Performance target sharing for machine learning models

Abstract

According to some embodiments, a method is for two-sided machine learning (ML) model sequential training. The two-sided ML model comprises a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model. The method is performed by the second node and comprises: receiving a ML model training dataset or ML model training parameter from a network node for training the second node-side ML model; receiving one or more performance targets associated with the second node-side ML model from the network node, wherein the one or more performance targets are based on training of the first node-side ML model using the testing dataset; and training the second node-side ML model based on the training dataset, a testing dataset, and the one or more performance targets.

Description

PERFORMANCE TARGET SHARING FOR MACHINE LEARNING MODELSTECHNICAL FIELD

[0001] Embodiments of the present disclosure are directed to wireless communications and, more particularly to performance target sharing for machine learning (ML) models.BACKGROUND

[0002] Artificial intelligence (Al) and machine learning (ML) have been investigated, both in academia and industry, as promising tools to optimize the air interface design in wireless communication networks. One example of AI / ML implementation for the physical layer (Al PHY) is using an AI / ML autoencoder to improve the channel compression accuracy and / or to reduce the channel state information (CSI) feedback overhead. This topic has been discussed throughout Third Generation Partnership Project (3GPP) Release 18 and continues to be discussed in Release 19.

[0003] In the legacy mechanism (i.e., non-A I / ML-bascd CSI compression), the user equipment (UE) can be configured to report a suggested precoder to the network. The precoder is identified by a precoder matrix indicator (PMI). The PMI is based on a measured CSI reference signal (CSI-RS) and sent to the network as a CSI report, based on a certain mechanism, the codebook. The codebook generally defines how the UE arranges the reported bits based on (the number of) beams and taps selected by the UE to be reported to the network and how the UE quantizes the precoding matrix.

[0004] In AI / ML-based CSI compression, an AI / ML-based autoencoder (AE) replaces, at least part of, the legacy mechanism. AEs can have different architectures. For example, AEs can be based on dense neural networks (NNs), multi-dimensional convolution NNs, variational, recurrent NNs, transformer networks, or any combination thereof. However, all AE architectures possess an encoder-bottleneck-decoder structure illustrated in FIGURE 1.

[0005] FIGURE 1 illustrates a fully connected autoencoder. The codeword’s size (denoted by Y in FIGURE 1) of an AE is smaller than the input data’s size (X in FIGURE 1). The AE encoder thus reduces the dimensionality of the input features X down to Y . The decoder part of the AE tries to invert the encoder and reconstruct X with minimal error, according to a predefined loss function, also known at target function in the general optimization literature.

[0006] FIGURE 2 illustrates how an AE might be used for AEML-enhanced CSI reporting in New Radio (NR). The UE measures the channel in the downlink using CSI-RS. The UEP113381WO01 PCT APPLICATION 2 of 51estimates the channel for each subcarrier (SC) from each base station TX antenna to each UE RX antenna. The estimate can be viewed as a three-dimensional (3D) channel matrix. The 3D channel matrix represents the multiple-input multiple-output (MIMO) channel estimated over several SCs and is input to the encoder. However, there are different architectures where a processed version of the 3D MIMO channel, or a processed subset of the information, e.g., singular vectors of the MIMO channel, is input to the encoder.

[0007] The AE encoder is implemented in the UE, and the AE decoder is implemented in the network, denoted BS for base station in FIGURE 2. The output of the AE encoder is signaled from the UE to the network over the uplink. The codeword can be viewed as a learned latent representation of the channel. Properties of the data (e.g., CSI-RS channel estimates), the channel size, uplink feedback rate, and hardware limitations of the encoder and decoder need to be considered when optimizing the AE’s architecture.

[0008] The weights and biases of an AE (with a fixed architecture) are trained to minimize the reconstruction error (the error between the input X and output X) on training datasets. For example, the weights and biases can be trained to minimize the mean squared error (MSE) (mean). To achieve good performance during live operation, the training data set should represent the actual data the AE will encounter during live operation.

[0009] In two-sided CSI compression, the output of the UE-side encoder needs to be communicated over the air interface to the gNB decoder with the assigned CSI reporting payload and, therefore, needs to be quantized to a finite number of bits (e.g., 1-4 bits per encoder output’s neuron) to obtain an efficient transmission, as shown in FIGURE 3. FIGURE 3 illustrates a quantization operation at the output of the encoder to fit the CSI payload over the air interface.

[0010] Accordingly, a quantization layer is connected to the output of the encoder or directly included in the encoder. In an example, the quantization layer may implement scalar quantization, which quantizes the output of each neuron of the encoder output layer (the bottleneck layer of AE) to generate bits to fit the CSI reporting payload in the uplink control information (UCI). Other quantization methods, e.g., vector quantization, may also be used.

[0011] A proper pre-processing on the input to the encoder can greatly reduce the size and complexity of designing and / or training an AI / ML model, and improve the scalability and transferability of the model. In addition, pre-processing may reduce the need for multipleP113381WO01 PCT APPLICATION 3 of 51models depending on bandwidth variation and variation in the number of antenna ports at the gNB. By using pre-processing, instead of directly compressing the channel (i.e., with dimensions of RX x TX x SC as in FIGURE 2, the channels are first processed into another representation.

[0012] FIGURE 4 illustrates pre-processing the channel into eigenvector before use as inputs to the encoder inputs. One example of the pre-processing may include transforming the channel into eigenvectors (FIGURE 4). Here, the UE may conduct the following steps:1. Compute the covariance matrix of the channel and extract the relevant eigenvectors.2. The covariance matrix is summed over 4 “f-units” to get 13 “sub-chunks” (subbands) in frequency.3. For each of the 13 averaged covariance matrices, compute an eigen-decomposition and extract the 4 eigenvectors corresponding to the 4 largest eigenvalues.4. Normalize the phase and magnitude of the eigenvectors.5. The number of eigenvectors to feedback should be the same for all sub-chunks and depends on the rank hypothesis testing with a value between 1 and 4.

[0013] In another example, pre-processing may include transforming the channel into the beam-delay domain. The feature extraction for beam-delay reduced eigenvector-based feedback is illustrated in FIGURE 5. This is also referred to as a “W2 compressor” because the feature extraction is similar to the standardized 3GPP procedures and the jargon for the matrix left to compress for the AI / ML is “W2”.

