Pierre De Vries, MSFT: Imagine that you've spent a fortune on WiFi gear, and you and your neighbors are doing wonderful things with it, like allowing distributed encrypted backup and other community applications. Then, one day, your neighbor buys a perfectly legal, Part-15 certified analog TV retransmitter, and the whole network goes down.
Or imagine a wireless dormnet with wonderful streaming material -- music, video, etc -- that you normally love but tonight, it's so full up that you can't get online to do basic Internet connectivity.
(Ed: I forgot my power supply, Aaron forgot to get his WiFi card registered for the Stanford network's whitelist. We're sitting side by side, sharing an AC adapter, our machines tethered with an Ethernet cable, and I'm receiving WiFi from Stanford, republishing it on the Ethernet, and Aaron's republishing it on his wireless card so that everyone else who forgot to get authenticated can get online. SSID: "aaronsw")
Is there any spectrum etiquette that we can agree upon? Can we ask regulators to enshrine it?
A year ago, I dreamed of a future where we could get online no matter where we are, handing off from WiFi to GPRS. Today, I want a world where every device has wireless built in -- a future where people expect to have data over unlicensed bands everywhere. This will require regulation to make wireless bands robust.
(Ed: Hmmmm, De Vries is a Trusted Computing shill -- what's his agenda?)
Victor Bahl, MSFT: (They're showing slides that half the room can't see -- hard to follow now)
How do you propose an etiquette that accomodates WiFi and other users of unlicensed spectrum? It might be best to exclude non-data services from the band. Also, let's not regulate receivers, only transmitters. The etiquette will not speak to bits, bytes and frames: rather, kind, frequency and power.
We can't come up with a great mechanism for transmit power-control without regulating receivers -- you can't ensure that you don't shout when you can whisper, though this would be useful.
We propose dynamic frequency selection: to avoid interference with radar -- you sense traffic in a channel and back off if it's there. We want listen-before-talk for devices entering a new space. It's well understood and it actually works.
This is simple and minimal. The rules we propose are straightforward and have been in the industry for a long time, and they are easy to implement. There's existence-proof, some of this has been mandated in Europe.
Because you can't transmit when you see unknown traffic in the band, it creates an incentive to understand other users of the band.
De Vries: This will make it uneconomic to build devices as cheaply as possible that blast across the band -- in these new bands. Legacy devices can talk in the old band.
FCC CTO: So it's not a commons -- only people who adhere to your etiquette may use it.
De Vries: Every Commons has rules. We're proposing additional, minimal rules that will allow more robust operation and acknowledges that devices can be built to be intelligent and respond to the environment.
FCC CTO: So this new band will be for WiFi and devices than can peacefully coexist with WiFi.
Lessig: But if I have a device that's completely stupid built in 1930 that would be interfered with by the latest data-network, under your definition data devices would have to back off.
Bahl: No, that device would be excluded from this band.
Bahl: This solves the hidden-receiver problem. You need a receiver to tell you if he's there. (Ed: I'm not really following this -- there's a lot of interjection from the panel and Bahl's presentation style is pretty meandering).
Reed: You're seeing that recievers don't need transmitters, but transmitters must have receivers.
Dewayne Hendricks: Is this for LANs or WANs or both?
Bahl: LAN and multihop. Short-range.
Tom Freeburg, Motorola: Canopy and 802.11a coexist very well even though they disagree about etiquette. Interference is a function of peak power, not average. Interference takes place in the receivers. Poor receivers suffer from interference much worse than good receivers.
Directional antennas are going to be more and more widely used, which magnifies the hidden terminal system.
Radios are not wires. Just because something works very well on a wire, doesn't mean it will work in the air.
You can get more radio spectrum -- that's what cellular systems do. They get more spectrum out of the same MHz. You can build a radio system where you get more throughput by getting more cells. It's not easy, but Canopy is a living, breathing example of this.
We can't make a perfect etiquette because we can't see the future perfectly. Our rules have to be subject to evolution and need to be very lightweight. We should not be writing rules to apply to situations that haven't occurred, because 99% of them won't and the one that turns out to be a real problem will be the one we've never thought of.
The biggest problem is insufficient resource -- if there's more demand than supply, then you have a problem. You need to allocate enough spectrum.
If you avoid the Tragedy of the Commons problem, you'll need a lot less spectrum, though.
You have to expect that your users will be robust -- you can't protect the weak at the expense of the average. You need to protect everyone from bullies, but not defend the most fragile case.
I propose a simple set of rules that will allow a common set of users to share a resource. These rules need to be simple and allow for innovation. It's not enough to have your etiquette accommodate two kinds of sensing mechanisms, because there are six more we haven't thought of.
We need to make sure that no user takes more from the resource-pool than he actually uses (Ed: "from each according to his ability...")
De Vries: How do you test whether your rules accommodate any use in the future?
Freeburg: You don't -- you change the rules.
Lessig: If your rules would impose a heavy burden on person x or a light burden on person y, then your rules should ensure that y solves the problem.
If the burden on me is extremely simple to solve, then it should be my responsibility to solve it. The ambiguity about who is responsible for solving the interference is the most important piece of the problem.
(Ed: All of the FCC people at this event are wearing RIM Blackberries -- hill-rat multitools. Most of them have good suits and bizdev-guy blue sailcloth shirts. A sizable number are wearing snappy suspenders.)
Freeburg: I'm afraid that the etiquette will become so numerous and detailed that the good sense behind them disappears -- we could end up with our throughput cut by 80% and not know how to fix it.
Complying with the system may not get the best result -- we'll lose the ability to go outside of it. We can't kill good neighborliness with rules.
De Vries: MSFT's proposal is as minimal as we can make it (Ed: Minimal is not the same as mutable).
Freeburg: I hate listen-before-talk even though I helped design it. Once we had it implemented and understood, it turned out hat when you really implement a listen-before-talk system, you throw away 85% of the real channel capability, through deferring to exchanging pairs whom you would not have bothered. It's like being at a cocktail party where you're not allowed to talk so long as anyone in the room is talking. Let's make everyone robust so they can withstand a little interference.
* Whether or not a communication is successful depends on signal:noise at the receiver. Depends on the antenna and the receiver. You can still communicate at 1000:1 SNR -- when the signal is buried in the noise. It sets aside the notion that we have to decide who can speak and when. This reduces the number of bits/second, but you can always go wider (if you don't have a regulatory constraint).
* Listen before talks rarely works. Reed is right. low-powered conversations between high-powered conversations don't harmfully interfere.
* 100,000 people at a football stadium can communicate with their neighbors in just a few KHz -- the din of the crowd never gets that bad. You can still even have an effective Public Address system. EM is even less susceptible to interference that acoustics -- we can upgrade our antenna, we can switch polarity, etc.
* Etiquette should allow for simultaneous communication (i.e., like a football stadium).
* The MSFT proposal puts the burden in the wrong place -- the receiver is where the interference takes place, so that's where you should regulate.
* Motorola paper is better: it's etiquette without listen-before-transmit. But this doesn't need to be law, Darwinian winnowing away of devices that don't work will solve this.
Reed: You can detect interference on a round-trip -- when the receiver acknowledges receipt, you can tell if there's degradation. This is more of a burden than the MSFT proposal (because it requires a closed loop), but it's still preferable to listen-before-talk.
(Ed: Just noticed that Aaron has a folder in his mailer called "Need to read" that shows 15,000+ unread messages)
Bahl: Listen-before-talk gets a bum rap. You can improve the backoff that dramatically improves it. It does work. Discuss