the active links in WR versus PMP networks, 200 simulation trials were
performed with the 50-device mesh case and the PMP case. For each case, the link
4
Note that this characteristic highlights a key problem in basing the micro-routing decisions (e.g., within an AirHood) on
geographic coordinates, rather than on the actual RF quality of the links between the devices.

performance (or “system gain”) was set according to the analysis performed in Section 2.5 to
ensure 99% coverage. This meant using a link performance degradation of –6 dB from the
baseline radio and path loss allowance for the WR case, and a +33 dB link performance
improvement for the PMP case. For example, this 39 dB difference can be accounted for by
the use of directional antennas at the PMP subscriber devices, by higher power and more
sophisticated PMP radios, and higher base station antenna towers.
The resulting path loss and distances for the links for all trials is shown in Figure 14. Because
the wireless routing network selects the best links available, the average path loss exponent
for the active links is considerably lower (2.49)5
than the simulation scenario’s mean path loss
exponent (3.0), and even further from the average path loss for the PMP links (3.31).6
Also,
the average distance for the active links in the mesh network was approximately half that for
the PMP network (608 versus 1069 meters).7
In actual deployments, coverage targets for individual AirHoods or base station cells can
typically be much less than 99% due to the additional chance of being connected to devices or
base stations in neighboring AirHoods or cells. However, 99% is used above to minimize the
skew in the path loss results caused by the subscriber devices with the greatest path loss links
which were not able to connect to the network, and thus were not included in the results.
Nevertheless, to ensure that the resulting relationships were also valid at lower coverage rates,
the simulation was repeated using an 85% coverage target, which required a link degradation
of -10 dB for the 50-device mesh scenario, and +19 dB for the PMP network. Figure 15
presents the path loss distribution results. The average path loss exponents were 2.45 for the
mesh network and 3.08 for the PMP network, and again, the average link distance for the
mesh network was approximately half that for the PMP network (551 versus 1031 meters).8
5
Except where noted, the average path loss exponent was computed using the average of the absolute path loss factors
rather than the average of the logarithmic (dB) values.
6
Note that this value is greater than the mean path loss setting (3.0) due to the way that the Log-Normal path loss
model distributes values around the mean using a standard deviation in units of dB rather than in absolute values. If
instead the path loss exponents of all of the data points for the PMP case are first computed individually, and then the
average of these exponents is computed, then the result is a path loss exponent of 2.987, which is only slightly less
than 3.0 due to the 99% (less than full) coverage, and the fact that the links to subscribers that were not reached had
the greatest path loss values.
7
Note that for the wireless routing scenario in , the path loss values are capped (at 124 dB) due to the use of
alternate links to obtain the 99% coverage target, rather than by requiring radios and installations with high link
performance. On the other hand, for the PMP scenario, much better link performance is required to achieve the 99%
coverage target and the cap is instead in the distance dimension, at the 1-mile radius from the base station.
Figure 14
8
In the PMP case in F , there is also an apparent cap in the path loss dimension, due to the 15% of devices
that can’t be reached.
igure 15

60
80
100
120
140
160
10 100 1000 10000
Distance for active mesh links (meters)
Pathloss(dB)
n=2
n=3
n=2.49
60
80
100
120
140
160
10 100 1000 10000
Distance for PMP links (meters)
Pathloss(dB)
n=2
n=3
Figure 14 – Path loss versus distance for the active links in a 50-device mesh network (left) and for the links in
a PMP network (right), normalized with link performance settings that provide 99% coverage for each case.
60
80
100
120
140
160
10 100 1000 10000
Distance for active mesh links (meters)
Pathloss(dB)
n=2
n=3
n=2.45
60
80
100
120
140
160
10 100 1000 10000
Distance for PMP links (meters)
Pathloss(dB)
n=2
n=3
Figure 15 – Path loss versus distance, with scenarios normalized to provide 85% coverage for each case.
3.2 Network Capacity in Large, Dense Networks
This subsection will further examine the consequences of the scaling advantages for wireless
routing networks discussed in section 3.1 above, particularly as they relate to the overall
capacity of large, dense networks. For this analysis, we will use the 4-channel deployment
model illustrated in Figure 16.

The entire network, out to the radio horizon, at
radius Rh. Outside of the local region, devices
(and AirHoods) are assumed to use the local
AirHood’s channel according to an average
density.
The “local region,” radius Rl, consisting of this
local AirHood plus a collection of surrounding
AirHoods, all of which use a channel other
than the one being used in the local AirHood.
3 4
1 2
4
2
4 43
The “AirHood” of radius Ra, including the
AirHead, within which transmissions are
coordinated to avoid collisions, and where, in a
dense network, a single channel is shared.
Figure 16 – Definition of the AirHood, the Local Region, and the rest of the regional network.
By having each device choose its active links amongst its nearest 8 neighbor devices (nearest
in an RF sense), and by using adaptive power control to ensure that each transmission
reaches the intended receiver at approximately a target receive level (adjusted by the regional
noise conditions), then the average signal to interference ratio (SIR) for any transmission can
be determined using the following equation (refer to Appendix A for the derivation):
DR
horizonvisiblehorizonradioR
DNR
RR
R
y
aD
TSIR
x
T
h
l
y
h
y
l
x
T
TB
dBmdBm
dBmdBm
p
p
p
30002
3/4~
/3000
11
10)2(2
1010log10
)2/(1
226
1010
10
×=
=
×=
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
-××
×-
×+-=
+-
--
where Rl is the radius of the local region (see Figure 18), RT is the distance corresponding to
the average transmit power, x is the average path loss exponent for the active links, y is the
mean path loss exponent for the region, TdBm is the target receive signal level, BdBm is the noise
floor, a is the average transmit duty cycle for each wireless router, D is the density of wireless
routers in devices per square km, and N-1 is the average diameter, in hops, of each AirHood.9
Using this SIR equation, Figure 17 presents an example of the modulation rates that can be
supported in a wireless routing network as the density of the network increases given the
following assumptions:
9
With frequency hopping radios, the analysis is similar, except that the interference from the neighboring AirHoods
can be handled as a statistical chance for collisions (resulting in occasional retransmissions), and the interference
outside of the local region is lowered significantly by setting the density parameter D to the actual device density
divided by the number of orthogonal hopping frequencies.

