Machine learning-based channel estimation

1. A first device, comprising:

at least one processor; and

at least one non-transitory memory storing instructions, that when executed by the at least one processor, cause the first device to:

determine a signal quality that is expected in transmission of a reference signal from a second device to the first device;

receive the reference signal from the second device,

wherein the reference signal comprises a sounding reference signal;

select a channel estimation model from a plurality of channel estimation models based on the expected signal quality, the plurality of channel estimation models having been trained for a plurality of candidate signal qualities for the reference signal; and

determine, using the selected channel estimation model and based on the received reference signal, channel state information of a communication channel from the first device to the second device;

wherein selecting the channel estimation model is based on

obtaining a mapping between the plurality of candidate signal qualities and a plurality of signal quality ranges associated with the plurality of channel estimation models, performance metrics of the plurality of channel estimation models in determining channel state information with respect to the plurality of signal quality ranges satisfying a performance threshold;

determining, based on the mapping, one of the plurality of signal quality ranges into which the expected signal quality falls; and

selecting the channel estimation model that is associated with the plurality of signal quality ranges.

2. The device of claim 1 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to determine the expected signal quality with:

receiving a null signal from the second device;

determining at least one of a power of noise or interference in the transmission based on a received power of the null signal; and

determining the expected signal quality based on at least one of the power of noise or interference and a transmitted power of the reference signal, the transmitted power of the reference signal being indicated by the second device.

3. The device of claim 2 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to determine the expected signal quality with:

determining a transmitted signal-to-noise ratio of the reference signal based on at least one of a ratio of the transmitted power of the reference signal to the power of noise or interference.

4. The device of claim 1 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to receive the first number of transmissions of the reference signal with:

receiving the first number of transmissions of the reference signal with a second number of predetermined beams of the first device, respectively, the second number being equal to the first number.

5. The device of claim 4 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to select the channel estimation model with:

selecting, from the plurality of channel estimation models, the channel estimation model that has been trained with a set of training received reference signals generated based on the second number of predetermined beams.

6. The device of claim 4 , wherein the second number of predetermined beams comprises one or two predetermined beams of the first device.

7. The device of claim 1 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to:

receive the reference signal based on:

determining a first number of transmissions of the reference signal, the first number being equal to or larger than one;

transmitting an indication of the first number to the second device; and

receiving the first number of transmissions of the reference signal from the second device.

8. The device of claim 1 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to cause the first device to determine, for a given channel estimation model of the plurality of channel estimation models, the associated signal quality range with:

obtaining a plurality of test received reference signals, the plurality of test received reference signals generated based on a plurality of test signal qualities;

applying the given channel estimation model to determine test channel state information based on the plurality of test received reference signals;

determining a plurality of performance metrics of the given channel estimation model for the plurality of test signal qualities based on the test channel state information and ground-truth channel state information;

selecting a test signal quality from the plurality of test signal qualities with comparing the performance metrics with the performance threshold; and

determining the associated signal quality range based on the selected test signal quality and the candidate signal quality for which the given channel estimation model has been trained.

9. The device of claim 1 , wherein the at least one non-transitory memory is storing instructions executed by the at least one processor, to further cause the first device to:

determine a beamforming configuration for transmission from the first device to the second device based on the channel estimation information; and

transmit data over the communication channel from the first device to the second device based on the beamforming configuration.

10. A method comprising:

determining a signal quality that is expected in transmission of a reference signal from a second device to a first device;

receiving the reference signal from the second device,

wherein the reference signal comprises a sounding reference signal;

selecting a channel estimation model from a plurality of channel estimation models based on the expected signal quality, the plurality of channel estimation models having been trained for a plurality of candidate signal qualities for the reference signal; and

determining, using the selected channel estimation model and based on the received reference signal, channel state information of a communication channel from the first device to the second device;

wherein selecting the channel estimation model is based on

obtaining a mapping between the plurality of candidate signal qualities and a plurality of signal quality ranges associated with the plurality of channel estimation models, performance metrics of the plurality of channel estimation models in determining channel state information with respect to the plurality of signal quality ranges satisfying a performance threshold;

determining, based on the mapping, one of the plurality of signal quality ranges into which the expected signal quality falls; and

selecting the channel estimation model that is associated with the plurality of signal quality ranges.

11. A non-transitory program storage device readable by a machine, tangibly embodying a program of instructions executable by the machine for performing operations, the operations comprising the method as claimed in claim 10 .