Wiffract is based on a way of interpreting these signals to detect the edges of objects and their orientation
The news was released that a team of researchers from the University of California in Santa Barbara has developed a method for determining the contours of stationary objects behind a wall analyzing Wi-Fi signal distortion.
The method, called Wiffract is based on detecting changes in the signal that occur owed to the interaction of electromagnetic waves emanating from a Wi-Fi transmitter with the edges of objects.
“Imaging fixed landscapes with WiFi is a considerable challenge due to the lack of movement,” said Mostofi, professor of electrical and computer engineering. "We then took a completely different approach to tackling this difficult problem, focusing on tracing the edges of objects." The proposed methodology and experimental results appeared in the Proceedings of the 2023 IEEE National Radar Conference (RadarConf) on June 21, 2023.
The researchers explain that when a radio frequency wave (RF) of Wifi finds an edge point, generates a cone of outgoing rays known as "Keller's cone" guided by the principles of geometric diffraction theory (GTD).
It is mentioned that the mathematical model of Wiffract can capture the edges of stationary objects using GTD theory and the corresponding Keller cones. Once it identifies “high-confidence edge points,” Wiffract can reconstruct object shapes while further improving the resulting edge map using advanced computer vision techniques.
The mathematical apparatus used by the researchers is based on the geometric theory of diffraction GTD, which describes the effects that occur when an electromagnetic wave surrounds obstacles.
Wiffract Demo
In GTD, energy is assumed to propagate along the rays and the wave field is considered as the sum of the ray type fields. In addition to incident, refracted and reflected rays, GDT theory introduces the concept of diffracted rays, which occur when lightning strikes a sharp edge or point on the surface of an object.
If the beam hits an edge, the diffracted rays form the surface of a Keller cone whose opening angle is equal to twice the angle between the incident beam and the tangent to the surface of the edge at the point of diffraction. If the incident ray is perpendicular to the tangent to the edge, the cone becomes a plane, and if it hits the tip of the vertex, the diffracted rays diverge uniformly in all directions.
"When a given wave hits an edge point, a cone of outgoing rays emerges according to Keller's Geometric Theory of Diffraction (GTD), called the Keller cone," explained Mostofi. The researchers note that this interaction is not limited to visibly sharp edges, but applies to a broader set of surfaces with sufficiently small curvature.
“Depending on the orientation of the edge, the cone leaves different traces (i.e., conical sections) on a given receiving grating. “We then developed a mathematical framework that uses these conical traces as signatures to infer the orientation of the edges, thus creating an edge map of the scene,” Mostofi continued.
The proposed method does not require preliminary training of the neural network and is not limited to only identifying the objects covered during machine learning. Instead, the neural network attempts to recreate the contours of arbitrary objects by following their edges.
A signal analyzer that emulates a set of Wi-Fi receiver antennas takes into account changes in signal power at individual points on a two-dimensional plane. In the signal that reaches the analyzer, the neural network detects characteristic distortions of the diffracted waves that are produced when a wave impinges on an edge and recreates the spatial position of the edges.
As a demonstration of the method, the researchers organized the detection of mock-ups of letters of the English alphabet placed behind a wall, using three typical wireless signal transmitters operating on Wi-Fi frequencies.
To receive the signal, a scanning cart was created with several Wi-Fi receivers that move back and forth emulating a set of antennas. It should be noted that the method works not only for objects with visible sharp edges, but is also applicable to objects with a slight level of surface curvature.
finally if you are interested in knowing more about it, you can check the details in the following link