Ø 802.11a (unimproved) radios using the signal-to-noise values listed in Table 1, and
“improved” radios with signal-to-noise values which are 3 dB lower than those in the
table;
Ø Transmit power control limitations are ignored;
Ø Use of four, 802.11a channels throughout the network;
Ø Noise floor value of -95 dBm;
Ø Mean path loss exponents of 3.0 for the region and 2.5 for the active links (see Section
3.1);10
Ø 50-km radius to radio horizon (total of about 4 million devices at 500 per sq-km); and
Ø 50-device AirHoods with diameters of 3 hops, each busy with transmission activity 75%
of the time.
0
10
20
30
40
50
60
0 100 200 300 400 500
Wireless routers per square km
Linkdatarates-Mbps
802.11a radio
802.11a improved by 3 dB
0
50
100
150
200
250
300
0 100 200 300 400 500
Wireless routers per square km
Mbpspersquarekm
802.11a radio
802.11a improved by 3 dB
Figure 17 – Reliable wireless routing link data rates (left) and capacity per square km (right)
for 802.11a-style radios as network density increases.
Because of its ability to adaptively shrink the “microcells” around each wireless router using
power control and adaptive neighbor and routing algorithms, and by keeping the number of
wireless routers per AirHood roughly constant (by injecting new backhaul links into new
AirHead sites as the density increases), the total capacity per unit area is able to increase at a
nearly linear rate with increasing density.
Note that this analysis also suggests that in regions with few buildings, trees, hills or other
“clutter” between neighborhoods, the wireless router antennas should be mounted relatively
low (i.e., not raised on a high mast), to ensure a significant path-loss exponent between itself
and devices in remote AirHoods, while still maintaining connectivity to devices in its own
AirHood.
10
When computed from the interference power received, the average path loss exponent for nearby interference sources
will actually be greater than 3.0 (see footnote 6); however, this average exponent will approach 3.0 for remote sources due
to the diminishing influence of the random variable in the Log-Normal path loss formula with increasing distance.

3.3 Planning AirHead Sites
To consolidate backhaul bandwidth and installation costs, AirHeads can be grouped at
AirHead sites. Figure 18 presents a typical scenario for the 4-channel deployment model,
where each AirHead site is equipped with a wired or point-to-point wireless backhaul link and 4
AirHeads with sector antennas directed toward the four supported AirHoods. Sector antennas
are normally used at the AirHead site to avoid adjacent-channel interference among the co-
located AirHeads, and to more effectively reach the 1-hop subscriber devices of the
corresponding AirHood.
Note that this deployment model is also consistent with Figure 16, with AirHeads grouped at
the AirHood corners. Also note that the capacity analysis performed in Section 3.2 assumes
the use of omni-directional antennas for all of the wireless routers, so that analysis is actually
pessimistic regarding the interference received from remote AirHeads, and the vulnerability of
AirHeads from remote interference.
21 21 1
3
1
43
21
43
21
21
33 4
21
34
1
Channel being used by this AirHood.
AirHead sites with 4 AirHeads (one per
channel) and one P2P backhaul link.
Figure 18 – Typical 4-channel wireless routing deployment.
3.4 Growing with Increasing Customer Density
The nature of wireless routing permits an economically attractive migration to increasingly
dense networks, rather than requiring the initial infrastructure build-out to be sized according to
the eventual, target subscriber density. This gives operators the flexibility to install an initial
infrastructure that’s sized appropriately for the initial subscriber density, and then, with the
benefit of revenue from the first subscribers, continue to build-out the network infrastructure as
the subscriber density increases. This benefit is due to the following WR characteristics:
Ø The WR protocol software can automatically reduce its transmit power and adapt to
using the nearest set of neighbor devices and best paths to the AirHead according to
the network density.
Ø As the network density increases, the operator can setup new AirHead sites, or install
additional AirHeads at current AirHead sites, and then re-associate some of the existing
subscriber devices with the new AirHeads as appropriate, or even have the subscriber
devices select the nearest AirHead automatically. In particular, no operator visit is
needed to re-orient the subscriber’s antenna.

Therefore, the microcells around each WR naturally scale down in size, and the AirHoods of
each neighborhood network are easily split, to permit efficient reuse of the RF spectrum
throughout the region even with dramatic increases in subscriber density.
3.5 Using Unlicensed Spectrum
The scaling and coverage characteristics of wireless routing networks described above also
have an important, indirect consequence of permitting the reliable use of “unlicensed”
frequency bands. In the U.S., these bands include the Industrial Scientific and Medical (ISM)
bands of 902 to 928 MHz, 2.4 to 2.485 GHz and 5.725 to 5.875, and the Unlicensed National
Information Infrastructure (U-NII) bands of 5.15 to 5.25 GHz (indoor only), 5.25 to 5.35 GHz,
and 5.725 to 5.825. WR characteristics that make possible the reliable use unlicensed
frequency bands includes the following.
Ø The shorter and lower-path-loss links in a WR network result in a reduced transmit
power requirement for each link. This enables the use of radios which comply with the
unlicensed transmit power regulations (e.g., 4 Watt effective isotropic radiated power
(EIRP) for the ISM and upper U-NII bands), in addition to decreasing the costs of the
RF amplifier.
Ø Because the WR antennas are mounted low, within the neighborhood “clutter” (no high
base station antennas), and because the active links in the WR network are intelligently
selected, the interference from outside sources (including other unlicensed products) is
attenuated at a much greater rate than that for the active links in the WR network (see
Section 3.1).
Of course, the ability to employ unlicensed frequency bands has a dramatic effect on cost and
flexibility of wireless broadband solutions due to:
Ø Zero cost for spectrum license;
Ø Low cost radio equipment by taking advantage of components developed for other
unlicensed applications (such as wireless LAN);
Ø After equipment certification, no reliance on FCC approvals for roll-outs into each
market; and
Ø National or global spectrum availability, depending on the band.

