Home > Research > Spectrum Sensing & Geolocation

Spectrum Sensing & Geolocation

The goal of this effort is to design the hardware and algorithms necessary for an ultra-low-cost ($1), low-power (100's milliwatts) distributed sensor network for electromagnetic emitter geolocation and pervasive spectrum sensing. This work is funded through a collaborative agreement with the Army Research Laboratory. We also have a very strong collaboration within the Notre Dame Wireless Institute developing the RadioHound project.

In collaboration with the Army Research Laboratory we are developing a collection of 18 portable, weather-proof spectrum sensor packages interconnected through MANET (mobile ad-hoc network) radios. These sensors will facilitate rapidly deployed, mobile experiments both indoor and outdoor. Sensors are based on the RadioHound sensor node which is capable of scanning from 25 MHz to 6 GHz. It is based on a Raspberry Pi embedded computer, RTL-SDR receive-only software defined radio, and a custom RF down-converter. The RF front-end design conforms to the Raspberry Pi HAT protocol. This version costs around $70 and consumes around 3 watts at full load. The frequency scan is from 25 MHz to 6 GHz. Future revisions of this sensor node will directly sample in the frequency domain obviating the need for an RTL-SDR sampler with FFT computations for spectral estimation. It is anticipated this sensor will scan at rates between 1-10 GHz/second, cost around $5, and consume about 750 milliwatts.

ARL distributed sensor package - student 3

In the limit of sensor simplicity we are targeting a $1, 100 milliwatt sensor based on a 1-bit RSSI (power) meter. The sensor omits all components except the fundamental frequency conversion operation to save cost and power and instead relies upon increased density of deployment made possible by such "disposable" sensors. The figure below is a simulation of our simple spectrum sensor vision. A network of one-bit RSSI-based sensors combined with contour-map-based processing is able to effectively localize emitters with a modest increase in sensor density over high dynamic range sensor hardware but is able to save orders of magnitude in cost and power. The below figure shows a map of the San Jose valley with six emitters (red 'X'), iso-power contours based on a network of one-bit sensors, and the estimated emitter location based on centroid computations. Because the sensor is ~$1 and ~100 milliwatts, a sensor density can be 3x the density of ideal infinite dynamic range sensors but the total network is much lower cost and much lower power. Each sensor can be considered "disposable" at this level. The sensors in the simulation omit LNAs, use starved-LO single-diode mixers, permit image folding, and use single-bit threshold detectors. Backhaul connection (data exfiltration) is considered a separate system.

One of the challenges and advantages of distributes spectrum sensing is how best to combine the data from each sensor node. In addition to the iso-power contour approach described above we are also exploring cross-correlation methods to increase dynamic range, decrease noise, and simplify computational complexity.