[0014] FIGURE 5 illustrates pre-processing the channel into beam delay domain (W2-like format) before use as inputs to the encoder. The steps are as follows:1. The UE selects an orthogonal basis, one of a set of oversampled / rotated, spatial domain, discrete Fourier transform (DFT) bases. In that basis the UE selects L vectors. These vectors represent the spatial domain (SD) basis DFT. This basis is wideband applicable, i.e., valid for all subbands, and applicable for both polarizations. This selection may happen jointly, e.g., the UE does a spatial domain DFT on the 32x4 (TX x RX) matrix, per resource block (RB), and selects the L strongest beams out of 16 (for one polarization). The beam-space channel is computed by multiplying the channel with the selected SD basis. The covariance of the beam-space channel is summed over, e.g., 4 RBs to produce a covariance matrix for each subband.P113381WO01 PCT APPLICATION 4 of 512. For each covariance matrix (per subband) the UE extracts a number of eigenvectors and may select the rank, i.e. number of layers.3. The UE does a frequency domain DFT per layer, transforming to a delay domain, whereafter it selects the M strongest taps. The resulting tensor of dimensions 2L x number of layers x M is referred to as the linear combination coefficients, in 3GPP jargon “W2”. The W2 matrix can be used to reconstruct the UE suggested precoding matrices.4. The data is used as input in the AI / ML model. This could be the raw linear combination coefficients, or it could be enhanced with information about the selected beams and taps, noise levels, etc.

[0015] One mechanism to train a two-sided model is sequential training. Using a two-sided Al CSI compression model as an example, sequential training may start from one side, e.g., the network (NW) side, and be sequentially followed by another side (e.g., the UE side). Such a mechanism is referred to as NW-first training and may take the following steps.

[0016] The NW trains a NW nominal encoder and a NW actual decoder jointly with a first dataset. Here, the nominal encoder is only used / obtained during the training and is not used during the actual inference. Note that the nominal encoder may also be used for other life cycle management (LCM) purposes, e.g., for performance monitoring.

[0017] The NW sends the first dataset to the UE with information that enables the UE to train its encoder. The information, for example, may be at least one of: a second dataset, where the second dataset is the output of the NW nominal encoder associated with the first dataset; the NW nominal encoder structure; the NW nominal encoder parameters; and additional information such as NW actual decoder architecture, loss function, hyperparameters used for the training, performance target, etc.

[0018] The UE trains its encoder based on the information obtained from the NW. For example, the following may be done by the UE.

[0019] The UE may train the encoder based on the first dataset and the second dataset. The second dataset may be obtained from the NW or by inputting the first dataset to the NW nominal encoder.

[0020] The UE may train a UE nominal decoder using the second dataset as inputs and the first dataset as the target. The UE then trains an actual encoder based on the UE nominal decoder and the first dataset.P113381WO01 PCT APPLICATION 5 of 51

[0021] The first and the second datasets used by the UE for training may be in a modified version. For example, the second dataset may be the quantized version of the NW nominal encoder.

[0022] The sequential training mentioned above shall serve as examples, and other training mechanisms shall apply as well. For example, the UE may receive, on top of the first dataset, information on the NW actual decoder, etc. NW-first training shall also be used as an example, but shall apply to other sequential training mechanisms, e.g., UE-first training.

[0023] There currently exist certain challenges. For example, as mentioned above, one of the additional information that may be transmitted from a first node (e.g., the NW) to a second node (e.g., the UE) is information on the performance target. To make the information useful, however, the details of the design of the performance target sharing shall be described. Such details are currently not available.SUMMARY

[0024] As described above, certain challenges currently exist with performance target sharing for artificial intelligence (AI) / machine learning (ML) models. Certain aspects of the present disclosure and their embodiments may provide solutions to these or other challenges. For example, particular embodiments include mechanisms and configurations for the performance target sharing from a first node to a second node.

[0025] As an example, particular embodiments describe several aspects of the performance target sharing in the two-sided AI / ML-based channels state information (CSI) reporting with sequential training. Some embodiments include association between the performance target with other information shared by a first node to a second node, such as the training dataset, the encoder output associated with the training dataset, and / or the model parameters (and additionally the model structure) of the UE-part model(s) or the NW-part model(s). Some embodiments include the format / definition for the performance target, performance target(s) for multiple configurations, and / or relation between the performance target and performance monitoring.

[0026] According to some embodiments, a method is for two-sided ML model sequential training. The two-sided ML model comprises a first node-side ML model and a second nodeside ML model where the output of the second node-side ML model is input to the first nodeside ML model. The method is performed by the second node and comprises: receiving a MLP113381WO01 PCT APPLICATION 6 of 51model training dataset or ML model training parameter from a network node for training the second node-side ML model; receiving one or more performance targets associated with the second node-side ML model from the network node, wherein the one or more performance targets are based on training of the first node-side ML model using the testing dataset; and training the second node-side ML model based on the training dataset, a testing dataset, and the one or more performance targets.

[0027] In particular embodiments, the method further comprises receiving the testing dataset from the network node fortesting the second node-side ML model.

[0028] In particular embodiments, the one or more performance targets comprise one or more of a second node-side output performance target and an end-to-end performance target. In particular embodiments, the one or more performance targets comprise one or more pairs of a second node-side output performance target and an end-to-end performance target associated with the paired second node-side output performance target.

[0029] In particular embodiments, the one or more performance targets comprise multiple values representing statistics across samples of the testing dataset. For example, the multiple values representing statistics across samples of the testing dataset may comprise wherein the multiple values representing statistics across samples of the testing dataset comprise a mean value and one or more values of one or more specified percentiles.

[0030] In particular embodiments, the testing dataset may comprise multiple sub-datasets and each of the sub-datasets is associated with one or more performance targets.

[0031] In particular embodiments, the training dataset is associated with one or more configuration parameters and each of the one or more performance targets may be associated with a particular configuration parameter or parameter combination.

[0032] In particular embodiments, each of the one or more performance targets may be associated with a particular capability and the second node selects the performance targets that align with a capability of the second node.

[0033] In particular embodiments, receiving the ML model training dataset or the ML model testing dataset comprises receiving an identifier or parameter associated with the dataset and obtaining the data for the dataset based on the identifier or parameter.

[0034] In particular embodiments, the first node comprises a base station and the second node comprises a UE. In particular embodiments, the first node comprises a first UE and the second node comprises a second UE.P113381WO01 PCT APPLICATION 7 of 51

[0035] In particular embodiments, the two-sided ML model comprises an autoencoder / decoder ML model.

