Anatomy of a Radio LAN

3 Anatomy of a radio LAN
A radio network is a collection of nodes communicating together through radio devices,
using radio waves to carry the information exchanged (obvious, isn't it ?). It is sometime
called a radio Ethernet, by analogy of the wired technology. Most radio devices are a
card (IA, !cmcia) to "lug in a !# (or wor$station), and interact directly with the
standard networ$ing stac$ on it (no need of !!! or any s"ecific "rotocol stac$).
3.1 The radio modem
A radio device is com"osed of two main "arts. %he first is the radio modem. %his is the
"art transmitting (modulating) the data onto the fre&uency and receiving other
transmissions. It is com"osed of antenna(s), amplificators, frequency synthesisers, filters
and other bits of magic. %hese are mainly analog "arts, and a bit of digital (in an AI#,
the Baseband).
'sually, you can't see all those analog bits (and the cleverness of the board layout)
because all the modem is enca"sulated in a metal shield to "rotect your !# from those
high fre&uency radiations.
%he modem main characteristics are the frequency band, the signalling rate, the
modulation and the transmitted power. !eo"le building modems are also tal$ing a lot of
() and d*...
3.2 The MAC controller
%he second "art of the radio device is the MAC controller, res"onsible to run the MA#
"rotocol. %his is im"lemented mainly in an AI# and+or a microcontroler on the card, but
some functionalities of the MA# may be as well in the driver on the !#. %he card also
includes some memory for the MA# controller to store incoming and outgoing "ac$ets
(buffers) and other data (configuration, statistics).
Most of the time the few most time critical "arts are handled in the radio modem AI#
(the baseband), the bul$ of the MA# in a microcontroller and only some management
functionality in the driver. *ut, the different manufacturers "lace the boundary between
the different functionalities differently (cost+"erformance tradeoff), and some have
im"lemented driver only MA#s for lower cost.
%he main characteristics of the MA# are the packet format (si,e, headers), the channel
access mechanisms and the network management features. %he amount of on-board
memory is also im"ortant, because the MA# may need a significant number of buffers to
com"ensate the !# and interface latencies.
Functional diagram of a Wireless device :
3.3 The host interface
%he card interface to the !# through one of its buses (!"# $%# $cmcia...) or
communication "orts (serial# parallel# &!B or Ethernet). %his interface allows the
software (mostly the driver) to communicate with the MA# controller and most of the
time directly to the on board memory (the software writes "ac$ets to a s"ecific location
of it, then the controller reads them and sends them).
%he main characteristic of the interface is mainly the s"eed (i+o, shared memory or .MA)
and the ability to "rocess re&uests in "arallel. %he flexibility and functionality of it are
usually more a concern for the "erson writing the driver /-)
3.4 The driver
0ith all modern o"erating systems, the end a""lication doesn't access directly the
hardware but use a standard A!I. %he o"erating system needs a driver to interface the
hardware to the networ$ stac$ ('%$($# )etBeui# $*...). %he main function of the driver
is to manage the hardware and to answer its re&uest (to service interru"ts). In most of the
0ireless 1A(s, the driver also im"lements some "arts of the MA# "rotocol.
%he main characteristic of the driver is the bugs /-(
3.5 ireless LAN or not
0ireless 1A(s are not the only devices to ma$e use of wireless technology, and it's easy
to get confused between the different "roducts (es"ecially that sometimes they call
themselves incorrectly wireless networ$s). ome exam"le are wireless bridges, wireless
distribution systems and cable replacement, and they are &uite different from local area
networ$ing. %here is also wide area wireless networ$ "roducts, which are again &uite
different from 1A(s.
ireless !rid"es are used to connect two different 1A( segments via radio, for exam"le
between two buildings across the street. ireless distri#$tion systems is what are used
by I! to connect multi"le inde"endant customers to a base station, li$e houses in a
neighbourhood. Ca#le re%lacement is mostly li$e Ir.A (Infrared data lin$) to transfer
data between two com"uters without a serial or "arallel cable.
ometimes those "roducts use standard 0ireless 1A( modules, and most of the time
they are based on the same technologies as 0ireless 1A(s but with restricted
functionality (li$e no broadcasting) and only allow a set of "oint to "oint lin$s (so, no
native %#!+I! to"ology). %hey interface to the serial "ort (cable re"lacement) or ethernet
"ort (wireless bridges, wireless distribution system).
In this document we mostly restrict ourselves to true wireless 1A(s, because what
doesn't run natively %#!+I! is not 2fun2 /-)
3.& 'rofessional and (ome ireless LANs
(ow that 0ireless 1A(s are getting towards lower "rice, 0ireless 1A( manufacturers
are no longer targeting mobile commercial users only but also the home mar$et. ome
vendors, such as !roxim, offer two distinct line of "roduct based on the same technology
(and same "rotocol), the )ange1an3 for "rofessionals and ym"hony for home users.
As the business version of those 0ireless 1A(s are more ex"ensive than the home
"roducts, one might wonder what 4ustify the "rice difference a"art from the "ac$aging,
the mar$eting and software bundle.
%he radio modems may "resent different %erformances. %he modem is usually the most
ex"ensive "art of the device, and re"lacing analog "arts by less "erformant ones may
reduce the "rice. %he result may be a lower sensitivity, or less filtering of the ad4acent
bands or channels, which may reduce range and "erformance, es"ecially for high number
of nodes or collocated networ$s (which matter most for business environment).
%he host interface may be different. %he business line may offer more o"tions, such as
5thernet, erial and !#I, whereas home version may offer '*. %he home line may also
lac$ security (through encry"tion) or "ower management.
*ut in most cases, the hardware between the two lines is exactly the same. In fact, most
of the differences usually reside in the Access 'oints. %his is why 1ucent offer 6
different Access !oints de"ending on usage and targeted at different $ind of users, but
only one ty"e of card for all ty"es of users.
Access !oints for home users are mostly designed to interface with a "hone line (or
I.(, .1 or cable modem) and "rovide a "roxy or mas&uerading feature, allowing the
user to share its I! access between the nodes of the networ$.
7n the other hand, Access !oints for businesses connect directly to the 1A( via 5thernet
or act as wireless re"eaters, with o"timised bridge functionality, higher "erformance,
offer a wide range of management features (diagnostic, statistics, access control...)
roaming and out of range forwarding (see chapter +,-,.).
o, before investing your money, you have to as$ yourself what networ$ configuration
you are really after and which features you really do need...
3.) *i"ital radios and chan"in" the %rotocol
7ne &uestion "o""ing u" in my mailbox is the ability of doing "rotocol 89' (%.MA,
0ireless A%M) with device 8:' (a well $nown 0ireless 1A(). A variant of this &uestion
is "eo"le trying to im"lement a s"ecific scheme or o"timisation in the ;<3.== "rotocol.
%his is usually not "ossible. As we have seen above, most of the MA# "rotocol is
actually embedded in the device and only a few non "erformance critical functions are
handled by the driver on the host. 'sually, manufacturers don't tell you how to re"rogram
the firmware of their devices, but even if it was "ossible, it would not be enough.
%he very low "art of the MA# "rotocol, which is time critical, is im"lemented in the
baseband AI#, so &uite a challenge to change. >or exam"le the carrier sense and MA#
ac$nowledgement need reaction in the order of a do,en microseconds, so are "rime
candidate for the AI#. 'nfortunately, these are "recisely the functions that those "eo"le
want to change.
In fact, many "eo"le have been thin$ing of universal radios, which can be sim"ly
re"rogrammed to receive (and transmit) any radio standard. %he main idea is having a big
bloc$ of re"rogrammable logic on the card and to download a new configuration for each
"rotocol that the system wants to use, ma$ing it a fully di"ital radio.
%o achieve that goal, we need to go one ste" further down, and be able to ada"t to any
modulation and bit rate. Most im"lementations of common 0ireless 1A(s use fixed
analog com"onents in the modem, so are not suitable. o, a digital radio needs to digitise
(with a fast Ato.) the whole bandwidth and to feed that the a fast su"er .! or 5!1.
(5lectric !rogramable 1ogic .evice, li$e a 9ilinx or Altera) and to wor$ entirely in the
digital domain to demodulate (and modulate) the signal. 'nfortunately this is not really
cost effective and doesn't wor$ that well at the fre&uency we are tal$ing about (?@,).
4 The radio modem +%hysical layer,
%his section of the document deals with all the issues related to the "hysical layer
(bottom of the "ile, 7I wise /-), or in our case the radio modem.
4.1 -.M fre/$ency #ands +011 M(2 3 2.4 4(2,
In every country, the use of the radio s"ectrum is re"$lated by some organisations. %his
is the F%% for (orth America and the E'! for 5uro"e. %hese regulators define the
allocation of each radio fre&uency bandwidth / for %A and radio broadcasting, for the
telecommunication o"erators, for the army... 'sually, to use a fre&uency band, you must
negotiate with these bodies, register your architecture and buy the right to use the
fre&uency.
%hese organisations, aware of the "ros"ects of local radio communications for individual
users, have allocated some s"ecific fre&uency bands to be used in a more flexible way.
%he oldest and most commonly used ones are located at B<< M@, and 3.6 ?@, and
called the -.M #ands (ndustrial# !cientific and /edical). %he main characteristic of
these bands is that they are $nlicensed, this means that the user is free to use them
without having to register or to "ay anything (a"art from the radio hardware).
7f course, to avoid abuses, these organisations have im"osed a set of rules for these
fre&uency bands and only the "roducts certified to conform to those rules are allowed to
emit in the bands. %hese rules s"ecify at least the maximum "ower transmitted in the
band and the out of band emissions (to not "ollute ad4acent bands). %he IM bands rules
s"ecify as well that .%read .%ectr$m has to be used (either 0irect !equence or
Frequency 1opping, see chapter 2,-), and how the channels are defined, to allow the
"eaceful cohabitation of different systems (that's the theory).
%he "read "ectrum rules mandate 0irect !equence systems must s"read their signal at
least == times, and that Frequency 1opping systems stay on a channel a maximum of <.6
s and use CD channels at minimum in each E< s "eriod. *ut, don't trust me, chec$ the
exact wording of the rules...
%hese rules may vary de"ending on the country / the >## allocates both the B<< M@,
and 3.6 ?@, band with = 0 maximum "ower, whereas the 5%I allocates only the 3.6
?@, band with =<< m0 maximum "ower (B<< M@, is used for ?M cell "hones in
5uro"e). %he 3.6 ?@, band is available worldwide and the regulations are mostly
com"atible between the different authorities (usually ;< M@, of bandwidth between 3.6
?@, and 3.6; ?@,). %he main exce"tion is Fa"an which has some additional constraints.
%he "read "ectrum rules originally allowed around 3 Mb+s maximum bit rate (both >@
and .), but the .irect e&uence "eo"le managed to find a loo"hole and now offer ==
Mb+s systems (see chapter 2,3,-).
*ecause these bands are 2free2, they may be heavily %oll$ted by other unlicensed
systems. %he 3.6 ?@, band also suffers from the microwave oven radiations (this
ex"lains why it was given for free).
!lease note that the regulation for unlicensed bands is &uite different from the bands
reserved for radio amateurs (@AM). @AM "eo"le are not ha""y because their regulations
are much more strict (they have to "ass an examination including morse code and follow
stricter eti&uette) and the bandwidth available to them much more scarce.
4.2 5 4(2 fre/$ency #ands +(i%erLan and 5N-- #and,
%he D ?@, unlicensed bands are another very com"licated story.