4. MULTIMEDIA, BROADBAND SERVICE
Although the coverage and scalability benefits of wireless mesh networks are often reasonably
evident, the quality of service capabilities and benefits for handling multimedia broadband
traffic are often not as apparent. In particular, a couple of common questions are:
1. “Doesn’t the act of forwarding traffic over multiple wireless hops necessarily result in lower
end-to-end throughput?” and
2. “Aren’t wireless mesh networks too chaotic to provide the bandwidth and delay quality-of-
service assurances needed for multimedia traffic like voice?”
This section reviews possible sources for these questions and discusses characteristics and
design approaches for mesh networks that enable them to meet and exceed the demands of
high-speed multimedia traffic.
4.1 Throughput over Multiple Wireless Hops
Rather than sending packets directly from a subscriber unit to the base station (or AirHead)
over a direct link, mesh networks tend to route the packets through some number of (e.g., 1 to
3) intermediate subscriber devices before reaching the AirHead. Therefore, multiple
transmissions are required for each packet before it has arrived at the AirHead, rather than the
single transmission needed in a point-to-multipoint system. It is then natural to predict that
these multiple transmissions will result in a lower overall end-to-end throughput.11
However,
this prediction ignores one of the key features of wireless networks – their ability to adapt the
modulation rate based on the individual quality of each link. With this feature, smart wireless
routers can automatically increase the modulation rate to take advantage of the lower path loss
of the shorter links involved in multihop paths.
As an example
using standard
802.11a-style
radios12
(see
[802.11a]) with the
signal-to-noise
specifications in
Table 1, and each
with identical,
omni-directional antennas, consider the choice of routing over a single direct link from a
subscriber device to the base station (or AirHead) versus routing through an intermediate
device that’s centered between the other two devices. For the case of the routed path, two
transmissions must be made for every packet rather than one if the direct link is used.
However, assuming that equal transmit powers are used in each case, equal noise
environments at the receivers, and a 1/r2
path loss factor, then there will be a 6-dB advantage
in the signal-to-noise ratio for each of the transmissions over the routed path versus that for
the direct link. For 802.11a radios, this 6 dB advantage can translate to being able to use the
18 Mbps rather than 6 Mbps modulation rate for those transmissions. Therefore, for this case,
the effective potential throughput over the routed path would actually be 50% greater than that
for the direct link (9 Mbps effective throughput versus 6 Mbps), while generating 33% less total
Figure 19 – Choice of using direct link versus routed path.
Intermediate device
AirHead Subscriber
11
This discussion also assumes that single-transceiver wireless routers are used for the subscriber devices to
minimize their cost.
12
Note that this only refers to the 802.11a physical layer specifications, and not the 802.11 MAC protocol (see inset
titled “802.11 ‘ad hoc’ doesn’t mean multihop” for a related discussion on this).

co-channel interference power into other neighborhoods due to the reduced total time required
for the two transmissions.
For path loss exponents greater than 2, the advantage for the multihop route is even greater.
For instance, using a 1/r3
path loss factor in the above example, and with the distances
reduced to allow comparison against the same 6-Mbps data rate over the direct link, then the
multihop links would be able to use the 24 Mbps waveform resulting in a 100% potential
throughput advantage (12 versus 6 Mbps) and 50% reduction in interference generated to
other neighborhoods. Table 2 compares the effective throughput of using the direct link versus
using a multihop route for a variety of path loss exponents and number of equally spaced
intermediate forwarding devices.
Table 2 – Throughput comparison for direct links versus multihop routes
Using 802.11a-compliant radios.
able 2
Case Path
loss
exponent
Direct link
modulation
rate
13
Number of
intermediate
forwarders
Multihop
modulation
rates
Effective
multihop
throughput
14
Multihop
throughput
advantage
1 2 6 Mbps 1 18 Mbps 9 Mbps 50%
2 2 6 Mbps 2 24 Mbps 8 Mbps 33%
3 2 6 Mbps 3 36 Mbps 9 Mbps 50%
4 3 6 Mbps 1 24 Mbps 12 Mbps 100%
5 3 6 Mbps 2 36 Mbps 12 Mbps 100%
6 3 6 Mbps 3 54 Mbps 13.5 Mbps 125%
7 4 6 Mbps 1 36 Mbps 18 Mbps 200%
8 4 6 Mbps 2 54 Mbps 18 Mbps 200%
9 4 6 Mbps 3 54 Mbps 13.5 Mbps 125%
15
Figure 20 compares the performance benefit of selecting the two-hop route versus the direct
link for different values of the path loss exponent. Notice that the performance improvement
for the two-hop path increases at nearly the rate of the path loss exponent squared. For
example, as the path loss exponent doubles from 2.2 to 4.4, the performance improvement
increases by a factor of 4.
13
To ease comparison, the absolute link distances are modified so that the modulation rate achievable over the direct
link is 6 Mbps for each case. Across the range of the standard 6 and 54 Mbps 802.11a modulation rates, the increase
in data rate per dB in receive signal is fairly consistent, at approximately 2.7 Mbps per dB of signal improvement.
14
This table assumes the communication of relatively large (e.g., 1000-byte) packets, so that per-packet overhead
such as framing and MAC headers can be ignored. Also, fast-forwarding hardware or software is assumed which is
capable of forwarding IP packets at a rate that’s faster than they can be transmitted or received over the radio
interface.
15
The multihop throughput of case 9 in T is limited by the maximum 802.11a data rate of 54-Mbps, and thus has
an additional 7dB of margin over each link compared with the other cases.

0%
50%
100%
150%
200%
250%
300%
350%
400%
2 2.5 3 3.5 4 4.5 5 5.5 6
Path loss exponent
Throughputadvantageof2-hoppath
Figure 20 – Throughput advantage of the two-hop path versus the direct link for increasing values of the path
loss exponent using 802.11a-compliant radios.
Note that the following additional considerations have not been included in this analysis:
1. In highly obstructed regions, and with proper scheduling among sufficiently isolated
devices, mesh networks can also support simultaneous transmissions over multiple links
within single AirHoods. Such transmissions can of course result in a further increase in
performance for the mesh approach, within the limits of the capacity at any traffic
concentration point (in particular, at the AirHead).
2. Because wireless routers are able to select the neighbor routers to use for forwarding
packets based on the best RF-quality links available -- both at the link layer when selecting
neighbor links, and at the routing layer when selecting alternate paths to and from the
AirHead -- these multihop paths will tend to use links with lower path-loss exponents and
better multipath fading conditions than the direct link to the AirHead, potentially leading to
even higher modulation rates. However, this improvement is balanced somewhat by the
fact that these paths will in general involve intermediate devices that are not along the
direct line between the subscriber and the AirHead.
Comparison with Point-to-Multipoint
In point-to-multipoint networks, the path loss (and associated performance disadvantage) for
the longer links is sometimes overcome by the use of directional antennas at the subscriber
end.16
Such use of directional antennas also reduces the amount of (and vulnerability to) noise
to (from) subscriber devices in other neighborhood cells. However, the link-range advantage
of directional antennas is either limited by the relatively low EIRP transmit power limits of
unlicensed bands, or requires the use of licensed bands, which have higher EIRP limits. Also,
use of directional antennas at the subscriber premise is accompanied by the disadvantages of
16
For example, for case 8 in (1/r
4
path loss factor comparing the direct link to a multihop path with two
forwarding nodes), the throughput of the direct link can be improved to equal the multihop throughput by using a
subscriber antenna with 5dB additional directional gain (e.g., using 13 dBi directional antennas versus 8 dBi gain omni-
directional antennas for the multihop case).
Table 2