[0036] According to some embodiments, a method is for two-sided ML model sequential training. The two-sided ML model comprises a first node-side ML model and a second nodeside ML model where the output of the second node-side ML model is input to the first nodeside ML model. The method is performed by the first node and comprises: training a first nodeside ML model using a ML model training dataset or ML model parameters, a ML model testing dataset, and a nominal second node side ML model; determining one or more performance targets based on training the first node-side ML model; and transmitting a ML model training dataset or ML model parameters for training the second node-side ML model and the one or more performance targets to the second node.

[0037] In particular embodiments, the method further comprises transmitting the ML model testing dataset fortesting the second node-side ML model to the second node.

[0038] According to some embodiments, a network node comprises processing circuitry operable to perform any of the methods of the first node and / or second node described above.

[0039] Also disclosed is a computer program product comprising a non-transitory computer readable medium storing computer readable program code, the computer readable program code operable, when executed by processing circuitry to perform any of the methods performed by the first and / or second node described above.

[0040] Certain embodiments may provide one or more of the following technical advantages. For example, in particular embodiments the performance target may properly serve as guidance for the second node to train its model part, so that the models trained separately at the first node and the second node may be compatible with each other. The performance target assists the second node (or a trainer-node / server associated with the second node) to assess if the trained model(s), e.g., UE-part model(s), have converged in such a way that it is expected to have a reasonable performance without having to test it against a model running on the first node (or a trainer-node / server associated with the first node), e.g., test against a NW-part model. This helps achieve better isolation between UE-side training and NW-side training.P113381WO01 PCT APPLICATION 8 of 51BRIEF DESCRIPTION OF THE DRAWINGS

[0041] For a more complete understanding of the disclosed embodiments and their features and advantages, reference is now made to the following description, taken in conjunction with the accompanying drawings, in which:FIGURE 1 illustrates a fully connected autoencoder;FIGURE 2 illustrates using autoencoder for CSI compression;FIGURE 3 illustrates a quantization operation at the output of the encoder to fit the CSI payload over the air interface;FIGURE 4 illustrates pre-processing the channel into eigenvector before use as inputs to the encode inputs;FIGURE 5 illustrates pre-processing the channel into beam delay domain (W2-like format) before use is as inputs to the encode inputs.;FIGURE 6 illustrates an example communication system, according to certain embodiments;FIGURE 7 illustrates an example user equipment (UE), according to certain embodiments;FIGURE 8 illustrates an example network node, according to certain embodiments; FIGURE 9 illustrates a block diagram of a host, according to certain embodiments; FIGURE 10 illustrates a virtualization environment in which functions implemented by some embodiments may be virtualized, according to certain embodiments;FIGURE 11 illustrates a host communicating via a network node with a UE over a partially wireless connection, according to certain embodiments;FIGURE 12 illustrates a method performed by a wireless device, according to certain embodiments; andFIGURE 13 illustrates a method performed by a network node, according to certain embodiments.DETAILED DESCRIPTION

[0042] As described above, certain challenges currently exist with performance target sharing for artificial intelligence (AI) / machine learning (ML) models. Certain aspects of the present disclosure and their embodiments may provide solutions to these or other challenges. ForP113381WO01 PCT APPLICATION 9 of 51example, particular embodiments include mechanisms and configurations for the performance target sharing from a first node to a second node.

[0043] Particular embodiments are described more fully with reference to the accompanying drawings. Other embodiments, however, are contained within the scope of the subject matter disclosed herein, the disclosed subject matter should not be construed as limited to only the embodiments set forth herein; rather, these embodiments are provided by way of example to convey the scope of the subject matter to those skilled in the art.

[0044] As used herein, the network is mostly assumed to be the first node, and the user equipment (UE) is assumed to be the second node. The part of the two-sided model at the first node for model inference is assumed to be the network (NW)-part model, and the part of the two-sided model at the second node for model inference is assumed to be the UE-part model. This shall be seen as an example and the embodiments shall also be applicable for the other scenarios, e.g., the NW being the second node and the UE node being the first node, or the first node is a UE node, and the second node is another UE node (e.g., for sidelink / device-to-device communication related use cases). In some embodiments, the nodes themselves are not doing the training, but have a training node / server associated with them.

[0045] In addition, the Al channel state information (CSI) compression is described herein as an example use case for particular embodiments. This shall also not limit the implementation of the embodiments to the other two-sided AI / ML model use cases, for example, two-sided AI-based channel coding, two-sided Al-based source coding, etc., and the derivation of such two-sided models such as, joint source channel coding and CSI compression, etc.

[0046] Many examples and embodiments use encoder and decoder as examples for the UE-part model and the NW-part model. This shall not limit the applicability of the embodiments to other UE-part models and NW-part models.

[0047] The term training dataset shall be viewed as a general term for a dataset used during the training. This may include the dataset used for actual training and the dataset for training validation.

[0048] The concept of ‘network (NW)’ and / or a gNB may be understood as a generic network node, a NW-side training entity, a NW-side server, gNB, base station, unit within the base station, relay node, core network node, a core network node, or a device supporting device-to-device (D2D) communication. The node may be deployed in a fifth generation (5G), a sixth generation (6G), or any future network. The concept of “UE” may be understood as a genericP113381WO01 PCT APPLICATION 10 of 51device node, a UE-side training entity, a UE-side server, or a relay node. Moreover, in the description related to compressing with an AI / ML model, although the term AI / ML model uses a single form, it should be understood that it does not prevent the implementation of more than one AI / ML model for the same use case. The UE may either be configured or autonomously switch between the models depending on certain conditions and / or proprietary implementations.

[0049] The following are example embodiments at the UE side. The NW side embodiments may be mirrored from the UE side embodiments.

[0050] Some embodiments include methods implemented in a UE capable of performing two-sided Al CSI compression with sequential training. The method comprises receiving, from a network (NW), at least one of: a first set of information, where the first set of information consists of ground truth, e.g., target-CSIs; a second set of information, where the second set of information consists of associated UE-part model output from a NW-side nominal UE-part model, e.g., encoder outputs from aNW-side nominal encoder; athird set of information, where the third set of information consists of NW-side nominal UE-part model structure and / or model parameters, e.g., NW nominal encoder model structure and / or model parameters; and a fourth set of information, where the fourth set of information consists of NW actual NW-part model structure and / or model parameters, e.g., NW actual decoder model structure and / or model parameters.

