Federated learning across UE and RAN

1. An apparatus for a 5 th generation NodeB (gNB), the apparatus comprising:

processing circuitry to configure the gNB to:

transmit, to a user equipment (UE), Artificial Intelligence/Machine Learning (AI/ML) service information via at least one of a system information block (SIB) or dedicated radio resource control (RRC) signaling;

receive, from the UE after transmission of the AI/ML service information, a service request for an AI/ML model and local AI/ML capability of the UE;

determine whether the AI/ML model meets pre-defined performance criteria of an AI/ML use case;

in response to a determination that the AI/ML model meets the pre-defined performance criteria, transmit, to the UE, the AI/ML model;

receive, from the UE, training model updated parameters based on local training by the UE after reception of the AI/ML model;

aggregate the training model updated parameters from a plurality of UEs; and

update the AI/ML model based on the aggregated training model updated parameters; and

a memory configured to store the AI/ML model.

2. The apparatus of claim 1 , wherein the processing circuitry further configures the gNB to:

select the UE as a candidate UE for federated learning for joint training of the AI/ML model by the gNB and the UE; and

transmit, to the UE, a federated learning request, reception of the service request in response to acceptance by the UE of the federated learning request.

3. The apparatus of claim 1 , wherein the AI/ML service information includes a list of use cases and AI/ML models and minimum AI/ML capabilities to use each of the AI/ML models.

4. The apparatus of claim 1 , wherein the processing circuitry is further configured to select and train a suitable AI/ML model with stored historical data and past local AI/ML models reported previously by other UEs, as well as performance criteria for each supported use case of each of the AI/ML models.

5. The apparatus of claim 1 , wherein to transmit the AI/ML model to the UE, the processing circuitry further configures the gNB to:

determine whether the UE has previously received the AI/ML model; and

in response to a determination that the UE has previously received the AI/ML model, transmit, to the UE, updated model parameters without an entire AI/ML model package.

6. The apparatus of claim 1 , wherein:

the plurality of UEs include cell-centered UEs and cell-edge UEs, and

the processing circuitry is disposed in at least one of a gNB distributed unit or a gNB central unit.

7. The apparatus of claim 1 , wherein the processing circuitry is configured to:

determine beam management for the UE through a combination of time traces of measurements of a serving cell of the UE and neighbor cell Reference Signal Received Power (RSRP), with time traces of UE-side beam direction, mobility of the UE, and orientation information of the UE; and

aggregate AI/ML-based beam tracking models from the plurality of UEs to construct an AI/ML beam tracking model that is trained using past moving trajectories of the plurality of UEs within the serving cell.

8. An apparatus for a user equipment (UE), the apparatus comprising:

processing circuitry configured to:

receive, from a central server in a 5 th generation (5G) network, AI/ML service information via at least one of a system information block (SIB) or dedicated radio resource control (RRC) signaling;

transmit, to the central server after reception of the AI/ML service information, a service request for an Artificial Intelligence/Machine Learning (AI/ML) model and local AI/ML capability of the UE;

receive, from the central server in response to transmission of the service request and local AI/ML capability, the AI/ML model, the AI/ML model based on input from a plurality of UEs served by the central server;

obtain data for training the AI/ML model and train the AI/ML model based on the data; and

transmit, to the central server, training model updated parameters for the AI/ML model based on the training for the central server to update the AI/ML model; and

a memory configured to store the AI/ML model.

9. The apparatus of claim 8 , wherein

prior to transmission of the service request, the processing circuitry further configures the UE to select, from the AI/ML service information, the AI/ML model to request using the service request or a service subscription, and

the AI/ML service information includes a list of use cases and AI/ML models and minimum AI/ML capabilities to use each of the AI/ML models.

10. The apparatus of claim 9 , wherein the processing circuitry further configures the UE to:

determine whether the UE is able to meet a minimum AI/ML capability for each AI/ML model; and

transmit the service request or service subscription to the central server in response to a determination that the UE is able to meet the minimum AI/ML capability for the AI/ML model provided in the service request.

11. The apparatus of claim 8 , wherein the processing circuitry is further configured to:

receive, from the central server, a federated learning request for joint training of the AI/ML model by the central server and the UE; and

transmit the service request to the central server in response to acceptance of the federated learning request.

12. The apparatus of claim 8 , wherein the processing circuitry is further configured to:

train the AI/ML model when in an RRC_IDLE or RRC_INACTIVE state; and

after training the AI/ML model, report updated parameters of the AI/ML model when in an RRC_CONNCTED state.

13. The apparatus of claim 8 , wherein the processing circuitry further configures the UE to receive from the central server in response to the service request:

an entire AI/ML model package if the UE has not previously received the AI/ML model, or

updated model parameters of the AI/ML model without the entire AI/ML model package if the UE has previously received the AI/ML model.

14. The apparatus of claim 8 , wherein:

the central server comprises a location management function (LMF);

the AI/ML model is configured to predict positioning of the UE; and

the processing circuitry is configured to:

train the AI/ML model based on at least one of Reference Signal Received Power (RSRP) or Reference Signal Time Difference (RSTD) measurement of signals from a 5 th generation NodeB (gNB), and

use the AI/ML model in addition to a positioning algorithm calculation to determine an estimate of location of the UE.

15. The apparatus of claim 8 , wherein:

the central server comprises a 5 th generation NodeB (gNB);

the AI/ML model is configured to predict channel estimation and link adaption for the UE; and

the processing circuitry is configured to adjust communication parameters based on the AI/ML model in response to a sudden change in channel conditions.

16. The apparatus of claim 15 , wherein the sudden change in channel conditions occurs as a result of a scenario that has not previously occurred to the UE but have occurred to at least some of the plurality of UEs.

17. A non-transitory computer-readable storage medium that stores instructions for execution by one or more processors of a 5 th generation NodeB (gNB), the one or more processors to configure the gNB to, when the instructions are executed:

transmit, to a user equipment (UE), Artificial Intelligence/Machine Learning (AI/ML) service information via at least one of a system information block (SIB) or dedicated radio resource control (RRC) signaling;

receive, from the UE after transmission of the AI/ML service information, a service request for an AI/ML model and local AI/ML capability of the UE;

receive, from a user equipment (UE), a service request for an Artificial Intelligence/Machine Learning (AI/ML) model and local AI/ML capability of the UE;

transmit, to the UE, the AI/ML model;

receive, from the UE, training model updated parameters based on local training by the UE after reception of the AI/ML model;

aggregate the training model updated parameters from a plurality of UEs;

update the AI/ML model based on the aggregated training model updated parameters; and

transmit, to the UE, the updated AI/ML model.

18. The medium of claim 17 , wherein the instructions, when executed, further cause the one or more processors to configure the gNB to:

select the UE as a candidate UE for federated learning for joint training of the AI/ML model by the gNB and the UE; and

transmit, to the UE, a federated learning request, reception of the service request in response to acceptance by the UE of the federated learning request.

19. The medium of claim 17 , wherein the instructions, when executed, further cause the one or more processors to configure the gNB to:

select the UE as a candidate UE for federated learning for joint training of the AI/ML model by the gNB and the UE; and

transmit, to the UE, a federated learning request, reception of the service request in response to acceptance by the UE of the federated learning request.

20. The medium of claim 17 , wherein the instructions, when executed, further cause the one or more processors to select and train a suitable AI/ML model with stored historical data and past local AI/ML models reported previously by other UEs, as well as performance criteria for each supported use case of each of the AI/ML models.