The Less Obvious Hard Parts about Designing Systems with Radios

The Communications Technology Lab at Intel has a charter that matches its name – we are supposed to look at all the issues around communications for Intel’s future platforms. That’s a pretty broad charter since it means we worry about everything from how security works in Internet protocols, to how to process the packets flying off a fiber in 10Gbps and higher Ethernets into our servers, to future possibilities for integrating photonics and silicon, to how to build radios.

We tend to talk a lot about some of the more glamorous items like Silicon Lasers and CMOS radios but today I thought it worthwhile to focus on some of the possibly less glamorous but just as important issues around how you actually can build small platforms that integrate lots of radios and have them really work. It turns out that this is harder than you might think unless you are a radio person (who can probably stop reading now). Adding communications to a PC used to be as simple as choosing an Ethernet chip to add to the motherboard and writing a driver. Not that this was that easy but at least the problem was contained to the issues around making the Ethernet chip work and talk to the operating system. Radio is a bit more magical and is a much more holistic systems problem – one that is substantially different from what PC OEMs traditionally dealt with.

First of all you have the problem space – an almost boundless and growing set of radio standards. The well dressed platform of the next few years probably comes equipped with a large subset of WiFi, WiMax, Bluetooth, GPS, TV (of a number of sorts), UWB, and various legacy cellular radios. As if the number of raw standards weren’t enough, these radios will be utilizing radio frequencies from a few hundred megahertz (TV) up through nearly 6 gigahertz (WiFi) or even 10GHz (UWB). And then there is the alluring 60GHz band that is the next likely frontier for short range communication. The frequencies make a difference because antennas are generally designed to be specific to bands and a lot of bands could mean a lot of antennas. What is worse is that more and more of these radio systems are MIMO based which means that they need more than one antenna per band – at least 2 and maybe 4. Of course at the same time that we are growing the need for antennas we are also finding users looking for smaller, thinner, lighter devices – all this meaning that we are having to cram more stuff into ever smaller spaces. Finally, there is the complication that antennas don’t always like to live too close to their neighbors. They can interfere with one another and, in the case of MIMO systems, they need to be spread apart enough to get distinct samples of the signals. There is yet one more problem with squeezing all these radios into a small space – interference. One radio can interfere with another radio. Also, all the other circuits in a high speed modern PC tend to have RF noise as a side effect and can interfere with the radios. So how do we address all these issues?

Let’s start with the antenna problem. One could design very broadband antennas so that a single antenna could deal with a very large band of frequencies but this isn’t that effective. First of all designing such antennas with good response across all the bands is pretty hard. Worse, the broader the antenna bandwidth the more noise one captures along with the signal. So the performance of the radio using the antenna tends to be poorer. The other obvious alternative is to design an antenna for each band but this means lots of antennas to squeeze in, especially when the MIMO systems will need a couple of these per band. We have been focusing on the third alternative which is to build antennas that can be reconfigured to tune for a set of different bands. If we can do this we get the noise immunity that comes from a specific band antenna while being able to use one antenna for a number of different bands. Designing such antennas is difficult in itself since antenna design still has a lot of “art” in it in addition to the science. It also requires some very good switches to connect parts of the antenna to change its shape and hence sweet spot. We demonstrated an example of this approach though at our recent IDF in Beijing.

Another aspect of the antenna problem is isolation – particularly when two radios have to operate at the same time where one is transmitting and the other receiving. We like to talk about radios as though they operate in discreet channels – e.g., we talk about WiFi in channel 6 in the 2.4GHz band. However, in reality radio transmissions put out energy across all frequencies, usually more in nearby frequencies and less further away, and our designs simply focus the vast majority of that energy into the target band. Since rather little energy gets emitted outside the “channel” we don’t normally think much about it – but in a small device the transmitting antenna is very close to other receiving antennas so even the small amount of “out of band” emission can be a big problem. As an example of the difficulty consider that WiFi transmits in the 2.4GHz band while WiMax in the U.S. can operate in the 2.5GHz region. Even a very well behaved WiFi radio can cause fits for a WiMax receiver in that band on the same platform. While there are other mitigation approaches to this problem, creating antenna designs that maximize the isolation of one antenna from another can help a lot. We’ve also demonstrated this sort of design in the recent past. By the way – lest you think that you’d never need to have both such a WiFi and WiMax radio operating at the same time consider a carrier who wanted to use WiMax for wide area coverage but augment that with WiFi in hotspots to better utilize their precious licensed spectrum. In roaming from one to the other the client system would need to make a new connection before the old connection broke which would require simultaneous operation.

There are other ways to deal with radio to radio interference. One apparently simple approach would be to simply not transmit on one band while you were receiving on another. Unfortunately this is easier said than done because the MAC protocols of the various radio systems were not designed in general for such coordination. Also, the radios for the various systems may be entirely different chips with no knowledge of one another so trying to coordinate them can be difficult. Nevertheless, our lab has been working on mechanisms to work within the defined standards to use such approaches to mitigate interference when possible and it is a promising approach for many cases.

The last part of this design problem I wanted to touch on is the problem of interference from the rest of the system. There are lots of components in a PC that are switching on and off at very high frequencies – think 2+ GHz processors and the like. All these parts generate RF noise that while low in absolute terms can be really loud from the perspective of a radio that is trying to listen to a very faint signal at about the same frequency. This can cause the radio to be much less sensitive than it would otherwise be – rather like the problem of listening to someone speak while standing next to some loud piece of machinery. We have measured the real “noise” in various bands in real systems and found that is can be very non-uniform and spread out with the result that there is often no “quiet” spot to mount a radio component. In one system we measured for example, the LCD put a lot of noise right on top of the upper channels in the 2.4GHz WiFi band with the result that the system listening range on those channels would have been cut by about half. These kinds of effects can explain why two systems that use identical radio components may appear to behave very differently in how well they receive an access point’s signal. Solutions to these problems can involve shielding the aggressor components or shielding the victim radios with metal or other shields. Of course such shielding can add undesired bulk to small designs. It can also involve tweaking operating frequencies of other components – for example in the LCD case I mentioned even a small adjustment in the screen operating frequency was enough to move the offending noise to a place where it wouldn’t bother the radios.

Of course, all these issues (and more) are a lot more complex and difficult that I have been able to indicate here – luckily for me I have a lab full of brilliant engineers who specialize in this stuff. However, I hope that some of you now have at least a somewhat better appreciation for the difficulty involved in putting together the pieces that give you that magical internet access without wires!