[0051] The method further comprises receiving, from a NW, a set of performance targets and performing training of at least the UE-part model based on the received information.

[0052] In particular embodiments, a part of at least the first set of information is a training dataset and the remaining part is a testing dataset. The testing dataset is a dataset used by the NW to obtain the set of performance targets.

[0053] In particular embodiments, the separation of the training dataset and the testing dataset is predetermined in the standard text. In particular embodiments, the separation of the training dataset and the testing dataset is NW configurable.

[0054] In particular embodiments, the performance target is at least one of UE-part model performance target and end-to-end performance target. The set of performance targets may comprise one or more performance target statistics.P113381WO01 PCT APPLICATION 11 of51

[0055] In particular embodiments, the set of performance targets comprise one or more performance targets of the testing dataset associated with or included in the first, second, third, and / or forth set of information.

[0056] In particular embodiments, the set of performance targets comprise one or more performance targets of one or more additional conditions included in the first, second, third, and / or forth set of information.

[0057] In particular embodiments, the set of performance targets is associated with the first, second, third, and / or forth set of information.

[0058] In particular embodiments, the set of performance targets is associated with the testing dataset.

[0059] In particular embodiments, the set of performance targets is associated with the additional conditions.

[0060] Some embodiments are related to the testing dataset determination. To obtain the performance targets, the node applies the trained model (or part of the model) to a testing dataset. It is preferable that the testing dataset between the NW side and the UE side is aligned.

[0061] In some embodiments, a training dataset and a testing dataset are transmitted in different containers. The association between the training dataset and the testing dataset may be achieved, e.g., via ID. For example, the testing dataset ID shall contain the training dataset ID.

[0062] In some embodiments, the training dataset and the testing dataset are transmitted in the same container. In such an approach, the UE needs to separate the datasets into the training (and validation) dataset and the testing dataset.

[0063] In some embodiments, the determination of whether a sample is a training dataset or a testing dataset may follow a predetermined rule. For example, the last X percent of the dataset may serve as the testing dataset and the remaining samples serve as the training (and validation) dataset. In an example, the value of X may be predetermined in the standard or may be configurable by the NW.

[0064] In some embodiments, a label may be assigned to each dataset to determine whether a sample is a training dataset or a testing dataset. For example, one bit indication for each sample may be used, where a first value (e.g., 0) represents the training dataset and a second value (e.g., 1) represents the testing dataset. When the validation dataset is also explicitlyP113381WO01 PCT APPLICATION 12 of 51configured / predetermined, two-bit indications may be used, e.g., 00, 01, and 10 may represent a training, validation, and testing dataset, respectively.

[0065] In some embodiments, a training dataset and a testing dataset are generated from channel models that are parametrized in some way. The transfer of datasets for training and testing may then be implicit, in that the NW only signals those parameters to the UE, and the UE (or UE-side training node / server) has to generate the data samples from the channel model or collect the data samples from the field. In a special case, the parameters for generating / collecting the training dataset and testing dataset are signaled from the NW to UE in the form of an ID, which is used for the initial model training collaboration or offline alignment between NW and UE.

[0066] In another set of examples, a reference UE-part model structure and / or a reference NW-part model structure is / are standardized, or a reference UE-part model structure and / or a reference NW-part model structure is / are pre-aligned between the NW and the UE. Thus, a set of UE-part model parameters and / or NW-part model parameters based on the reference UE-part model structure and / or the reference NW-part model structure are transmitted from NW to UE together with a testing dataset to assist the UE in training its UE-part model(s).

[0067] In some embodiments, a set of model parameters and a set of testing datasets are transmitted from the NW (the first node) to the UE (the second node) in different containers. The association between the model parameter set and the testing dataset may be achieved, e.g., via ID. For example, the testing dataset ID shall contain the model parameter set ID.

[0068] In some embodiments, a set of model parameters and a set of testing datasets are transmitted from the NW (the first node) to the UE (the second node) in the same container.

[0069] In some embodiments, a testing dataset is generated from channel models or collected from the field, that are parametrized in some way. The transfer of datasets for testing may then be implicit, in that the NW only signals those parameters to the UE and the UE (or UE-side training node / server) generates the data samples from the channel model or collects the data samples from the field. In a special case, the parameters for generating / collecting the testing dataset are signaled from the NW to UE in the form of an ID, which is used for initial model training collaboration or offline alignment between the NW and UE.

[0070] Some embodiments are related to performance target’s input / output. In two-sided AI / ML models, performances may come in two flavors, i.e., the UE-part model output performance and the end-to-end two-sided model pair performance. A UE-part model outputP113381WO01 PCT APPLICATION 13 of 51performance (e.g., encoder output performance for the CSI compression use case shown in Figure 3) is measured directly from the UE-part model outputs. The end-to-end performance (e.g., the encoder-decoder model pair performance for the CSI compression use case shown in Figure 3), on the other hand, is measured from the performance of the UE-part model and NW-part model chains. Some embodiments include variations. For example, the encoder output performance for the CSI compression use case may be calculated from the quantized version of the encoder outputs. Similarly, the end-to-end performance may also include quantization after the encoder output, resulting in the encoder-quantization-decoder chain. Similarly, the performance target may also be in two flavors, i.e., the encoder output performance target and the end-to-end performance target.

[0071] The following examples use encoder as an example of the UE-part model and decoder as an example of the NW-part model to describe the embodiments.

[0072] Some embodiments include an encoder output performance target. For the encoder output performance target, the performance target, in one embodiment, may be derived by comparing the nominal encoder output every P epoch to a final version of the nominal encoder, which is obtained, e.g., in the nth epoch. Here, the final version may refer to the final version of one training session. The nominal encoder may be updated / retrained, e.g., when a new dataset is available, etc. In one example, the NW may conduct a model training to obtain a nominal encoder and an actual decoder. During the training, the NW may save the encoder parameters every P epochs. The NW may then calculate the similarity between those encoder outputs to the encoder outputs in epoch n. The NW may then also calculate the similarity between the decoder output (i.e., the reconstructed CSI) to the encoder input (i.e., the target-CSI) between a chain of encoder n - decoder n to a chain of encoder n - decoder n, where n e kP, where k is an integer. The NW may then derive the encoder output similarity that still can be accommodated by the NW to obtain a certain end-to-end performance threshold, e.g., performance of which it is still beneficial for the UE to be configured with the AI / ML-based CSI compression. Such an encoder output similarity value is then used as the performance target and is shared to the UE.

