``Wireless'' is epitomized by Bluetooth---it is replacing a wire without wires. Bluetooth is not so much a great leap forward as it is a thrifty optimization. All modern wired signaling standards send radio over the wires, which means we already have radios on both ends of the wire. Whether we run these radio signals over wires or over the air really doesn't make as much cost difference as one might expect. Over-the-air signalling might eventually become cheaper since it replaces costly components like plugs, jacks, and wires with a free antenna made from PCB traces.
Modern wire signaling like Fast Ethernet, Firewire, and USB 2.0 actually requires two chips: one ASIC for the MAC layer and DMA, and another chip called a ``PHY'' which is basically a radio. The PHY uses a different fabrication process than the controller ASIC because it handles analog radio signals rather than digital logic. It's possible to make ``mixed-mode'' chips that incorporate both analog radios and digital logic, but the fabrication process is more expensive than making single-mode chips, so I think sometimes it is cheaper to do the job with two chips than one chip. I suspect that we will find over time that making PHYs to work over the air rather than over wires does not add much cost, especially in very large quantities.
This idea hits someone in the shower. He shares it with Marketing. Marketing invents ``body-area networks'' and starts blathering about ``convenience,'' when convenience is not central to the Bluetooth idea. Rather, the important development is to converge everyone on the same PHY or Bluetooth radio so that the quantities will be gigantic enough to ammortize the radio design cost.
so, that is ``wireless.'' It is, quite literally, eliminating wires.
``Network'' means that many devices can communicate, and that features exist so the network can scale to very large numbers of devices and areas of coverage. Bluetooth and 802.11 are both wireless, but they are not networks.
This is where the 802.11 zealots step up and complain that they don't like my definitions of ``wireless'' and ``network.'' You will draw a series of analogies to wired networks that try to redefine these words. Spare me, okay? I crafted the definitions deliberately to make a point, so if you don't like my definitions then rather than arguing with them, accept them temporarily and step up one level so you can see the argument I am making with them.
I want to point out IrDA, Bluetooth, and 802.11 are wireless. All three are potentially LAN access mediums, but none of them are wireless networks. 802.11, IrDA, and Bluetooth similar to each other, and they are different from wireless networks like Ricochet, CDPD, 3G, GPRS, amateur packet radio, and i-Mode.
IrDA, Bluetooth, and 802.11 can be sledgehammered into the role of a wireless network by kludging protocols borrowed from wired networks on top of them---such as STP bridging or Mobile IP. However, if one were designing a wireless network from scratch, one would not do this, because it works poorly!
The modern cellular standards incorporate ``soft handoff'' by talking to two base stations at once during a handoff. Ricochet's mesh architecture means you are always talking to about 2 - 4 repeaters.
802.11 associates with one access point at a time, so it goes dead during a handoff. Kludging Ethernet bridging or Mobile IP on top of 802.11 will not solve the fundamental deficiency of 802.11's roaming capability. Indeed, nothing can solve this deficiency without absorbing routing into the wireless network standard. Ricochet's routing algorithms are so tightly integrated with the radio that terminals will send a packet through Repeater B if Repeater A is in-range but already talking to another terminal. Ricochet uses something more complicated than Ethernet's CSMA/CD or 802.11's CSMA/CA to multiplex its radios, and the Ricochet routing algorithm is integrated with this broadcast-sharing algorithm.
802.11 will never work from a car. Working in cars is supposedly a big challenge of omnidirectional systems---PHS and first-generation Ricochet couldn't do it, but current second-generation Ricochet claims it will work at 70mph. Ricochet does a variant of soft-handoff by always associating multiple repeaters. PHS takes 0.5 - 2 seconds to hand off between base stations. 802.11b's kludgy upper-layer wired-network protocols take tens of seconds to accomplish a handoff---it doesn't even work well when you're walking, much less driving.
Such statements will no doubt provoke outrage from the huge audience of 802.11 zealots: ``802.11 changed my life and empowered me, and the next version will be even better! I won't have this wonderful technology with unlimited potential diminished by your feeble linguistic argument.'' Yeah, whatever. Go suck an egg. Come back and thank me when someone tries to pass off Mea as ``4G,'' and people start debating the relative merits of Bluetooth and 802.11 LAN access points.
``Access points.'' Heh.
Supposedly, devices on the ISM bands have to share the band with spread-spectrum. When properly implemented, the point of CDMA when used as spread-spectrum is to allow incompatible consumer devices to share the band without understanding each other's modulation. Foreign devices degrade the channel's quality gracefully, just like natural background noise would.
