Fixed wireless

This is, for example, a valid criticism of Hughes DirecWay. They could have a ``nomadic'' system with nearly global coverage within their ever-expanding worldwide Empire. It would be the sort of thing that works at camp sites, but at costs that make a raving mockery of Inmarsat's similar terminals---if they would only resolve the interference problems of mispointing their satellite uplink. Instead, they aim to match low-end cable and DSL Interweb access. You're thinking about signing up for DirecTV DBS, but you can only get a fair price on cable Internet if you also buy cable TV? No problem. You can get Internet from DirecTV, too! They probably want customers to ignore the wireless advantages so that they will think as little as possible about how the ``package'' differs from cable. so the nomadic aspect is swept under the rug (and, for the moment, explicitly forbidden---only AUTHORIZED tradesmen may aim the uplink dish).

However, Airfiber and Radiant are terrestrial companies that use bands unsutiable for mobile or nomadic use. The mobile (celfone) and nomadic (camp site) advantages of wireless do not pan out with these two companies. Airfiber uses red light beams, so I don't think that needs much elaboration---the beamwidths for FSO (``free-space optical'') are 3 - 6 mRad, and there are shadows. Radiant uses the 28GHz and 40GHz ``LMDS band'' (local multipoint distribution, a stupid TLA that means ``short-haul narrow-beam 28GHz''). The Radiant links are also narrow beams that cast sharp shadows---the 3dB beamwidth is 4.5 degrees in the horizontal plane and 9 degrees in the vertical plane---so neither will generalize to mobile radios. The wavelengths are too short.

Apparently there is some dialog about ``picocellular'' mobile service at 30 or 60 GHz, but I think this is dumb.

Radiant Networks uses radio at 28 or 40 GHz (about 10000000nm or 7500000nm), and Airfiber uses red light (at 785nm), yet they have several eerie similarities. Both use antennas on rooftops (not in windows). Both use antenna modules that contain 2 - 4 antennas which are directional and independently rotatable over 360 degrees. Both have maximum link lengths of about 1km (although they fall decisively on either side of 1km). Both wire customer nodes into a mesh by putting ATM switches in their rooftop units: traffic hops from one customer node to another, and the number of non-customer repeaters is ideally kept to a minimum. The ATM switch in the rooftop node commits them to a specific kind of QoS, and means they can provide voice and data services the same way that the telephone company provides them, putting them in direct service-for-service competition with fixed wired networks.

They both face challenges with bad weather (rain for Radiant, fog for Airfiber), and they both address these challenges with basically the same argument: they use transmit power modulation, they artificially limit themselves to short links (0.5km for OptiMesh, 2.8km for Radiant) even though the good-weather distance is longer, and they use weather pattern data to show they will attain some hard figure of either 99.99% or 99.999% reliability even though the link does go out in particularly bad weather, and this is competitive with wired reliability (which also goes out in particularly bad weather). I like this argument. I find it convincing. Really only Radiant makes the argument convincingly, but this is AirFiber's basic plan, too.

Airfiber makes an additional argument: you can back up their laser links with unlicensed 60GHz microwave. This strategy doesn't appeal to me for several reasons. They differ in a few ways that may turn out to be significant. Lightpointe, an FSO company, has a nice summary of FSO vs. LMDS that's general to the two types of physical links. To expand Lightpointe's summary to the complete Radiant and Airfiber networks rather than the general link types, here are the differences between the OptiMesh and Radiant implementations, as far as I can tell. I may have made mistakes:

• Cost. Radiant's system is under $10,000 per node, which they hope to reduce to about $1000 per node although that has not happened yet.

  OptiMesh claims, ``the cost to deploy OptiMesh averages about $20,000 per building. That total includes site acquisition and preparation, equipment and overhead such as the development of a network operations center and roof rights.''