[0073] In some embodiments, the NW may also train a decoder with multiple encoders. For example, the NW may train a decoder with a preferred encoder and a basic encoder. In an example, the preferred encoder is the encoder that is deemed to be the most suitable with the decoder architecture. The basic encoder may be, for example, an encoder with a low complexityP113381WO01 PCT APPLICATION 14 of 51and is expected to perform lower compared to the preferred encoder (but still may be accommodated by the decoder to obtain sufficient end-to-end performance). The NW may then calculate the difference between the basic encoder output and the preferred encoder output and set it as the performance target for the UE. In one example, the preferred encoder is the NW nominal encoder that is trained jointly with the NW actual decoder, and the basic encoder is trained in a sequential manner against the frozen (i.e., non-updating) NW actual decoder. In some embodiments, instead of training a basic encoder, the NW may train an encoder with the same / similar model structure as the preferred encoder but is trained under different hyperparameters, such as the learning rate, initialization, drop-out rate, under different numbers of training dataset samples, etc.

[0074] In some embodiments, the encoder output performance target is derived based on the error / noise level. The NW may train a decoder with a nominal encoder first, which gives a first end-to-end performance. Then, the NW adds errors / noises on the nominal encoder output samples, where the errors / noises follow a certain distribution, e.g., Gaussian, and the distorted nominal encoder output samples are fed to the decoder to generate the corresponding “distorted” decoder output samples. The NW derives a second end-to-end performance by comparing the “distorted” decoder output samples with the ground-truth target CSI samples. The performance degradation due to “distorted” / imperfect encoder output can be calculated as the difference between the first end-to-end performance and the second end-to-end performance. By doing so, the NW derives an error / noise level for the nominal encoder output that can be tolerated so that the end-to-end performance degradation is within an acceptable range.

[0075] In the above embodiments, the encoder output performance target serves as guidance for the UE to train an encoder for achieving a certain performance threshold. To obtain the optimum performance, the UE shall simply strive to train its encoder to obtain a similarity value of 0 (when normalized mean square error (NMSE) is used) or 1 (e.g., if squared generalized cosine similarity (SGCS) is used).

[0076] Some embodiments include an end-to-end performance target. For end-to-end performance target, the performance target, in one embodiment, may be obtained from the performance of the chain of NW nominal UE-part model, e.g., nominal encoder, and NW actual NW-part model, e.g., NW actual decoder, for a given testing dataset. In some embodiments, the NW may add a tolerance factor, ft. to obtain the performance target. For example, theP113381WO01 PCT APPLICATION 15 of 51performance target, T, may be defined as E — {J. where E is the (expected) performance of the NW nominal UE-part model, e.g., nominal encoder, and NW actual NW-part model, e.g., NW actual decoder, for a given testing dataset.

[0077] In one example, the tolerance factor ft may be derived by adding errors / noise to the input of the NW nominal UE-part model and / or to the input of the NW actual NW-part model. The noise may follow some given distribution or error model, e.g., Gaussian or Wishart. In a related example, the tolerance factor ft may be derived by adding errors / noise to the trained AI / ML weights of the NW nominal UE-part model and / or the NW actual NW-part model.

[0078] In another example, ft may also be derived by evaluating the end-to-end performance of the chain of an encoder trained in the nth epoch and the NW actual decoder, or a basic encoder and the NW actual decoder, or an encoder trained with different hyperparameters and the NW actual decoder, or an encoder trained with a different number of training data samples, and the NW actual decoder.

[0079] In some embodiments, both T and E are received by the UE from the NW.

[0080] Some embodiments include multiple performance target types. In an alternative, e.g., to offer flexibility toward UE to train its encoder, multiple performance target types may be supported (i.e., both the UE-part model output performance target and the end-to-end performance target).

[0081] In some embodiments, the performance target for a given testing dataset consists of a plurality of UE-part model output performance targets and their associated end-to-end performance targets (e.g., AL pairs of targets). For example, the following may be informed to the UE for a given testing dataset.testingDatcisettarget _1 {UEPartOutTarget 1 , end2endTarget_l},target_2 {UEPartOutTarget 2, end2endTarget_2},target M {UEPartOutTarget M, end2endTarget_M},

[0082] The testing dataset (and its corresponding performance target) may not necessarily be in a standalone container. Rather, the testing dataset may come from the same container as the dataset that was shared from the NW (the first node) to UE (the second node) and shall be used by the UE (the second node) for assisting its UE-part model training. Receiving information on the performance targets, the UE may then determine to what extent the UE shall conductP113381WO01 PCT APPLICATION 16 of 51the training. For example, the UE may train its UE-part model, e.g., encoder, to achieve at least the end2endTarget under target_2. As the UE trains its UE-part model directly (trains from the encoder inputs and expected encoder outputs pairs and does not first train a nominal NW-part model, e.g., decoder), the UE could not estimate / predict the end-to-end performance by the UE-part model training alone. Here, the UE may check the UE-part model output target associated with the end-to-end target that the UE is aiming for.

[0083] In some embodiments, the UE may request any one of the performance targets from the NW, but not a pair of targets.

[0084] Some embodiments are related to performance target statistics. In AI / ML, the testing dataset used to test the AI / ML model consists of multiple samples. In some embodiments, a single performance target across testing data samples is shared with the UE for a given testing dataset. The single performance target may be in terms of mean value, median value, or an X-th percentile of one or more performance metrics across data samples in the testing dataset.

[0085] In some embodiments, the performance targets consist of multiple values representing statistics across samples. For example, the performance targets may comprise the mean value and the X-th percentile of one or more performance metrics across data samples in the testing dataset. In another example, the performance targets comprise multiple X-th percentiles for different values of X. In yet another example, the performance targets comprise the percentage of samples whose corresponding performance metric value is within a certain value range, e.g., 5% of testing data samples have SGCS values below 0.6, 85% of testing data samples have SGCS values between 0.6 and 0.8, and 10% of testing data samples have SGCS values above 0.8.

[0086] In some embodiments, the testing dataset may be divided into multiple sub-dataset where each of the sub-dataset is associated with a performance target. For example, the grouping to the subsets may depend on the SGCS of the samples when the NW applies the NW nominal encoder and the NW actual decoder. Based on the SGCS value, the NW may group the samples into two (or more) subsets where the first subset contains samples that have a lower SGCS (e.g., because is hard to be compressed) and the second subset contains samples that have higher SGCS.