5%I was the first to o"en the D ?@, band, and so far, the D.3 ?@, band is dedicated to
(i%erLan (see chapter 4,-), and the D.6 ?@, band reserved for (i%erLan -- (alias
*)A(, see chapter 4,2). As they have done for 5!/ and 0E%', only systems that fully
conform to those standards (!hy and MA#) may o"erate in the band.
In the tates, the >## has allocated the band between D.3 and D.; ?@, (5N-- #and) with
some very liberal rules (no s"read "ectrum mandated, no channels allocated). %o limit
systems, they have introduced com"licated "ower rules, ma$ing the use of around 3<
M@, bandwidth o"timal (system using less bandwidth can transmit less "ower, system
using more bandwidth don't get more "ower), and divided the band in E chun$s, for low
"ower systems (D.3 ?@,), medium "ower (D.6 ?@,) and high "ower (D.G ?@,). ome
"eo"le have tried to come u" with some 2eti&uette2 for the '(II band (stricter set of
rules) but they couldn't accommodate the conflicting re&uirement of all "arties.
In the D ?@, band, because of the availability of more bandwidth, higher s"eed are
"ossible (=< to 6< Mb+s). *ut, o"erating in a higher fre&uency band increases the noise
level, obstacles and walls are more o"a&ue to transmissions (see chapter 2,6,2), and a
higher bit rate re&uire more () (ignal (oise )atio - see chapter 2,4,2), which means a
reduced range com"ared to 3.6 ?@H "roducts, which is bad news.
In summary, in 5uro"e it's @i"er1an or nothing. In the 'A, the low "ower chun$ of the
'(II band (D.3 ?@,) is li$ely to be used by ;<3.== at D ?@, (see chapter 4,.) and
@i"er1an, and "eo"le are unli$ely to "ro"ose yet another standard. %he high "ower chun$
will be used by wireless distribution systems, and both ty"e of system will fight for the
medium "ower chun$...
4.3 .%read .%ectr$m techni/$es
.%read s%ectr$m is a techni&ue (mainly "ioneered by the army) trading bandwidth for
reliability. %he goal is to use more bandwidth than the system really needs for
transmission to reduce the im"act of localised interferences (bad fre&uencies) on the
system. "read s"ectrum, as it "revents one system to use the full bandwidth ca"acity,
also force inde"endant systems to share the bandwidth (in a mostly fair way). In the 3.6
?@, band, the regulation s"ecifies that systems have to use one of the two main s"read
s"ectrum techni&ue / 0irect !equence or Frequency 1opping.
0hich one is better ? %his is the main technical war between the radio 1A( vendors.
5verybody, of course, argue that its own technology is better. >or now, no one has come
with some decisive arguments about the com"arative "erformance and robustness of
these two technologies (estimating "erformance of radio systems is a tric$y 4ob). 7f
course, com"aring "roducts doesn't ma$e sense because the "erformance of a system
de"end on many other com"onents (the MA# "rotocol, the signalling rate), the
o"timisation chosen ("erformance versus reliability versus cost) and the actual
im"lementation (hum, hum...).
4.3.1 *irect .e/$ence
%he "rinci"le of *irect .e/$ence is to spread the signal on a larger band by multi"lexing
it with a signature (the code), to minimise localised interference and bac$ground noise.
%he system wor$s over a fixed large channel. %o s"read the signal, each bit of the "ac$et
to transmit is sur-modulated by a code (a fast re"etitive "attern). In the receiver, the
original signal is recovered by receiving the whole s"read channel (averaging effect) and
demodulating by the same code ("rocessing gain). >or a 3 Mb+s signalling rate modulated
by a == chi"s code (li$e the 0avelan), the result is a signal s"read over 33 M@, of
bandwidth.
Any narrowband interferer, because it uses only a small "art of the total bandwidth used
by the system, will a""ear much wea$er to the .irect e&uence system (I thin$ it will be
much clearer if you loo$ at the "icture below). Moreover, the demodulator use the same
code as the transmitter to match the received signal, which decrease further signals not
modulated by the code (this is called the "rocessing gain of the code, == chi"s as used in
;<3.== gives in theory a =< d* "rocessing gain).
0irect !equence :
7riginal signal !pread signal 0ecoded signal
.irect e&uence is also the "rinci"le used by %0/" (#ode .ivision Multi"le Access -
one of the cellular "hone techni&ue), but in #.MA each individual "hone channel is
given a different code on the same fre&uency. *y having each channel having a
orthogonal code and the same received "ower (so, using "ower control), it is "ossible to
recover every #.MA channel using its code. %he only limit of the scheme is that the
noise is "ro"ortional of the number of channels (so the degradation with increased
ca"acity is graceful). %he configuration also needs to be a star to"ology (to use "ower
control), which doesn't suit well 0ireless 1A(.
%he s"reading with the code "roduces a faster modulation, therefore a . modem is &uite
com"licated (it usually re&uire faster circuits and a .! or e&uivalent logic for the
s"reading). 7ne the other hand, the fact of having one single fixed channel (as o""osed to
>re&uency @o""ing) eases the tas$ of the higher layers (MA#).
*ecause it uses a large channel, a .irect e&uence system has only a few channels
available in the bandwidth (E for the Wavelan - on different fre&uencies). %hose channels
are totally se"arate (they don't generate interferences on each other). .irect e&uence
also offers the "ossibility to use "artially overla""ing channels for systems in ad4acent
areas, increasing slightly the number of channels. *ut this last solution tends to increase
the noise and decrease the "erformance of the system, because all those systems usually
o"erate with the same code (and not one code "er fre&uency).
4.3.2 6re/$ency (o%%in"
6re/$ency (o%%in" uses a set of narrow channels and wal$ through all of them in
se&uence. >or exam"le, the 3.6 ?@, IM band is divided in CB channels of = M@,.
!eriodically (every 3< to 6<< ms usually), the system hop to a new channel, following a
"redetermined cyclic hopping pattern.
%he system avoids interferences by never staying on the same channel / if a channel is
bad, the system might not be able to use it and 4ust waits for the next good channel. As
the "attern ma$es the whole networ$ ho" through all the bandwidth available, the system
average the effect of bad channels over the time.
%his is where >re&uency @o""ing has a slight advantage over .irect e&uence / in the
very s"ecific case of strong narrow-band interferer "resent in the band, >re&uency
@o""ing loose some ho"s but will manage to get some ho"s on good fre&uencies. 7n the
other hand, if the noise is stronger than the received signal, there is not much that the
.irect e&uence node can do. *ut, for most interferers at common "ower levels, it's not
totally clear which will give the highest "erformance (it de"ends).
Frequency 1opping :
7n the other hand, >re&uency @o""ing introduces more com"lications at the MA# level /
scanning to find the networ$ at the initialisation (a moving target), $ee"ing the
synchronisation of the nodes, managing the ho"s.
%his com"lexity of the MA# has a "rice in term of "erformance, and the >re&uency
@o""ing mechanism has some overhead. %here is management overhead to manage the
synchronisation, and there is some dead time in the transmission when the system ho". In
theory, this can be $e"t to a minimum.
Also, the >re&uency @o""ing system have to include a "rocess called whitening, to
conform to radio transmission constraints, inserting some regular stuff bits in each
"ac$ets (to avoid long strings of < or =), adding more overhead (on the other a .irect
e&uence signal is withened by the .irect e&uence "rocess).
%he >re&uency @o""ing techni&ue can accommodate many more inde"endent systems
collocated in the same area than the .irect e&uence techni&ue by using different
ho""ing "attern (u" to =D for the 8ange9an.). 7n the other hand, the different ho""ing
"atterns of >re&uency @o""ing will 2collide2 on the same (or ad4acent) fre&uency from
time to time. %he collisions of the >re&uency @o""ing "atterns may reduce the
through"ut significantly / the systems 2colliding2 on the same (or an ad4acent) fre&uency
will have to share the bandwidth between them (see discussions on aggregate through"ut
in chapter +,2,4).
4.3.3 Com%arison...
In term of com"lexity, the .irect e&uence modem is more com"licated than the
>re&uency @o""ing one, and the .irect e&uence has a sim"ler MA# "rotocol. 0ith the
increasing integration of digital hardware, it doesn't cost much more to im"lement the
s"ecific MA# functionalities re&uired for the >re&uency @o""ing system, and as the
"rice of the modem is a big "ortion of a radio 1A( and doesn't follow the same cost
reduction trends, >re&uency @o""ing systems will tend to be chea"er.
In term of bandwidth sharing, the two technologies "erform really differently. %he same
is true in term of resistance to interferences (it de"end on the strength and "attern of the
interferer). .irect e&uence systems tend also to have a lower overhead on the air.
In summary, most vendors are going to >re&uency @o""ing because of the lower cost and
try to convince "eo"le that it is better, and vendors having heavily invested in .irect
e&uence try to "ush their raw "erformance advantage (es"ecially now with ;<3.== @),
see chapter 4,.), so it is still a $ind of religion war.
4.4 *iversity
*iversity is a generic conce"t of introducing redundancy in the system to overcome
noise and to increase the reliability of the system. >or exam"le, spread spectrum is a ty"e
of fre&uency diversity, using more bandwidth than necessary to avoid bad "arts of the
s"ectrum. 8etransmission is a very usual tem"oral diversity. FE% (>orward 5rror
#orrection) is another $ind of tem"oral diversity. Aery often, 2diversity2 is associated
with antenna diversity only. Antenna diversity is only one form of diversity (a s"acial
diversity).
Antenna diversity means that the radio device has two (or more) antennas. %he
transmission conditions on the channel vary a lot over the time. %he channel tends to fade
in and fade out (see chapter 2,6,:), so the device has moment of good rece"tion and
moment of bad rece"tion. *ut, these conditions are also de"endant on the s"acial
"osition. *y having two antennas, even &uite close (a few cm), the condition at each
antenna is very often totally different. 7ne antenna may give a "oor signal and the other a
good one, and a few ms later it might be the reverse. o, before receiving each "ac$et, the
receiver chooses the best antenna of the two by com"aring the signal strengths, and so
can avoid most of the fade out "eriods.
4.5 *irectional antennas
Most wireless 1A(s use omnidirectional antennas, but may offer directional antennas
in o"tion. Instead of receiving in every directions, the directional antenna favour
rece"tion in a more or less narrow angle. %he narrower the angle is, the higher the gain is
(and the range), because you get rid of more unwanted emissions and bac$ground noise
in the other directions.
0ith directional antennas, it is &uite common to have a few $ilometres of range in line of
sight with "roducts in the IM band. %he first "roblem is that you must of course "oint
each antenna towards the node you intend to communicate with (de"ending on the angle
this needs to be more or less "recise). %he second "roblem is that very directional
antennas tend to be &uite big.
%his is why directional antennas are only suited for fixed "oint to "oint lin$s ("roducts
li$e ireless !rid"es). >or most networ$s where nodes need to tal$ to different other
nodes in different directions and might need to move, omnidirectional antennas are much
more "ractical.
.ectored antennas are very similar to directional antennas, and heavily used in cellular
"hone base stations. A set of wide angle directional antenna are assembled on a vertical
"ole, each one covering one "ortion of the hori,on (a sector, for exam"le E antennas =3<
degrees wide). 0hen tal$ing to a s"ecific node, the base station 4ust select the sector of
the sectored antenna that cover this node, giving the benefit of directionality without
sacrificing the coverage.