increased installation costs, increased vulnerability to environmental changes that may
obstruct a subscriber’s link, and scaling limitations due to the need to re-steer subscriber
antennas to associate them with different base stations when splitting cells or balancing
capacities.
These scenarios also assume that the noise level at each of the receivers is the same.
However, to meet coverage requirements, point-to-multipoint architectures typically require
that the base station be mounted on relatively high towers. For coverage purposes, this has
the desired effect of reducing the path loss to near free-space characteristics for a portion of
the path toward the subscriber devices (thus increasing the receive signal strength for the up-
and down-stream links); however, this is accompanied by the undesirable effect of also
reducing the path loss from, and increasing the effect of, interfering transmissions in the region
(in addition to the increased cost and logistics in locating, leasing, obtaining needed approvals
for, and installing the base station towers). This interference problem must then be overcome
by more careful channel planning and/or by the use of more sophisticated radio equipment.
Lastly, because base station towers with omni-directional or sector antennas are not required
for mesh networks, fully adaptive TDD/TDMA channel scheduling can be used to optimally
match the up- and down-stream traffic requirements of each AirHood. However, PMP
networks must often resort to FDD or synchronized TDD methods (synchronized across all
base stations within a region) to avoid base station-to-base station interference, essentially
fixing the up- and down-stream bandwidths (by spectral bandwidth in the FDD case and by
time in the TDD case). See Section 3.1 for more details on this.
4.2 Efficient, Fair and Reliable Channel Access
The wireless channel in multihop mesh networks comes with unique challenges to fair, efficient
and reliable transmission scheduling and link-layer service. Unlike point-to-point wireless or
wired links, each transmission must be coordinated among multiple neighbor devices; unlike
other shared networks such as Ethernet or fully-connected wireless networks, the set of
devices affected by each transmission differs greatly depending on which device is
transmitting; and unlike point-to-multipoint wireless networks, there is no master controller
within the range of every device to coordinate scheduling (also, see the inset box titled “802.11
‘ad hoc’ doesn’t mean multihop”). Furthermore, transmissions by pairs of devices that are
unable to successfully communicate can result in collisions at the receiver(s) (known as the
“hidden terminal problem”), and for further complication, devices that are within hearing range
of each other are still sometimes able to successfully transmit packets simultaneously to
neighbor devices that are isolated from the other transmitter, but would be restrained from
doing so using traditional channel access protocols.
These challenges have required the development of new channel access methods, designed
specifically for multihop mesh networks. With these new methods, combined with Internet
DiffServ-compliant packet handling, wireless mesh networks can become seamless extensions
of the wired Internet, able to handle the service demands of multimedia broadband traffic as
well or better than the rest of the Internet, only with wireless links between the routers rather
than wired links.
In order to provide the reader with a concrete design approach able to meet the service
challenges of multihop mesh networks, the following sections outline the key characteristics of
the mesh-mode channel access extensions currently included in the IEEE’s draft 802.16 MAC
standard. Section 4.3 then discusses how these channel access methods can be extended by

QoS-sensitive classification and packet handling methods to ensure high-quality service for
multimedia traffic.17
Network Timing Structure
First, the network uses a distributed protocol to ensure that all devices in the network are
synchronized to the same time base. Time is measured in microseconds, slots, frames, and
super-frames. Slots are 2x
microseconds in duration, there are 256 slots in a frame, and there
are 2y
frames in a super-frame. The first 2z
slots in each frame are reserved for control
packets, with the remainder of the frame being used primarily for data packets. (x, y & z are
integers.) The first control portion of each super-frame is reserved for “network configuration”
packets, while the control portion of the other frames is used for “data scheduling” control
packets. Figure 21 presents an example timing structure using the values: 32 ms per slot, 32
slots in the control portion of each frame, and 8 frames per super-frame.
65,536-ms super-frame
Network Configuration
Control
Data Scheduling
Control
Control
224-slot (7,168-ms) data portion
256-slot (8,192-ms) frame
32-slot (1,024-ms) control portion
32-ms slot
Figure 21 -- Example timing structure for multihop mesh network.
The first control portion in each super-frame is used for network configuration packets (the first
~1 ms in every ~65-ms super-frame in the above example). The main purposes for these
packets are:
· Initialization and maintenance of network synchronization;
· Support for secure network entry of new devices;
· Detection of new or lost links between devices;
· Measurement of the link qualities and path loss between devices; and
17
The channel access and QoS methods described are also those used in the Nokia RoofTop Wireless Routing
product family.

· Dissemination of the information needed by the distributed scheduling algorithm used
for these network configuration packet transmissions.
Data portion of frame
Dist. sched.A g 21 g A
Distributed scheduling2 g 3Channel 2 â
Channel 1 â
A
AirHead
2
Subscriber 2
3
Subscriber 3
1
Subscriber 1
Figure 22 -- Simple example network and data transmission schedule.
Because of their “boot-strapping” role, the scheduling of these network configuration packets
uses a fair, robust, and completely distributed algorithm, which ensures collision-free
transmissions among the devices in each local area. Also, by including a phantom device in
the scheduling, periodic measurements of the background noise can be made. With the noise
level and path loss information per link, the transmit power and modulation rate can be set
appropriately according to the network density and signal quality of each link.
Data Transmission Scheduling
The remaining control portions of each frame are used for transmitting data scheduling control
packets. The transmission scheduling for these control packets can be efficiently determined
by the device being used as an Internet access point (i.e., the AirHead). For example, super-
frames can be further grouped into “scheduling epochs.” During the first half of each
scheduling epoch, the current persistent traffic demands at each wireless router within an
AirHood are reported up to the nearest AirHead. The AirHead then computes a revised
AirHood schedule and, during the second half of the scheduling epoch, distributes this
schedule throughout the AirHood to become effective at the next scheduling epoch boundary.
The AirHood schedule consists of a list of active links, each associated with a {slot range,
channel assignment, frame number offset, and frame occurrence rate} set, to be used during
that scheduling epoch. In addition, the AirHood schedule can include a list of unscheduled slot
ranges, which can be used by the distributed scheduling method described below. These
assignments are then used in each frame to transmit data across the active links in a collision-
free manner. Figure 22 presents a simple example network and the snapshot of the schedule
for a particular frame which supports two active traffic flows, one down-stream from the
Internet to subscriber 3, and one up-stream flow from subscriber 1 to the Internet, and which
assumes availability of two channels for use by this AirHood.
With this scheduling method, and with proper subscriber provisioning (see Section 4.3), the
per-hop latency for at least for the high priority traffic will be bounded. For instance, with active
links that are allocated slot ranges in each frame, the per-hop latency for the high priority traffic
will be limited to about 8 ms using the timing structure of Figure 21. Note also that data
packets should be fragmented (and reassembled) at the link-layer to efficiently fill the slot
range in each frame for each active link. Also, link-layer acknowledgments with selective
retransmissions should be used to ensure reliable delivery of each packet (or packet fragment)
across each link; therefore, each active link must also be assigned at least a minimally-sized

