Nicholas Weaver is a researcher at the International Computer Science Institute in California.
Radars -- especially a high frequency ones -- are very precise. But they can't see around walls and buildings. Which means if your radar is in a plane looking down into a city, you actually can't see all that much -- buildings hide most of the good stuff.
A radar beam doesn't just reflect off objects, though. It can also scatter, bouncing around the landscape before returning to the radar (or a separate sensor). Normally, this "multipathing," as its called, is a bad thing, confusing the radar. But DARPA is hoping to take advantage of the scattering. Even if a radar can't see between two buildings, it could bounce a beam off of one building into the gap between them, looking for cars on the hidden street below. It's like an electromagnetic bank shot.
For example, if such a radar was available today, 4 or 5 aircraft or drones over Baghdad could provide an "Automobile AWACS": allowing the US military to know where all cars in the city are and track their movements, which would be a huge boon for counterinsurgency and intelligence work (tracing a car-bomb as it travels through the city, tracing back from an attack, or even attempting to create a 'graph of association' of who meets with who) as well as providing significant early warning for troops on the ground. A conference to kick-start this "Multipath Exploitation Radar" project is being held this Wednesday, in Virginia.
Actually building such a radar, on the other hand, is a "DARPA-hard" problem in the classic sense -- a project tough enough for the Pentagon's way-out research arm to handle. It will require breakthroughs and integration in mapping, LADAR, RADAR, computer graphics algorithms, digital signal processing, and computation. And the DARPA proposal gives a lot of insight into how to do it and the interesting problems involved.
Begin by creating a highly detailed three dimensional model of the city and all its buildings by flying over it with a LADAR/RADAR combination. Not only does this map need to be very precise (accomplished by the LADAR), but it also needs estimates on how individual buildings (and even surfaces on buildings) will scatter the radar signal.
This map is now used by a flying radar, operating at 10,000 to 15,000 feet, in order to track cars on the streets below. If the radar has an unoccluded view of the street, it just operates normally: reflecting signals off the cars in the street to track where they are and how fast they are moving. But if its view of the street would be blocked by a building it uses multipathing.
When looking at a street in multipath mode, the radar needs to calculate where the beam should be reflected and scattered off the building behind the street. This problem is effectively raytracing, a graphics technique used to create some highly photo-realistic images by predicting how light rays will move through space and bounce off objects in the scene. A truly massive amount of computing, as well as possibly new algorithms in what are effectively computer graphics and digital signal processing, will be necessary for the radar to operate.
The project itself begins with an existing high-frequency radar which will be modified to include a huge amount of data collection hardware to record the raw information generated by the radar. Then tests will be conducted in an urban environment by first making a detailed LADAR model and then using the radar for several minutes to focus on a 1/2 km by 1/2 km grid within the city, with all the data from the highly instrumented radar recorded to files. These gigabyte data traces would be used in a subsequent second phase to develop the algorithms and computing systems necessary to interpret the radar results.
It won't solve the multipath radar problem, even if it works out perfectly. But it will but create the dataset necessary to test possible algorithms in a follow-on project.
But if you are waiting for such a dataset, you better be involved in some related defense research. I'm almost certain the dataset will at least be tagged For Official Use Only, at least. If the radar used was classified, the data will probably be at least at that level of classification. That's because the raw data would not only be a huge boon to researchers hoping to enhance our radars, but to anyone interested in evading our radars as well.