!eo"le are also investigating #eam formin" antennas. %his is an ada"tive directional
antenna, using a set of unidirectional antennas and interferometry to enhance the signal.
*asically, by adding all the signal of the different antennas with s"ecific offset (to
com"ensate "ro"agation delay), it is "ossible to aim the system towards a s"ecific
direction and have the same benefit as directional antenna. As this system is ada"tive and
dynamic, it could be used for 0ireless 1A(s
4.& 7an"e iss$es
%he %ro%a"ation of radio transmissions is influenced by many factors. 0alls and floors
tend to decrease and reflect the signal, and bac$ground noise ma$es it more difficult to
demodulate. In a ty"ical environment, all the shadows due to obstacles and reflections on
the walls create a very un"redictable &uality of transmission for each s"ecific location.
%he channel &uality also vary &uite a lot over the time (fading, see chapter 2,6,:) because
the environment is not static.
*ecause of the way radio transmissions are affected by the environment in such a
com"lex way, it is &uite difficult to "redict the com"ortment of the system and to define a
ran"e. :ou will have some good, fair and bad area+"eriod, the closer the two devices are
the more li$ely they are to be in a good one.
Most vendors attem"t to define a range for their "roducts, which is the average maximum
distance in usual o"erating conditions between two nodes (diameter of a cell - radio
neighbourhood). ome even give different ranges for different ty"ical environments. >or
exam"le / o"en environment (no obstacles), semi-o"en (cubicles) and closed (real walls).
*ut there is no standard and common o"erating "rocedure to measure a range (exce"t in
free s"ace, but this is useless), so we can't really com"are the different "roducts from the
ranges as indicated in their data-sheets, and you must ta$e these values with a bit of
caution.
If you want to com"are "roducts in term of range "erformance, you must loo$ closely at
the transmitted power and sensitivity values. %hese are some measurable characteristics
of the hardware which indicate the "erformance of the "roduct in that res"ect. In fact, I
would also recommend to do some benchmar$ of different "roducts in your own
environment to get a better idea of what coverage you can ex"ect.
4.&.1 Transmitted %ower
%he transmitted %ower is the strength of the emissions measured in 0atts (or
milli0atts). 0e have already seen that the regulations limit this "ower (see chapter 2,:).
!roducts having a high transmit "ower will also be li$ely to drain the batteries faster. *ut,
having a high transmit "ower will hel" to emit signals stronger than the interferers in the
band (and other systems).
@aving a strong transmitted "ower has some drawbac$ for frequency reuse. %his means
that if you want to "ut many different networ$s in areas close to each other, they will tend
to "ollute each other. 0ith less transmitted "ower you can ma$e smaller cells. %his is
why some "roduct may allow to select different transmitted "owers.
4.&.2 .ensitivity
%he sensitivity is the measure of the wea$est signal that may be reliably heard on the
channel by the receiver (it is able to read the bits from the antenna with a low error
"robability). %his indicates the "erformance of the receiver, and the lower the value the
better the hardware (higher in absolute value). %he figure is given in d*m, the magic
formula to transform "ower in 0atts to d*m is / $ dBm ; -< = :<,log($ W). 'sual
values are around -;< d*m (the lowest, the better, for exam"le -B< d*m is better).
7ne "roblem is that all manufacturer and standards use the same reference to define
sensitivity. ;<3.== s"ecify the sensitivity as the "oint when the system suffer from E I of
"ac$ets losses (for "ac$ets of 6<< *ytes in a ?aussian channel). ome "roducts use D< I
"ac$et losses as the definition of sensitivity, which of course gives a better number. %he
use of a ?aussian channel also gives better numbers (the use of a )ayleigh >ading
channel with antenna diversity would give results a""roximately C d* worse).
4.&.3 Atten$ation
Jnowing those two values, you may calculate the maximum "ossible atten$ation of the
"ac$ets (this is the difference between the two values, in d*). %he larger the maximum
"ossible attenuation, the larger the range. >or a =<< m0 system with a -;< d*m
sensitivity, we have =<< d* maximum attenuation.
%he attenuation is the decrease of signal strength between the transmitter and the receiver.
In the air, the attenuation is sim"ly "ro"ortional to the s&uare of the distance. If you $now
exactly the com"osition of the signal "aths between the two nodes (distance in the air,
ty"e of obstacles, reflections...), you may calculate the attenuation. *ut usually it is &uite
tric$y to determine the attenuation as a function of the distance, es"ecially that the signal
may be the com"osite from different "ro"agation "aths (see chapter 2,6,2). Moreover, the
variation in the environment ma$e the attenuation change over the time (see
chapter 2,6,:).
*ecause of this non straightforward relationshi", $nowing the maximum "ossible
attenuation won't give you the maximum range, but 4ust a feeling. %he only safe thing is
that "roducts with a greater maximum "ossible attenuation are very li$ely to have a larger
range.
$ropagation and 8ange :
4.&.4 .i"nal to noise ratio +.N7,
In the case of multirate systems, I've been tal$ing of .i"nal to Noise ratio (()). %he
sensitivity is in fact closely lin$ed to the minimum () of the modem. %he () defines
the difference of "ower in the receiver between a valid signal and a noise. %o be able to
decode successfully the received signal, the receiver needs a minimum () (i.e. the
signal not too much "olluted by the noise). %his minimum () de"ends on the &uality of
the receiver hardware and the modulation chosen (see chapter 2,3,: on multi rate
systems).
o, the lin$ between sensitivity and minimum () is &uite obvious. If you add the
minimum () to the bac$ground noise in the receiver (hardware noise and bac$ground
noise on the channel), you will find the sensitivity. o, having a low sensitivity means
also a low minimum (), so the ability to receive reliably "ac$ets with "otentially
higher interference strength, which ex"lain why the sensitivity is such an im"ortant
"erformance characteristic.
4.) Mod$lations
%he main 4ob of the radio modem is to transform bits into modulations of the radio
waves, but there is many way to do that. Most systems use a carrier (a base fre&uency)
and modulate it. %he sim"lest way is to modulate the strength of the signal (Am"litude
Modulation), but as the attenuation of the channel is usually not constant (see
chapter 2,4,-), this lead to "oor "erformance. Most modern systems modulate either the
fre&uency of the signal or the "hase of the signal (fre&uency offset), which gives much
greater "erformance.
4.).1 M$lti8rate systems
If you want a better through"ut, the most sim"le way is to use more bandwidth. %he
"roblem is that the IM s"read s"ectrum regulations limits the amount of bandwidth
usable (= M@, channels for >re&uency @o""ing). Also, in most hardware the filters used
to recover the signal are fixed, so the channel width is fixed. %his limit the rate of
symbols that you can use (= Mbauds for >re&uency @o""ing).
o, how could some >re&uency @o""ing systems offer E Mb+s in = M@, channels ? %he
use of more com%le9 mod$lation schemes allows to overcome this limitation. >or
exam"le, the standard 3>J allows to "ut = bit "er symbol, whereas 6>J allows 3 bits
"er symbols, doubling the signalling rate.
7f course, there is a drawbac$ / a more com"lex modulation scheme is less robust and
will re&uire a higher received ignal to (oise )atio to wor$ (() - see chapter 2,4,2).
0hen going from 3>J to 6>J, each time the receiver reads a symbol, instead of
having to distinguish two fairly se"arated values, now it has to distinguish 6 closer to
each other (see chapter 2,3,.). More com"lex modulations stuff even more values in the
same s"ace, but then the slightest "erturbation of the signal (noises) will ma$e the
receiver reads the wrong value for the symbol.
o, we have the choice between a high s"eed modulation which re&uires strong received
signal and a slower modulation which wor$s even on wea$ signals. In other words, the
higher the signalling rate, the shorter the range.
*ecause users want both range and s"eed, some vendors have build some systems using
multi"le levels of modulations, changing automatically from the fast modulation to the
robust one de"ending on the channel conditions (when a "ac$et fail, the rate is
automatically reduced). %his introduces a bit of overhead and com"lexity, but the system
offer a much better "erformance characteristic (range or s"eed).
4.).2 26.: and 46.:
26.: (>re&uency hift Jeying) is the sim"lest form of fre&uency modulation. *asically,
the system use two different fre&uencies for the values < and = of each bit. >or exam"le,
if B is the base fre&uency (the carrier) and d the carrier deviation, each time the system
want to transmit a < it creates a waveform of fre&uency B>d (a symbol), and each time it
want to transmit a = it creates a waveform of fre&uency B=d. %he receiver 4ust need to
measure the deviation of the signal to the reference fre&uency B to $now which value of
the bit was transmitted.
Frequency /odulation (.F!?) :
Measuring this deviation is not easy, because each symbol is very short in time / the
transmitter change it for every bit to transmit at the s"eed given by the baudrate. %he
receiver needs of course to $now when the bits are transmitted, which re&uire timing
synchronisation on the received signal. %he carrier deviation has to be chosen carefully to
enable enough differentiation between the two symbols but to have the signal generated
fitting in the band allocated to it (usually around one hundred $@, for a = M@, channel at
3.6 ?@,).
As mentioned above, it is "ossible to "ut more than one bit "er symbol (see
chapter 2,3,:), li$e using 46.:. 6>J use 6 different symbols having 6 different carrier
deviation, B=:(.d# B>:(.d# B=-(.d and B>-(.d, each symbol is ma""ed to a combination
of two bits (<<, <=, =<, ==).
(ote that the difference in fre&uency between each symbol for 6>J is smaller than for
3>J, to allow the signal to fit in roughly the same channel width. *etween each symbol,
the difference is only d for 3>J, instead of .d for 6>J, which ex"lains why 6>J is
more sensitive and re&uires a better () (see chapter 2,4,2).
4.).3 ;12.11 (7 +11 M#<s,
0hen ;<3.== was eventually released, = and 3 Mb+s was no longer considered as decent
s"eed for 0ireless 1A( and "eo"le were already tal$ing of using the D ?@, band for
higher through"ut (@i"er1an and ;<3.== at D ?@,). @owever, the migration from 3.6
?@, to D ?@, re&uires to change all nodes and doesn't "rovide bac$ward com"atibility
(it's mot the same fre&uency band, so a new modem is necessary).
%herefore, "eo"le "roducing 3.6 ?@, "roducts tried to find way to extend the life of their
technology (mostly @arris and 1ucent). %hey cheated with the "read "ectrum rules,
and got away with it, enabling them to offer D and == Mb+s systems.
*asically, a . system generate signal which occu"y around 33 M@, of bandwidth. %hey
designed their == Mb+s system to generate signal similar to a standard . system. %hen,
they went to the ># and claimed that as their new system was generating the same ty"e of
signal as a . system, it's im"act on other systems in the band was the same, so it should
be authorised as well. After a bit of negociation, the >## did acce"t this extension of the
rule. (ote that some >@ vendors also tried to get D M@, >@ channels in the 3.6 ?@,
band but failed to obtain it.
9ucent came u" with the sim"lest solution, !!M (!ulse !osition Modulation), which is
included in their 2%urbo2 line of "roducts, offering D and =< Mb+s. !!M sim"ly shift the
code used in the . modem, each "osition can encode some more bits. !!M is sim"le,
chea", but low "erformance.
1arris tried M*7J (M-ary *i-7rthogonal Jeying), offering D.D Mb+s and == Mb+s,
which is a more com"lex modulation than !!M, so more ex"ensive and more robust. %he
signal "roduced by the transmitter is also less similar to a . signal.
