[LINK] RFC: Shannon's Law and RFID

[rchirgwin at ozemail.com.au]

Linkers,

After last week's long RFID thread, I decided to write the article 
pasted below the signature.

Before I publish it, I'm soliciting Linkers' comments on the technical 
accuracy of the piece. I am trying to explain Shannon's Law to a lay 
audience, which will be a trick, and I want to make sure I haven't 
strayed into the sort of technical inaccuracies which I hate in others!

Replies probably offlist so as not to trigger more theology!

Cheers,
Richard Chirgwin

RFID Capacity and Reach

Why Shannon’s Law Matters

A couple of weeks ago, partly because of the misconceptions I observed 
even in technically-savvy people, I decided I should write a series of 
features in which I looked in-depth, and independently, at RFID 
technology. That way, I might walk a fine line between the Philip K 
Dick-like fears of its opponents, and the tech-utopian promotions of its 
advocates.

However, the more I discussed, debated, and analysed peoples’ 
understanding of RFID, the clearer it became that there are other things 
which have to be explained along the way. RFID has a great many 
interactions with very fundamental communications concepts - and one of 
them is Shannon’s Law.

I suspect one reason that RFID is so feared on one side, and so revered 
on the other, is that a great many people believe in a technology 
development curve that will go upwards forever - and, because they’re 
not mathematicians or technologists, they compare the unknown (like 
RFID) to the familiar (like Moore’s Law). Moore’s Law (which predicted 
the way the growth in processing power would stay in step with falling 
prices) hasn’t run out of steam in processors, RFID is just another 
application of processors, therefore RFID’s capabilities will be as 
limitless as the development of the microprocessor, right?

Well, yes - if we’re just talking about the processor itself. I’m quite 
happy to believe that Moore’s Law will work just as well in the silicon 
we build for RFID chips as it works for Intel CPUs. It is, in fact, one 
of the reasons that processors are now small enough to fit on 
millimetre-by-millimetre tags.

But the microprocessor isn’t the only constraint an RFID tag has to 
respect. The tags communicate over radio channels - and in radio 
communications, an older principle applies.

Shannon’s Law

My guess is that people with radio communications training don’t need to 
have the maths explained to them; and people without that background 
don’t want the maths. So as far as possible, I’m going to follow 
Hawking’s principle and avoid setting down formulae for people to memorise.

Fortunately Shannon’s Law is easy to put into words. Stated by one of 
the founders of the discipline known as information theory, Claude 
Shannon, it predicts the useful capacity of any communications channel.

In a perfect world, the “bandwidth” of a communications channel and its 
throughput would be identical.

If you count /everything /occupying the channel, that’s true. The 
problem for the user is that their data traffic isn’t the only thing 
using a communication channel.

There’s also noise. Whatever space the noise takes up in your channel, 
there’s less room for “payload”. The more noise, the more the “carrying 
capacity” of a channel approaches zero.

The equation itself can be found here (a Wikipedia link). 
http://en.wikipedia.org/wiki/Shannon%27s_theorem

So it’s simple: channel capacity falls as signal-to-noise (SNR) ratio 
falls. In a perfect world, where SNR is 1, the channel capacity and the 
bandwidth are identical.

One of the aims, for all radio communications designs, is to find better 
ways to separate the signal from the noise - to preserve as much 
bandwidth as possible for the payload.

Workarounds

In fact, Shannon’s Law doesn’t just work for radio communications - it’s 
applicable for wireless, wired, even fibre-optic communications. The 
system with the least noise will always preserve the greatest amount of 
the “raw” channel capacity for user communications.

If you have a “raw” capacity of 10M bits/s, the wireless medium might 
get a throughput of 1M bits/s. Copper wire, however, experiences less 
noise, so it might preserve 4M bits/s of that channel for the end user. 
Fibre - even starting just with that 10M bits/s channel - gets the best 
throughput (maybe approaching 10M bps), because it suffers the least 
noise. An optical fibre doesn’t couple with external signals - outside 
light doesn’t get in.

And this example completely ignores the difference between “raw” 
capacity - in signal-to-noise terms alone, the hierarchy puts light as 
the best, copper in the middle, and wireless at the bottom.

So what can we do to “work around” Shannon? Only two things: improve the 
“raw” capacity of the channel, or improve its signal-to-noise ratio.

Capacity, Power, Compression

Working at higher frequencies is one way to improve a channel’s raw 
capacity - if you start with a carrier frequency in the Gigahertz range, 
you’re starting well ahead of someone whose carrier frequency is in the 
Megahertz spectrum.

The second is to use a “wider” channel.

This needs a little explanation. Most people are accustomed to thinking 
of radio signals as occupying a single frequency - for example, ABC 702 
in Sydney just uses the 702kHz frequency, right?

Wrong. The “702kHz” designation refers only to its carrier wave 
frequency. The carrier is a pure sinewave, but as soon as you add the 
voice of Angela Catterns or Richard Glover, you end up with a complex 
mix of many frequencies starting at 702kHz - X, and ending at 702kHz + X.

For someone talking on AM radio, “X” is only around 12kHz or so - and 
one of the reasons we have spectrum management is so that the next radio 
transmitter doesn’t send a signal which your radio confuses with 702.

If you are sending (for example) 10MHz of data, you will occupy a 
“wider” channel (the emerging next-generation Wireless Ethernet 
proposals need a 40MHz channel to achieve predicted throughput of 100M 
bps using a 500M bps-plus “raw” speed).

“Raw” speed isn’t the only way to improve throughput: you could also 
work to improve the signal-to-noise ratio, and the easiest way to do 
this is to pump up the power. Apart from human-generated interference, 
the amount of noise appearing on a radio channel is nice and predictable 
- so a 1,000-watt transmitter gives you a lot better signal-to-noise 
ratio than a 1-watt transmitter, and lets you keep more of your “raw” 
capacity.