[0087] In some embodiments, the performance target information may be shared with the UE for every sample of the testing dataset.P113381WO01 PCT APPLICATION 17 of 51

[0088] Examples of a performance metric include the accuracy of an end-to-end output sample (e.g., SGCS calculated between an end-to-end model output sample and the corresponding ground-truth label, and / or NMSE calculated between an end-to-end model output sample and the corresponding ground-truth label), the accuracy of an UE-part model output sample (e.g., NMSE calculated between an UE-part model output sample and the corresponding groundtruth label), the data distribution of UE-part model output samples generated using the testing dataset, and the data distribution of end-to-end model output samples generated using the testing dataset.

[0089] Considering that a performance target may consist of multiple values, the term single / a performance target as used herein shall also be extendable to a single set of performance targets. For example, a single set of performance targets may consist of the mean value and X-th percentile if such statistics are supported. A single set may also consist of UE-part model output performance target and end-to-end performance if such performance target types are supported.

[0090] Some embodiments are related to performance targets with multiple additional conditions. The performance of the task to which a two-sided model is applied, e.g., CSI compression, may vary depending on configurations and / or additional conditions. For example, for CSI compression, the performance for multiple-input multiple -output (MIMO) layer 1 is typically better than the performance for MIMO layer 4 (assuming layers are ordered with the strongest layer first). In another example, the performance of Al CSI compression with a larger payload size is typically better than the performance for a lower payload size.

[0091] In some embodiments, a single set of performance target(s) is associated with a single training dataset (or training dataset ID). As different configurations / conditions may result in noticeably different model performances, the NW may label / group the training dataset according to the configurations / conditions associated with the data samples. The number of dataset labels / identifiers (ID) is then equal to the number of possible configurations / conditions supported by the NW. For example, for CSI compression, if the NW supports three possibilities on the number of subbands, two possibilities on the number of TX ports, and two possible MIMO layers, there are 12 sets of performance targets associated with 12 dataset IDs. Note that a single training dataset may refer to a set of data that is used during one session of training . Multiple training datasets may actually be included in a super set of training dataset. For example a dataset may consist of samples with number of subbands = 13. This dataset may beP113381WO01 PCT APPLICATION 18 of 51considered during the training as a super set of training dataset with number of subbands = 1, 2, ... , 13. The exact term of ID may refer to the dataset, super set of dataset, etc.

[0092] In a straightforward approach, one training dataset may be associated with one testing dataset (which then corresponds to one set of performance targets). As mentioned previously, however, one (super set of) training dataset may consist of multiple training sub-datasets. In the case of one super training dataset, it may be up to the UE on how to split such a training dataset into multiple sub-datasets to conduct the training. Taking the number of subbands dimension as an example, to support multiple different numbers of subbands, the UE may train an encoder by using the largest possible number of subbands as the input of the model and perform a masking for certain subbands to accommodate training with a smaller number of subbands. In another mechanism, the UE may treat different numbers of subbands as different datasets and thus treat all possible numbers of subbands as the input of the encoder for the training. In yet another mechanism, the UE may consider training different UE-part models (e.g., encoders) for different numbers of subbands.

[0093] For testing purposes, however, it is beneficial to have an aligned testing dataset. Therefore, in one embodiment, a set of performance targets may be associated with one testing dataset. Here, one training dataset may consist of one or more testing datasets. Taking the number of subbands as an example, one testing dataset may be associated with each possible number of subbands.

[0094] In some embodiments, the samples of the testing dataset are explicitly marked by having an associated set of performance targets. The samples of the training dataset (may also include the validation dataset) do not have an associated set of performance targets.

[0095] In some embodiments, the samples of the testing dataset are explicitly marked by having an associated set of performance targets. The set of model parameters does not have an associated set of performance targets.

[0096] In some embodiments, a set of performance targets may be associated with the configurations / conditions. In an example, a set of performance targets is associated with each combination of (supported) configurations / conditions. In such an approach, the relation between the performance target to the testing dataset and / or the training dataset may be left as UE proprietary. In another example, the association of a set of performance targets to a configuration / condition is a sub-association to the association toward the testing dataset.P113381WO01 PCT APPLICATION 19 of 51

[0097] In one example where different configurations / conditions apply to the same dataset, the UE (the second node) may get multiple targets associated with a single testing dataset from the NW (the first node). For example, the set may be associated with C different configurations / conditions and thus C different sets of performance targets.testingDatcisetConfiguration / condition l : Target _1,Configuration / condition_2: Target _2,Configuration / condition C: Target C,

[0098] For example, for CSI compression, if the dataset consists of channel matrices, configurations / conditions may be the rank of the CSI report (which may be up to UE implementation to decide) and different performance targets for different MIMO layers. Other configurations / conditions may be Al model complexity, which may be linked to device capability, number of payload sizes, quantization granularity, etc.

[0099] Some embodiments are related to the relation of performance targets with performance monitoring. Performance targets may serve as information on the NW suggested or achievable performance (e.g., performance of the pair of the NW nominal encoder and NW actual decoder). During the inference, this information may be used as a reference in determining the performance monitoring results. In one embodiment, the UE (and / or the NW) may derive a performance threshold during the performance monitoring based on the performance target shared by the NW. Here, the performance threshold is referred to as a threshold where the model is deemed to have an acceptable / unacceptable performance. A performance (during the inference) below the threshold may indicate that the performance of the model (in a given CSI report-related configuration) is unacceptable, while a performance above the threshold may indicate that the performance of the model is acceptable. Here, the term unacceptable, in one example, may refer to a performance in which it is no longer beneficial for the UE to be configured with Al-based CSI compression (e.g., due to lower intermediate key performance indicator (KPI) compared to the legacy non-AI-based CSI compression) or indicate that there is data drift and the UE-part model (or NW-part model) may need to be re-trained. In such an example, the determination of the acceptability of the Al-based CSI compression may be determined per each CSI-report (and CSI-RS) related configuration (e.g., payload, rank, layer),P113381WO01 PCT APPLICATION 20 of 51or by considering the performances across different CSI-report (and CSI-RS) related configurations.

[0100] In some embodiments, a factor, alpha, is introduced to represent the gap between the performance target and performance that still can be tolerated by the NW (or by the UE). In other words, a performance of T — a, where T is the performance target and a is a tolerance factor, may still be considered as acceptable. In an example embodiment, the value of a may be configured by the NW, e.g., is included in the performance target information or is included in the performance monitoring configuration. In another example, such a value may be predetermined in the standard text. Alternatively, such a value may be a proprietary value set by the UE.

