Wireless Networks: A Cost-Effective Way to the Internet
Danton NUNES <firstname.lastname@example.org>
Access to the Internet in many places is either difficult or expensiveor a perverse blend of both, due to poor public communicationinfrastructure and/or the high cost of existing services. Thispaper discusses an alternative to leased or switched lines basedon wireless technology.
Two case studies are discussed: a college campus network and asmall wireless backbone that brings Internet access to businessusers in five neighboring towns on the Paraiba's Valley, in thestate of São Paulo, Brazil.
The University of Taubaté (UNITAU) is not concentratedon a delimited campus; rather, its buildings are spread over thetown. Connecting all buildings with leased lines exceeded theacceptable budget and yielded unacceptable performance, so theschool decided to go wireless. UNITAU's wireless network operatesat 2.4 GHz direct sequence spread spectrum. The network topologyis a star in a point-to-multipoint configuration. The centralstation is a router equipped with an omnidirection active antenna,and the satellite stations are equipped with high-gain directionalantennas. The central station was installed in a clock tower thatis visible from almost anywhere in the town, so straight-linevisibility is not a hurdle. The routers on all stations are old386 PCs rescued from the cemetery of obsolete machines. They areuseless as desktop personal computers because they can't efficientlyrun the new graphic user interfaces with plenty of whistles andbells, but they are still useful as routers. UNITAU's wirelessnetwork is extensible in the sense that new satellite stationscan be added easily. If the need arises, the topology can evolveto a tree, adding a new central station.
UNITAU's network design was based on InterNexo's own wirelessnetwork. InterNexo is an Internet service provider, and the wirelessnetwork is its main medium to offer Internet access to businessusers. At the time of this writing its configuration comprisesthree nodes (central points with omnidirectional antennas) pluseight satellite client stations. Two nodes are located in SãoJosé dos Campos, and the third one is in Taubaté,39 km away. Communication between São José and Taubatéis carried through a dedicated 2.4 GHz link with active high-gaindirectional antennas. The system operates at 915 MHz in SãoJosé and 2.4 GHz in Taubaté. The original plan wasto operate at 915 MHz there too, but strong interference causedby a radiocall (pager) operator made the 915 MHz ISM band unusable.Radiocall operates at 200 W or more, while ISM services are limitedby Brazilian regulations to 1 W. The routers are PCs running Linux.They accumulate a number of functions besides routing, such asserving Web pages, time reference, and diagnostic tools. Thissystem began operating in November 1996. The only outages werecaused by the already-mentioned interference and a malfunctioningamplifier. The uptime is comparable to the much more expensiveleased lines.
One year of experience running wireless networks at ISM bandsshows that this technology is a good choice wherever cost, availability,or reliability of ground-based alternatives is problematic. Interferencebetween different spread-spectrum systems is far from being critical;however, interference caused by other sources is, especially inthe 915 MHz band, where powerful neighbors dump their spectralgarbage.
Our company is a small Internet service provider established inSão José dos Campos, in the valley of the riverParaíba, one of the most industrialized areas of the stateof São Paulo, Brazil. Our goal is to provide high-qualityInternet access services to business users in the region. We startedwith leased lines but at that time they were very expensive andperformed poorly. They are a little less expensive and work alittle better nowadays but price and performance are still burdens.In the beginning of 1996 we were looking for alternatives to theleased lines. Packet switching networks based either on X.25 orframe relay were available but even more expensive than leasedlines, considering the expected volume of traffic. ISDN was notoffered by the local telephone company. The wireless option waspromising so we decided to spend some time and money on it.
We were really looking for
The answer to the first wish was Lucent's (formerly NCR's) WaveLanfamily of products on the hardware side and Linux, a hi-tech Unixclone, distributed freely under GPL (GNU Public License).
The second wish was fulfilled by the Brazilian Ministry of Communications,which issued a regulation on the use of the former ISM (Industrial,Scientific, and Medical) bands, waiving the need of license providedthat only approved equipment was used (WaveLan is approved) andstrict operational limits on frequency, modulation, power, andfield intensity were observed.
We eventually learned about Brazilian manufacturers of antennasand ancillary equipment that allowed us not only to cover mostof the town but also to establish a 39Km link at very low power.That was the third wish.
After some five minutes of very deep analysis we realized thatthe best topology for the wireless access provider is a star withthe customers at the tips and the gateway to the Internet in thecenter. In order to reach all actual and potential customers,the central station should be equipped with an omnidirectionalantenna. An amplifier was added to compensate for its low gain.The standard customer equipment included a high gain Yagi antenna,with or without amplifier according to distance and link conditions.