Of course, if you endlesslly “pump up the volume”, you’re going to 
eventually run afoul of the regulator; not to mention you’ll eventually 
start to notice that your skin burns whenever you’re too near the 
transmitter. Boosting the power is only viable up to a point.

The third option is to reduce your demands - compress the data so that 
the user experiences throughput of 10M bits/s but is only actually 
sending 1M bits/s.

But Aren’t Radios Smarter?

Yes.

But they don’t somehow magically “break” Shannon’s Law. Techniques like 
spread spectrum - and in fact all of the computationally-expensive new 
modulation techniques - give you better throughput by combining the 
above workarounds: they start with a “wider” communication channel (so 
regardless of SNR, the starting point is more “raw” capacity); or they 
use their “smarts” to tell the difference between “signal” and “noise” 
(thus improving the SNR); or they compress the data traversing the radio 
channel, so as to give the user more apparent throughput than the 
channel offers.

They might even adjust the power of the transmitter to compensate for a 
noisy environment.

Smarter radios don’t break Shannon’s Law: they adjust capacity and 
signal-to-It noise ratio to get the best possible throughput.

By the way, this also applies to copper-based communication. The history 
of broadband on copper networks is one of using more intelligent, more 
powerful, and faster devices to adjust the Shannon’s Law parameters of 
the connection.

Let’s apply all this in, for example, WiFi, which uses spread spectrum 
radio technology to get really fast communications.

A full explanation of spread spectrum is another article entirely, but 
if you combine a radio that’s able to transmit on multiple carriers at 
once with a really smart processor at each end, you get a wider radio 
channel than if you just used one carrier frequency. So the “raw” 
capacity improves.

Moreover, because the communication is using multiple frequencies, 
negotiated between the devices at each end, the processor has a better 
chance of distinguishing between signal and noise - so the SNR improves.

Up to a point, WiFi devices can tell whether a device in the same 
“footprint” is part of their network or not - my access point can decide 
that someone else’s access point is “noise” without losing performance.

It all comes unstuck, however, if there are too many overlapping devices 
on different networks - which is why there have already been reports of 
hotspot “wars” in which people just keep ramping up the output power of 
their access point.

There is, of course, one last way to separate signal from noise: use a 
highly directional antenna. That way, nearby transmitters don’t 
contribute their own transmissions to “your” channel as “noise” - and 
you get to keep more of the raw capacity. Once again, this doesn’t 
change Shannon’s Law, it merely changes the value of one of the 
Shannon’s Law variables.

RFID

Back in he world of RFID, what we’re talking about is really small 
devices, with limited transmitter power, limited processing power, with 
regulatory limits on the radio systems.

Active RFID devices - those which have their own battery to provide 
power - are far less constrained than the passive tags, but since it’s 
the passive devices that get all the attention in retail settings, I’m 
going to stick with them. Active devices have one very important 
characteristic for this debate: they’re expensive, so retailers don’t 
like giving active tags away with products. Of course, there are 
intrusive applications of active tags - for example, the possible 
data-matching provided by tollway tags is a handy example - but it’s 
consumer goods that worry consumers.

And it’s also, frankly, the retail application that’s getting hyped to 
the skies. Just about everywhere you turn, there’s a vendor promising 
things like completely automated supermarket checkouts, really cool CRM 
applications based on the tag in the shoe, and so on.

I’ll give this more detail in a future article, but I’d at least like to 
set down a general principle: Shannon’s Law, not Moore’s Law, is the one 
that’s going to set the long-term constraints on just how effective RFID 
can become in retail settings.

Your power budget in the device is extremely limited: you start by 
getting power “over the air” (the devices are powered up by what they 
get from the radio signal sent by the tag reader). They start with 
microwatts or nanowatts; some of that is needed to power-up the 
processor, some to run the radio transmitter, and the rest is wasted by 
inefficiency.

There’s been talk in the local media that the RFID business associations 
would like the Australian Communications Authority to lift the 
transponder power limit from the current 1W to 4W. I understand why - 
but even that’s not going to make a huge difference to capability. At 
best, it’s going to mean systems that are only marginally reliable might 
become nearly-feasible, but it’s not going to give the world a pervasive 
RFID scheme that actually works.

Because the tags have so little power, there’s none to spare for fancy 
radio modulation schemes. Instead of something like spread spectrum, 
which least improves the signal-to-noise ratio, passive tags use an AM 
modulation scheme, and they all use the same frequency.

The tag reader has a moderately directional antenna - but the tag 
itself, which is transmitting on tiny power levels, has to spray its 
communications in all directions. It doesn’t know where the transponder 
is located, and you can’t guarantee that someone is handling the tag to 
make sure it’s “pointed” the right way. So a smarter antenna scheme 
doesn’t help much.

All of these, as I said, fall under Shannon’s Law. There is one place 
where Moore’s Law is going to help the tags: their electronics will get 
smarter and use less power. Perhaps one day the microprocessor on the 
RFID tag will have enough grunt to add compression to its capabilities, 
and increase throughput that way - but that’s about all.

Retailers looking at RFID should be asking the vendors about the real, 
practical constraints on the technology. And consumer advocates 
preaching excessive fear would do well to learn just a little bit of 
maths, get a handle on Shannon’s Law, and focus instead on the back-end 
data applications that already represent dangerous privacy invasions.

If you’re not interested in RFID at all, try thinking about Shannon’s 
Law and wonder whether wireless will truly destroy Telstra.

The second article in this series will look in more detail at passive 
RFID tag technology.