slot range in the reverse direction for transmission of the acknowledgment packet. (These
small reverse slot ranges are not included in the figure above for clarity.)
Between schedule assignments, and to handle sporadic traffic needs on otherwise inactive
links, distributed data scheduling can be used. Distributed schedule control packets are
transmitted during the unscheduled portions of the current schedule, and are used to quickly
establish new, transient slot ranges for links. These transient slot ranges are valid only for the
number of frames indicated in the distributed schedule establishment packets (based on the
instantaneous traffic demand for the corresponding link), and any conflict with a new AirHood
schedule when it becomes effective (at a scheduling epoch boundary) is resolved by
terminating the conflicting slots established by the distributed schedule.
This approach to the channel access methods allows for the efficient transmission scheduling
of mesh wireless routing networks in both a distributed manner with the arrival of brief traffic
bursts, and for “persistent” flows lasting about a half second or more in a manner that provides
latency and throughput assurances for the duration of the traffic flow.
Comparison with Point-to-Multipoint
The methods described in this section to synchronize the network and establish AirHood
schedules have similarities with the methods specified in the draft 802.16 standard for point-to-
multipoint networks. However, for multihop mesh networks, these 802.16 PMP methods
required extensions to perform the AirHood scheduling across multiple hops from the AirHead,
to allow for distributed scheduling between scheduling epochs, and to provide distributed
scheduling methods for the network configuration and network entry packets.
On the other hand, the 802.11 MAC layer operates very differently using a prioritized
contention-based protocol. Even in point-to-multipoint wireless LAN networks, this protocol
does not provide reliable assurances for service-sensitive traffic under heavy traffic loads.
However, the 802.11e task group (see [802.11e]) is currently working on extensions to this
MAC protocol to substantially enhance its QoS support in point-to-multipoint WLAN networks.
4.3 Quality-of-Service Support for Multimedia Traffic
Section 4.2 described a MAC-layer approach, compliant with the draft 802.16 standard, for
wireless routing networks capable of providing latency and throughput assurances for user
traffic flows, while efficiently adapting to changing traffic demands. This section presents
wireless routing quality-of-service (QoS) methods that utilize this MAC-layer to ensure that its
assurances are applied to the end-to-end traffic flows that require them. (Note that given the
MAC approach outlined in Section 4.2, the methods in this section are to a large extent
applicable to any IP network product, particularly near the subscriber “edge.”)
These QoS methods involve the following primary components:
1. Regulation and classification of packets at the entry (or “ingress”) points to the wireless
routing network.
2. Prioritized scheduling, retransmissions, and intelligent packet dropping over each link in
the network according to the selected “class,” “drop precedence,” and “reliability” for
each packet.
3. Appropriate operator provisioning of per-subscriber service levels.

This section will review the role that each of these plays in order to provide the Internet
DiffServ-compliant QoS assurances needed for multimedia broadband service. (Also, see
Internet references [RFC2597], [RFC2598], [RFC2474], and [RFC2475].)
Ingress Regulation and Classification
At the boundary of the wireless routing network, packets are classified according to the type of
application and the level of service provided to a particular user. Figure 23 outlines the ingress
packet processing. First, the packet is classified into one of four target “assured forwarding”
classes (AF1 to AF4),18
and also marked with a (e.g., 2-bit) reliability value. As described in the
next subsection, packets belonging to the AF1 class receive the most favorable per-hop
treatment, and the reliability marking is used to determine how many retransmissions are
attempted for each link. One other possible outcome of the classification process is the “Filter
Drop” instruction, which results for packets that match the “firewall” rules configured into the
classification lookup. These packets are simply dropped without further processing. The class
and reliability classification can be done either by accepting the DiffServ Codepoint (DSCP) of
the received packet’s Type-of-Service (TOS) field, or by classifying the packet according to a
lookup based on the packet’s source and destination IP addresses, protocol, TCP or UDP port
numbers, and TOS-field settings.19
Packet for
forwarding
marked with
{AF class,
reliability & drop
precedence}
Regulate and
mark drop
precedence
Police
per-class
flows
Filter
Classify and
mark AF class
& reliability
Ingress
Packet
Filter Drop Regulator Drop
Figure 23 -- Ingress Packet Classification and Regulation.
The next step polices the level of traffic in each AF class being injected into the network by this
subscriber (or destined for a particular subscriber at the AirHead). Therefore, the traffic activity
for each AF class is monitored, and whenever a subscriber’s average rate exceeds that
subscriber’s limit for a particular class, AFx, then a given microflow that was assigned to that
class is downgraded to class AFx+1. (In this policing phase, there is no limit on the level of AF4
traffic.) A microflow is defined as a particular {source address, destination address, IP
protocol, source port, destination port} quintuplet. An entire microflow is downgraded, rather
than, for example, selecting packets at random among multiple microflows to avoid packet
reordering in the network (which can cause disruptions in end-to-end transport protocols or
applications). Over time, as the measured traffic rates for each AF class permit, microflows
can also be upgraded back to their target class.
Finally, the overall subscriber flows are compared against the peak, average, and guaranteed
minimum flows provisioned for this subscriber. Traffic within the peak flow limit is accepted
into the network, but marked with a “drop precedence” setting according to whether it is above
the average limit (DP set to 2), below the average but above the guaranteed minimum (DP set
18
Actually, there may also be other classes such as the Expedited Forwarding (EF) and Best Effort (BE) classes;
however, these are not included for clarity of this discussion.
19
The methods used to configure the data structures involved in this lookup are outside the scope of this document.
However, these methods could combine statically configured entries to recognize certain port ranges for example, with
dynamically updated entries based on an explicit flow establishment protocol or on detection of higher-layer flow setup
packets as they are forwarded by the ingress routers.

