Radar signal interference mitigation with generative networks

1 . An automotive radar system comprising:

at least one transmitter and at least one receiver, wherein the at least one transmitter and the at least one receiver are configured to transmit and receive radar signals, wherein the at least one transmitter and the at least one receiver are coupled to a vehicle; and

an automotive radar processor configured to:

cause the at least one transmitter to transmit, at a first time, a first radar signal,

receive, using the at least one receiver, a received signal y, wherein the received signal y is of the form y=x s +x i +n, wherein x s represents a desired signal component, x i represents an interference signal component, and n represents a noise component of the received signal y,

determine a desired signal prior p θ1 , wherein the desired signal prior p θ1 defines a first data distribution expected in the desired signal component x i of the received signal y and is determined using generative modeling,

determine an interference signal prior p θ2 , wherein the interference signal prior p θ2 defines a second data distribution expected in the interference signal component and is determined using generative modeling,

execute an iterative procedure using the desired signal prior p θ1 and the interference signal prior p θ2 and to determine an estimate of the interference signal component x i and an estimate of the desired signal component x i , wherein each iteration of the iterative procedure comprises a data consistency operation to increase a data consistency attribute between the received signal y, the estimate of the interference signal component x i , and the estimate of the desired signal component x s , and

determine a desired signal using the estimate of desired signal component x s , and

process the desired signal to determine an attribute of a target object in a field of vision of the automotive radar system.

2 . The automotive radar system of claim 1 , wherein the automotive radar processor is configured to determine the desired signal by subtracting the estimate of the interference signal component x i from the received signal y to determine the desired signal.

3 . The automotive radar system of claim 1 , wherein the interference signal prior p θ2 is determined by performing denoising score matching on a training data set, wherein the training data set includes simulated time-domain interference signals.

4 . The automotive radar system of claim 3 , wherein values in the training data set are modified by adding Gaussian noise values to the values in the training data set to generate a modified training data set, wherein Gaussian noise values added to the values in the training data set have a range of standard deviations.

5 . The automotive radar system of claim 4 , wherein the generative modeling is configured to estimate gradients of the values in the modified training data set by mapping the values in the modified training data set to corresponding values in the training data set.

6 . The automotive radar system of claim 4 , wherein the Gaussian noise values are determined by a finite number of noise levels.

7 . The automotive radar system of claim 4 , wherein the Gaussian noise values are determined according to a diffusion process, wherein the diffusion process is defined using a stochastic differential equation.

8 . The automotive radar system of claim 1 , wherein the received signal includes a signal encoded as a frequency-modulated continuous wave radar signal.

9 . A system comprising:

a transmitter configured to transmit a radar signal towards a target object;

a receiver configured to receive a received signal in response to the transmitted radar signal; and

a processor, configured to:

receive, using the receiver, a received signal,

execute an iterative procedure using a desired signal prior including a first data distribution expected in a desired signal component of the received signal and an interference signal prior including a second data distribution expected in an the interference signal component of the received signal and to determine an estimate of an interference signal component of the received signal and an estimate of the desired signal component of the received signal, wherein the interference signal prior is determined using generative modeling and each iteration of the iterative procedure executes a data consistency operation,

use the estimate of the interference signal component and the received signal to determine a desired signal, and

processing the desired signal to determine an attribute of the target object.

10 . The system of claim 9 , wherein the desired signal prior is a sparse prior.

11 . The system of claim 9 , wherein the interference signal prior is determined by performing denoising score matching on a training data set, wherein the training data set includes at least one of simulated time-domain interference signals and measured time-domain interference signals.

12 . The system of claim 11 , wherein values in the training data set are modified by adding Gaussian noise values to the values in the training data set to generate a modified training data set, wherein Gaussian noise values added to the values in the training data set have a range of standard deviations.

13 . The system of claim 12 , wherein the generative modeling is configured to estimate gradients of the values in the modified training data set by mapping the values in the modified training data set to corresponding values in the training data set.

14 . The system of claim 12 , wherein the Gaussian noise values are determined by a finite number of noise levels.

15 . The system of claim 12 , wherein the Gaussian noise values are determined according to a diffusion process, wherein the diffusion process is defined using a stochastic differential equation.

16 . The system of claim 9 , wherein the received signal includes a signal encoded as a frequency-modulated continuous wave radar signal.

17 . A method comprising:

receiving a received signal y, wherein the received signal y is of the form y=x s +x i +n, wherein x s represents a desired signal component, x i represents an interference signal component, and n represents a noise component of the received signal y;

determining a desired signal prior p θ1 , wherein the desired signal prior p θ1 defines a first data distribution expected in the desired signal component x i of the received signal y and is determined using a generative network;

determining an interference signal prior p θ2 , wherein the interference signal prior p θ2 defines a second data distribution expected in the interference signal component and is determined using generative modeling,

executing an iterative procedure using the desired signal prior poi and the interference signal prior p θ2 and to determine an estimate of the interference signal component x i and an estimate of the desired signal component x s , wherein each iteration of the iterative procedure executes a data consistency operation to increase a data consistency attribute between the estimate of the interference signal component x i and the estimate of the desired signal component x s ,

determining a desired signal using the estimate of desired signal component x s , and

processing the desired signal to determine an attribute of a target object in a field of vision of an automotive radar system.

18 . The method of claim 17 , wherein determining the desired signal further comprises subtracting the estimate of the interference signal component x i from the received signal y to determine the desired signal.

19 . The method of claim 17 , further comprising determining the interference signal prior p θ2 by performing denoising score matching on a training data set, wherein the training data set includes at least one of simulated time-domain interference signals and measured time-domain interference signals.

20 . The method of claim 19 , further comprising modifying values in the training data set by adding Gaussian noise values to the values in the training data set to generate a modified training data set.