A Clever Second Use for A Radar

In May of 2015, Google introduced a miniature radar chip called Soli to it's developer's conference.  Soli is a specially designed system that's intended to allow gesture controlled interfaces to the computer.  There are a few of these ideas being pushed to enable this technology.  We talked about one in this space back in July, a wearable armband called Myo that embodied accelerometers to sense the user's movement.  Other approaches appear to be based on cameras.

Soli departs from those approaches and goes to radar.  Conventional radars don't have the time/space resolution to detect subtle movements, but Soli does. 

Imagining gesture interfaces on everyday objects is particularly intriguing: ATAP used the example of an analog radio where gestures control the volume and station. But it could be applied to any number of use cases. Soli's sensors can detect motion at a range of about two to three feet, Schwesig says, so any device you use within that range stands to benefit. Imagine dismissing smartphone notification with the wave of a hand or pressing your fingers together to play music from a bluetooth speaker.

Soli isn't real hardware yet, but Infineon, the semiconductor manufacturer that partnered with Google to build the chip, has said that it expects samples to be available in the first half of 2017 and production devices to go on sale in the second half.

This, as they say, is just about the end of the beginning.  Prototype hardware has been available for a while, and goup of experimenters at St. Andrews University in Scotland have almost stumbled into some interesting and unanticipated uses for Soli. They've found ways to use it to instantly recognize objects like metals or peoples' body parts and distinguish between them, according to a paper presented last month at the Symposium on User Interface Software and Technology.

The system, also known as Radar Categorization for Input & Interaction or RadarCat, trains itself with machine learning algorithms to read those signatures and assign them to an object. It has been shown to instantly identify things like sponges and smartphones, differentiate between copper and steel, and tell if a glass of water is empty.

What's going on here?  I have no details, but just based on what I know about radars, here's a couple of guesses.  There are two ways to get finer position resolution with a radar: use of higher frequencies for their shorter wavelengths and use of shorter pulses.   Higher frequencies have only gotten easier every year since just about forever.  There's a WiFi protocol called WiGig or 802.11ad that uses frequencies around 60 GHz for ultrabroadband WiFi.  Conventional WiFi is at either 2.4 or 5 GHz.  You might get bit rates around 100 Megabits/second with your WiFi network.  WiGig offers a couple of channels available at 6-8 Gigabits/second. 

Back in 2014, my mind was blown by finding that over a million WiGig modems had shipped (mostly in Dell laptops) in their first year of production.  Today, projections are for a billion WiGig chipsets to ship in 2021, not by 2021, just five years from now.  I had worked on a 60 GHz modem at Major Southeastern Defense Contractor before I left in '96; probably in the '94 time frame, and it was exotic, excruciatingly expensive technology (although kind of fun to work with!).

WiGig chipsets, if flexible in how they work, could fit into the range where the wavelength is small enough to see individual fingertips with adequate resolution.  A quarter wavelength at 60.0 GHz is .047", so it's very easy to get many wavelengths over a finger's width.  The other way of increasing resolution is by using shorter pulses, or pulse compression in the radar.  Shorter pulses are probably accommodated by WiGig frequency range hardware, too.  They offer just short of 10 GBPS data, so let's say a pulse is just one-ten billionth of a second long. That's 100 picoseconds and 6 full cycles of the 60GHz RF.  In 100 picseconds, the radar pulse travels 3 mm, 0.118". 

This is not to say they're using the WiGig band, I don't even know if they'd be allowed, but the point is that the hardware isn't that excruciatingly expensive and exotic technology any more.

As they say in the video, the reflection from various objects is different, and they let the system self-program to learn things (machine learning).  The more objects it's exposed to, the more it will be able to identify.   A really interesting side application of something never intended for that use.