Processing timeline considerations for channel state information
A first wireless device, such as a user equipment, generates a message indicating a processing time for at least one of training a neural network for channel state information (CSI) derivation or for reporting the CSI based on a trained neural network. The first wireless device transmits the message indicating the processing time to a second wireless device. The second wireless device may be a network entity, such as a base station, a transmission reception point, or another UE.
1 . An apparatus for wireless communication at a user equipment (UE), comprising:
memory; and
at least one processor coupled to the memory and configured to:
determine, at the UE, a processing time to train a machine learning (ML) or neural network (NN) model at the UE for channel state information (CSI) derivation; and
transmit a message that includes an indication of the determined processing time to a first network entity, wherein the at least one processor is further configured to:
report a first processing time and a second processing time;
receive a configuration to use the first processing time or the second processing time; and
apply the first processing time or the second processing time based on a power saving feature of the UE.
2 . The apparatus of claim 1 , wherein the first network entity is a base station, a transmission reception point (TRP), or another UE and the apparatus further includes:
at least one antenna; and
a transceiver coupled to the at least one antenna and the at least one processor.
3 . The apparatus of claim 1 , wherein the processing time corresponds to an amount of time between a first time associated with a reference signal used for training in the ML or NN model and a second time at which the UE has successfully trained the ML or NN model.
4 . The apparatus of claim 3 , wherein the ML or NN model is successfully trained when the UE is able to report back the CSI trained by the ML or NN model, or is able to use trained weights of the ML or NN model to achieve an accuracy or a quality of service (QoS).
5 . The apparatus of claim 1 , wherein the processing time corresponds to a time between reception of a command to train the ML or NN model and completion of the training of the ML or NN model.
6 . The apparatus of claim 1 , wherein the at least one processor is further configured to:
receive a configuration from a second network entity at least based on the processing time; and
transmit channel state information (CSI) to the second network entity based on the configuration.
7 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on at least one of:
a number of layers in the ML or NN model,
a number of weights in the ML or NN model, or
a type of one or more layers of the ML or NN model.
8 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on at least one of quasi co-location (QCL) information or a model state indication from a previously trained model.
9 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on an amount of models or layers to be trained.
10 . The apparatus of claim 9 , wherein the processing time is further based on whether a single ML or NN model or multiple models are to be trained simultaneously.
11 . The apparatus of claim 10 , wherein the processing time is based on training the multiple models are trained simultaneously based on concurrent training in at least one of:
a same component carrier,
a same band,
a same bandwidth part,
a same band combination,
a same frequency range,
a same slot,
a same subframe, or
a same frame,
wherein the multiple models are trained simultaneously until the UE responds with a complete training message.
12 . The apparatus of claim 10 , wherein the processing time is further based on whether a single layer or multiple layers of the ML or NN model are to be trained simultaneously.
13 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on a sequence order of multiple layers of the ML or NN model.
14 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on a type of wireless signal procedure performed by the ML or NN model, the type of the wireless signal procedure including at least one of:
channel state information determination,
demodulation,
positioning determination, or
waveform determination.
15 . The apparatus of claim 1 , wherein the processing time for training the ML or NN model is based on an accuracy level.
16 . The apparatus of claim 1 , wherein the message further indicates another processing time for reporting the CSI that is based on at least one of:
an encoder output vector,
an encoder input vector,
one of more vectors determined at the ML or NN model,
a number of layers in the ML or NN model,
a first number of elements in an input of the ML or NN model,
a second number of elements in an output of the ML or NN model,
a third number of elements in an intermediate vector of the ML or NN model,
a layer type of one or more layers of the ML or NN model,
an amount of models for overlapped reporting, or
a sequence order of multiple layers of the ML or NN model.
17 . The apparatus of claim 1 , wherein the processing time is for at least one of:
a bandwidth part,
a numerology,
a component carrier,
a band,
a band combination,
a frequency range, or
one or more timeline factor, and
wherein the one or more timeline factor includes at least one of:
a layer,
a layer type,
a combination of layers,
an input vector length,
an output vector length,
an intermediate vector length,
a number of layers, or
a sequence of layers.