%hey both went bac$ to the ;<3.== grou", but neither wanted to ado"t the system of the
other. o, they settled down on yet another modulation, CC: (#om"lementary #ode
Jeying), which eventually got ado"ted for the ;<3.== @) standard and a""roved by the
>##. ##J is the most com"lex of the E modulations, offering better "erformance, but
higher cost, and signals even less similar to the original . signals.
;12.11 (7 offer == and D.D Mb+s rate (using the ##J modulation) and is bac$ward
com"atible with original ;<3.== . systems. @owever, the higher bit rate re&uire a
higher (), which reduce the range significantly. (ote as well that because of bac$ward
com"atibility most of the underlying "rotocol is still designed for the = Mb+s standard
(headers and management frames are = Mb+s, contention window si,e is still based on =
Mb+s systems), which mean that at higher rate the overhead of the system is much higher.
4.).4 =6*M
!eo"le building high s"eed system li$e 1iper9an were com"laining that adding to their
"roducts an 5&ualiser necessary to combat delay s"read (see chapter 2,6,2) was a ma4or
cost. o, they invented a new techni&ue to get similar or better "erformance at lower cost,
called =6*M (7rthogonal >re&uency .ivision Multi"lex).
'sing e&ualisation is a "ost-"rocessing techni&ue, which tries to overcome delay s"read
by brute force. 7>.M is a "re-"rocessing techni&ue, where the signal transmitted on the
band is "re"ared in such a way that the im"act of delay s"read is reduced.
.elay s"read is damaging because the symbol time is very short, so 7>.M will only use
large symbol time. @owever, by increasing the symbol time we reduce the bit-rate. %o
overcome this constraint, 7>.M transmit the symbols no longer serially but in "arallel K
%his way, we have very high bit rate with large symbol time.
7>.M use a set of subcarrier fre&uencies, the fre&uencies being orthogonal. 5ach
subcarrier is modulated individually, the bit rate and signal strength of each subcarrier
can be ada"ted to get maximum "erformance of the system (we "ut more bits on the good
subcarriers and less on the bad ones). %hen, the system s"lits the bits to transmit between
the subcarriers, each subcarrier is modulated and then combined to "roduce the
transmitted signal (using a >ast >ourrier %ransform).
%he main drawbac$ of 7>.M is that it re&uire a greater fre&uency accuracy (we traded
timing accuracy to fre&uency accuracy). As the 7>.M signal contains many subcarrier
very close to each other in fre&uency, the system must be very accurate to match all of
them.
%he first use of 7>.M was in the @i"er1an II standard (see chapter 4,2), but since
;<3.== at D ?@, has ada"ted a very similar modulation (see chapter 4,.).
4.; -nterferences and noises
In the "revious section we have examined what does affect the range "erformance of a
system. 'nfortunately, other "henomenon on the radio waves affect the "erformance of a
system (even if they may not reduce the range), and all $ind of interferences and
bac$ground radio noises will im"act the system.
4.;.1 6adin"
6adin" defines all the tem"oral variations of the signal attenuation due to its "ro"agation
in a real environment li$e an office or a house. %he radio signal interact in various way
with the environment (see chapter 2,4 and chapter 2,6,2), so vary a lot with the
environment configuration. Moving a few centimetres can ma$e a big different in signal
&uality (see chapter 2,2).
Moreover, the environment is not static, humans are moving, things are moving, and the
nodes may be moving themselves. All these small movements may "roduce im"ortant
variations in time in the attenuation of the signal. >or exam"le the "ro"agation between
two nodes may alternate from "oor to good on a "ac$et basis.
!eo"le usually describe the "attern of attenuation with a 8ayleigh fading model (case
where there is no line of sight) or a 8icean model (line of sight L additional "aths). %he
main conse&uence is that transmission errors on the channel tend to be clustered and are
anything but following a ?aussian distribution.
>ading cause transmissions errors that need to be overcome by the system. 7f course,
recovering from these error will add overhead. %he greater the range the greater will be
the im"act of the fading and the system will degrade with higher range until it loose
communication.
%he most efficient techni&ue to overcome the effect of fading is antenna diversity (see
chapter 2,2).
4.;.2 Microwave oven and other interferers
As we have mentioned earlier, 0ireless 1A(s tend to be im"lemented in the unlicensed
bands, which adds more constraints. %he vast ma4ority of the 0ireless systems (cellular
"hone, telecoms, aviation, military...) are designed for dedicated radio bands, so benefit
from an absence of interferers in the band they are using. %his is not the case for 0ireless
1A(s, they have to co"e with the emissions of other systems.
%he de"loyment of unlicensed systems is totally uncoordinated. o, other radio systems
o"erating in the area do create interferences. %his includes other 0ireless 1A(s, cordless
"hones (B<< M@, and now 3.6 ?@,) and other communication systems.
%he 3.6 ?@, band is also the fre&uency where water molecules resonate, so is used for
microwave oven. .omestic microwave oven (the one used to heat food in the $itchen)
generates a limited amount of interferences, the various regulations limit the "ower of the
radiation they can lea$ to less than =0, they emit "eriodic short bursts and "ollute only a
limited "ortion of the 3.6 ?@, band. #ommercial microwave ovens (for exam"le a huge
dryer in a "a"er factory) generate much more interferences.
%he result of interferences is that "ac$ets collide with interference signal and can be
received corru"ted. If the () between the "ac$et and the interferer is high enough (see
chapter 2,4,2), the receiver can 2ca"ture2 the "ac$et, otherwise it is corru"ted.
Most 0ireless 1A(s co"e very well with interferers, in fact usually much better than
cordless "hones, but interferences do reduce "erformance.
4.;.3 6>C +6orward >rror Correction,
%he most obvious way to overcome transmission errors is to use 6>C. >5# goes further
than #)# which 4ust detects errors, >5# adds in every transmission some additional
redundancy bits. .e"ending on the number of bits added and the >5# code used (the
strength of the code), this allows to re"air a certain number of errors in the transmission.
>5# has been used with success in many systems, and the T$r#o Codes are "robably the
most efficient one / they are very close to the hannon limit in a ?aussian channel. In
other world, if the error follow ?aussian distribution (and the "arameters are $nown),
there is a turbo code nearly o"timal giving the highest through"ut in this channel.
'nfortunately for us, errors on a radio channel (for 0ireless 1A() follow a fading model
and are clustered. %his means that most of the time the signal is strong, so the "ac$et is
error free, but when the signal is wea$ the "ac$et contains lots of error. Interferences has
roughly the same effect as fading, either the "ac$et is collision free so intact, or when a
collision occur most of the "ac$et is corru"ted.
%o correct all those errors in corru"ted "ac$ets, it would re&uire a very strong >5# code.
'nfortunately, this code would add lots of redundancy bits, so lots of overhead. A normal
>5# code would add less overhead, but be useless with the correct "ac$ets and inefficient
with the highly corru"ted "ac$ets.
o, for 0ireless 1A(s, using >5# tends to be ineffective against fading and interferers,
and no 0ireless 1A( do im"lement >5#. A much better solution is to use
retransmissions (4ust retransmit the original "ac$et in case of errors - some form of
"ac$et scheduling and retransmission has been "roven to be nearly o"timal in )ayleigh
fading channels). %his is usually im"lemented at the MA# level (see chapter +,.,:).
@owever, in a few case >5# might be needed in 0ireless 1A(s. ome receivers, either
due to "oor im"lementation or s"ecific design (li$e having an 5&ualiser), generate
random (?aussian) errors, and might benefit from >5#.
4.;.4 M$lti%ath and delay s%read
)adio waves reflect or diffract on obstacles, and are attenuated differently by different
materials. %his is exactly li$e light, which goes through glass, is reflected by mirrors and
sto" by most obstacles, exce"t that much more materials are trans"arent or reflector to
radio than to light.
In a real environment li$e an office or a house, there is a lot of surface reflecting radio
(walls, ceilings, metal), being semi-trans"arent to radio (walls, ceilings, humans) or
o"a&ue to radio (metal). %his gives trouble estimating the range of the system (see
chapter 2,4). %his also mean that the signal received at a node may come from different
directions (de"ending on reflections on the environment) with different strength
(de"ending on attenuations), and the receiver sees only the combinations of all these
reflections. %his "henomenon is called m$lti%ath.
Most of the time, multi"ath is good, because the addition of all the reflections of the
signal increase its strength. %he main effect of multi"ath is that range is very difficult to
evaluate (see chapter 2,4,-) and the receiver ex"eriences fading (see chapter 2,6,:).
*ut, the main "roblem of multi"ath is that it creates delay s%read. .e"ending on the
number of reflections and the "ro"agation s"eed in different signals, all these signals
don't arrive exactly at the same time at the receiver. It's li$e the 2echo2 you may hear in
the mountains, the signal going directly will be faster than one reflecting twice on the
walls.
7f course, as radio "ro"agate at the s"eed of light, those difference are very small (below
the microsecond). *ut, when the bitrate of the system increases, those time differences
becomes significant with regards to the symbol time (see chapter 2,3,.), to the "oint of
creating destructive interferences (the current symbol will be corru"ted by the echo of the
"revious symbols).
*it rate lower than = Mb+s are relatively immune to delay s"read "roblems (the symbol
time is = Ms and higher), but as the bit rate increase above = Mb+s the effect of delay
s"read increases. It is considered that systems faster than D M+s should have some
techni&ue to overcome delay s"read.
/ultipath and 0elay !pread :
%he main techni&ue to overcome delay s"read is using an >/$aliser. An e&ualiser is a big
digital circuit that try to estimate the different com"onents of the signals. %he e&ualiser
need to be trained ("ac$ets includes a s"ecific well $nown training se&uence) to
determine what are the different "ath, their relative timings and strength. %hen, the
e&ualiser se"arate the different com"onents of the signal and recalculate the signal
removing the delay s"read.
%he main disadvantage of 5&ualiser is that they are ex"ensive. )ecently, some standards
are starting to use =6*M (see chapter 2,3,2), which is a clever modulation techni&ue
minimising the im"act of delay s"read.
5 The MAC level +link layer,
%his section of the document focus on the next layer u", the lin$ layer. %his mostly
com"rise the MAC (Medium Access #ontrol) "rotocol. .ifferent MA# "rotocols and
techni&ues are "resented.
5.1 Main channel access mechanisms
%he main 4ob of the MA# "rotocol is to regulate the usage of the medium, and this is
done through a channel access mechanism. A channel access mechanism is a way to
divide the main resource between nodes, the radio channel, by regulating the use of it. It
tells each node when it can transmit and when it is ex"ected to receive data. %he channel
access mechanism is the core of the /"% protocol. In this section, we describe '0/"#
%!/" and polling which are the E main classes of channel access mechanisms for radio.
5.1.1 T*MA
In this cha"ter, we discuss %.MA as a channel access mechanism and not its a""lications
and "rotocols based on it.
T*MA (%ime .ivision Multi"lex Access) is very sim"le. A s"ecific node, the #ase
station, has the res"onsibility to coordinate the nodes of the networ$. %he time on the
channel is divided into time slots, which are generally of fixed si,e. 5ach node of the
networ$ is allocated a certain number of slots where it can transmit. lots are usually
organised in a frame, which is re"eated on a regular basis.
%he base station s"ecify in the beacon (a management frame) the organisation of the
frame. 5ach node 4ust needs to follow blindly the instruction of the base station. Aery
often, the frame is organised as downlin$ (base station to node) and u"lin$ (node to base
station) slots, and all the communications goes through the base station. A service slot
allows a node to re&uest the allocation of a connection, by sending a connection re&uest
message in it (see chapter +,.,2). In some standards, u"lin$ and downlin$ frames are one
different fre&uencies, and the service slots might also be a se"arate channel.
