Neural components for differentiable ray tracing of radio propagation
Apparatuses, systems, and techniques relate to neural components for differentiable ray tracing of radio propagation. Differentiable ray tracing may be used to refine the scene geometry of the physical environment, to learn or optimize the scene properties of objects in the scene, to learn or optimize the scene properties of antennas, and to learn or optimize antenna patterns, array geometries, and orientations and positions of transmitters and receivers. Once scene properties have been learned or optimized, the differentiable ray tracer may further be used to simulate the performance of different configurations of the transmitters, receivers, and scene geometry. In an embodiment, one or more of the scene geometry, scene properties, and antenna characteristics are computed by a differentiable parametric function, such as a neural network, etc. and parameters of the differentiable parametric function are learned using the differentiable ray tracing.
1 . A computer-implemented method, comprising:
computing, by a differentiable ray tracer, simulated radio characteristics for a three-dimensional (3D) scene based on configured parameters and at least one trainable parameter corresponding to a scene property, wherein for each ray intersection point in the 3D scene, a neural component estimates the at least one trainable parameter; and
updating weights applied by the neural component to estimate the at least one trainable parameter using gradient-based optimization to minimize a loss function of the simulated radio characteristics and reference radio characteristics, wherein the at least one trainable parameter comprises a scattering pattern that is modeled as a linear combination of components on a hemisphere.
2 . The computer-implemented method of claim 1 , wherein the neural component generates an embedding vector used to estimate the at least one trainable parameter.
3 . The computer-implemented method of claim 1 , wherein each ray intersection point is encoded into a higher dimension space for input to the neural component.
4 . The computer-implemented method of claim 1 , wherein each ray intersection point is encoded using a multiresolution hash grid for input to the neural component.
5 . The computer-implemented method of claim 1 , wherein each ray intersection point is normalized to a unit cube and encoded into a higher dimension space for input to the neural component.
6 . The computer-implemented method of claim 1 , wherein each ray intersection point comprises at least one angle relative to a surface in the 3D scene at the intersection point.
7 . The computer-implemented method of claim 1 , further comprising updating the weights to modify at least one of a meta material, a reconfigurable intelligent surface, an antenna pattern, an antenna orientation, and an antenna position.
8 . The computer-implemented method of claim 1 , wherein the configured parameters or the at least one trainable parameter include one or more of scene geometry, configuration of reconfigurable intelligent surfaces and meta materials, antenna patterns, array geometries, and transmitter and receiver orientations and positions.
9 . The computer-implemented method of claim 1 , wherein the scene property comprises at least one of relative permittivity, reflection coefficients, transmission coefficients, conductivity, effective roughness, and permeability of object surfaces and scattering functions.
10 . The computer-implemented method of claim 1 , wherein the at least one trainable parameter comprises an antenna pattern that is modeled as a mixture of spherical Gaussian distributions.
11 . The computer-implemented method of claim 1 , wherein the differentiable ray tracer computes paths of electromagnetic waves.
12 . The computer-implemented method of claim 1 , wherein the simulated radio characteristics estimate qualities of a transmitted electromagnetic wave at a receiver.
13 . The computer-implemented method of claim 1 , wherein the simulated radio characteristics comprise one or more of channel impulse responses, channel frequency responses, path delays, path losses, angles of arrival, angles of departure, amplitudes, powers, delay spread, Doppler spread, angular spread, power-delay-angular profile, and a number of paths.
14 . The computer-implemented method of claim 1 , wherein the reference radio characteristics are measurements taken at different locations in the 3D scene.
15 . The computer-implemented method of claim 1 , wherein at least one of the steps of computing or updating is performed on a server or in a data center and the simulated radio characteristics or an image generated from the simulated radio characteristics is streamed to a user device.
16 . The computer-implemented method of claim 1 , wherein at least one of the steps of computing or updating is performed within a cloud computing environment.
17 . The computer-implemented method of claim 1 , wherein at least one of the steps of computing or updating is performed for training, testing, or certifying a neural network employed in a machine, robot, or autonomous vehicle.
18 . The computer-implemented method of claim 1 , wherein at least one of the steps of computing or updating is performed on a virtual machine comprising a portion of a graphics processing unit.
19 . A system, comprising:
a memory that stores reference radio characteristics; and
a processor that is connected to the memory, wherein the processor is configured to
produce simulated radio characteristics for a three-dimensional scene (3D) by:
computing, by a differentiable ray tracer, the simulated radio characteristics for the 3D scene based on configured parameters and at least one trainable parameter corresponding to a scene property, wherein for each ray intersection point in the 3D scene, a neural component estimates the at least one trainable parameter; and
updating weights applied by the neural component to estimate the at least one trainable parameter using gradient-based optimization to minimize a loss function of the simulated radio characteristics and the reference radio characteristics, wherein the at least one trainable parameter comprises a scattering pattern that is modeled as a linear combination of components on a hemisphere.
20 . The system of claim 19 , wherein the differentiable ray tracer computes paths of electromagnetic waves.
21 . A non-transitory computer-readable mediummedia storing computer instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:
computing, by a differentiable ray tracer, simulated radio characteristics for a three-dimensional (3D) scene based on configured parameters and at least one trainable parameter corresponding to a scene property, wherein for each ray intersection point in the 3D scene, a neural component estimates the at least one trainable parameter; and
updating weights applied by the neural component to estimate the at least one trainable parameter using gradient-based optimization to minimize a loss function of the simulated radio characteristics and reference radio characteristics, wherein the at least one trainable parameter comprises a scattering pattern that is modeled as a linear combination of components on a hemisphere.
22 . The non-transitory computer-readable medium of claim 21 , wherein the differentiable ray tracer computes paths of electromagnetic waves.