Time gaps for artificial intelligence and machine learning models in wireless communication
Aspects of the present disclosure provide apparatuses and methods for providing time gaps that can be used for training, verifying, compiling, and/or switching artificial intelligence (AI)/machine learning (ML) models for use in wireless communication. In the time gaps, a wireless apparatus can deprioritize certain normally or routinely performed processes and functions to spare processing power and/or resources for performing AI/ML model related functions. In one example, an apparatus can provide one or more time gaps associated an AI/ML model used for communication with a network entity. The apparatus can deprioritize, in the one or more time gaps, at least one of uplink (UL) communication or downlink (DL) communication with the network entity. The apparatus can perform, in the one or more time gaps, one or more AI/ML model related processes.
1 . A method of wireless communication at a user equipment (UE), comprising:
providing one or more time gaps configured to enable one or more processes related to an artificial intelligence and machine learning (AI/ML) model to be performed by the UE, the AI/ML model configured to operate with an air interface between a network entity and the UE;
increasing available resources for performing the one or more processes related to the AI/ML model by deprioritizing, in the one or more time gaps, at least one of uplink (UL) communication or downlink (DL) communication with the network entity, wherein the deprioritizing comprises at least one of:
deactivating, in the one or more time gaps, data transmission with one or more secondary serving cells (SCells); or
switching, in the one or more time gaps, an active bandwidth part (BWP) to a dormant BWP or a narrower BWP to reduce at least one of processing or resource usage of the UE; and
performing, in the one or more time gaps, the one or more processes comprising at least one of model training, model inference, model validation, model testing, model compiling, model activation, model deactivation, or model switching.
2 . The method of claim 1 , further comprising at least one of:
performing channel measurement and reporting using the AI/ML model;
performing beam management using the AI/ML model; or
performing positioning using the AI/ML model.
3 . The method of claim 1 , wherein the one or more processes comprise:
training the AI/ML model based on at least one of a physical downlink control channel (PDCCH) transmission, a physical downlink shared channel (PDSCH) transmission, or a downlink reference signal;
refraining from transmitting an acknowledgement of the PDCCH transmission or the PDSCH transmission that is configured for the one or more processes related to the AI/ML model;
refraining from transmitting a measurement report based on the downlink reference signal that is configured for the one or more processes related to the AI/ML model; or
transmitting a report indicating at least one of a selection of an AI/ML model or a reliability of an AI/ML model.
4 . The method of claim 1 , wherein the one or more time gaps are associated with a subset of currently configured carriers or bandwidth parts of the UE.
5 . The method of claim 1 , further comprising at least one of:
receiving a query of the one or more time gaps; or
sending a request of the one or more time gaps.
6 . The method of claim 1 , further comprising:
receiving a radio resource control (RRC) configuration associated with the one or more time gaps, wherein the RRC configuration comprises at least one of DRX configuration, MIMO configuration, or measurement and report configuration.
7 . A user equipment (UE) for wireless communication, comprising:
one or more memories stored with executable code; and
one or more processors coupled to the one or more memories, wherein the one or more processors are configured by the executable code to cause the UE to:
provide one or more time gaps configured to enable one or more processes related to an artificial intelligence and machine learning (AI/ML) model to be performed by the UE, the AI/ML model configured to operate with an air interface between a network entity and the UE;
increase available resources for performing the one or more processes related to the AI/ML model by deprioritizing, in the one or more time gaps, at least one of uplink (UL) communication or downlink (DL) communication with the network entity, wherein the deprioritizing comprises at least one of:
deactivating, in the one or more time gaps, data transmission with one or more secondary serving cells (SCells); or
switching, in the one or more time gaps, an active bandwidth part (BWP) to a dormant BWP or a narrower BWP to reduce at least one of processing or resource usage of the UE; and
perform, in the one or more time gaps, the one or more processes comprising at least one of model training, model inference, model validation, model testing, model compiling, model activation, model deactivation, or model switching.
8 . The UE of claim 7 , wherein the one or more processors are further configured to cause the UE to, at least one of:
perform channel measurement and reporting using the AI/ML model;
perform beam management using the AI/ML model; or
perform positioning using the AI/ML model.
9 . The UE of claim 7 , wherein to deprioritize the at least one of UL communication or DL communication, the one or more processors are further configured to cause the UE to, at least one of:
deactivate, in the one or more time gaps, data transmission with one or more secondary serving cells (SCells); or
switch, in the one or more time gaps, an active bandwidth part (BWP) to a dormant BWP or a narrower BWP to reduce at least one of processing or resource usage of the UE.
10 . The UE of claim 7 , wherein the one or more processes comprise, at least one of:
training the AI/ML model based on at least one of a physical downlink control channel (PDCCH) transmission, a physical downlink shared channel (PDSCH) transmission, or a downlink reference signal;
refraining from transmitting an acknowledgement of the PDCCH transmission or the PDSCH transmission that is configured for the one or more processes related to the AI/ML model;
refraining from transmitting a measurement report based on the downlink reference signal that is configured for the one or more processes related to the AI/ML model; or
transmitting a report indicating at least one of a selection of an AI/ML model or a reliability of an AI/ML model.
