Channel estimation using machine learning

1. A method comprising:

generating, using a first machine-learning model, an initial set of estimated channel information from a first input set of channel information obtained from a set of radio-frequency signals, wherein the first input set of channel information corresponds to measurements from a first plurality of radio-frequency chains, and wherein the estimated set of channel information corresponds to the first plurality of radio-frequency chains and a second plurality of radio-frequency chains;

generating, using a second machine-learning model, a set of estimated channel phases from the initial set of estimated channel information and a second input set of channel information, wherein the second set of input channel information corresponds to measurements from the second plurality of radio-frequency chains; and

combining the initial set of estimated channel information and the set of estimated channel phases to generate an enhanced set of estimated channel information, wherein the enhanced set of estimated channel information corresponds to the first plurality of radio-frequency chains and the second plurality of radio-frequency chains.

2. The method of claim 1 , wherein the first plurality of radio frequency chains comprises a plurality of high-resolution radio frequency chains, and the second plurality of radio frequency chains comprises a plurality of low-resolution radio frequency chains.

3. The method of claim 2 , wherein the first machine-learning model is a neural network comprising one or more convolutional layers.

4. The method of claim 1 , wherein the first machine-learning model comprises a plurality of down-sampling blocks followed with a plurality of up-sampling blocks.

5. The method of claim 4 , wherein the first machine-learning model comprises one or more skip-connections, the skip connection being between a down-sampling block and a corresponding up-sampling block.

6. The method of claim 1 , wherein the second machine-learning model is a long short-term memory model.

7. The method of claim 6 , wherein the initial set of estimated channel information is converted to time domain information prior to input into the long short-term memory model.

8. The method of claim 1 , wherein the second input set of channel information comprises estimated phases corresponding to the second plurality of radio-frequency chains, and wherein generating the set of estimated channel phases comprises:

determining further phase information from the initial set of estimated channel information; and

inputting the estimated phases corresponding to the second plurality of radio-frequency chains and the further phase information to the second machine-learning model.

9. The method of claim 1 , wherein combining the initial set of estimated channel information and the set of estimated channel phases to generate an enhanced set of estimated channel information comprises:

combining the estimated channel phase with an absolute value of a corresponding element of the initial set of estimated channel information.

10. The method of claim 1 , further comprising:

determining an additional set of channel information from the initial set of estimated channel information and the second input set of channel information using an alternative channel estimation method;

receiving further channel information corresponding to a measurement of an antenna element associated with a constrained-resolution radio-frequency chain, wherein the further channel information has been measured using a high-resolution radio-frequency chain;

determining a first error value, the first error value comparing the further channel information to a corresponding element of channel information in the enhanced set of estimated channel information;

determining a second error value, the second error value comparing the further channel information to a corresponding element of channel information in the additional set of channel information; and

if the first error is lower than the second error, outputting the enhanced set of estimated channel information.

11. A non-transitory program storage device readable with an apparatus, tangibly embodying a program of instructions executable with the apparatus for performing the method of claim 1 .

12. A base station, comprising:

a plurality of high-resolution radio frequency chains;

a plurality of low-resolution radio frequency chains;

one or more processors; and

one or more non transitory memories storing instructions that, when executed with the one or more processors, cause the base station to perform:

generating, using a first machine-learning model, an initial set of estimated channel information from a first input set of channel information obtained from a set of radio-frequency signals, wherein the first input set of channel information corresponds to measurements from a first plurality of radio-frequency chains, and wherein the estimated set of channel information corresponds to the first plurality of radio-frequency chains and a second plurality of radio-frequency chains;

generating, using a second machine-learning model, a set of estimated channel phases from the initial set of estimated channel information and a second input set of channel information, wherein the second set of input channel information corresponds to measurements from the second plurality of radio-frequency chains; and

combining the initial set of estimated channel information and the set of estimated channel phases to generate an enhanced set of estimated channel information, wherein the enhanced set of estimated channel information corresponds to the first plurality of radio-frequency chains and the second plurality of radio-frequency chains.