We choose to operate in the low ISM band (915Mhz), for, at thattime, UHF equipment was cheaper and maximum legal power was higherthan for 2.4GHz. This decision proved later to be a mistake. Thelow band is very sensitive to interference from ill-behaved equipmentoperating in neighbor bands. On the other hand, the prices ofSHF equipment fell down practically to the same level of UHF.
São José dos Campos viewed from Central StationOne.
The next little problem to be solved was the physical installationof Central Station One. Our office is on the sixth floor and theantenna is on the roof, above the eighteenth floor. There wasno way to install a low loss coaxial cable from our office tothe roof due to its diameter and mechanical characteristics; thebuilding was just not designed for this. We decided then to putthe computer there on the roof and send it power and LAN accessthrough smaller and much more flexible cables than the coax waveguide.A special box with passive thermal control was built to housethe equipment. The "pigeon loft," as we call it, wasattached to the wall that receives the least direct sunlight allyear round. In November 1996 the whole system was ready for betatesting.
The pigeon loft.
Tests were conducted with a link to our most faraway customer,a BBS and dial-up access provider. The results were pretty good;we finally got LAN class access almost 6Km from the hub. The resultswere received enthusiastically by our users and in a few monthsmany converted to the wireless system. We managed to keep theprices of the new service below the price of leased lines, soit was attractive from both price and performance viewpoints.
First São José, then, the world! Thanks to the successof Central Station One and under the pressure of some customers,we started plans to have another point of presence in Taubaté,an important economic and cultural center of the Valley, some40 Km away. The idea was to use the same technology to establisha link between Central Station One and the yet-to-be Central StationTwo. The 915MHz band showed not to be feasible, for the antennaswould need huge (and expensive) reflectors, so we moved to 2.4GHz.After some back-of-the-envelope calculations, we found out thatsomething better than 55dBi plus amplifiers at both ends wouldbe enough. Station One was endowed with a new 2.4GHz WaveLan board,an amplifier, and an antenna of 24dBi. Station Two was installedon an FM radio tower with a big 34dBi drum antenna. After sometrouble with the alignment of the drum antenna, we managed toput the link into operation. The distance between both ends asmeasured by a GPS receiver is 39Km. Central Station Two also hadan omnidirectional 915MHz antenna to convey the signal to thecustomers in Taubaté. Soon we would have to change it to2.4GHz due to interference from a pager repeater (read the interference dramabelow). The first ping to Taubaté was issued at 4:00 PM,19 May 1997. A few weeks later, regular operation began.
24dBi antenna with amplifier
Another important town in the region is Jacareí, some 15Kmfrom São José. We could not reach it because a buildingin São José downtown eclipses it completely. Asartillery was out of the question, we looked for a place to installanother station in São José with a view to Jacareí,our office building, and to the few areas in São Joséwe could not cover from Station One. So Station Three was born.It was equipped with an omnidirectional antenna (as always), plusa Yagi to enhance the gain in the direction of Jacareí.It entered on duty in December 1997.
Our wireless backbone now extends from Jacareí to Taubatéwith three points of presence. Within the range and line of sightof any of the central stations and with favorable conditions toinstall antennas, we are able to put any network on the Internetin a matter of hours.
Taubaté is a midsized town in the heart of the Valley ofParaíba, about 40Km from our offices in São Josédos Campos and some 140Km from the big city of São Paulo.It hosts a complex of high education, research, and communityservice facilities, the University of Taubaté, best knownby the acronym UNITAU. UNITAU is not concentrated on a delimitedcampus; its buildings are spread over the town instead. Howevergood this may be, for it allows best integration of the universityand the community, communication between the many university buildingsbecomes a problem. Connecting all the buildings with leased lineswas far beyond acceptable budget and below acceptable performance.Fiber optics was out of the question in spite of its relativelylow cost and remarkable performance. The telephone company doesnot allow passage of foreign cables in its galleries and diggingnew galleries would be very expensive, require special permitsfrom City administration, and cause a lot of trouble to the citizens.
The solution was to go wireless. We presented a project basedon our own wireless network and we won the bid. The initial goalwas to connect the local area networks at six buildings. Due tophysical constraints there was no way to make an antenna at oneof the buildings see the other five, so we decided in favor ofa star configuration. A seventh point was added to work as a centralrelay station. This station was installed in a high clock towerthat is visible from almost everywhere in the town, including,of course, our six buildings. The radios selected were Lucent'sWaveLan in the 2.4GHz band. We decided against the 915MHz banddue to a previous experience with interference caused by a pagerrepeater on our own system (read the story below in this paper).