'0/" channel access mechanism :
%.MA suits very well "hone a""lications, because those a""lication have very
"redictable needs (fixed and identical bit rate). 5ach handset is allocated a downlin$ and
a u"lin$ slot of a fixed si,e (the si,e of the voice data for the duration of the frame). %his
is no sur"rise why %.MA is used into all cellular "hone standards (?M in 5uro"e,
%.MA and !# in the 'A) and cordless "hone standards (.5#% in 5uro"e). %.MA is
also very good to achieve low latency and guarantee of bandwidth (where #MA+#A is
&uite bad).
%.MA is not well suited for data networ$ing a""lications, because it is very strict and
inflexible. I! is connectionless and generates bursty traffic which is very un"redictable by
nature, while %.MA is connection oriented (so it has to suffer the overhead of creating
connections for single I! "ac$ets). %.MA use fixed si,e "ac$ets and usually symmetrical
lin$, which doesn't suit I! that well (variable si,e "ac$ets).
%.MA is very much de"endant of the &uality of the fre&uency band. In a dedicated clean
band, as it is the case for cellular "hone standard, %.MA is fine. *ut, because of it's
inflexibility, and because it doesn't really ta$e care of what's ha""ening on the channel,
%.MA can't co"e and ada"t to the bursty interference sources found in the unlicensed
bands (unless a retry mechanism is "ut on to" of it).
5.1.2 C.MA<CA
C.MA<CA (#arrier ense Multi"le Access+#ollision Avoidance) is the channel access
mechanism used by most wireless 1A(s in the IM bands. A channel access mechanism
is the "art of the protocol which s"ecifies how the node uses the medium / when to listen,
when to transmit...
%he basic "rinci"les of #MA+#A are listen before talk and contention. %his is an
asynchronous message "assing mechanism (connectionless), delivering a best effort
service, but no bandwidth and latency guarantee (you are still following ?). It's main
advantages are that it is suited for networ$ "rotocols such as %#!+I!, ada"ts &uite well
with the variable condition of traffic and is &uite robust against interferences.
#MA+#A is fundamentally different from the channel access mechanism used by
cellular "hone systems (see '0/" in chapter +,:,:).
#MA+#A is derived from #MA+#. (#ollision .etection), which is the base of
Ethernet. %he main difference is the collision avoidance / on a wire, the transceiver has
the ability to listen while transmitting and so to detect collisions (with a wire all
transmissions have a""roximately the same strength). *ut, even if a radio node could
listen on the channel while transmitting, the strength of its own transmissions would
mas$ all other signals on the air. o, the "rotocol can't directly detect collisions li$e with
Ethernet and only tries to avoid them.
%!/"(%" channel "ccess /echanisms :
%he "rotocol starts by listening on the channel (this is called carrier sense), and if it is
found to be idle, it sends the first "ac$et in the transmit &ueue. If it is busy (either another
node transmission or interference), the node waits the end of the current transmission and
then starts the contention (wait a random amount of time). 0hen its contention timer
ex"ires, if the channel is still idle, the node sends the "ac$et. %he node having chosen the
shortest contention delay wins and transmits its "ac$et. %he other nodes 4ust wait for the
next contention (at the end of this "ac$et). *ecause the contention is a random number
and done for every "ac$ets, each node is given an e&ual chance to access the channel (on
average - it is statistic).
As we have mentioned, we can't detect collisions on the radio, and because the radio
needs time to switch from receive to transmit, this contention is usually slotted (a
transmission may start only at the beginning of a slot / D< Ms in ;<3.== >@ and 3< Ms in
;<3.== .). %his ma$es the average contention delay larger, but reduces significantly the
collisions (we can't totally avoid them).
5.1.3 'ollin" MAC
'ollin" is the third ma4or channel access mechanism, after '0/" and %!/"(%" (see
chapter +,:,: and chapter +,:,. res"ectively - %here exist also %o$en )ing, but I guess
that nobody would be cra,y enough to im"lement it on a radio lin$). %he most successful
networ$ing standard using "olling is =<<vg (I555 ;<3.=3), but some wireless standard
are also using it. >or exam"le, 6<.,:: offers a "olling channel access mechanism (!oint
#oordination >unction) in addition to the #MA+#A one.
!olling is in fact in between %.MA and #MA+#A. %he base station retains total control
over the channel, but the frame content is no more fixed, allowing variable si,e "ac$ets to
be sent. %he base station sends a s"ecific "ac$et (a "oll "ac$et) to trigger the transmission
by the node. %he node 4ust wait to receive a "oll "ac$et, and u"on rece"tion sends what it
has to transmit.
!olling can be im"lemented as a connection oriented service (very much li$e %.MA, but
with higher flexibility in "ac$et si,e) or connection less-service (asynchronous "ac$et
based). %he base station can either "oll "ermanently all the nodes of the networ$ 4ust to
chec$ if they have something to send (that is wor$able only with a very limited number
of nodes), or the "rotocol use reservation slots (see chapter +,.,2) where each node can
re&uest a connection or to transmit a "ac$et (de"ending is the MA# "rotocol is
connection oriented or not).
$olling channel "ccess /echanisms :
In the case of =<<vg, the "olling mechanism doesn't use any bandwidth (it's done out of
band through tones), leading to a very efficient use of the channel (over BG I user
through"ut). >or ;<3.== and wireless 1A(, all the "olling "ac$ets have to be transmitted
over the air, generating much more overhead. More recent system use reservation slots,
which is more flexible but still re&uire significant overhead.
As #MA+#A offers ad-hoc networ$ing (no need of a base station) and similar
"erformance, it is usually "referred in most wireless 1A(s. >or exam"le, most ;<3.==
vendors "refer to use the distributed mode (#MA+#A) over the coordinated mode
("olling).
5.1.4 7eservation %rotocols and ATM
%he most interesting feature of "rotocols based on %.MA or !olling mechanism is that
the *ase tation has absolute control of the traffic and can guarantee bandwidth and
latency for a""lications that re&uire it. ce"tics might wonder what can be guaranteed
anyway in an environment o"en to interferers and without de"loyment control (see
chapter 2,:), but that's another to"ic of discussions.
%he guarantee of bandwidth is essential for "eo"le de"loying 0ireless .istributions
ystems (also called 1ast Mile .elivery ystems), li$e re"lacing the cable between your
house and your I! with wireless. %hose "eo"le want to be able to restrict and segregate
users and guarantee fairness. tandards such as @i"er1an II (*roadband )adio Access
(etwor$ "ro4ect - see chapter 4,2) is aiming at those usages.
%he basic idea is to "ut A%M (Asynchronous %ransfer Mode) over radio, as A%M
im"lement all the Nuality 7f ervice features that they are dreaming off. %he networ$ is
centrally managed (so uses %.MA or !olling mechanism with reservation slots), the base
station im"lement a call admission control (acce"t or re4ect new A%M circuits) and
scheduler ("rioritise and send A%M cells) to guarantee the &uality of service re&uested.
7n to" of the MA#, all the usual A%M layers are needed (virtual circuits,
segmentation+reassembly, I! ada"tation...), as well as some s"ecific mobile features (to
manage roaming).
'nfortunately, radio transmission has a lot of overhead (li$e large synchronisation field
and headers) which is somewhat incom"atible with the small A%M cells. %he main
benefit of A%M small cells is to allow very efficient switching, but this is not needed over
radio. At the end of the day, 0A%M doesn't resemble at all to A%M O A%M uses individual
channel for each node and is asynchronous, whereas 0A%M uses a shared medium and is
totally synchronous.
5.2 MAC techni/$es
0e have described the main "rinci"le of #MA+#A (see chapter +,:,.), but most MA#
"rotocols use additional techni&ues to im"rove the "erformance of #MA+#A.
5.2.1 MAC retransmissions
As we have seen in the "revious cha"ter, the main "roblem of the %!/"(%" protocol is
that the transmitter can't detect collisions on the medium. %here is also a higher error rate
on the air than on a wire (see chapter 2,6), so a higher chance of "ac$ets being corru"ted.
%#! doesn't li$e very much "ac$et losses at the /"% layer (see %#! and "ac$et losses
"roblem - chapter +,2,+). *ecause of that, most MA# "rotocols also im"lement %ositive
acknowled"ement and MAC level retransmissions to avoid losing "ac$ets on the air.
%he "rinci"le is &uite sim"le / each time a node receives a "ac$et, it sends bac$
immediately a short message (an ac$) to the transmitter to indicate that it has successfully
received the "ac$et without errors. If after sending a "ac$et the transmitter doesn't receive
an ac$, it $nows that the "ac$et was lost, so it will retransmit the "ac$et (after contending
again for the medium, li$e in 5thernet).
Most MA# "rotocols use a sto" and go mechanism, they transmit the next "ac$et of the
&ueue only if the current "ac$et has been "ro"erly ac$nowledged (no sliding window
mechanism li$e in %#!). %he rationale is that it ma$es the "rotocol sim"ler, minimise
latency and avoid desen&uencing "ac$ets (something that %#! doesn't li$e as well).
/"% retransmissions in %!/"(%" :
%he ac$s are 2embedded2 in the MA# "rotocol, so they are guaranteed not to collide (the
contention starts after the ac$ - see figure). %hese ac$s are very different from the %#!
ac$s, which wor$ at a different level (and on a different time frame). 7f course, broadcast
and multicast "ac$ets are not ac$nowledged, so they are more li$ely to fail...
If all modern 0ireless 1A( "rotocols im"lement this essential feature, some old "roducts
may lac$ it. 0ireless 0A( "rotocols (li$e satellite lin$s) don't im"lement that either,
because the round tri" delay in their case is so long that by the time they would receive
the ac$ they could have sent another "ac$et. If your 0ireless 1A( doesn't im"lement
MA# level retransmissions, all is not lost / students of *er$eley have created a "rotocol
called snoop (see at ft"/++daedalus.cs.ber$eley.edu+"ub+snoo"+) which filters the %#! ac$s
and retransmits the lost "ac$ets before %#! even notices that they are lost (this is still a
lin$ level retransmission, but done 4ust over the MA#).
5.2.2 6ra"mentation
%he radio medium has a higher error rate than a wire. 0e have ex"lained in the "revious
cha"ter that it was why most "roducts were including MA# level retransmissions to
avoid losing "ac$ets.
MA# level retransmissions solve this "roblem, but is not really "erformant. If the "ac$et
to transmit is long and contains only one error, the node needs to retransmit it entirely. If
the error rate is significantly high, we could come to some situation were the "robability
of error in large "ac$et is dangerously close to = (we can't fit a "ac$et between the bursts
of errors due to fading or interferers), so we can't get "ac$et through.
%his is why some "roducts use fra"mentation. >ragmentation is sending the big "ac$ets
in small "ieces over the medium. 7f course, this adds some overhead, because it
du"licates "ac$et headers in every fragments. 5ach fragment is individually chec$ed and
retransmitted if necessary. %he first advantage is that in case of error, the node needs only
to retransmit one small fragment, so it is faster. %he second advantage is that if the
medium is very noisy, a small "ac$et has a higher "robability to get through without
errors, so the node increases its chance of success in bad conditions.
5.2.3 7T.<CT.
In the cha"ter about range (chapter 2,4), we have seen that the main effect of
transmission on radio waves is the attenuation of the signal. *ecause of this attenuation,
we have very commonly a "roblem of hidden nodes.
