WiFi Doppler Imaging

---

- Comment Resolved, 18 December 2021 (UTC)

Example of Doppler shifting due to source motion.

WiFi Doppler Imaging is an imaging modality that leverages the Doppler effect to detect human gestures and motions. By quantifying shifts in frequency content in reflected WiFi waves, several predetermined actions and motions can be distinguished from each other. Currently, the technology is in its infancy and has not been widely commercialized outside of the research setting.

Existing WiFi protocols use microwaves in the 2.4 to 6 GHz frequency range (5-12 cm wavelength). This energy is non-ionizing and has very high signal coverage in indoor locations (houses, office spaces, etc.). WiFi imaging leverages this high spatial range by analyzing Doppler shifting in the signals reflected back to a WiFi router by its surroundings. In this case, Doppler shifting results from human targets moving in relation to the stationary WiFi router. The human targets are echoing WiFi signals back towards the router, acting as non-stationary transmitters. Thus, a frequency shift is observed at the receiving end of the router.

Current technologies

Current technologies to detect human gestures and motions in indoor settings primarily involve visible-light cameras. A continuous video feed of the target is streamed into a machine learning algorithm, which annotates the target's pose in a frame-by-frame processing sequence. This method is highly accurate in determining the target's actions, but comes at the cost of user privacy, especially in a home environment. Another drawback of visual detection is that the user can be obstructed from a camera's view by furniture and walls, requiring multiple camera angles for this technology to be feasible in a simple environment. Other methods use low power IR transmitter/receiver pairs to differentiate rudimentary actions, such as a person standing still versus walking. While this enables higher user privacy, the scope of possible actions that can be detected is extremely low.

Importance and objective

WiFi Doppler imaging has four key principles. 1) The technology is easily accessible and uses a readily available energy source. WiFi devices are naturally spread out in the environment. This means that there need be no additional signal inputs, and cost could also be significantly lower. 2) WiFi is relatively long range, travelling up to 22 meters. 3) There are limited privacy concerns in comparison to modalities such as camera systems utilizing image processing techniques. 4) Imaging can be done in real time. The applications for WiFi Doppler are broad. It can be used to monitor elders in their homes, detecting falls or other user gestures that indicate distress. Whole-house gesture detection for smart device control has also been implemented for a more seamless method of interaction that has extremely large coverage when compared to conventional approaches.

Energy format

WiFi uses microwaves, a type of electromagnetic radiation, in the 2.4 to 6 GHz frequency bands. A narrow band of frequencies around a central one (e.g. 2.4 GHz or 5 GHz) are modulated to encode data. Higher frequencies have greater imaging resolution, but come with the tradeoff of decreased range due to increased attenuation. There is an exponential decay in signal strength as distance increases, and addition of walls in the detector-target path accelerates this decay according to the path loss seen here. Out of common household materials, it is important to note that WiFi is easily reflected by metal and is highly absorbed by concrete. Thus, imaging with these materials in close proximity may lead to artifacting or poor signal strength at the detector.

Energy source and detector

Example of a 3-antenna WiFi router made by Netgear.

The energy source used for imaging are WiFi routers that are in range of the object. This makes the energy source low cost as existing technology is harnessed with no additional device implementation. WiFi routers generally produce different frequency channels for close and long ranges. Thus, the trade-off between range and image resolution can be balanced by switching to higher frequency channels when a target is nearby. Under normal transmission conditions, there is a complex frequency modulation/demodulation process to encode/decode data packets into the WiFi signal. However, these data encoding steps are not required for WiFi Doppler imaging, and simply add complexity in deciphering the user's actions. Consequently, current WiFi Doppler techniques assume that a portion of the router's frequency band is allocated exclusively for imaging.

Since WiFi routers need to support both data download and upload, their antennas are capable of transmitting and receiving data. Thus, the reflected WiFi signals can be measured using the same router that transmitted the original waves. Most commercial routers have anywhere from 1 to 7 mounted antennae. Having more antennae increases the imaging resolution and signal-to-noise ratio of Doppler detection. An example case study by Huang et al. (2014) found that a 2.4 GHz router with 5 antennae yielded Doppler resolution of 2 Hz, which is well below average human movement speeds of 8-150 Hz. This indicates that such a setup would be sufficient for WiFi Doppler imaging of human targets.