CDMA achieves this in a particularly elegant way. For example, if two CDMA devices on the same frequency band are interfering with each other, one reasonable approach to solving the problem would be to double the power output of both devices. Even FHSS does not work that way!
It sounds to me as though 802.11b isn't really CDMA at all. It's not even spread-spectrum. It's just an old-fashioned fixed-frequency service that separates itself from other services by running on a different frequency. Surely it borrows some modulation schemes from CDMA research, but it doesn't look, walk, or quack like CDMA.
In the lariat case, another 900 MHz device obliterates an (older, pre-802.11b) 900 MHz DSSS 802.11's ability to transact data. A proper CDMA device should merely suffer a reduction in range-per-watt. And what is that nonsense where you have to assign your 802.11b access point to a ``channel,'' meaning a frequency sub-band within 2.4GHz, and if another nearby access point is using the same frequency they will interfere? That's not CDMA! A proper CDMA device should share the band with just about anything, including, when applicable, itself. 802.11b devices can't even use CDMA to share the band among themselves---they can only share among themselves using CSMA/CA. This is the method employed by ancient fixed-frequency links like Alohanet, not spread spectrum which inherently permits multiple access. Perhaps this is why 802.11b transmitters are always hilariously underpowered. Perhaps the low transmit powers of 1/10th - 1/30th the legal limit helped 802.11b proponents squeeze this fraud past the FCC.
Why would they do such a thing? Because PeeCee morons only care about SPEED. Sure, 802.11's designers could have worried about sharing the band with other devices, or about congested areas with more than three administratively independent 802.11 nets, but why? Since 802.11 proponents are selling a wireless toy to eager PeeCee users rather than an wireless network service which must scale, these interference and scalability problems don't concern them. Instead, they must offer the absolute fastest point-to-point speed achievable to a single user, no matter what the cost.
Now, lariat's example does not match with current 802.11b folk-wisdom, and for good reason. The old pre-802.11b NCR/Lucent Wavelans they were using (1) operate in a different band and (2) do MAC-layer acknowledgement to work around TCP's assumption that packet loss indicates router output queue congestion and not radio interference. And (3) there is little or no Ricochet traffic in many markets, where Aerie has kept the network dark or is marketing only to government insiders like police, fire, and that mathematician-dude in Independence Day. But some of the other architectural weaknesses that lariat points out remain relevant. 802.11b's band sharing is better-suited to a band reserved for 802.11b only, not protocol-heterogeneous ISM. Actually, it's more suited to a single-carrier network that can coordinate access point antenna patterns and channel assignments, than an open band where anyone can set up an AP. Well, with only three channels it's not even good for single-carrier two-dimensional deployments---it's architected for deployment in a single-carrier 1-dimensional underground subway tunnel where one can alternate between two channels along the tunnel's length.
Ricochet takes an approach which matches much better with the ISM rules about ``sharing'' the band. The speed available to a pair of Ricochet users peaks at 400kbit/s, but the total speed available to many Ricochet users in a congested area is higher. This matches the ISM rules: one device can transmit at most one watt, but ten devices can transmit ten watts. If you don't like this, then I'm guessing you're one guy with one device running one link.
I'm not the only one saying this stuff. NTT East is competing with 802.11b carriers for hotspot and fixed-wireless last mile service:
Being able to handle the massive user demand on an outdoor hot spot network is one reason NTT East uses AWA technology, a European standard with different functionality than 802.11. According to Kobayashi, IEEE 802.11 is not suitable for public use because there's a mass collision of data and interference problems.
How could 802.11b have dealt with interference better than providing three non-overlapping channels and synchronizing terminals through ``access points''? Here are a few example ideas.
First-generation Ricochet had some way of aligning FDMA timeslots across the mesh by propogating a shared clock. While 802.11b is not FDMA, it still might use a self-forming mesh to deal with interference since the problems of mesh routing and interference are conceptually related: routing---using a neighbor to reach a third node that is too far away to signal directly; interference---not obstructing your neighbor's ability to receive transmissions from a third node that is too far away for you to hear. Radiant uses such a scheme to assign the channels and timeslots of its nodes so that a node's multiple neighbors will always transmit to a node one-at-a-time, but Radiant has fixed nodes, directional antennas, and a Central Planning machine that knows the entire network's topology. Radiant's interference challenge is not exactly the same as that of 802.11, a slowly-mobile network where administratively-distinct neighbors share a single band.
Anyway, a band-sharing strategy doesn't need to be that difficult to beat what 802.11b offers. 802.11b could also simply use more CDMA receiver gain instead of ``channels.'' Easiest of all, it could use the minimum necessary transmission power, like every other wireless network since AMPS. These simple strategies are particularly appropriate since they help share the band with non-802.11b devices, rather than just preventing the network from collapsing under excessive collisions if the ``access point'' density gets too high.