%he hidden node "roblem comes from the fact that all nodes may not hear each other
because the attenuation is too strong between them. *ecause transmissions are based on
the carrier sense mechanism, those nodes ignore each other and may transmit at the same
time. 'sually, this is a good thing because it allows frequency reuse (they are effectively
in different cells).
*ut, for a node "laced in between, these simultaneous transmissions have a com"arable
strength and so collide (in its receiver). %his node could be im"ossible to reach because
of these collisions.
%he fundamental "roblem with carrier sense only is that the transmitter tries to estimate if
the channel is free at the receiver with only local information. %he situation might be
&uite different between those two locations.
An sim"le and elegant solution to this "roblem ("ro"osed by !hil Jarn in his MA#A
"rotocol for A9.3D) is to use 7T.<CT. ()e&uest %o end+#lear %o end). )%+#% is a
handshaking / before sending a "ac$et, the transmitter sends a )% and wait for a #%
from the receiver (see figure below). %he rece"tion of a #% indicates that the receiver is
able to receive the )%, so the "ac$et (the channel is clear in its area).
At the same time, every node in the range of the receiver hears the #% (even if it doesn't
hear the )%), so understands that a transmission is going on. %he nodes hearing the #%
are the nodes that could "otentially create collisions in the receiver (assuming a
symmetric channel). *ecause these nodes may not hear the data transmission, the )%
and #% messages contain the si,e of the ex"ected transmission (to $now how long the
transmission will last). %his is the collision avoidance feature of the )%+#%
mechanism (also called virtual carrier sense) / all nodes avoid accessing the channel after
hearing the #% even if their carrier sense indicate that the medium is free.
8'!(%'! and hidden nodes in %!/"(%" :
)%+#% has another advantage / it lowers the overhead of a collision on the medium
(collisions are much shorter in time). If two nodes attem"t to transmit in the same slot of
the contention window, their )% collide and they don't receive any #%, so they loose
only a )%, whereas in the normal scenario they would have lost a whole "ac$et.
*ecause the )%+#% handsha$ing adds a significant overhead, usually it is not used for
small "ac$ets or lightly loaded networ$s.
5.2.4 7eservation and service slots
7ne of the main "roblem of %.MA and !olling "rotocol is for the base station to $now
when the nodes want to transmit. In #MA+#A, each node sim"ly waits to win a
contention, so this "roblem doesn't exist. @owever, %.MA and !olling usually re&uire a
service slot or reservation slot mechanism.
%he idea is to offer a "eriod of time where nodes can contend (com"ete) and send to the
base station some information about their traffic re&uirements (a reservation re&uest
"ac$et), this "eriod of time coming at regular interval (the remaining of the time, nodes
4ust obey the base station normally). %he base station feeds the reservation re&uests to its
scheduling algorithm and decides the main frame structure (when each node will
transmit). %his "eriod of time for sending reservation re&uests is either called service slot
(if it is use for more "ur"ose li$e cell location and roaming) or reservation slot (if it is use
only to re&uest a transmission or connection).
If the MA# is connection oriented, the rate of new connection is low, so usually a single
service slot is enough (see figure in chapter +,:,:). If the MA# is "ac$et oriented, the rate
of re&uests is higher, so usually the "rotocol offer many reservation slots together (see
chapter +,:,-). (odes use a sim"le "loha protocol in the slots / they transmit, and if it fail
(collision with other re&uests or medium errors) they bac$off a random number of slots
before retrying.
!rotocols which use many different channels, such as cellular "hone, can even have a
dedicated service channel se"arate from other transmissions, instead of multi"lexing
service re&uests with the data traffic.
5.3 Network to%olo"y
%he to"ology of 0ireless 1A( is very different from traditional 1A(s. %he connectivity
is limited by the range, so we usually don't have com"lete coverage (some node may not
see each other). %his brea$s some assum"tions of higher layers. %o overcome this, either
the networ$ is divided in cells managed by an "ccess $oint, or the networ$ use /"%
level forwarding.
5.3.1 Ad8hoc network
Ad-hoc networ$ is the sim"lest form of 0ireless 1A( is a networ$ com"osed of a few
nodes without any bridging or forwarding ca"ability. All nodes are e&ual and may 4oin or
leave at any time, and have e&ual right to the medium. In fact, it's very much li$e an
5thernet, where you may add or remove node at discretion. %his is the $ind of radio
networ$s de"loyed in homes of small offices.
7f course, for this to wor$ all nodes must be able to see all the other nodes of the
networ$, to be able to establish communication with them. 0hen a nodes goes out of
range, he 4ust loose connection with the rest of the ad-hoc networ$. 5ffectively, this is a
single cell networ$.
7ne of the node of the ad-hoc networ$ may "rovide routing or "roxying to communicate
to the rest of the wor$, but nodes are still confined to the area within that cell.
5.3.2 Access 'oints and 7oamin"
0ireless networ$s are sometime isolated networ$s (called ad-hoc), but most of the time
they need to be connected to the rest of the world (and the Internet /-). %his is usually
done through Access 'oints.
In fact, an Access !oint is sim"ly a #rid"e, connected on one side to the radio networ$
and on the other side to Ethernet (usually), forwarding "ac$ets between the two
networ$s. A bridge wor$s at the MA# level, 4ust loo$ing through the MA# headers to
ma$e its decisions (filtering) and changing MA# headers according to the MA# "rotocol
used. %his means that )etBeui and $* wor$ across the access "oint, and that the nodes
connected to the radio must use the same '%$($ subnet as the 5thernet segment the
access "oint is connected to.
*ecause of the interactions with MA# level ac$nowledgement, most of the time bridging
on 0ireless 1A( is not as sim"le and trans"arent as on 5thernet, and a s"ecific scheme is
designed in the MA# "rotocol. 0hen a node sends a "ac$et, the source address must be
his to "ro"erly receive the MA# level ac$ coming bac$ (and vice versa). In theory, if the
MA# and the driver are carefully im"lemented it could be "ossible to su""ort
trans"arently 5thernet bridges (li$e in a 1inux box), but most manufacturers don't bother
(es"ecially that they want you to buy an Access !oint).
'sing Access !oints allows to divide the networ$ in cells. 5ach Access !oint is at the
centre of a cell and is given a different channel (fre&uency, ho""ing "attern... - the goal is
for each cell to interferer the least with the others). *y careful de"loyment of those
Access !oint, it is "ossible to give networ$ access in all "arts of large areas.
In fact, most radio access "oints "rovide more than this sim"le bridging functionality.
Most of them "rovide access control (to "revent any unwanted radio node to access the
networ$), roaming and out of range forwarding.
%he use of the last two features re&uires that all the access "oints that are used to cover
the desired area are connected on the same wired segment (I! subnet). 5ach node needs
to register to one of the access "oint (to avoid confusion between the A!s), the nearest
one, usually (in fact, more li$ely the one having the strongest signal, which might not be
the nearest). If the node moves, it will automatically switch from one access "oint to
another to retain its access to the wired networ$ (that is roamin"). If a node wants to
communicate with a node which is not in its reach, its access "oint forwards the "ac$ets
through the wired networ$ and via the access "oint where the destination is registered
(that is o$t of ran"e forwardin").
A few systems use as well the access "oint as a networ$ central coordinator of the
channel access mechanism (%.MA and "olling mode). %his is a bad idea, because it
decreases the overall reliability and flexibility of the system / every node must be able to
communicate at any time the access "oint in order to wor$, even if it wants to
communicate with a close neighbour.
"ccess $oints# roaming and radio /"% forwarding :
8oaming @ "ccess $oints 8adio /"% forwarding
5.3.3 7adio MAC forwardin"
%he forwarding mechanism designed around "ccess $oints (see chapter +,-,.) re&uires a
fixed wired infrastructure to lin$ the Access !oint. %his might be satisfactory for most
usages, but is not ade&uate for ad-hoc networ$s.
ome MA# "rotocol (such as @i"er1an - see chapter 4,-) "rovide a MAC level
forwardin", where every node of the networ$ can be used to relay the message on the air
to the destination. %he "rotocol doesn't rely any more on a fixed infrastructure, but on all
the wireless nodes on the "ath.
o, how do we found the o"timal "ath through the nodes to the correct destination ? %his
forwarding mechanism use management message to "ro"agate networ$ changes and
to"ology information, and from those messages nodes can com"ute the o"timal
forwarding tables. (odes must im"lement the forwarding ca"ability and "ro"agate
message based on those routing tables. In fact, each node of the networ$ acts as a ad-hoc
wireless bridge.
*roadcast and multicast messages are a bit of a "roblem (they have always been on
bridging technologies) / all nodes 4ust re"eat them and the strategy is to flood the networ$
with them (that's the only way to ma$e sure they reach all "ossible destinations).
ome access points also offer the "ossibility to be configured as ireless 7e%eaters,
which "rovide the same $ind of radio forwarding but in a managed way.
)adio MA# forwarding is elegant and interesting, but all the forwarding consume some
more radio bandwidth, which is already limited to start with.
5.4 .ome thro$"h%$t considerations
If the "hysical layer "eo"le are mostly tal$ing range and d*, MA# layer "eo"le are (or
should be) concerned about the through"ut of the system.
5.4.1 !it8rate vers$s ma9im$m $ser thro$"h%$t
1i$e for wired "roducts, most radio 1A( vendors indicate only the #it8rate of their
"roducts (also called signalling rate). >or exam"le, Ethernet is =< Mb+s, =<< Mb+s or =
?b+s, and most radio 1A( "roducts between <.D and E Mb+s (higher rate li$e =< Mb+s are
slowly coming to the mar$et). %he signalling rate is the s"eed at which bits are
transmitted over the medium, but, because of the many overheads of the "rotocols used to
communicate, the user through"ut is usually less (note also that they use decimal
multi"licators, so for them = Mb+s is =<
G
b+s K). %he 0ireless 1A(s "rotocols have
usually a hi"her overhead than their wired counter"art (such as 5thernet). %his is due to
different factors /
%he first is the radio technology / radio receivers re&uire large synchronisation fields
(receiver training, antenna selection...) O the radio itself is slow to react (switch from
receive to transmit), so needs large slots in the contention window and between "ac$ets.
%he second is the addition of the features necessary for the radio "rotocol which ma$es
the "ac$et MA# headers larger (fields for networ$ id, encry"tion "arameters...) or
introduces new management "ac$ets (synchronisation, authentication, access "oint
registration).
%he third is that some trade>offs are made to im"rove the reliability. >or exam"le, we
might s"lit big "ac$ets into small inde"endent fragments to decrease the error "robability
(see chapter +,.,. on fragmentation). Ac$s and )%+#% add also some overhead.
@aving a slotted contention decreases the collisions but ma$es the average contention
delay larger as well.
0hen you add all this, it starts to ma$e a significant difference. If in the case of 5thernet
you may ho"e to reach ;<-BD I of the signalling rate, for most radio "roducts, des"ite
being slower, the user through"ut is usually between D< and C< I of the signalling rate
(or even less...).
5.4.2 M$ltirate system considerations
Most vendors offer multirate systems (see chapter 2,3,:), the lower rate allowing a
greater coverage and the higher rate allowing greater through"ut at lower range, and offer
a mechanism for each node to ada"t the bit-rate de"ending on channel conditions.
*asically, when "ac$ets start to fail, the node reduce the rate.
