Bo Tan, University College London
Proving the concept that local WiFi signals can be used to monitor moving objects and bodies that are otherwise visually obscured. Although fundamentally similar to traditional radar systems, our novel approach is entirely passive—utilizing the wireless signals that already swamp our urban airways. We could use this technology, for example, for efficient and undetectable public safety surveillance.
Producing a demonstrative passive radar system using multiple NI Universal Software Radio Peripherals (USRPs) and LabVIEW to fuel our cutting-edge research. The USRPs, which can operate over wide frequency bands, acquire radio frequency (RF) signals on several channels. We used LabVIEW to design and iterate the advanced signal processing used to detect minute Doppler shifts in the acquired wireless signals to sense movement.
With dedication and a creative approach, University College London (UCL) research is helping to address the world's most urgent problems. Whether designing healthier cities or grappling with issues such as global health and climate change, the challenges of daily life inspire UCL students and academics. Based at UCL, our team of electrical engineering researchers is investigating passive radar technologies that can see through walls using WiFi radio waves.
Our novel research required a real-time, passive (noncooperative) wireless target detection demonstration system capable of tracking moving bodies through walls and obstacles. Much like traditional radar systems, our approach still relies on detecting the Doppler shifts in radio waves as they reflect off moving objects. However, unlike traditional radar systems that actively transmit radio waves, our passive system relies on the existing WiFi signals that already swamp our airwaves. The complete lack of spectrum occupation and power emission ensures our radar is undetectable, making it ideal for military or security surveillance in urban settings.
Aside from public defense applications, our passive detection could be applied in a broad range of scenarios, including crowd and traffic monitoring and human-machine interfacing. Different types of wireless signals can be applied to different situations. For example, our system could acquire IEEE 802.11x (b, g, n, ac) signals to detect indoor moving targets for security purposes, such as hostage situations. Alternatively, the same system could monitor cellular signals, such as Global System for Mobile Communications (GSM) or Long-Term Evolution (LTE), to detect direction and velocity of moving vehicles before triggering an appropriate machine response to the detected movement.
Maximizing the versatility of our devised radar system requires multiple channels for compatibility with multiple frequency bands. The system should be flexible enough to work with almost any type of WiFi signal (IEEE 802.11 b, g, n, ac), as well as FM and cellular signals. This relies on flexible RF hardware that can accommodate wide frequency ranges, in additional to easily reconfigured signal-processing software.
In order to accurately capture target movement, we required at least two receiver channels for frequency-time processing, known as ambiguity analysis. One channel locks onto the base radio signal from the direct path to a local wireless signal transmitter (such as a WiFi router)—this becomes the reference channel. The other receiver channel measures the reference signal as it reflects off a moving target—this is the surveillance channel. At the simplest level, the reference and surveillance signals can be compared to ascertain velocity and position of a detected target. However, in reality, this requires advanced ambiguity analysis, cross-correlation, Fourier transformation, and intelligent error detection.
For our research, we successfully built a two-channel demonstration system that used any available WiFi (IEEE 802.11x) signal to detect moving objects or bodies behind closed doors.
At the heart of our system were two USRP-2921 RF transceivers used to receive the reference and surveillance signals. Not only did the USRPs meet our accuracy and frequency range requirements, but their software-defined nature helped us rapidly iterate our algorithm designs.
From a software perspective, we chose LabVIEW. However, we did initially investigate other tools, such as GNU Radio, for processing data with C++. As our ambiguity analysis, which includes in-depth vector calculations and visualization, required complex, multithreaded processing operations, which would be difficult to implement in traditional textual languages. As LabVIEW is an inherently multithreading development tool, it naturally reduced our code complexity. This, combined with other features of LabVIEW, including intuitive graphical programming and built-in design patterns, reduced our development time by weeks.
