Radar system and method for performing direction of arrival estimation
According to various embodiments, a radar system is described comprising a radar receiver configured to receive radio signals, wherein each radio signal is associated with a channel of a plurality of channels, a peak detector configured to perform peak detection using the received radio signals, wherein each detected peak corresponds to a detected object and a direction of arrival estimator configured to, for a detected peak, generate a vector having, for each of the channels, an entry specifying a Doppler Fourier transform result for the channel, supply the vector to a machine learning model trained to output, for each of one or more additional channels, an entry specifying a predicted Doppler Fourier transform result corresponding to the additional channel and perform direction-of-arrival estimation using an output from the machine learning model which the machine learning model outputs in response to being supplied with the vector.
1. A radar system, comprising:
A radar receiver configured to receive radio signals, wherein each radio signal is associated with a channel of a plurality of channels, each channel corresponding to a pair of transmit antenna and receive antenna;
A peak detector configured to perform peak detection using the received radio signals, wherein each detected peak corresponds to a detected object; and
A direction of arrival estimator configured to, for a detected peak,
generate a vector having, for each of the channels, an entry specifying a Doppler Fourier transform result for the channel calculated from radio signals transmitted from the transmit antenna and received by the receive antenna of the pair to which the channel corresponds;
Supply the vector to a machine learning model trained to output, for each of one or more additional channels, an entry specifying a predicted Doppler Fourier transform result corresponding to the additional channel; and
perform direction-of-arrival estimation using an output from the machine learning model which the machine learning model outputs in response to being supplied with the vector.
2. The radar system of claim 1 , comprising a radar processor configured to generate, for each channel, a range-Doppler map from radio signals associated with the channel and wherein, for each channel, the entry is a Doppler Fourier transform result value that the range-Doppler map generated for the channel contains for a range bin and a velocity bin of the detected peak.
3. The radar system of claim 2 , wherein, for each of the one or more additional channels, the predicted Doppler Fourier transform result is a Doppler Fourier transform result value for a range bin and a velocity bin of the detected peak for the additional channel.
4. The radar system of claim 2 , wherein the peak detector is configured to perform the peak detection by integrating the range-Doppler maps generated for the channels to an aggregate range-Doppler map and identify peaks in the aggregate range-Doppler map.
5. The radar system of claim 1 , wherein each channel and each additional channel is associated with a respective position in an array of receive antennas and wherein the machine learning model is configured to generate, for each additional channel, the predicted Doppler Fourier transform result as if the predicted Doppler Fourier transform result was calculated from receive signals received via a receive antenna at the position associated with the additional channel.
6. The radar system of claim 5 , wherein at least some of the additional channels are associated with positions between positions associated with channels of the plurality of channels.
7. The radar system of claim 5 , wherein at least some of the additional channels are associated with positions at one side of the positions associated with the channels of the plurality of channels.
8. The radar system of claim 5 , wherein the array of receive antennas is a uniform array.
9. The radar system of claim 5 , wherein the radar system is a multiple input multiple output radar system and wherein the array of receive antennas comprises virtual receive antennas.
10. The radar system of claim 1 , wherein the vector comprises zero entries for the one or more additional channels and the machine learning model is trained to fill the zero entries with entries specifying a Doppler Fourier transform result of the peak predicted to be calculated from radio signals transmitted via the additional channels.
11. The radar system of claim 1 , wherein the machine learning model is a machine learning model trained to update the entries of the vector for the plurality of channels and the direction of arrival estimator is configured to perform direction-of-arrival estimation using the updated entries.
12. The radar system of claim 1 , comprising a training device configured to train the machine learning model.
13. The radar system of claim 12 , wherein the training device is configured to train the machine learning model by supervised learning wherein the training device is configured to generate each of a multiplicity of ground truth vectors by simulation of received radio signals and processing it to a determination-of-arrival input vector or simulating a determination-of-arrival input vector.
14. The radar system of claim 13 , wherein the ground truth vector comprises, for each of the one or more additional channels, a ground truth entry specifying a predicted Doppler Fourier transform result for a detected peak with which the ground truth vector is associated.
15. The radar system of claim 14 , wherein the ground truth vector additionally comprises, for each of the one or more channels, a ground truth entry specifying a predicted Doppler Fourier transform result for a detected peak with which the ground truth vector is associated.
16. The radar system of claim 13 , wherein the training device is configured to generate a multiplicity of ground truth vectors by varying at least one of a number of targets of a scene for which the ground truth vectors are generated, locations of the respective number of targets in the scene, radar cross sections of the targets in the scene, signal-to-noise ratios of the targets in the scene and channel phase errors to train the machine learning model using the multiplicity of ground truth vectors.
17. The radar system of claim 1 , wherein the machine learning model is a neural network.
18. The radar system of claim 17 , wherein the machine learning model comprises at least one of a multi-layer perceptron and an autoencoder.
19. A method for performing direction of arrival estimation in a radar system, the method comprising:
receiving radio signals, wherein each radio signal is associated with a channel of a plurality of channels, each channel corresponding to a pair of transmit antenna and receive antenna;
perform peak detection using the received radio signals, wherein each detected peak corresponds to a detected object; and
for a detected peak,
generating a vector having, for each of the channels, an entry specifying a Doppler Fourier transform result for the channel calculated from radio signals transmitted from the transmit antenna and received by the receive antenna of the pair to which the channel corresponds;
supplying the vector to a machine learning model trained to output, for each of one or more additional channels, an entry specifying a predicted Doppler Fourier transform result corresponding to the additional channel; and
performing direction-of-arrival estimation using an output from the machine learning model which the machine learning model outputs in response to being supplied with the vector.
20. The method of claim 19 , further comprising training the machine learning model.