to 1), or below the guaranteed minimum (DP set to 0). Traffic in excess of the provisioned peak
flow limit for this subscriber is dropped.
Per-Hop Packet Handling
Following the above ingress operation, each packet traversing the wireless network is tagged
with a service-handling triplet: {Assured Forwarding (AF) Class, Drop Precedence (DP),
Reliability}. This information allows each wireless router along the packet’s path to provide
service-specific handling to control congestion intelligently, to set link parameters according to
the needed packet reliability, and to schedule packets according to their urgency. In particular:
· Each wireless router maintains a set of queues for each link corresponding to the
various traffic classes, AF1 to AF4. When a link becomes available for transmission of
new packets, the packets are selected from the classes according to an algorithm
which gives the highest priority to the AF1 queue, the next highest priority to the AF2
queue, and so on.20
As an example consequence, as long as the bandwidth scheduled for a given link is
sufficient to transmit the packets that may arrive in the link’s AF1 queue during a frame
period, then those packets tagged with AF1 will experience a single-frame delay, at
most, across each hop (which is about 8 ms using the example described in Section
4.2).
· When the link queues in a wireless router become too large (requiring many frames to
empty them and threatening exhaustion of the router’s memory resources), then
packets are dropped using a random early dropping algorithm weighted according to
the Drop Precedence tags on the packets.
Therefore, for example, when the network encounters congestion, packets belonging to
subscribers that are above their provisioned average flow rates will be the first to be
dropped.
· Lastly, the packet’s reliability setting is used to indicate how many retransmissions
should be attempted before the packet is discarded or returned to the routing layer. For
example: “0” for a single transmission, “1” to attempt up to one retransmission, “2” to
attempt up to three retransmissions, and “3” to attempt up to six retransmissions.
The classifier’s selection of the reliability setting will depend both on the timeliness and
the importance of the application data. For wireless networks, it is typically more
efficient to retransmit packets at the link layer rather than rely on retransmissions using
end-to-end protocols. However, for data that quickly becomes stale (e.g., voice), only a
small number of transmission attempts are appropriate.
20
To ensure that no class is “starved” of bandwidth regardless of the traffic load, this packet selection algorithm could
also include a “bandwidth” component whereby each queue is ensured of at least some percentage of the link
bandwidth, given that the queue has packets to transmit.

802.11 “ad hoc” doesn’t mean multihop
A point of confusion in the wireless networking industry is due to the multiple uses of the phrase “ad hoc.” In
some contexts, such as in the MANET (Mobile Ad hoc Networks) working group of the Internet Engineering Task
Force (IETF), “ad hoc” is used to describe multihop mesh networks due to their distributed, self-configuring
nature. However, “ad hoc” has also been used in the specification of the 802.11 MAC protocol to describe a
channel access mode that can be used in “a network composed solely of stations within mutual communication
range of each other.” In other words, the 802.11 “ad hoc” mode is designed for use in a wireless network where
all of the nodes have a direct one-hop link to each other. However, because of the prevalence of 802.11
devices, it is tempting to use them, complete with their MAC protocol, in multihop mesh prototypes. In small
networks, such prototypes are able to pass “proof-of-concept” demonstrations. However, as the size, density, or
traffic load increases, these networks quickly encounter the limitations of this MAC protocol applied to an
environment for which it wasn’t designed.
The 802.11 MAC operates in one of two primary modes: one mode is designed for the traditional point-to-
multipoint WLAN architecture where the Access Point controls access to the channel; another “ad hoc” mode
allows nodes to schedule transmissions in a distributed, contention-based manner using a request-to-send
(RTS), clear-to-send (CTS) setup handshake before each packet burst. For small, mesh networks in which every
node is a direct neighbor of either or both of the nodes of every link in the network, and relatively tame traffic
conditions, the 802.11 ad hoc channel access mode can perform reasonably well because neighbor nodes that
overhear the setup handshake will avoid making interfering transmissions for the duration of the packet burst,
and because the carrier sensing is more likely to indicate a condition for which it is indeed better to remain silent.
However, as the size, density, or traffic demands of the network increases, the performance of the 802.11 MAC
degrades substantially due to issues such as: collisions by “hidden terminals,” particularly among the RTS and
CTS packets; insufficient support for fair treatment of multiple traffic streams; and unwise decisions based on
carrier sensing. In particular, the following problems surface:
· Low throughput – The 802.11 MAC protocols are ill equipped to efficiently coordinate the transmissions
of multiple nodes with packets to send in a multihop mesh network. One result is low end-to-end
throughput for the user traffic. The problems encountered include collisions among simultaneous
transmissions, and wasted airtime during which transmissions could have been made successfully.
· High latency and latency variance – Because the 802.11 MAC protocol is contention-based, combined
with other problems mentioned above for multihop mesh networks, it is unable to provide the latency
guarantees needed for services like voice, or useful for improving TCP performance. Task Group E of
the IEEE 802.11 standards group is working on extensions that should improve its quality of service
performance in point-to-multipoint networks (see [802.11e]).
· Extreme unfairness – The 802.11 MAC protocol has no guaranteed method for informing neighbor nodes
of the traffic demands for the links to them. This leads to extreme unfairness in the servicing of multiple
simultaneous traffic streams that flow through shared network nodes.
· Unreliable link quality assessment – The 802.11 MAC protocol has no mechanism to reliably monitor the
quality of the links among neighboring nodes. Therefore, higher layer protocols must rely on measures
such as the success rate of periodic probe packets, which are themselves unreliable due to problems
identified above. The consequences include: false indications of link failures (causing unneeded routing
perturbations); delayed detection of new links; and overly coarse approximations of the link qualities
resulting in poor decisions for the transmit power and/or modulation rates to be used for each link.
For further information on some of the performance problems with using the 802.11 MAC protocol for multihop
mesh networks, see [Xu01] and [UCSC-MAC].

Network Service Provisioning
With the MAC-layer and QoS mechanisms described above, it becomes possible to provision
the services of an AirHood according to subscriber service agreements and while providing the
appropriate level of per-subscriber assurances for: 21
Ø Latency – for latency-sensitive traffic (e.g., some number of voice “lines”), and
Ø Bandwidth – specified as guaranteed, average and peak bandwidth levels.
For example, given an AirHood with 10 small business and 40 residential subscribers with a
relatively spread-out distribution pattern for the distance from the AirHead (25%, 50%, 15%,
10% for 1-, 2-, 3-, and 4-hop residential subscribers, and 50%, 50% for 1- and 2-hop business
subscribers; see Section 2), that all links are using the 802.11a 24-Mbps modulation rate or
better, that the total network overhead is 25% (so effective link speeds are 18 Mbps), that 16-
kbps encoding is used for the voice streams, that only a single channel is available to the
AirHood, and assuming the worst case that only a single transmission at a time is possible
within the AirHood, then the following represents a reasonable provisioning using a 10-to-1
statistical multiplexing ratio for the average bandwidth computation.
10 small businesses Number of assured voice calls 422
Guaranteed bandwidth 768 kbps23
Average bandwidth 3 Mbps
Peak bandwidth 6 Mbps
40 residential subscribers Number of assured voice calls 2
Average bandwidth 1.5 Mbps
Peak bandwidth 3 Mbps
As another example, an all-residential AirHood with 100 subscribers could be provisioned with
the same level of service provided for the residential subscribers in the above table by using
the same assumptions except for relaxing the statistical multiplexing ratio for the average
bandwidth computation to 18-to-1.
21
This AirHood provisioning could be done by computing the capacity limits of the AirHood given its connectivity and
link data rates, or by following conservative guidelines based on the size of the AirHood and typical link data rates in
the target environment.
22
In this example, all voice calls will be classified as AF1 traffic (highest priority user traffic), and the number of voice
calls with assured service is used to limit the amount of traffic classified as AF1 for a given subscriber. E.g., for a
subscriber provisioned for up to four 16-kbps voice streams, the AF1 flow limit will be set to 64 kbps.
23
Because the network is fully adaptive to the up- and down-stream needs, bandwidth is provisioned as the total
allowance for up- and down-stream traffic rates.