• I think OptiMesh is faster. It's hard to know for sure because
  • Radiant performs an analysis and concludes peak- and sustained-bandwidth per subscriber, at a given subscriber density. OptiMesh performs no such analysis.
  • OptiMesh quotes OC-12 (622Mbit/s) in both directions, per link. For Radiant, ``peak data is determined by the physical interfaces provided for the user, which for the Radiant Mesh is limited to 25Mbit/s. . . . [that is,] 50Mbit/s [shared between link directions] (25Mbit/s duplex for TDD . . .)'' The 60GHz point-to-point radios from Telaxis and Harmonix are faster than 1Gbit/s, so I think this is a cost-saving thing, not a technical limitation of the radio link type. but it is a pretty big deal, making Radiant appropriate only for individual users in the suburbs, not large apartment buildings much less office towers where many humans will share a single rooftop unit.
  • Radiant perturbs the network topology, using special node types near the wired interconnect point. OptiMesh uses only one node type. This may mean that OptiMesh has different interconnect-bottleneck parameters than Radiant, but OptiMesh might be able to achieve an equivalent effect by putting more than one node on rooftops which are at and one-hop-away-from the wired interconnect.
• OptiMesh doesn't have a plan for installing ``seed nodes'' on streetlights. All nodes are subscriber nodes on rooftops.
• Radiant's rotating antennas rotate 360 degrees in one plane, in 1-degree increments. From the photos of the hoods on Airfiber's web page it looks like OptiMesh mirrors rotate in two planes, but only over a narrow angle. However, this marketing verbage claims otherwise:

  Each optical link has a field of view of 360 degress horizontally and +/- 20 degrees vertically. Each optical link automatically tracks out any sub hertz movement of the building (e.g., movement caused by wind loading or thermal building expansion).

  so I guess they do. Another photo makes this look plausible. (Why doesn't Airfiber photograph that product for their marketing instead of this other ugly thing?) It's important to resolve this issue decisively. The question is whether the NOC management tool uses antenna rotation to reconfigure the mesh, breaking old connections and making new ones when, for example, new customers subscribe or obstructions appear (as Radiant does), or whether the rotation is only used to track building sway.

  In any case, OptiMesh wins with their 20-degree vertical rotation rather than 9.5-degree vertical beamwidth, but I'm not sure how important this is to typical roofscapes.

• Radiant's nodes interfere with themselves a lot more. In a given timeslice, a Radiant node can either transmit or receive on only one of the four antennas. I think OptiMesh nodes can transact on all eight (link, direction)s at once.

  Radio and light are different enough over a single link that I don't think it proper to consider this difference in itself a liability of Radiant---rather it should be absorbed into the fact that Radiant's links are slower than OptiMesh's links. However, Radiant's interesting claims about mesh networks were made with these parameters in mind, so their mesh analysis may not generalize perfectly to the optical mesh. What interests me about these products is their mesh architecture, so this is a particularly interesting difference even though I'm not sure it has any fundamental influence on the finished system.

  I think TDD (time-division duplex) has unpleasant political implications because it gives carriers a way to sell asymmetric links. The network itself doesn't make transmitting data more expensive than receiving data like Digital Television does, but carriers like to discriminate between the price of ``upstream'' and ``downstream'' bandwidth, charging less from subscribers that are allowed to consume like stupid sheep but not allowed to publish. I think the subsidy, if any, should be in exactly the opposite direction. Radiant's system makes asymmetric links more cost-effective than OptiMesh's system, where intuitively it seems like the OptiMesh system would leave bandwidth unsold if carriers artificially capped ``upstream'' data rates. Although a purely technical perspective might consider this a liability of OptiMesh, a political perspective considers it an advantage that the carrier's hands are tied in this way.

• I could be convinced otherwise, but for now I am more confident in Radiant's ability to deal with rain than OptiMesh's ability to deal with fog. The short version of this difference is: OptiMesh links that match Radiant links in reliability could easily be as little as 1/10th - 1/20th as long.

  Radiant considers rain in their analytical pitch:

  link budget analysis shows . . . Maximum link ranges calculated show between 2,800m and 750m depending on modulation scheme, assuming 99.99% availability in 32mm/hr of rainfall.

  What they mean is, rainfall exceeds 32mm/hr only 0.01% of the time in Northern Europe. while OptiMesh simply claims they have achieved ``carrier-class reliability on 0.2 - 0.5km links,'' without actually calculating the downtime that a specific city's MAN would experience.

  OptiMesh links are at most 19 mW at 785 nm. A trade publication surveys similar equipment in this band and power level suggests that 4-9s (P(link up) = 0.9999) is a ridiculous fantasy for this type of FSO, and that 0.99 - 0.999 is more realistic for 0.5km links. The survey bases this claim on human visibility records from La Guardia airport, which is pretty shoddy because to use the La Guardia weather data they need to measure what power level of FSO has equivalent penetrative power to human eyesight. Even such a measurement isn't a perfect way to scale the data for various FSO systems because, as they point out, fog causes problems for vision through sunlight scattering washing out the image, while for FSO the problem is almost entirely attenuation.