7f course, "eo"le are li$ely to benchmar$ nodes in relatively close "roximity (two nodes
on the table), when the system will use the highest rate, but the real advantage of 0ireless
1A(s is usually given at higher range (in the garden, moving around), and in this case
the system is li$ely to select the lower rate (and maybe suffer from "ac$et losses and
retransmission due to range), so the "erformance will be less.
@owever, those rate ada"tation schemes are not always the most clever. 0hen there is an
interferer in the band, reducing the rate may increase marginally chances of "ac$ets to get
through, but most of the time having longer transmission time 4ust increase the
"robability of collision. In cases where there is lot's of contention (lot's of nodes with lot's
of traffic), some "roducts do reduce the rate which doesn't hel" to reduce to congestion
(I've seen that "ersonally). In those "articular cases, you may want to fix the rate yourself
to the highest and disable the rate reduction feature.
@aving a multi-rate system also im"act the overhead of the system, es"ecially at high
rates. All the basic "art of the "rotocol (headers, management messages, contention) is
designed for the slowest rate, so when going to higher rate their relative si,e increase
(their duration remain the same while the "ayload duration decreases).
>or exam"le, when sending the same =D<< * "ac$et at 6>J instead of 3>J with
;<3.==, the overhead of the contention window double, the overhead of the MA# level
ac$nowledgement and )%+#% double and the overhead of the header increases by
3; I. I've heard that the overhead for ;<3.== @) at == Mb+s was significantly noticeable
com"ared to = and 3 Mb+s s"eeds, and 1ucent claims that increasing the bit rate from 3 to
=< Mb+s (1ucent turbo !!M . modulation), the effective through"ut (user level) is
increased only by a factor E.
5.4.3 .hared thro$"h%$t vers$s individ$al thro$"h%$t
In the "revious cha"ter, we have examined the overhead added by the "rotocol and tal$ed
about the maximum user through"ut usable by the 0ireless 1A(. *ut, sometimes, even
in a clear channel, the maximum node to node thro$"h%$t may be even less than that.
%his is usually caused by im"lementation "roblems.
%he most obvious is for exam"le a slow interface between the !# and the 0ireless
device. A serial or "arallel interface is slower than an IA or !cmcia bus and may be a
bottlenec$.
%he second exam"le is devices im"lementing only one transmit buffer. %his saves some
cost (memory, com"lexity), but, as the buffer may be either written by the driver or
transmitted over the air but not both at the same time, this creates dead time over the air
while the driver refills the buffer and reduces the available through"ut. %his was one of
the "erformance gain between the first and the second generation of 5thernet cards in the
old days.
%he "rotocol might also "erforms better when many node are active than when only one
of them transmits. >or exam"le, the contention window in #MA+#A (number of
contention slots) im"act the "erformance O a larger contention window will decreases the
collisions but when there is a few nodes, those will wait on average longer to access the
channel (the common ;<3.== "arameters gives better "erformance for 3 active nodes than
for =). A "olling "rotocol which uses a round robin scheduling mechanism (as$s each
node in turn if it has a "ac$et to transmit) "erforms better is every node has something to
send than only one node (in this case, between each "ac$et of this node the "rotocol has
to "ool all the other nodes of the networ$ for nothing).
1astly, in the case of MA#s being connection oriented (%.MA and some im"lementation
of "ooling), an individual node may not be able to use the full lin$ ca"acity, limiting its
"erformance. >or exam"le, if a %.MA system has =< slots "er frame, some "hysical layer
or MA# layer constraints may "revent a node to use more than one slot in each frame,
even if the B other slots of the frame are free. If the node im"lementation can only
manage one slot, the node individual through"ut is only =+=<th of the shared through"ut.
>or the individual through"ut to be the maximum through"ut, the node must be able to
manage multi"le slots and multi"lex data between these slots.
5.4.4 Contention and con"estion
In the "revious cha"ter we examine why the shared through"ut could be higher than the
individual through"ut. *ut, the reverse can also be true (and is actually more li$ely for
#MA+#A systems).
0hen there is many nodes sending "ac$ets on the networ$, the "robability of having two
nodes choosing the same slot in the contention window increases. 0hen two nodes
choose the same slot (and they are first), their "ac$ets collide and are lost. %his mean that
when the level of contention increases, the number of retry increases as well, so the
"erformance of the networ$ dro" u" to the "oint of congestion.
In fact, ;<3.== has a relatively short contention window (=G slots but with a memory
effect), and is very sensitive to contention. 'nfortunately, it's very easy for any $ind of
device to generate enough traffic to saturate the wireless lin$, es"ecially those which
assume being on an 5thernet. I have "ersonally seen a nodes com"osed of E nodes and =
access "oint (;<3.==) where the number of retransmissions was higher than the number of
"ac$ets sent (each "ac$et transmitted on average more than twice).
A solution to this "roblem is to use )%+#% (see chapter +,.,-), because )%+#%
ma$es each collision much shorter. In fact, with )%+#% enabled, ;<3.== can su""ort
more than a do,en active nodes without significant reduction in "erformance due to
contention (a"art that those nodes have to share the bandwidth). As the )%+#%
handsha$e is usually done at the basic rates, its benefit tends to decrease for the highest
transmission rates.
5.4.5 TC' and %acket losses %ro#lem
%#! has been develo"ed for wired 1A(s, where "ac$et losses are minimal. If a "ac$et is
lost, %#! assumes that it is dro""ed in a router or a bridge because of con"estion. %o try
to reduce the congestion, %#! slows down drastically.
7n the radio medium, collisions can't be detected and the error rate is higher, so there is
more "ac$et losses (if we don't do anything about it). %#! sees that as congestion and
reduces its through"ut, and so doesn't use all the available bandwidth.
In modern 0ireless 1A(, MAC level retransmissions (see chapter +,:,-) solve totally
this "roblem by detecting and eliminating "ac$et losses due to errors and collisions (and
also avoid dese&uencing "ac$ets), so %#! sees a reliable channel and has no reason to
slow down (exce"t if MA# level retransmissions are "oorly im"lemented).
5.4.& A""re"ate thro$"h%$t
It's &uite common "ractice for vendors to advertise for their "roducts something called
a""re"ate thro$"h%$t. %his figure indicates the maximum through"ut that it is "ossible
to transmit in the full bandwidth by having different ad4acent and inde"endent networ$s
on different fre&uencies or ho""ing "atterns.
7f course, the user of the 0ireless 1A( will never see such a through"ut, and it is a bit
li$e advocating that by having =< Ethernet :<base' cables you are able to have a
=<< Mb+s through"ut... *ut, it gives an indication of how well overla""ing cells will
share the bandwidth.
>or exam"le, with a >re&uency @o""ing system having =.G Mb+s user through"ut, by
"utting =D networ$s, each on a different ho""ing "attern, we should have in theory a
36 Mb+s aggregate through"ut. In fact, because the different >re&uency @o""ing "atterns
2collide2 on the same fre&uency (and also suffer from co-channel interference) from time
to time, the actual aggregate through"ut is less, and is in this exam"le only =D Mb+s.
These collisions of the ho%%in" %atterns is why
6re/$ency (o%%in" can?t offer $% to )0 networks on
the )0 channels +#$t only $% to 15 in this case,... &
.ome ireless LAN standards
A short gallery of the most famous 0ireless 1A( standard (but unfortunately not
necessarily the most wides"read...).
&.1 ->>> ;12.11
%he main "roblem of radio networ$s acce"tance in the mar$et "lace is that there is not
one $ni/$e standard li$e 5thernet with a guaranteed com"atibility between all devices,
but many "ro"rietary standards "ushed by each inde"endent vendor and incom"atible
between themselves. *ecause cor"orate customers re&uire an established uni&ue
standard, most of the vendors have 4oined the I555 in a effort to create a standard for
radio 1A(s. %his is ->>> ;12.11 (li$e Ethernet is EEE 6<.,-, 'oken 8ing is EEE
6<.,+ and :<<vg is EEE 6<.,:.).
7f course, once in the ;<3.== committee, each vendor has "ushed its own technologies
and s"ecificities in the standard to try to ma$e the standard closer to its "roduct. %he
result is a standard which too$ far too much time to com"lete, which is overcom"licated
and bloated with features, and might be obsoleted before "roducts come to mar$et by
newer technologies. *ut it is a standard based on ex"erience, versatile and well designed
and including all of the o"timisations and clever techni&ues develo"ed by the different
vendors.
%he ;<3.== standard s"ecifies one MA# "rotocol and E "hysical layers / >re&uency
@o""ing = Mb+s (only), .irect e&uence = and 3 Mb+s and diffuse infrared (can we really
call it a 2standard2 when in includes E incom"atible "hysical layers ?). ince then, it has
been extended to su""ort 3 Mb+s for >re&uency @o""ing and D.D and == Mb+s for .irect
e&uence (;<3.==b). %he MA# has two main standards of o"eration, a distributed mode
(#MA+#A), and a coordinated mode ("olling mode - not much used in "ractice). ;<3.==
of course uses MA# level retransmissions, and also )%+#% and fragmentation.
%he o"tional "ower management features are &uite com"lex. %he ;<3.== MA# "rotocol
also includes o"tional authentication and encry"tion (using the 05!, 0ired 5&uivalent
!rivacy, which is )#6 6< bits - some vendors do offer =3; bits )#6 as well). 7n the
other hand, ;<3.== lac$s to defines some area (multirate, roaming, inter A!
communication...), that might be covered by future develo"ments of the standard or
com"lementary standards. ome ;<3.== "roducts also im"lement "ro"rietary extensions
(bit-rate ada"tation, additional modulation schemes, stronger encry"tion...), those
extensions may or may not be added to the standard over time.
0hen ;<3.== was finalised (se"tember BC), most vendors were slow to im"lement ;<3.==
"roducts because of the com"lexity of the standard and the number of mandatory features
(and in some cases they also need to "rovide bac$ward com"atibility with their own
"revious line of "roducts). ome of the o"tional features (encry"tion and "ower saving)
did only a""ear months after the initial release of the "roduct. *ut things seem to be
sorted out and we now have fully featured "roducts on the mar$et. %he com"lexity of the
s"ecification, the tightness of the re&uirements and the level of investment re&uired made
;<3.== "roducts ex"ensive com"ared to the "revious generation of wireless 1A(s, but
because of the higher standardisation and higher volumes, "rices are now dro""ing.
5ven if vendors eventually have launched ;<3.== "roducts, the standard doesn't fully
guarantee inter-o"erability / the "roducts have to use at least the same "hysical layer, the
same bit rate and the same mode of o"eration (and there is so many other little im"ortant
details...). %he most coo"erative vendors have been busy lately sorting out
intero"erability issues with inde"endent testing labs, but it is still a touchy sub4ect...
&.2 ;12.118# and ;12.118a +;12.11 at 5 4(2,
After C years of arguing in sub-committees ma$ing ;<3.==, you would thin$ that most
"eo"le would had enough of it. In fact no, the ;<3.== committee is now busy "ushing a
new standard at D ?@,, and also higher s"eed at 3.6 ?@, (by twea$ing the .irect
e&uence "hysical layer). *oth standard ma$es changes only to the "hysical layer, so that
the ;<3.== MA# can be reused totally unmodified, saving costs.
;12.118a (;<3.== at D ?@,) was standardised first (s"ring BB), based on 7F0/ (see
chapter 2,3,2), and using the '(II band (see chapter 2,. - so it won't be available in
5uro"e and Fa"an). %he 7>.M "hysical layer is a very close co"y of the one used in
1iper9an (so they might be some sort of com"atibility - see chapter 4,2), using D3
subcarriers in a 3< M@, channel, offering G, =3 and 36 Mb+s and o"tional B, =;, EG, 6;
and D6 Mb+s bit-rates. (o "roducts are yet on the mar$et.