5. CONCLUSION
The purpose of this document is to present the fundamental core characteristics of wireless
routing (mesh) networks with attention to the cost, coverage, scaling, and service benefits it
provides for wireless broadband access networks. Example benefits include:
Ø Full coverage of residential neighborhoods with substantially lower system gain than
required for point-to-multipoint architectures, translating directly to lower equipment
costs, increased installation flexibility, etc. For example, to reach an 85% to 99%
coverage target, 50-device wireless routing networks require 35 to 45 dB (respectively)
lower system gain than is required for point-to-multipoint networks, using the scenario
presented in Section 2.5 (cells or AirHoods 1-mile in radius, mean path loss exponent
of 3, etc.).
Ø Natural scaling to large, dense networks, and the reliable use of unlicensed frequency
bands, which are enabled by a favorable relationship between the propagation
characteristics of the active links in the network and the paths followed by the
interference, versus point-to-multipoint networks which must cope with an unfavorable
relationship between these signals (see Sections 3.1 and 3.2).
Ø Automatic selection of routing links combined with adaptive setting of the modulation
rate used for each link to actually increase the end-to-end bandwidth of subscriber
traffic streams by routing over multiple hops, rather than attempting to use the longer,
possibly more obstructed, direct links (see Section 4.1).
Ø The automatic use of alternate paths when links fail, which substantially reduces the
need for operator “repair” visits and reduces the fading margin needed for the system
(again translating to lower equipment costs, etc.). Specifically, approximately 6 dB of
decreased shadow fading margin is required compared with point-to-multipoint (see
Section 2.4).
Of course, complete wireless router products require a full complement of other capabilities to
address the needs both of an infrastructure router, and as an access device for the end
subscriber. These requirements can include:
· Full Internet Protocol support to permit the wireless mesh networks to serve as
seamless extensions of the Internet;
· Home gateway functions (such as the Internet’s DHCP and NAT protocols) to ease the
connection and permit self-configuration of the home networking devices;
· An efficient, Internet-compatible routing protocol able to integrate link quality and link
modulation rate into routing decisions;
· Network synchronization, link quality management and scheduling algorithm support for
the MAC-layer framework described in Section 4.2; and
· Support for network management, subscriber provisioning, network security, and
remote, “over-the-air” firmware upgrades.
6. ACKNOWLEDGMENTS
This paper benefited greatly by the helpful review and discussions with Darren Lancaster,
Greg Desbrisay, Andy Kelm, JJ Garcia-Luna-Aceves, and Nico VanWaes, and also by editorial
help from Jo Albers and graphical contributions from Darci Hermann.

A. APPENDIX A – SIGNAL-TO-INTERFERENCE CAPACITY SCALING DERIVATION24
A.1 Local Neighborhood
1. The network deployment illustrated in Figure 16 is used for this analysis.
2. A node’s local neighborhood is defined as the radius (Rn) required for the node to reach its
8 nearest neighbors (RF-wise). A node’s “local region” of radius (Rl) is then computed as N
(e.g., 2 to 5) times the neighborhood radius. For a wireless router network, Rl is roughly 1
to 1½ times the distance between AirHoods that are using the same channel. For nodes
scattered randomly over a plane at density D nodes/km2
, the average radius Rl equals:
metersDR
DRNRR
meters
D
NR
n
nnl
l
p
p
p
10003
9
3000
2
×=
==
×=
.
Rn
i Ra
j
Rl
3. Transmission scheduling and channel assignment are coordinated such that there will be
only a single transmission (that from node i to node j) within radius lR around receiver j.
4. Nodes use adaptive power control to ensure that the signal arrives at the intended receiver
at some target level, TdBm.
A.2 Receive Power of Desired Signal
1. The receive power (P) into the antenna port of a receiving node from an arbitrary
transmitting node can be approximated with the following formula where y is the path loss
exponent, r is the distance between the nodes (in meters), X is the transmit power used, Ax
is the transmitter antenna gain minus connector losses, Ar is the receive antenna gain
minus connector losses, and C accounts for the reference far-field path loss.
)(log10 10 ryCAAXP rxdBm --++=
2. However, local path-loss characteristics are highly variable, dependent mainly on the
presence and type of any obstructions in the path. The selection of links chosen by a
mesh network to use for routing packets is not random; they are the best links available
between nearby nodes. Therefore, estimating the path loss for these links requires the use
of a significantly lower average path loss exponent (x).
24
This analysis is similar to the one performed by T. Shepard in [Shepard95]. Some key differences are that Shepard
used only a single path-loss exponent for both the signal and the interference, and assumed multi-transceiver radios
with a high spreading gain modulation which simplifies channel assignment and transmission scheduling
considerations.

3. Given that the receive power at the intended receiver is TdBm (ensured by adapting the
emitted power at the transmitter appropriately), the propagation formula between
transmitting node i and the intended receiver j is given by:
)(log10)( 10 rxCAArXT rxdBm --++=
4. Using the average transmit power X, associated link distance RT (derived below), and
solving for C:
)(log10 10 TdBmrx RxTAAXC --++=
A.3 Average Transmit Power and Associated Link Distance
1. The transmit power for any link can be derived from above:
x
CAAT
mw
rxCAAT
mw
rxdBmdBm
rrX
rX
rxCAATrX
rxdBm
rxdBm
10
10
)(log10
10
10)(
10)(
)(log10)(
10
+--
++--
=
=
++--=
2. Assuming an equal likelihood of transmitting to each of a node’s routing neighbors, then
the average transmit power (X), and associated link distance (RT), can be computed by
integrating the transmit power required to reach each of the neighbors up to this average
distance, and setting that to equal the same integration from that distance to the edge of
the nodes neighborhood area, at radius Rn.
DR
orRR
RRR
x
r
x
r
drrdrr
drrrdrrr
drrrXDdrrrXD
x
T
n
x
T
TnT
R
R
xRx
R
R
x
R
x
R
R
x
CAAT
R
x
CAAT
R
R
R
xxx
n
T
T
n
T
T
n
T
rxdBm
T
rxdBm
n
T
T
p
pp
30002
,2
22
1010
)(2)(2
)2/(1
)2/(1
2
0
2
1
0
1
10
0
10
0
222
×=
=
-=
+
=
+
=
=
=
+-
+-
++
++
+--+--
+++
òò
òò
òò