Energy tissue interaction

WiFi imaging leverages Doppler shifting in the signals reflected back to a WiFi router by its surroundings. In particular, "micro-Dopplers", or frequency shifts resulting from varying body part movements in relation to the stationary WiFi router, are combined to form Doppler signatures. The Doppler shift can be quantified as:

$\mathrm{\Delta }f\propto \frac{2vcos(\theta )}{c}f$

where $\mathrm{\Delta }f$ is the Doppler shift, $v$ is the target's velocity relative to the source, $c$ is the speed of light and $\theta$ is the angle between the target and source.

The strength of the reflected waves will be largely determined by three factors: object material, object size, and diffraction and interference effects. The object material will determine the level of attenuation and reflection. Metallic surfaces reflect more wireless signals than plastic. The type of reflection is also determined by object material. Some materials scatter incident signals in many directions to create diffuse reflection while others reflect waves orderly, according to the laws of reflection, like a mirror to create specular reflection. Larger objects will reflect more signals than smaller objects. Lastly, diffraction in reflected signals creates blur in the image. This means that image formation will be restricted by the angular resolution. Similar to the lateral resolution in ultrasound imaging, the angular resolution θ here according to Rayleigh criterion is:

$\theta \approx 1.220\frac{\lambda }{D}$

where λ is the signal wavelength and D is the length of the router's antenna array. Therefore, resolution and blur can be improved by utilizing higher frequency and energy signals and increasing antenna array length.

Image formation

Gesture sketches

The image that is created to detect motion and actions are Doppler spectrograms, or frequency-time Doppler profiles, with time on the x axis and frequencies on the y axis. This spectrogram displays the range of Doppler frequencies generated by moving objects over time.

The foundation of the image formation is the receiver computing the Fast Fourier Transform (FFT) of the WiFi signals over time. FFT is computed over samples within an interval. The duration of the interval provides the Doppler resolution. (e.g. 2 second intervals give a 0.5 Hertz resolution). Then, the receiver shifts forward by an offset to compute the FFT of the next overlapping interval. This offset gives the time resolution, as actions faster than this offset will not be detected by the receiver. The cycle of FFT calculations and shifting are repeated to obtain a frequency-time profile and the real-time spectrogram.

Frequency-time Doppler profiles of the gestures

As different movements are performed, the spectrogram will change. The change in frequency-time profile is unique to each action, allowing movements to be classified. Classification algorithms are trained on a set of predetermined user actions (running, punching, etc.), associating particular spectrogram patterns with each action. Later, any captured spectrogram can be classified based on its similarity to each of the trained actions. The temporal alternation between positive and negative Doppler shifts was found to be the most significant predictor of what action a spectrogram corresponded with. A significant drawback of this image formation approach is the need for constant retraining of the classification algorithms when the router's environment changes.

Future

A major limitation of existing WiFi Doppler imaging technologies is multi-target discrimination. The presence of multiple moving users makes it extremely difficult for clustering algorithms to map the observed spectrograms to actions. Pu et al. found a 40% decrease in classification accuracy after 4 total users were present. Future work to determine methods of isolating system users would help increase viability in larger households. Furthermore, a major drawback of existing WiFi Doppler imaging is that the classification algorithm must be retrained for each new type of router and is not robust to changes in its environment. Moving furniture or bringing a cell phone nearby may cause the reflective behaviors of the router's surroundings to change, requiring the classification system to be retrained. It would be beneficial to create a more generalizable classification pipeline that doesn't require constant retraining by the end user. This would enhance the user experience, but poses a challenge due to variability across different household router setups.

Conclusion

In conclusion, WiFi can be leveraged to image motions and actions of living organisms. The guiding principle of this WiFi imaging modality is the Doppler effect. Because microwaves are reflected upon meeting an object and its frequency contents shift due to the Doppler effect, frequency-time profiles could be created to determine and predict motions.

References

This article "WiFi Doppler Imaging" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:WiFi Doppler Imaging. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.

---