  I understand AirFiber claims their mesh provides ``redundnacy'' that other FSO systems lack, but fog attenuation is not distributed as randomly over network links as it is over days of the year---entire buildings or even the entire mesh is likely to be engulfed in fog at once. The only thing OptiMesh has done to deal with fog is to shorten the link length. Is this an effective way to deal with fog? As we'll see later, it definitely is. The trade publication says fog is 30 - 60dB / km. This means the 3dB attenuation of fog is 50 - 100m: for the 30dB fog, every additional 100m that the link must penetrate requires double transmission power. Reality check: that means an extra 1km requires 1000x the transmitter power. Absolutely, link length is a big deal.

  Terabeam uses extremely high power (1 watt) which is only eye-safe at their 1550nm band, along with huge apertures, to minimize the effects of weather, but from the trade paper above I think they may still have problems with fog. Presumably the incorporation of a Giant ``Laser'' also makes their endpoints much more expensive, reducing the appeal of FSO over RF. Unfortunately, Terabeam has a really lousy web site---their markedroids seem to have chosen press releases over the detailed technical arguments that Radiant makes. One thing that the press says but the website doesn't, is that Terabeam sells service subscriptions, not infrastructure. It's hard to tell to what extent anyone is selling anything at this stage---a lot of the products seem ``experimental,'' if such a term exists at these high levels of inter-corporate cooperation.

  Is 1 watt enough to blast through fog? Well, let's figure it out. OptiMesh says their ``typical'' broadcast power is 9.5 mW, and the paper above says fog attenuation is 30 - 60 dB / km. My fog table is pretty crude because it ought to consider other kinds of attenuation besides fog---in my model, fogless 9.5mW links could be infinitely long---but I think this is a reasonable approximation since fog is so intense compared to everything else. It also assumes fog attenuation is about the same at 785nm and 1550nm, which I think is true---the difference between the two bands is mostly that 1550nm is eye-safe at brighter intensities, but if you switch your 785nm system to 1550nm you don't immediately gain anything until you pay extra for the more expensive brighter transmitter. At a given cost, I don't see the point of changing bands.

  Amount of fog	no fog	200m of fog	500m of fog	1km of fog	1.5km of fog	2km of fog
  30dB fog attenuation	1	0.251	0.0316	0.00100	0.0000316	0.00000100
  power for 30dB fog	9.5mW	38mW	300mW	9.5W	300W	9500W
  power for 60dB fog	9.5mW	150mW	9.5W	9500W	9500kW	9500MW

  The point of this table is that Terabeam's 1W transmitters are only about 50x brighter than OptiMesh's 19.5mW transmitters. Fifty times brighter sounds like a lot, but look at it this way. However far OptiMesh can go in 30dB fog (even in the pessimistic hypothetical that this distance is nearly zero), Terabeam can go at most 0.57km farther (or 0.28km farther in 60dB fog).

  Maybe it's not immediately intuitive, but that's the way attenuation works, so get used to it.

  Is Terabeam's increased power an effective way to deal with fog? well, for 0.3 - 0.5km links it definitely is. Beyond that distance, I don't know. The much bigger receiver aperture of that Magna monstrosity probably helps a lot, too, but remember the effects are of the same order: using the same received-power analysis, increasing the receiver diameter 10x will get you 0.33 - 0.67km further fog penetration.

  The problem with my table is that, unlike Radiant's analysis, it doesn't include statistical data about how often fog of a given thickness occurs in some specific subjectively foggy area. The interesting plot is NOT the power-over-distance that I've done, but rather distance-against-reliability at a given power level. If we had this elusive distance-over-reliability plot for a certain power level, one could easily estimate similar plots at other power levels. The line drawn on such a plot represents some fixed amount of fog attenuation, l, at which that particular system stops working. One must know this amount---let's say it turns out to be l = 10dB. If we double the transmitter power, the system will now work at l' = 13dB. So long as l and l' are expressed in the log scale of dB, we can make a new distance-over-reliability plot for the more powerful system by scaling (multiplying) the distance axis by l' / l. In this example, if the old system delivered 0.9999 reliability at 0.1km, the new twice-as-powerful system will deliver 0.9999 reliability at 0.13km.