Aery soon after, ;<3.== did standardise ;12.118# (;<3.== @)), based on a modified .
"hysical layer (see chapter 2,3,-). %he goal was to extend the life of the 3.6 ?@, band by
overcoming the ma4or drawbac$ / low s"eed. 7n to" of the original ;<3.==-. standard,
;<3.==-b offer additional D.D Mb+s and == Mb+s bit rates. It was a""roved by the >## and
they are now "roducts on the mar$et (which are &uite "o"ular).
&.3 (i%erLan
(i%erLan is the total o""osite of 6<.,::. %his standard has been designed by a
committee of researcher within the >T.-, without strong vendors influence, and is &uite
different from existing "roducts. %he standard is &uite sim"le, uses some advanced
features, and has already been ratified a while ago (summer BG - we are now only waiting
for the "roducts).
%he first main advantage of @i"erlan is that it wor$s in a dedicated bandwidth (D.= to
D.E ?@,, allocated only in 5uro"e), and so doesn't have to include s"read s"ectrum. %he
signalling rate is 3E.D Mb+s, and D fixed channels are defined. %he "rotocol uses a variant
of #MA+#A based on "ac$et time to live and "riority, and MA# level retransmissions.
%he "rotocol includes o"tional encry"tion (no algorythm mandated) and "ower saving.
%he nicest feature of @i"erlan (a"art from the high s"eed) is the ad-hoc routing / if your
destination is out of reach, intermediate nodes will automatically forward it through the
o"timal route within the @i"erlan networ$ (the routes are regularly automatically
recalculated). @i"erlan is also totally ad-hoc, re&uiring no configuration and no central
controller.
%he main deficiency of @i"erlan standard is that it doesn't "rovide real isochronous
services (but comes &uite close with time to live and "riority), doesn't fully s"ecify the
access "oint mechanisms and hasn't really been "roved to wor$ on a large scale in the real
world. 7verhead tends also to be &uite large (really big "ac$et headers).
@i"er1an suffers from the same disease as ;<3.== / the re&uirements are tight and the
"rotocol com"lex, ma$ing it very ex"ensive.
&.4 (i%erLan --
(i%erLan -- is the total o""osite of 1iper9an (see above O-). %he first @i"er1an was
designed to build ad-hoc networ$s, the second @i"er1an was designed for managed
infrastructure and wireless distribution systems. %he only similarities is the @i"er1an II is
being s"ecified by the 5%I (*roadband )adio Access (etwor$ grou"), o"erate at D ?@,
(D.6 to D.C ?@,) and the band is dedicated in euro"e.
@i"er1an II was the first standard to be based on =6*M modulation (see chapter 2,3,2).
5ach sub-carrier may be modulated by different modulations (and use different
convolutional code, a sort of >5#), which allow to offer multi"le bit-rates (G, B, =3, =;,
3C and EG Mb+s, with o"tional D6 Mb+s), with li$ely "erformance around 3D Mb+s bit-
rate. %he channel width is 3< M@, and includes 6; 7>.M carriers used to carry data and
6 additional are used as references ("ilot carriers - total is D3 carriers, E=3.D $@, s"acing).
@i"er1an II is a ireless ATM system (see chapter +,:,2), and the MA# "rotocol is a
%.MA scheme centrally coordinated with reservation slots. 5ach slot has a D6 * "ayload,
and the MA# "rovide A) (segmentation and reassembly - fragment large "ac$ets into
D6 * cells, see chapter +,.,.) and A)N (Automatic )e&uest - MA# retransmissions, see
chapter +,.,:). %he scheduler (in the central coordinator) is flexible and ada"tive, with a
call admission control, and the content of the %.MA frame change on a frame basis to
accommodate traffic needs. @i"er1an II also defines "ower saving and security features.
@i"er1an II is designed to carry A%M cells, but also I! "ac$ets, >irewire "ac$ets (I555
=EB6) and digital voice (from cellular "hones). %he main advantage of @i"er1an II is that
it can offer better &uality of service (low latency) and differentiated &uality of service
(guarantee of bandwidth), which is what "eo"le de"loying wireless distribution system
want. 7n the other hand, I'm worried about the "rotocol overhead, es"ecially for I!
traffic.
&.5 =%enAir
7"enAir is the "ro"rietary "rotocol from 'ro9im. As !roxim is one of the largest
0ireless 1A( manufacturer (if not the largest, but it de"ends which numbers you are
loo$ing at), they are trying to "ush 7"enAir as an alternative to ;<3.== through the
L-6 (0ireless 1A( Intero"erability >orum). !roxim is the only one having all the
detailed informations on 7"enAir, and strangely enough all the 7"enAir "roducts are
based on !roxim's module.
7"enAir is a "re-;<3.== "rotocol, using >re&uency @o""ing and <.; and =.G Mb+s bit rate
(3>J and 6>J). %he radio turnaround (si,e of contention slots and between "ac$ets) is
much larger than in ;<3.==, which allow a chea"er im"lementation but reduces
"erformance.
%he 7"enAir MA# "rotocol is #MA+#A with MA# retransmissions, and heavily based
on )%+#%, each contention slot contains a full )%+#% exchange, which offer good
robustness but some overhead. A nice feature of the "rotocol is that the access "oint can
send all its traffic contention free at the beginning of each dwell and then switch the
channel bac$ to contention access mode.
7"enAir doesn't im"lement any encry"tion at the MA# layer, but generates (etwor$ I.
based on a "assword (ecurity I.). %his "rovide some security only because !roxim
controls the way all the im"lementation behave (they don't "rovide a way to synchronise
to any networ$ as ;<3.== manufacturers do). 7"enAir also "rovide coarse "ower saving.
&.& (ome76 3 .A'
)7'E : this chapter was written when was finishing writing the !W"$ :,< specification
in 0ecember A6, "fter left the 1ome8F# a lot of big political game did happen# which
triggered some critical changes to the specification (!W"$ :,:), donBt really know how
much of it is still accurate# but believe that the standard is no longer as open and
vendor neutral as it was and that performance has been dramatically reduced,
%he (ome76 is a grou" of big com"anies from different bac$ground formed to "ush the
usage of 0ireless 1A( in the home and the small office. %his grou" is develo"ing and
"romoting a new )adio 1an standard / .A'.
%he @ome is a good mar$et for 0ireless 1A( because very few houses are nowadays
cabled with 5thernet wire between the different rooms, and because mobility in the home
is desired (browse the web on the sofa). %he use of the 3.6 ?@, band allows a free
worldwide de"loyment of the system.
%he @ome)> has decided to tac$le the main obstacle "reventing the de"loyment of
0ireless 1A( / the cost. Most users 4ust can't afford to s"end the money re&uired to buy
a cou"le of )adio 1A( cards to connect their !#s (without tal$ing of the access "oint).
%he main cost of a radio 1A( is the modem. As this is analog and high "ower
electronics, it doesn't follows /ooreBs law (the mar$et trend that allow you to buy a %ray
at the "rice of a calculator after a few years) and modems tend to be fairly stable in "rice.
Frequency 1opping modems tend to be less ex"ensive, but the 6<.,:: s"ecification
im"ose tight constraints on the modem (timing and filtering), ma$ing it high cost. %he
.A' s"ecification, by releasing slightly those constraints, allows for a much chea"er
im"lementation, but still $ee"s a good "erformance.
%he MA# "rotocol is im"lemented in software and digital, so doesn't contribute that
much to the final cost of the "roduct (exce"t in term of develo"ment cost). )eleasing
some hardware constraints "revented the use of the ;<3.==, which anyway was much too
com"lex and including too many features not necessary for the tas$.
%he main $iller a""lication that the @ome)> grou" envisages is the integration of digital
cordless tele"hony and the com"uting word, allowing the !# to reroute the "hone calls in
the home or to offer voice services to the users.
A new MA# "rotocol has been designed, much sim"ler, combining the best feature of
.5#% (an 5%I digital cordless "hone standard) and I555 ;<3.== / a digital cordless
"hone and ad-hoc data networ$, integrated together.
%he voice service is carried over a classical '0/" "rotocol (with interference "rotection,
as the band is unlicensed) and reuse the standard .5#% architecture and voice codec. %he
data "art use a %!/"(%" access mechanism similar to ;<3.== (with MA# level
retransmission, fragmentation...) to offer a service very similar to 5thernet.
%he = Mb+s >re&uency @o""ing "hysical layer (with o"tional 3 Mb+s using 6>J) allows
G voice connections and enough data through"ut for most users in the @ome. %he voice
&uality should be e&uivalent to .5#% in 5uro"e and much better than any current digital
"hone in the '. .ata "erformance should be slightly lower than ;<3.==. %he MA#
"rotocol has also been designed in a very flexible way, allowing to develo" very chea"
handset or data terminals and high "erformance multimedia cards for !#s...
%he .A' s"ecification is an o"en standard (in fact, more o"en than ;<3.==, because
there should be no royalty or "atent issues), &uite sim"le and straightforward. In fact, the
combination of voice and data gets already most mar$eting "eo"le drooling K %he only
drawbac$ is that you will have to wait a bit before seeing 0A! "roducts in your
favourite su"ermar$et...
&.) !l$eTooth
*lue%ooth should not even be mentioned in this document, but "eo"le $ee" thin$ing that
*lue%ooth is a 0ireless 1A(. *lue%ooth is a ca#le re%lacement technology mostly
develo"ed and "romoted by >ricsson with the hel" of -ntel, offering "oint to "oint lin$s
and no native su""ort for I! (need to use !!!). It may be good for some a""lications, but
not for 0ireless 1A(s.
I "ersonally read the *lue%ooth s"ecification, and I was not im"ressed, ex"ect by the si,e
of the thing (more than =D<< "ages K). My ta$e is that *lue%ooth offers the functionality
of a ireless 5.!, and in fact loo$ing into the huge s"ecification we can see some
similarities in the design.
*lue%ooth offers the "ossibility to create a set of "oint to "oint wireless serial "i"es
(8f%omm) between a master and u" to G slaves, with a "rotocol (!0$) to bind those "i"es
to a s"ecific a""lication or driver. %he *lue%ooth mindset is very vertical, with various
"rofiles defining every details from bit level to a""lication level. %#!+I! is only one
"rofile, im"lemented through !!! in a s"ecific "i"e. %here are other "i"es for audio,
7bex... 0ith *lue%ooth, nodes need to be ex"licitely connected, but they remember
bindings from one time to another.
%his is miles away from the current wireless 1A( a""roach (connectionless broadcast
interface, native I! su""ort, cellular de"loyement, hori,ontal "lay), so *lue%ooth doesn't
fit %#!+I! and wireless 1A( a""lications too well. 7n the other hand, as a wireless '*,
it fulfil a role that regular wireless 1A(s can't, because %#!+I! discovery and binding
"rotocols are more heavyweight.
#urrently, *lue%ooth is moving very slow (my first reading of the s"ec was autumn BC -
then called M#-1in$) due to its com"lexity and the inherent limits due to the "rotocol
design ("eo"le are learning how to wor$around 2features2), but eventually some "roducts
should reach the mar$et and later on software su""ort should come...
In summary, if all you want is to run %#!+I!, you may find it chea"er and more effective
to (7% wait for *lue%ooth and live without the hy"e.