Omnidirectional boosted antenna
The routers on all stations are old 386 PCs rescued from the cemeteryof obsolete machines. They are useless as desktop personal computersbecause they can't efficiently run the new graphic user interfaceswith plenty of whistles and bells, but they are still useful asrouters. Linux was installed and configured on a master disk.The master disk was copied seven times so at the end we had adisk for each station plus a backup unit. All disks are identicalexcept for one file containing just the system name. The routerswere assembled and tested at the laboratory of UNITAU's InformationScience Department.
The central station received an omnidirectional antenna boostedby a HyperLink amplifier. Each satellite station received a highgain (24dBi) parabolic antenna and no amplifier at all. This antennacould be a little oversized for some of the points but it wasadopted in all points for the sake of uniformity. All stationswere endowed with uninterruptible power supplies (UPS) for tworeasons: the UPS works as a power stabilizer and makes the systemstill usable during power outages.
At the time of this writing, UNITAU's wireless network has beenoperating painlessly for six months. Its star topology may notbe very efficient in the sense that each data package is transmittedtwice, from a satellite station to central station and back to(some other) satellite station, but it is not strongly constrainedby city landscape and it is extensible by adding new satellitestations. If need arises, a new central station in another pointcan be also added to the system. The use of old computers, off-the-shelfcomponents, and free software (Linux) made the whole project fitthe low budget of the university.
The key element of the wireless networks described here is a clusterof radio stations with one single central station endowed withan omnidirectional antenna and a number of satellite stationswith high gain directional antennas oriented towards the centralstation. At each station there is a PC with a radio interfacethat behaves pretty much like a standard ethernet card. The cluster,however, does not work like a standard ethernet LAN because eachsatellite station can see only the central station unless, againstall odds, two opposite satellites and the central stations lieon an approximately straight line. On a well-configured ethernetLAN, each station can see each other.
If a data package must go from a satellite station A to anothersatellite station B, it must be relayed through the central station.There are at least two ways to implement the relay mechanism:the stations may be configured to work like bridges or IP routers.Linux running on the PC supports both. Which is better dependsmostly on the application of the wireless network. If it is tomerge several LANs into a big virtual LAN covering different protocolfamilies, bridging is certainly the right choice. On the otherhand, if it is an IP-only network and the connecting LANs mustkeep a certain degree of independence, IP routing performs best,for the routers may support packet filtering, IP translation,and other firewall features. In both cases studied here, IP routingwas the relay method of choice. The wireless network of the Universityof Taubaté is IP-only and connects a number of differentdepartments, each with its own access and security policies. Thewireless Internet access provider has different commercial enterprisesat each satellite station, so the differences in access and securitypolicies are even more dramatic.
When writing down the routing tables of the satellite stations,one must take into account that while the radio interface lookslike multiple access it actually works as if it were point-to-point.This peculiar behavior also makes dynamic route updating (e.g.,by RIP or OSPF) something of a nightmare. If the point-to-pointcharacter of the radio interface is not properly accounted for,the satellite station will not be able to talk to other satellitestations although other hosts on its LAN will.
When we decided to install a point of presence on our wirelessInternet access service in Taubaté, we designed the centralstation with two radios: one at 915 MHz and an omnidirectionalantenna to feed our customers and another at 2.4.Ghz and a veryhigh gain dish antenna to link to our main office in SãoJosé dos Campos 39 km away. We choose 2.46hz for the long-distancelink because the size and cost of the antennas for 915mhz wereprohibitive. The link to the customers was chosen to be at 915MHz due to our previous successful experience in São José.
Three months after the Taubaté station went on duty, thenightmare began. From time to time, randomly, the 915 MHz radiowent deaf for 20 to 30 seconds. When it was back to sanity, itreported good link conditions to all customers, very low backgroundnoise, and an astoundingly high count of rejected packets dueto unrecognized network IDs. The malfunction happened more oftenin daylight, especially during business hours, and almost vanishedlate in the evening.
Using a spectrum analyzer we found out that there was a strongsignal at 921MHz in the air during the deafness crisis of ourradio. Such signal was inside the spread spectrum band, from 902to 928 MHz. This could explain the high count of rejected packets.Finally, after some time roaming the neighborhood carrying weirdequipment, like Ghostbusters, we managed to single out the sourceof interference: a repeater of pager operator, licensed to workat 931MHz, 150 Watts, against our mere 1 Watt. The 921 MHz signal,was -- still is, quite likely -- due to a parasite 10 MHz amplitudemodulation on their FM carrier. This kind of spectral contaminationis prohibited by Brazilian telecommunication regulations. To makethings worse, the repeater antenna was just some 30 meters fromours. (The place is indeed overcrowded with antennas due to itssuperb view of the whole town.)