Of course, such telephone service wouldn't work while you're walking, and if anyone else in the area is using a 2.4GHz device your telephone would go nuts. But that's not the only problem.
I actually make a similar claim about packet-based phone service over Ricochet, but I immediately admit to the serious potentially show-stopping challenge that could prevent constructing Ricochet phones.
In short, VoIP has serious problems on wireless packet networks because the way the wireless spectrum is shared. 802.11 shares spectrum with CSMA/CA, and Ricochet uses a more complicated but still, as telco-fascists say, ``non-deterministic'' algorithm. This makes QoS hard to implement. The VoIP I have seen so far working in offices is based on switched VLANs. Technically it runs over Ethernet, but this is only Ethernet to a literalist. It is modern ``full-duplex'' Ethernet, which is not CSMA/CD like traditional Ethernet, and that helps the VoIP phones work. (That was not a typo. Full-duplex Ethernet is not CSMA/CD.)
Existing VoIP phones running over switched Ethernet experience a switched network end-to-end. It is still ``non-deterministic,'' but they get a big help because the switches can enforce that VoIP packets always get sent before all other packets. Sharing the network with computers actually helps the VoIP phones because the computers justify a network bulging with overcapacity. The VoIP phones always get to go first, so they experience this overdesigned network as if they were the only users.
If the network were appropriately-sized and composed of exclusively VoIP phones, it would collapse under the primitive priority-based QoS that current Ethernet switches use. If the shared network were real CSMA/CD Ethernet without switches to enforce priority, it would also collapse. However, a CSMA/CD Ethernet network, overdesigned similarly to the shared switched network, yet reserved for VoIP traffic only, would work fine.
Both problems---predominantly VoIP traffic mixture without overcapacity, and QoS on a broadcast network---are solvable. Better QoS algorithms like HFSC can tolerate a greater proportion of VoIP users even though the networks are still ``fundamentally non-deterministic''. Wireless QoS is an emerging area of study, which I think will probably involve replacing Ethernet's CSMA/CD or 802.11's CSMA/CA with a wireless-specific access method that incorporates QoS. There are obvious ways to do this, and probably some less-obvious ways, too.
It's possible that 802.11 will eventually solve the broadcast QoS challenge. In that case, it will still face roaming problems and interference problems. However, I think Ricochet is in a much better position to implement wireless QoS, because, as I mentioned earlier, Ricochet's broadcast-link-sharing strategy is already integrated with its packet routing. While 802.11's rather simple CSMA/CA could be replaced, this is such a fundamental change that I wonder if the resulting Frankenstein bears enough resemblance to 802.11 to deserve the same name. Set your hopes on enough hypothetical future developments, and eventually the prediction ``802.11 will challenge 3G'' reduces to ``something wireless will challenge 3G.''
An omnidirectional short-range telephone service already exists called PHS. It supports voice and data, and uses short-range hotspot bases like 802.11. Like 802.11, you can set up your own PHS base station at home and use the PHS handset like a cordless phone. It is very widely deployed in Japan. The base stations are inexpensive and short-range.
The PHS backers are having lots of problems making money off it because each base station needs an ISDN connection to NTT's wired PSTN, and the rates for so many ISDN loops dominates the service's cost. The problem is described here.
``The biggest problem in making low cost PHS services a reality was with reference to charges for interconnection between PHS service providers and NTT's PSTN.''
802.11 shares this problem of too many wired connections. The paper claiming 802.11 will challenge 3G considers only the cost of equipment, not the cost of wired links to that equipment which make up most of the infrastructure cost. Not only are 3G towers less numerous than PHS or 802.11 ``access points,'' but the large budget for a 3G tower means that it becomes reasonable to invest in connectivity to support the tower, rather than leasing connectivity from someone else at usurous prices.
Ricochet, of course, partially overcomes this limitation with ``repeaters,'' but unfortunately repeaters make QoS harder to implement.
It seems reasonable to predict that these ``deterministic'' standards will fall to their overdesign, their lack of detachment as they treat soulless objects with increasing hatred and arrogant condescention the further these objects get from the network's ``center,'' and of course the obscene cost and mediocre performance of the resulting services. Cellular will fall, just like X.25 and Telenet fell to the Internet and telnet, ATM with its idiotic 53-byte packets will fall to dark fiber and CBQ, and terminal multiplexers and serial leased lines fell to terminal servers and TCP/IP. It does seem reasonable. so, go ahead. Predict.
but 802.11 is not a contender.