  Anyway, saying ``the technology needs to deal with heavy fog'' is meaningless telco nay-saying. To compete with Radiant's (and layed fiber's) reliability, FSO need not deal with any of the fog that occurs during the foggiest 0.01% of the year. The problem is that no one seems to know how intense this 0.01% fog is in various urban areas. The information is available for rain, but not fog.

Granted, the speed and link-length factors remain. It's hard to know what to make of these, since the effective speed of a mesh doesn't depend on link rate alone---it also depends on intra-node interference and wired interconnect strategy. Likewise, a 10x shorter link length could, at worst, mean 100x as many ``seed nodes,'' but a practical situation might be much better.

The most interesting difference to me, since I'm not an investor, is that Radiant has a much nicer web site. In particular, they do a nice job of pitching the mesh aspect of their design, compared to competing traditional networks with Towers and Customers. I was a little baffled by their argument as I started to wade into it, because I kept thinking about their wired interconnects---don't these limit the mesh's capacity, just like an overloaded tower does with point-to-multipoint? As far as I can tell, this is true, but the mesh still has advantages:

• point-to-multipoint systems waste spectrum through interference when Customers are near to more than one Tower, which they usually are. Mesh systems can be efficient without the pesky problem of designing ``cells'' with sharp edges.
• it's hard to add a new Tower and split an existing point-to-multipoint cel that has become too crowded because many existing Terminal antennas have to be re-aimed at the new Tower. Radiant nodes have redundant motorized antennas meaning that unlike the cel, one can reconfigure the mesh without downtime and without a customer site visit, hopefully even with machine-made decisions about how to connect the mesh.
• the mesh decouples capacity from coverage, which is always important, but is especially important when the network is just getting started and has few subscribers.
• it matters very much where one puts Towers, but it doesn't matter too much where wired interconnects go within the mesh. This is a pretty big deal since it's kind of the whole point of the radio link---if you were able to declare exactly where you want the fiber ring to go for ideal Tower placement, you could simply declare instead that the fiber must go past all the customer sites, and there would be no ``last mile'' problem.
• in fact, the wired interconnects have directional antennas, so it's possible to put several of them in the same spot, and achieve the same bandwidth advantage as several Towers. I'm guessing this trick is already incorporated into Radiant's wired interconnect devices, and that's why their strategy has two layers of nonstandard nodes near the interconnect.

Anyway, I can't emphasize enough that the interesting characteristics of Radiant and OptiMesh are in the general properties of wireless meshes. It's interesting to see wireless over vastly different link types---omnidirectional 2.4GHz, directional 28GHz, and red light---all gravitating toward the mesh.

The old model for global networks was to use a mesh on the macro-scale, and interconnect the mesh to hub-and-spoke networks to accomplish the last mile. Radiant, OptiMesh, Ricochet, Mea are mounting an increasingly convincing challenge to this condescending telco dogma. The replacement model right now seems to be two independent, overlapping layers of mesh, with ``wired interconnects'' between the micro last-mile mesh and the macro global mesh.

Here is an interesting question: does mesh networking theory vindicate the usefulness of these ``wired interconnects'' and two-overlapping-layers arrangement, or do interconnects exist only as political demarcations between two administratively separate routing domains?

I'll end with my favorite quote from Radiant's copious website:

At the point where [Towers] are separated by 1.5km, the fibre to the [Tower], on average, goes within 350m of the subscribers themselves and thus it is likely to be more cost effective to simply deploy fibre to the home or the curb (once the [Tower] equipment and build costs are taken into account).

More up-to-date links:

  License-free:

    Gigabit:

      http://www.tessco.com/products/displayProductInfo.do?sku=485801
      http://www.meridianmicrowave.com/unlicensed-free_wireless_wans_60ghz.html

    Slower stuff, 50 - 100 Mbit/s, might be easier to get sponsored:
 
      http://www.connectronics.com/ceragon/#License_Free_Ethernet_Bridges -- see Fibeair 4858
   
    FSO:

      http://www.fsona.com/product.php?sec=155e
      http://www.lightpointe.com/products/fl_express.cfm
      http://www.usa.canon.com/html/industrial_canobeam/canobeam/canobeam120.html
      http://www.laserbit.net/products.php?pname=PRONTO
      http://www.mrv.com/products/line/terescope.php

  Licensed gigabit:

    http://www.elva-1.spb.ru/pdf/PPC-1000_flyer_Eng.pdf