3. The average transmit power is thus:
( )
)(log10
,)(log102log
2
10
210
210
10
1010
)2/(10
)2/(110
TrxdBmdBm
nrxdBmdBm
x
n
xx
CAAT
mw
x
n
x
CAAT
mw
RxCAATX
orRx
x
x
CAATX
RX
RX
rxdBm
rxdBm
++--=
+
+
-+--=
=
=
+-
+--
+-
+--
A.4 Receive Interference Power
1. The average interference power P received at node j from an arbitrary interfering
transmitter at distance r is:
)(log10)(log10
)()(log10
1010
10
ryRxTP
RrryCAAXP
TdBmdBm
lrxdBm
-+=
³--++=
2. Converting to mWatts:
y
x
T
T
mW
ryRxT
mW
ryRxT
mW
r
R
P
P
P
dBm
TdBm
TdBm
×=
××=
=
-
-+
10
10
)(log10
10
)(log10
10
10
)(log10)(log10
10
101010
10
1010
1010
3. Outside of the receiver’s local region, nodes will be using the same channel used by this
AirHood with a 25% probability due to the 4-channel architecture used for this analysis.
So, given that the density of nodes on the same channel is D/4 [or (D/4)x10-6
) nodes/sq-
meter] and given an average transmit duty cycle a for each node, then the total
interference power PT
received by node j from all interfering transmitters (from Rl out to the
radio horizon Rh) plus the thermal noise floor BdBm is:
r
j
Rl
÷÷
ø
ö
çç
è
æ
-××
×-
×+=
--
××
×
×××+=
úû
ù
êë
é
××
×
×××+=
×
×
×××+=
--
-
-ò
ò
226
1010
26
1010
16
1010
6
1010
11
10)2(2
1010
)2(
1
2
104
1010
1
2
104
1010
2
104
1010
y
h
y
l
x
T
TB
T
mW
R
R
y
x
T
TB
T
mW
R
R y
x
T
TB
T
mW
R
R y
x
T
TB
T
mW
RR
R
y
aD
P
ry
D
aRP
dr
r
D
aRP
drr
D
a
r
R
P
dBmdBm
h
l
dBmdBm
h
l
dBmdBm
h
l
dBmdBm
p
p
p
p
dr

4. Converting back to dBm:
÷
÷
ø
ö
ç
ç
è
æ
÷÷
ø
ö
çç
è
æ
-××
×-
×+= -- 226
1010
10
11
10)2(2
1010log10 y
h
y
l
x
T
TB
T
dBm
RR
R
y
aD
P
dBmdBm
p
.
A.5 Receive Signal-to-Interference Level
The receive signal-to-interference level is therefore given by:
DRorRR
horizonvisiblehorizonradioR
DNRwhere
RR
R
y
aD
TSIR
PTSIR
x
Tn
x
T
h
l
y
h
y
l
x
T
TB
dBmdBm
T
dBmdBmdBm
dBmdBm
p
p
p
300022
3/4~
/3000
11
10)2(2
1010log10
)2/(1)2/(1
226
1010
10
×==
=
×=
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
-××
×-
×+-=
-=
+-+-
--

REFERENCES
[802.11] IEEE, “Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) specifications.” IEEE Standard 802.11, June, 1999.
[802.11a] IEEE 802.11, Amendment 1, “High-speed Physical Layer (PHY) in the 5 GHz
band,” IEEE Standard 802.11a, June, 1999.
[802.11e] Task Group E of the IEEE 802.11 standards group, web address:
http://grouper.ieee.org/groups/802/11 .
[Beyer90] D.A. Beyer, “Accomplishments of the DARPA Survivable Adaptive Networks
(SURAN) Program,” Proceedings of MILCOM Conference (Monterey,
California: 1990).
[Beyer99] D.A. Beyer, M.D. Vestrich, and J.J. Garcia-Luna-Aceves, “The Rooftop
Community Network: Free, High-Speed Network Access for Communities,”
chapter in “The First 100 Feet: Options for Internet and Broadband Access,”
The MIT Press, Cambridge, Massachusetts, 1999, ISBN 0-262-58160-4.
[Garcia-Luna97] J.J. Garcia-Luna-Aceves, C.L. Fullmer, E. Madruga, D.A. Beyer and T.
Frivold, "Wireless Internet Gateways (WINGS)", Proc. IEEE MILCOM'97,
Monterey, California, November 2-5, 1997.
[Jubin87] J. Jubin and J. Tornow, “The DARPA Packet Radio Network Protocols,”
Proceedings of the IEEE (January, 1987).
[Kahn78] R.E. Kahn, S.A. Gronemeyer et al, “Advances in Packet Radio Technology,”
Proceedings of the IEEE (November, 1978).
[Rappaport96] T.S. Rappaport, “Wireless Communications: Principles & Practice,” Prentice
Hall, Inc., Upper Saddle River, NJ, 1996, ISBN 0-13-375536-3.
[Rappaport98] T.S. Rappaport, “RF Propagation and System Design Techniques for
Broadband Wireless Applications from 5 to 40 GHz,” 1998 Bellcore Horizons
Workshop, available via http://www.mprg.ee.vt.edu.
[RFC2597], [RFC2598], [RFC2474], and [RFC2475]
Internet “Request for Comments” (RFC) documents, available at
http://www.ietf.org/rfc.html .
[Seidel91] S.Y. Seidel, T.S. Rappaport, S. Jain, M. Lord, and R. Singh, “Path Loss,
Scattering and Multipath Delay Statistics in Four European Cities for Digital
Cellular and Microcellular Radiotelephone,” IEEE Transactions on Vehicular
Technology, Vol. 40, No. 4, pp. 721-730, November, 1991.
[Shepard95] T.J. Shepard, “Decentralized Channel Management in Scaleable Multihop
Spread-Spectrum Packet Radio Networks” (doctorial thesis, Massachusetts
Institute of Technology, MIT/LCS/TR-670, 1995).
[UCSC-MAC] “Media Access Control” section of the Computer Communications Research
Group’s web site at the University of California, Santa Cruz, maintained by
Prof. J.J. Garcia-Luna-Aceves: http://www.cse.ucsc.edu/research/ccrg .
[Xu01] Xu, S, Tarek, S, “Does IEEE 802.11 MAC Protocol Work Well in Multi-hop
Wireless Ad Hoc Networks?” IEEE Communications, June 2001.

Mesh_Fundamentals_14dec01

More Related Content

What's hot (19)

Viewers also liked (13)

Similar to Mesh_Fundamentals_14dec01 (20)

Mesh_Fundamentals_14dec01