RF Location
Because of our unique background in RF propagation modeling and radiometry tracking systems, positioning via RF ranging is a key core compentency at Stephens Labs.
Many different systems types have been engineered. One example below uses multiple receive points and frequency diversity to range and angulate a person with a transponding device.
Synthetic aperture radar (SAR)
While traditionally used almost exclusively in airborne applications, we have developed a system which uses ultrawideband radar on a ground vehicle, for mapping and recognition of either targets of opportunity or intentionally placed devices with unique return signatures.
The radar system was also designed by us from the ground up, using ns-level micro impulses on a carrier, and quadrature response captured and reduced via FPGA and DSP processing.
The first use of such a system is for pure mapping. Below is an example terrain with objects and structures...
... and here is the SAR map of the area, with return normalized to object radar cross-section. While this is just the circular-polarized co-polarization response, the system has been engineered to allow capture of both polarizations (either 2 linear or 2 circular)
Using the co-circular-polarized return is particularly interesting when trying to identify intentional reflectors such as the "horizontal corner cube" below
For normal radar return off single planar reflectors, the circular polarization has an almost complete handedness inversion. With a 2-bounce corner cube object, or any place where 2 walls meet at a right angle, the polarization is preserved. This allows the ability to isolate desired landmark features from a more general clutter.
As an example of SAR operation (forward motion allowing angular discrimination), the following simulation shows the detectability of corner cubes and their range/angle location accuracy. The first simulation below has no ground clutter and only receiver noise. It starts at rest, and slowly increases in speed, showing the dependence of angle determination on speed. This is an animated GIF, so click on the image to render the video
Now we repeat the simulation with ground clutter. Note that at rest, not only is angle unknown but ground clutter completely prevents detection of the passive reflectors. But at about 1 ft/s (0.3m/s), the clutter return has been distributed between angular bins, allowing sufficient SNR per angular bin to detect the passive reflectors.
Side use of radar: perception and safeguarding
While this radar was developed for positioning, it also serves its more traditional role of object detection and doppler measurement. For instance, the example below shows a vehicle approaching a robotic nav system and the corresponding radar signature.
While the SAR system (by design) is omnidirectional in response, the range, range-rate, and size of object is observable, thus serving as a "time to collision" system.
Transponder variation on SAR system
While polarization and/or sufficient motion allows discrimination between clutter and intentional landmarks, in certain environments the clutter is so pronounced that active landmarks are needed. Since the SAR system was designed with a high degree of flexibility (pulse train programmable, and decisions made in a DSP at 1KHz), it is compatible with active radar transponders which not only amplify the signal but modulate with a audio sequence (typically few hundred Hz). Thus, when the radar is at rest, all static ground clutter and passive landmarks return at 0Hz, but the active beacons show up at their modulation frequencies in the return stream.
We fabricated several different types of these beacons, customized for different environments and operating ranges. Here are a few prototype builds, in weatherized containers
This system was deployed for robotic mowing in a baseball stadium... for more information, see here.
Transceiver systems
We have also designed and prototyped narrow-band ranging systems, based on dedicated CDMA transceivers. Each transceiver broadcasts its sequence in a time slot, and listens for N-1 time slots to the other transceivers. The observed time of reception is a pseudorange, since each transceiver is operating on its own separate clock. But each transceiver then broadcasts these N-1 observed pseudoranges for all other transceivers to use. In this way, any transceiver has access to all observables, and pairs of pseudoranges can be added to cancel clock effects, which yields all pairwise ranges between transceivers in the system.
While designed to work within the ISM band at 5.8GHz and adjacent spectral masks, sufficient power exists at +/-1GHz for sharp CDMA chip edges and hence nearly mm accuracy ranging in free space, and to 100s of meters.
The dominant error source for this system is the error from multipath. While effort has been made to reject chip transitions with greater delay than about 1ns, ground bounces can easily fall in this range. The following simulation shows the effects of ground/wall multipath and object diffraction in a volume, showing positioning errors growing to almost 1cm at times.
RFID multi reception
Using the hardware developed at Mojix Inc, we created an RFID location system using its unique characteristics of separated tag excitation from multple receive points, but whose transmit and receive are still phase coherent.
While the hardware was designed with the traditional approach of a linear beamformer for tag angulation, an estimation theory approach allows consideration of receive elements in any configuration. This allows the distribution of the receivers in an operating volume, yielding greater path diversity and more precise location based on phase, and whose receiver separation is limited only by the ambiguity resolution of phases across the EPC frequency set.
The design also used signal strength as a measurand, so that sparse measurements still gave a coarse estimate of location before phase-based sequential ranging could be employed. This was implemented in a unique particle filter design with a seamless transition from RSSI to phase-based estimation without separate algorithms or modes being employed. This approach also allows any number of tag exciters or receive points to be used scalably, and even dynamically.
We have also developed the realtime software for tracking thousands of tags simultaneously. It was written in java but used high performance Intel IPP libraries, saved its results to a MySQL database and generated statistical reports for asset tracking.
An advanced feature of the estimation process was to allow external measurements and use case constraints to improve tag location. For instance, in a warehouse environment, door sensors were used to distinguish tags being loaded/unloaded on to a trailer vs those statically staged nearby. Also, prior knowledge of tag movement types was known, that allowed association of tag position regions and its likely velocity. For instance, certain regions of the warehouse were roadways, where tags were mostly likely not at rest and moving in one direction, as were loading ramps into the trucks. When used in a particle filter, a tag that is hypothesized to be in a roadway or loading lane is given a velocity from a prior distrbution, so such hypotheses should be observed to rapid move. These ideas of course can be generalized to any use case with constrained prior behaviors.
Radiator RSSI localization: WiFi and cell tower
[coming soon]