18 . A method of wireless communication at a user equipment (UE), comprising:
determining, at the UE, a processing time to train a machine learning (ML) or neural network (NN) model at the UE for channel state information (CSI) derivation; and
transmitting a message indicating the determined processing time to a first network entity,
wherein the method further comprises:
reporting a first processing time and a second processing time;
receiving a configuration to use the first processing time or the second processing time; and
applying the first processing time or the second processing time based on a power saving feature of the UE.
19 . An apparatus for wireless communication, comprising:
memory; and
at least one processor coupled to the memory and configured to:
receive a message that includes an indication of a processing time from a user equipment (UE), wherein the indication indicates the processing time to train a machine learning (ML) or neural network (NN) model at the UE for channel state information (CSI) derivation;
transmit a configuration to the UE based on the processing time; and
receive the CSI from the UE after transmission of the configuration,
wherein the at least one processor is further configured to:
receive a first processing time and a second processing time from the UE; and
configure the UE to use the first processing time or the second processing time.
20 . The apparatus of claim 19 , wherein the wireless communication is at a base station, a transmission reception point (TRP), or another UE, the apparatus further comprising:
at least one antenna; and
a transceiver coupled to the at least one antenna and the at least one processor.
21 . The apparatus of claim 19 , wherein the processing time corresponds to an amount of time between a first time associated with a reference signal used for training in the ML or NN model and a second time at which the UE has successfully trained the ML or NN model, and wherein the ML or NN model is successfully trained when the UE is able to report back the CSI trained by the ML or NN model, or is able to use trained weights of the ML or NN model to achieve an accuracy or a quality of service (QoS).
22 . The apparatus of claim 19 , wherein the processing time corresponds to a time between reception of a command to train the ML or NN model and completion of the training of the ML or NN model.
23 . The apparatus of claim 19 , wherein the processing time for training the ML or NN model is based on at least one of:
a number of layers in the ML or NN model,
a number of weights in the ML or NN model,
a type of one or more layers of the ML or NN model, quasi co-location (QCL) information from a previously trained ML or NN model, or
a ML or NN model state indication from the previously trained ML or NN model.
24 . The apparatus of claim 19 , wherein the processing time for training the ML or NN model is based on at least one of an amount of models or layers to be trained and whether a single ML or NN model or multiple models are to be trained simultaneously, wherein the multiple models s are trained simultaneously based on concurrent training in at least one of:
a same component carrier,
a same band,
a same bandwidth part,
a same band combination,
a same frequency range,
a same slot,
a same subframe, or
a same frame.
25 . The apparatus of claim 19 , wherein the processing time is based on at least one of whether a single layer or multiple layers of the ML or NN model are to be trained simultaneously, a sequence order of the multiple layers of the ML or NN model, a type of wireless signal procedure performed by the ML or NN model, or an accuracy level.
26 . The apparatus of claim 19 , wherein the message further indicates another processing time for reporting the CSI that is based on at least one of:
an encoder output vector,
an encoder input vector,
one of more vectors determined at the ML or NN model,
a number of layers in the ML or NN model,
a first number of elements in an input of the ML or NN model,
a second number of elements in an output of the ML or NN model,
a third number of elements in an intermediate vector of the ML or NN model,
a layer type of one or more layers of the ML or NN model,
an amount of models for overlapped reporting, or
a sequence order of multiple layers of the ML or NN model.
27 . The apparatus of claim 19 , wherein the processing time for at least one of:
a bandwidth part,
a numerology,
a component carrier,
a band,
a band combination,
a frequency range, or
one or more timeline factor, and, wherein the one or more timeline factor includes at least one of:
a layer,
a layer type,
a combination of layers,
an input vector length,
an output vector length,
an intermediate vector length,
a number of layers, or
a sequence of layers.
28 . A method of wireless communication, comprising:
receiving a message that includes an indication of a processing time from a user equipment (UE), wherein the indication indicates the processing time to train a machine learning (ML) or neural network (NN) model at the UE for channel state information (CSI) derivation;
transmitting a configuration to the UE based on the processing time; and
receiving the CSI from the UE after transmission of the configuration,
wherein the method further comprises:
receiving a first processing time and a second processing time from the UE; and
configuring the UE to use the first processing time or the second processing time.