[0101] Different types of UE may have different capabilities. A more capable UE may implement a more complex model, which (most likely) may be translated into better model performance. To accommodate UEs with different capabilities, in one embodiment, the UE may be configured with one of multiple performance targets by the NW. Here, the NW may derive the performance target by training multiple nominal encoders, e.g., for different encoder sizes. In some embodiments, the multiple performance targets may be associated with the multiple encoder / model sizes, structure, complexity (e.g., in terms of floating point operations (FLOPs)), number of parameters, processing time, etc. Note that information on the encoder / model sizes, structure, complexity, number of parameters, processing time, etc., may not necessarily be included explicitly in the container of the performance target. Rather, such information may be included in the UE capability information, and the performance target is then associated with the UE capability information.

[0102] FIGURE 6 illustrates an example of a communication system 100 in accordance with some embodiments. In the example, the communication system 100 includes a telecommunication network 102 that includes an access network 104, such as a radio access network (RAN), and a core network 106, which includes one or more core network nodes 108. The access network 104 includes one or more access network nodes, such as network nodes 110a and 110b (one or more of which may be generally referred to as network nodes 110), or any other similar 3rd Generation Partnership Project (3GPP) access node or non-3GPP access point. The network nodes 110 facilitate direct or indirect connection of user equipment (UE), such as by connecting UEs 112a, 112b, 112c, and 112d (one or more of which may be generally referred to as UEs 112) to the core network 106 over one or more wireless connections.P113381WO01 PCT APPLICATION 21 of51

[0103] Example wireless communications over a wireless connection include transmitting and / or receiving wireless signals using electromagnetic waves, radio waves, infrared waves, and / or other types of signals suitable for conveying information without the use of wires, cables, or other material conductors. Moreover, in different embodiments, the communication system 100 may include any number of wired or wireless networks, network nodes, UEs, and / or any other components or systems that may facilitate or participate in the communication of data and / or signals whether via wired or wireless connections. The communication system 100 may include and / or interface with any type of communication, telecommunication, data, cellular, radio network, and / or other similar type of system.

[0104] The UEs 112 may be any of a wide variety of communication devices, including wireless devices arranged, configured, and / or operable to communicate wirelessly with the network nodes 110 and other communication devices. Similarly, the network nodes 110 are arranged, capable, configured, and / or operable to communicate directly or indirectly with the UEs 112 and / or with other network nodes or equipment in the telecommunication network 102 to enable and / or provide network access, such as wireless network access, and / or to perform other functions, such as administration in the telecommunication network 102.

[0105] In the depicted example, the core network 106 connects the network nodes 110 to one or more hosts, such as host 116. These connections may be direct or indirect via one or more intermediary networks or devices. In other examples, network nodes may be directly coupled to hosts. The core network 106 includes one more core network nodes (e.g., core network node 108) that are structured with hardware and software components. Features ofthese components may be substantially similar to those described with respect to the UEs, network nodes, and / or hosts, such that the descriptions thereof are generally applicable to the corresponding components of the core network node 108. Example core network nodes include functions of one or more of a Mobile Switching Center (MSC), Mobility Management Entity (MME), Home Subscriber Server (HSS), Access and Mobility Management Function (AMF), Session Management Function (SMF), Authentication Server Function (AUSF), Subscription Identifier De-concealing function (SIDF), Unified Data Management (UDM), Security Edge Protection Proxy (SEPP), Network Exposure Function (NEF), and / or a User Plane Function (UPF).

[0106] The host 116 may be under the ownership or control of a service provider other than an operator or provider of the access network 104 and / or the telecommunication network 102 and may be operated by the service provider or on behalf of the service provider. The host 116 mayP113381WO01 PCT APPLICATION 22 of 51host a variety of applications to provide one or more services. Examples of such applications include live and pre-recorded audio / video content, data collection services such as retrieving and compiling data on various ambient conditions detected by a plurality of UEs, analytics functionality, social media, functions for controlling or otherwise interacting with remote devices, functions for an alarm and surveillance center, or any other such function performed by a server.

[0107] As a whole, the communication system 100 of FIGURE 6 enables connectivity between the UEs, network nodes, and hosts. In that sense, the communication system may be configured to operate according to predefined rules or procedures, such as specific standards that include, but are not limited to: Global System for Mobile Communications (GSM); Universal Mobile Telecommunications System (UMTS); Long Term Evolution (LTE), and / or other suitable 2G, 3G, 4G, 5G standards, or any applicable future generation standard (e.g., 6G); wireless local area network (WLAN) standards, such as the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standards (WiFi); and / or any other appropriate wireless communication standard, such as the Worldwide Interoperability for Microwave Access (WiMax), Bluetooth, Z-Wave, Near Field Communication (NFC) ZigBee, LiFi, and / or any low-power wide-area network (LPWAN) standards such as LoRa and Sigfox.

[0108] In some examples, the telecommunication network 102 is a cellular network that implements 3GPP standardized features. Accordingly, the telecommunications network 102 may support network slicing to provide different logical networks to different devices that are connected to the telecommunication network 102. For example, the telecommunications network 102 may provide Ultra Reliable Low Latency Communication (URLLC) services to some UEs, while providing Enhanced Mobile Broadband (eMBB) services to other UEs, and / or Massive Machine Type Communication (mMTC)ZMassive loT services to yet further UEs.