11 . The UE of claim 7 , wherein the one or more time gaps are associated with a subset of currently configured carriers or bandwidth parts of the UE.
12 . The UE of claim 7 , wherein the one or more processors are further configured to cause the UE to, at least one of:
receive a query of the one or more time gaps; or
send a request of the one or more time gaps.
13 . The UE of claim 7 , wherein the one or more processors are further configured to cause the UE to:
receive a radio resource control (RRC) configuration associated with the one or more time gaps, wherein the RRC configuration comprises at least one of DRX configuration, MIMO configuration, or measurement and report configuration.
14 . A method of wireless communication at a network entity, comprising:
providing one or more time gaps to a user equipment (UE), the one or more time gaps being configured to enable one or more processes related to an artificial intelligence and machine learning (AI/ML) model to be performed at the UE in a radio access network associated with the network entity, the AI/ML model configured to operate with an air interface between the network entity and the UE;
increasing available resources for performing the one or more processes related to the AI/ML model by deprioritizing, in the one or more time gaps, at least one of uplink (UL) communication or DL communication with the UE, wherein the deprioritizing comprises at least one of:
deactivating data transmission with the UE via one or more secondary serving cells (SCells); or
switching an active bandwidth part (BWP) of the UE to a dormant BWP or a narrower BWP to reduce at least one of processing or resource usage of the UE; and
performing, in the one or more time gaps, the one or more processes comprising at least one of model training, model inference, model validation, model testing, model compiling, model activation, model deactivation, or model switching.
15 . The method of claim 14 , wherein the one or more processes comprise, at least one of:
training the AI/ML model based on at least one of a physical downlink control channel (PDCCH) transmission, a physical downlink shared channel (PDSCH) transmission, or a downlink reference signal;
disregarding, from the UE, an acknowledgement of the PDCCH transmission or the PDSCH transmission related to the training of the AI/ML model;
disregarding, from the UE, a measurement report based on the downlink reference signal related to the training of the AI/ML model; or
receiving a report indicating at least one of a selection of an AI/ML model or a reliability of an AI/ML model.
16 . The method of claim 14 , wherein the one or more time gaps are associated with a subset of currently configured carriers or bandwidth parts of the UE.
17 . The method of claim 14 , further comprising:
transmitting a query of the one or more time gaps; or
receiving a request of the one or more time gaps.
18 . The method of claim 14 , further comprising:
transmitting a radio resource control (RRC) configuration associated with the one or more time gaps, wherein the RRC configuration comprises at least one of DRX configuration, MIMO configuration, or measurement and report configuration.
19 . A network entity for wireless communication, comprising:
one or more memories stored with executable code; and
one or more processors coupled to the memory, wherein the one or more processors are configured by the executable code to cause the network entity to:
provide one or more time gaps to a user equipment (UE), the one or more time gaps being configured to enable one or more processed related to an artificial intelligence and machine learning (AI/ML) model to be performed at the UE in a radio access network associated with the network entity, the AI/ML model configured to operate with an air interface between the network entity and the UE;
increase available resources for performing the one or more processes related to the AI/ML model by deprioritizing, in the one or more time gaps, at least one of uplink (UL) communication or downlink (DL) communication with the UE, wherein to deprioritize the at least one of UL communication or the DL communication, the one or more processors are further configured to cause the network entity to, at least one of:
deactivate data transmission with the UE via one or more secondary serving cells (SCells); or
switch an active bandwidth part (BWP) of the UE to a dormant BWP or a narrower BWP to reduce at least one of processing or resource usage of the UE; and
perform, in the one or more time gaps, the one or more processes comprising at least one of model training, model inference, model validation, model testing, model compiling, model activation, model deactivation, or model switching.
20 . The network entity of claim 19 , wherein the one or more processes comprise, at least one of:
training the AI/ML model based on at least one of a physical downlink control channel (PDCCH) transmission, a physical downlink shared channel (PDSCH) transmission, or a downlink reference signal;
disregarding, from the UE, an acknowledgement of the PDCCH transmission or the PDSCH transmission related to the training of the AI/ML model;
disregarding, from the UE, a measurement report based on the downlink reference signal related to the training of the AI/ML model; or
receiving a report indicating at least one of a selection of an AI/ML model or a reliability of an AI/ML model.
21 . The network entity of claim 19 , wherein the one or more time gaps are associated with a subset of currently configured carriers or bandwidth parts of the UE.
22 . The network entity of claim 19 , wherein the one or more processors are further configured to cause the network entity to:
transmit a query of the one or more time gaps; or
receive a request of the one or more time gaps.
23 . The network entity of claim 19 , wherein the one or more processors are further configured to cause the network entity to:
transmit a radio resource control (RRC) configuration associated with the one or more time gaps, wherein the RRC configuration comprises at least one of DRX configuration, MIMO configuration, or measurement and report configuration.