Granted Patent B2

US 12,414,146 · App. 17/960,144 · Granted Sep 9, 2025

Inventor: Yuan Fu Huang (New Taipei, TW)

Assignee: Wistron Corporation

H04W72/543G06F18/213G06N3/044H04B7/0667H04L25/03165

Abstract

Claims (62)

1. A scheduling method for beamforming, applicable to a network entity, the scheduling method comprising:

predicting a future location according to at least one past location of a user equipment (UE), wherein the at least one past location is a location of the user equipment of at least one past time point, and the future location is a location of the user equipment at a future time point;

determining a precoder according to the future location, wherein the precoder reflects a downlink channel state at the future time point; and

determining a direction of a beam of a base station at the future time point according to the precoder.

2. The scheduling method for beamforming according to claim 1 , wherein the step of predicting the future location according to the at least one past location of the user equipment comprises:

inputting the at least one past location to a location prediction model, wherein the location prediction model is based on a machine learning algorithm; and

predicting the future location through the location prediction model.

3. The scheduling method for beamforming according to claim 2 , wherein the location prediction model comprises a first recurrent long short-term memory (LSTM) layer and a second recurrent LSTM layer, the first recurrent LSTM layer and the second recurrent LSTM layer respectively comprise a plurality of LSTM model, and the step of predicting the future location through the location prediction model comprises:

transmitting an output to a first LSTM model in the second recurrent LSTM layer and a second LSTM model in the first recurrent LSTM layer through a first LSTM model in the first recurrent LSTM layer; and

transmitting an output to a second LSTM model in the second recurrent LSTM layer through the first LSTM model in the second recurrent LSTM layer.

4. The scheduling method for beamforming according to claim 3 , wherein the location prediction model further comprises a fully-connected (FC) layer, and the step of predicting the future location through the location prediction model comprises:

linearly combining an output of the second recurrent LSTM layer through the fully-connected layer to obtain the future location.

5. The scheduling method for beamforming according to claim 1 , wherein the step of determining the precoder according to the future location comprises:

inputting the future location to a precoding prediction model, wherein the precoding prediction model is based on a machine learning algorithm;

predicting a candidate precoder through the precoding prediction model; and

determining the precoder according to the candidate precoder.

6. The scheduling method for beamforming according to claim 5 , wherein the precoding prediction model comprises a feature extraction layer and a multilayer perceptron (MLP), and the step of predicting the precoder through the precoding prediction model comprises:

obtaining a plurality of random Fourier features (RFFs) from the future location through the feature extraction layer; and

determining the candidate precoder according to the random Fourier features through the multilayer perceptron.

7. The scheduling method for beamforming according to claim 5 , wherein the step of determining the precoder according to the candidate precoder comprises:

in response to having a codebook, generating the precoder according to a one with a highest correlation with the candidate precoder among a plurality of precoding matrices in the codebook; and

in response to not having the codebook, using the candidate precoder as the precoder.

8. A scheduling method for beamforming, applicable to a network entity, the scheduling method comprising:

predicting a future location and a future channel state according to at least one past location and at least one past channel state of a user equipment, wherein the at least one past location is a location of the user equipment of at least one past time point, the at least one past channel state is a downlink channel state at the at least one past time point, the future location is a location of the user equipment at a future time point, and the future channel state is a downlink channel state at the future time point;

determining a precoder according to the future location and the future channel state, wherein the precoder reflects a downlink channel situation at the future time point; and

determining a direction of a beam of a base station at the future time point according to the precoder.

9. The scheduling method for beamforming according to claim 8 , wherein the step of predicting the future location and the future channel state according to the at least one past location of the user equipment comprises:

inputting the at least one past location and the at least one past channel state to a location and channel prediction model, wherein the location and channel prediction model is based on a machine learning algorithm; and

predicting the future location and the future channel state through the location and channel prediction model.

10. The scheduling method for beamforming according to claim 9 , wherein the location and channel prediction model comprises a first recurrent long short-term memory (LSTM) layer and a second recurrent LSTM layer, the first recurrent LSTM layer and the second recurrent LSTM layer respectively comprise a plurality of LSTM models, and the step of predicting the future location through the location and channel prediction model comprises:

transmitting an output to a first LSTM model in the second recurrent LSTM layer and a second LSTM model in the first recurrent LSTM layer through a first LSTM model in the first recurrent LSTM layer; and

transmitting an output to a second LSTM model in the second recurrent LSTM layer through the first LSTM model in the second recurrent LSTM layer.

11. The scheduling method for beamforming according to claim 10 , wherein the location and channel prediction model further comprises a fully-connected layer, and the step of predicting the future location and the future channel state through the location and channel prediction model comprises:

linearly combining an output of the second recurrent LSTM layer through the fully-connected layer to obtain the future location and the future channel state.

12. The scheduling method for beamforming according to claim 8 , wherein the step of determining the precoder according to the future location and the future channel state comprises:

in response to having a codebook, generating the precoder according to a one with a highest correlation with the future channel state among a plurality of precoding matrices in the codebook; and

in response to not having the codebook, determining the precoder according to the future channel state.

13. A network entity, comprising:

a memory, used to store a code; and

a processor, coupled to the memory and configured to load and execute the code to:

predict a future location according to at least one past location of a user equipment, wherein the at least one past location is a location of the user equipment of at least one past time point, and the future location is a location of the user equipment at a future time point;

determine a precoder according to the future location, wherein the precoder reflects a downlink channel state at the future time point; and

determine a direction of a beam of a base station at the future time point according to the precoder.

14. The network entity according to claim 13 , wherein the processor is further configured to:

input the at least one past location to a location prediction model, wherein the location prediction model is based on a machine learning algorithm; and

predict the future location through the location prediction model.

15. The network entity according to claim 14 , wherein the location prediction model comprises a first recurrent long short-term memory (LSTM) layer and a second recurrent LSTM layer, the first recurrent LSTM layer and the second recurrent LSTM layer respectively comprise a plurality of LSTM models, and the processor is further configured to:

transmit an output to a first LSTM model in the second recurrent LSTM layer and a second LSTM model in the first recurrent LSTM layer through a first LSTM model in the first recurrent LSTM layer; and

transmit an output to a second LSTM model in the second recurrent LSTM layer through the first LSTM model in the second recurrent LSTM layer.

16. The network entity according to claim 15 , wherein the location prediction model further comprises a fully-connected layer, and the processor is further configured to:

linearly combine an output of the second recurrent LSTM layer through the fully-connected layer to obtain the future location.

17. The network entity according to claim 15 , wherein the processor is further configured to:

input the future location to a precoding prediction model, wherein the precoding prediction model is based on a machine learning algorithm;

predict a candidate precoder through the precoding prediction model; and

determine the precoder according to the candidate precoder.

18. The network entity according to claim 17 , wherein the precoding prediction model comprises a feature extraction layer and a multilayer perceptron (MLP), and the processor is further configured to:

obtain a plurality of random Fourier features (RFFs) from the future location through the feature extraction layer; and

determine the candidate precoder according to the random Fourier features through the multilayer perceptron.

19. The network entity according to claim 18 , wherein the processor is further configured to:

in response to having a codebook, generate the precoder according to a one with a highest correlation with the candidate precoder among a plurality of precoding matrices in the codebook; and

in response to not having the codebook, use the candidate precoder as the precoder.

20. The network entity according to claim 13 , wherein the network entity is implemented by at least one of a radio access network intelligent controller (RIC), an open radio access network central unit (O-CU), and an open radio access network distributed unit (O-DU) in an open radio access network (O-RAN).

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (13)