Deep learning for super resolution in a radar system

1. A method of using deep learning for super resolution in a radar system, the method comprising:

obtaining first-resolution time samples from reflections based on transmissions by a first-resolution radar system of multiple frequency-modulated signals, wherein the first-resolution radar system includes multiple transmit elements and multiple receive elements;

reducing resolution of the first-resolution time samples to obtain second-resolution time samples;

implementing a matched filter on the first-resolution time samples to obtain a first-resolution data cube and on the second-resolution time samples to obtain a second-resolution data cube;

processing the second-resolution data cube with a neural network to obtain a third-resolution data cube;

training the neural network based on a first loss obtained by comparing the first-resolution data cube with the third-resolution data cube; and

using the neural network with a second-resolution radar system to detect one or more objects.

2. The method according to claim 1 , further comprising disposing the second-resolution radar system in a vehicle.

3. The method according to claim 1 , further comprising performing detection of the one or more objects using the first-resolution data cube and using the third-resolution data cube.

4. The method according to claim 3 , wherein the training the neural network is based on a second loss obtained by comparing a result of the detection performed using the first-resolution data cube with a result of the detection performed using the third-resolution data cube.

5. The method according to claim 1 , further comprising inverting the third-resolution data cube to obtain third-resolution time samples.

6. The method according to claim 5 , wherein the training the neural network is based on a third loss obtained by comparing the first-resolution time samples with the third-resolution time samples.

7. The method according to claim 1 , wherein the reducing the resolution includes using all the first-resolution time samples resulting from only a sub-set of the multiple frequency-modulated signals.

8. The method according to claim 1 , wherein the reducing the resolution includes using only a sub-set of the first-resolution time samples resulting from all the multiple frequency-modulated signals.

9. The method according to claim 1 , wherein the reducing the resolution includes using the first-resolution time samples resulting from only a sub-set of the multiple transmit elements and the multiple receive elements of the first-resolution radar system.

10. The method according to claim 1 , wherein the reducing the resolution includes obtaining the second-resolution time samples at a same resolution as time samples provided by the second-resolution radar system.

11. A system to use deep learning for super resolution in a radar system, the system comprising:

a first-resolution radar system configured to obtain first-resolution time samples from reflections based on transmissions of multiple frequency-modulated signals, wherein the first-resolution radar system includes multiple transmit elements and multiple receive elements;

a processor configured to reduce resolution of the first-resolution time samples to obtain second-resolution time samples, to implement a matched filter on the first-resolution time samples to obtain a first-resolution data cube and on the second-resolution time samples to obtain a second-resolution data cube, to process the second-resolution data cube with a neural network to obtain a third-resolution data cube, and to train the neural network based on a first loss obtained by comparing the first-resolution data cube with the third-resolution data cube; and

a second-resolution radar system configured to use the neural network to detect one or more objects.

12. The system according to claim 11 , wherein the second-resolution radar system is in a vehicle.

13. The system according to claim 11 , wherein the processor is further configured to perform detection of the one or more objects using the first-resolution data cube and using the third-resolution data cube.

14. The system according to claim 13 , wherein the processor is configured to train the neural network based on a second loss obtained by comparing a result of the detection performed using the first-resolution data cube with a result of the detection performed using the third-resolution data cube.

15. The system according to claim 11 , wherein the processor is further configured to invert the third-resolution data cube to obtain third-resolution time samples.

16. The system according to claim 15 , wherein the processor is configured to train the neural network based on a third loss obtained by comparing the first-resolution time samples with the third-resolution time samples.

17. The system according to claim 11 , wherein the processor is configured to reduce the resolution of the first-resolution time samples by using all the first-resolution time samples resulting from only a sub-set of the multiple frequency-modulated signals.

18. The system according to claim 11 , wherein the processor is configured to reduce the resolution of the first-resolution time samples by using only a sub-set of the first-resolution time samples resulting from all the multiple frequency-modulated signals.

19. The system according to claim 11 , wherein the processor is configured to reduce the resolution of the first-resolution time samples by using the first-resolution time samples resulting from only a sub-set of the multiple transmit elements and the multiple receive elements of the first-resolution radar system.

20. The system according to claim 11 , wherein the processor is configured to reduce the resolution to obtain the second-resolution time samples at a same resolution as time samples provided by the second-resolution radar system.