After unsuccessful negotiations with the owner of the radio callrepeater and a never-answered e-mail to the liaison office ofthe Ministry of Communications in São Paulo, we reckonedthat legal action would cost us more and take longer than changingour whole system to 2.4GHz. So we did.
Quosque tandem interferentes abutere patientia nostra
Amplifiers for outdoor use are not as reliable as we would like;some flaws in the operation of our system were due to amplifierfailures. Cabling and connectors are also a source of problemsif they are not handled properly.
Installation of outdoor amplifiers requires some care with properprotection against water pouring into its case. The case coveris tightly closed with an o-ring, and all possible points wherewater could get through are sealed with silicon rubber. Neverthelesswe had an amplifier that died after nine months on duty. Therewas more than 2ml of water inside the case and some contacts weredeteriorated by oxidation. The remains of the device were sentback to the manufacturer for analysis.
Another amplifier failure that annoyed us a lot was a switchingfrom transmit to receive state before the transmitted signal wentoff. This problem was very hard to detect because it looked likeas if some weird source of interference were in the air. We werekind of paranoid about interference at that time and that wasour first hypothesis, quickly dismissed after tests with a spectrumanalyzer. The funny thing was that by reducing the maximum transmitunit of the interface from the usual 1500 octets to a very lownumber, say, 168, the problem was minimized. We could only findout that the troublemaker was the amplifier after swapping everycomponent on that line, from the radio board to cables, DC injector,and amplifier for their backup units.
The frequency response of the amplifiers goes far beyond the spreadspectrum range. This makes them sensible to signals in neighborbands. The symptoms are the same as those of interference butnothing is observable with a spectrum analyzer. The good newsis that the use of cavity bandpass filters between the antennaand the amplifier to make it more frequency selective. The badnews is that a cavity filter costs about the same as the amplifieritself and is a rather clumsy object to install outdoors. Noticethat a cavity filter would not avoid the kind of interferencedescribed above.
We experienced some failures due to cables with improperly assembledconnectors. At least in one case it was very hard to identifythe fault. Nowadays we test every cable and connector before installationdespite the fact that they feature beautiful logos and qualitycertificates of well-known, respected manufacturers.
We have been successfully operating a wireless network for Internetaccess since November 1996. It proved to be cost-effective: weoffer the service profitably at a price lower than that for leasedlines. The project for UNITAU showed that the same technologyfits the need for connectivity of distributed organizations inurban areas. Remote sites can be reached also at relatively lowcost. Another point to remark is that once an infrastructure ofbase stations is set up, installing satellite stations is quickand easy. A portable satellite station made up by a notebook PCwith a PCMCIA radio adapter may be useful in emergency situations.In areas where the conventional communication infrastructure isprecarious or not existing at all, systems like those describedhere may be the best -- if not the only -- solution to bring Internetaccess.
The good cost-performance of this technology depends stronglyon the availability of low-cost off-the-shelf components, likethe WaveLan boards, cables, amplifiers, and antennas. Such equipmentis not expensive due to mass-scale manufacturing, which is onlyfeasible because the ISM bands are free areas of the spectrum(although in a free area, some regulations must be obeyed so everybodycan coexist there). Look for radio equipment that does exactlywhat ours does but in another frequency range. It will certainlycost one order of magnitude or even more than if it were designedfor an ISM band.
Another key point for the success of our wireless networks isLinux operating system -- not just because it is free, but becauseit is distributed with the source code of every component. Theavailability of source code was instrumental to make it work witha previously unsupported new generation of radio boards. The talkto the developers of the code via e-mail was better than any commercialsoftware support we are aware of. Initiatives to produce freesoftware like Linux and the Free Software Foundation must be encouragedand supported.
Interference, especially in the low ISM band of 915MHz, is a realdrawback. Unfortunately this band is between analogue cellulartelephony and ground-based applications. Cellular telephony isnot a source of interference because its operation is very professionaland in accordance with regulations but we cannot say the sameabout the ground-based applications around 931MHz and above. ISMunlicensed operation is a secondary service so it is supposedto accept any interference, though the only legal argument againstthe operators of interference sources is to prove that they oftenviolate the regulations on side lobes and modulation. We neverobserved denial of service by interference in the 2.4GHz band.