The NI USRP platform is available on multiple frequency bands, covering 50 MHz to 5.9 GHz, so our passive radar system could cover a huge range of wireless signals, including FM, GSM, LTE, IEEE 802.11x, IEEE 802.16, and digital audio broadcasting (DAB) or digital video broadcasting (DVB). On each frequency band, we used a 20 MHz baseband I/Q bandwidth streaming at 25 MS/s for host-based processing with LabVIEW. The bandwidth is large enough to capture the widest communication signals used for the passive target detection demo.
Besides wide-frequency band coverage, another advantage of USRP is that it includes a dedicated port for daisy-chaining and synchronizing advanced multiple input, multiple output (MIMO) systems. This will be very useful as we extend the radar system for future research.
To program the USRP, LabVIEW provides an API that allowed us to quickly open, configure and initiate receiver sessions; set parameters such as centre frequency, IQ sampling rate, channel gain, and length of samples; and receive data from the air. The API offers complex double and half-precision floating-point data for adapting to different processing accuracy and speed requirements. Once acquired, ambiguity processing is applied to IQ data using the mathematics and signal processing tools in LabVIEW.
With USRP and LabVIEW, we built and tested the passive wireless detection demo very quickly. Using functions built into LabVIEW, we can efficiently implement a series of vector operations, such as array subset, indexing array, array reshaping and analysis, in a single block. Beside the mathematical functions, we were able to perform tailored fast Fourier transforms using built-in LabVIEW signal processing functions, saving us both computing and programming time.
Following time-frequency ambiguity analysis, we apply a threshold that dynamically changes with the environment to processed signals to determine whether the detected result is a real target or false alarm.
We used two detection scenarios to demonstrate the capabilities of the designed system. The first scenario is to detect a walking person using WiFi signal emissions from a common WiFi access point (AP) which has 15 dBm. In the experimental setup, a 25 cm-thick brick wall separated the reference and surveillance antenna from the person and the WiFi AP (see Figure 3). Both reference and surveillance signals are digitized by the USRPs and processed in LabVIEW.
The second scenario is to detect body gestures through the wall, using the same experimental environment. The difference between these two scenarios is type and magnitude of human target movement. In order to detect the small movement in the second scenario, different software processing parameters are used for longer integration time and lowered detection thresholding.
Figure 4 shows the detection results for scenario 1, where a person is walking back and forth. The LabVIEW front panel presents the instant Doppler surface results (upper left), determined target (upper middle), spectrum of the target range bin (upper right), Doppler record showing a 60-minute detection history (bottom left), and target intensity index record (bottom right). The threshold is applied on the target intensity index, so when a detected signal exceeds a certain level, the system will treat the current detection as a valid target. The Doppler record graph (bottom left) shows clear positive and negative Doppler shifts, which correspond to forward and backward walking directions.
Figure 5 shows the detection results of smaller body movements as a person transitions from squatting to standing stances. In this case, the system can recognize less than 1 Hz Doppler discrepancies caused by the small disturbance. Each periodic wave represents a detected squatting-standing gesture cycle, where a positive Doppler shift means that a certain body part is approaching the surveillance antenna. We have since progressed the radar system to detect even smaller movements, such as hand gestures.
Experimental results gained via our USRP-based radar system have definitely proven the concept of through-wall passive WiFi sensing. In addition, with the high sensitivity of the NI solution, we can detect smaller movements than we initially thought possible.
LabVIEW and NI USRP are an ideal choice for rapidly prototyping wireless signal transmission, reception and processing. The wide frequency bands and ready-made signal processing libraries helped speed up code development and real-world experimentation.
We are truly excited about how our novel approach to passive radar can be used in the future, including public security (hijack or hostage situations), eHealth (a monitoring system for the aged) and new human-machine interfacing (for both industry and entertainment).
Aside from being a strong proof of concept, our demonstrative passive detection system will be used as a highly engaging teaching platform for engineering students and a testbed for future passive detection algorithm development.
Bo Tan
University College London
1104 Roberts Building, Torrington Building, Department of Electronic and Electrical Engineering, University Collage London
London WCIE 7JE
b.tan@ee.ucl.ac.uk