[0109] In some examples, the UEs 112 are configured to transmit and / or receive information without direct human interaction. For instance, a UE may be designed to transmit information to the access network 104 on a predetermined schedule, when triggered by an internal or external event, or in response to requests from the access network 104. Additionally, a UE may be configured for operating in single- or multi-RAT or multi-standard mode. For example, a UE may operate with any one or combination of Wi-Fi, NR (New Radio) and LTE, i.e. being configured for multi-radio dual connectivity (MR-DC), such as E-UTRAN (Evolved-UMTS Terrestrial Radio Access Network) New Radio - Dual Connectivity (EN-DC).P113381WO01 PCT APPLICATION 23 of 51

[0110] In the example, the hub 114 communicates with the access network 104 to facilitate indirect communication between one or more UEs (e.g., UE 112c and / or 112d) and network nodes (e.g., network node 110b). In some examples, the hub 114 may be a controller, router, content source and analytics, or any of the other communication devices described herein regarding UEs. For example, the hub 114 may be a broadband router enabling access to the core network 106 for the UEs. As another example, the hub 114 may be a controller that sends commands or instructions to one or more actuators in the UEs. Commands or instructions may be received from the UEs, network nodes 110, or by executable code, script, process, or other instructions in the hub 114. As another example, the hub 114 may be a data collector that acts as temporary storage for UE data and, in some embodiments, may perform analysis or other processing of the data. As another example, the hub 114 may be a content source. For example, for a UE that is a VR headset, display, loudspeaker or other media delivery device, the hub 114 may retrieve VR assets, video, audio, or other media or data related to sensory information via a network node, which the hub 114 then provides to the UE either directly, after performing local processing, and / or after adding additional local content. In still another example, the hub 114 acts as a proxy server or orchestrator for the UEs, in particular if one or more of the UEs are low energy loT devices.[oni] The hub 114 may have a constant / persistent or intermittent connection to the network node 110b. The hub 114 may also allow for a different communication scheme and / or schedule between the hub 114 and UEs (e.g., UE 112c and / or 112d), and between the hub 114 and the core network 106. In other examples, the hub 114 is connected to the core network 106 and / or one or more UEs via a wired connection. Moreover, the hub 114 may be configured to connect to an M2M service provider over the access network 104 and / or to another UE over a direct connection. In some scenarios, UEs may establish a wireless connection with the network nodes 110 while still connected via the hub 114 via a wired or wireless connection. In some embodiments, the hub 114 may be a dedicated hub - that is, a hub whose primary function is to route communications to / from the UEs from / to the network node 110b. In other embodiments, the hub 114 may be a non-dedicated hub - that is, a device which is capable of operating to route communications between the UEs and network node 110b, but which is additionally capable of operating as a communication start and / or end point for certain data channels.P113381WO01 PCT APPLICATION 24 of 51

[0112] FIGURE 7 shows a UE 200 in accordance with some embodiments. As used herein, a UE refers to a device capable, configured, arranged and / or operable to communicate wirelessly with network nodes and / or other UEs. Examples of a UE include, but are not limited to, a smart phone, mobile phone, cell phone, voice over IP (VoIP) phone, wireless local loop phone, desktop computer, personal digital assistant (PDA), wireless cameras, gaming console or device, music storage device, playback appliance, wearable terminal device, wireless endpoint, mobile station, tablet, laptop, laptop-embedded equipment (LEE), laptop-mounted equipment (LME), smart device, wireless customer-premise equipment (CPE), vehicle-mounted or vehicle embedded / integrated wireless device, etc. Other examples include any UE identified by the 3rd Generation Partnership Project (3GPP), including a narrow band internet of things (NB-IoT) UE, a machine type communication (MTC) UE, and / or an enhanced MTC (eMTC) UE.

[0113] A UE may support device-to-device (D2D) communication, for example by implementing a 3GPP standard for sidelink communication, Dedicated Short-Range Communication (DSRC), vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), or vehicle-to-everything (V2X). In other examples, a UE may not necessarily have a user in the sense of a human user who owns and / or operates the relevant device. Instead, a UE may represent a device that is intended for sale to, or operation by, a human user but which may not, or which may not initially, be associated with a specific human user (e.g., a smart sprinkler controller). Alternatively, a UE may represent a device that is not intended for sale to, or operation by, an end user but which may be associated with or operated for the benefit of a user (e.g., a smart power meter).

[0114] The UE 200 includes processing circuitry 202 that is operatively coupled via a bus 204 to an input / output interface 206, a power source 208, a memory 210, a communication interface 212, and / or any other component, or any combination thereof. Certain UEs may utilize all or a subset of the components shown in FIGURE 7. The level of integration between the components may vary from one UE to another UE. Further, certain UEs may contain multiple instances of a component, such as multiple processors, memories, transceivers, transmitters, receivers, etc.

[0115] The processing circuitry 202 is configured to process instructions and data and may be configured to implement any sequential state machine operative to execute instructions stored as machine -readable computer programs in the memory 210. The processing circuitry 202 mayP113381WO01 PCT APPLICATION 25 of 51be implemented as one or more hardware-implemented state machines (e.g., in discrete logic, field-programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), etc.); programmable logic together with appropriate firmware; one or more stored computer programs, general-purpose processors, such as a microprocessor or digital signal processor (DSP), together with appropriate software; or any combination of the above. For example, the processing circuitry 202 may include multiple central processing units (CPUs).

[0116] In the example, the input / output interface 206 may be configured to provide an interface or interfaces to an input device, output device, or one or more input and / or output devices. Examples of an output device include a speaker, a sound card, a video card, a display, a monitor, a printer, an actuator, an emitter, a smartcard, another output device, or any combination thereof. An input device may allow a user to capture information into the UE 200. Examples of an input device include a touch-sensitive or presence-sensitive display, a camera (e.g., a digital camera, a digital video camera, a web camera, etc.), a microphone, a sensor, a mouse, a trackball, a directional pad, a trackpad, a scroll wheel, a smartcard, and the like. The presence-sensitive display may include a capacitive or resistive touch sensor to sense input from a user. A sensor may be, for instance, an accelerometer, a gyroscope, a tilt sensor, a force sensor, a magnetometer, an optical sensor, a proximity sensor, a biometric sensor, etc., or any combination thereof. An output device may use the same type of interface port as an input device. For example, a Universal Serial Bus (USB) port may be used to provide an input device and an output device.

[0117] In some embodiments, the power source 208 is structured as a battery or battery pack. Other types of power sources, such as an external power source (e.g., an electricity outlet), photovoltaic device, or power cell, may be used. The power source 208 may further include power circuitry for delivering power from the power source 208 itself, and / or an external power source, to the various parts of the UE 200 via input circuitry or an interface such as an electrical power cable. Delivering power may be, for example, for charging of the power source 208. Power circuitry may perform any formatting, converting, or other modification to the power from the power source 208 to make the power suitable for the respective components of the UE 200 to which power is supplied.

[0118] The memory 210 may be or be configured to include memory such as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmableP113381WO01 PCT APPLICATION 26 of 51read-only memory (EEPROM), magnetic disks, optical disks, hard disks, removable cartridges, flash drives, and so forth. In one example, the memory 210 includes one or more application programs 214, such as an operating system, web browser application, a widget, gadget engine, or other application, and corresponding data 216. The memory 210 may store, for use by the UE 200, any of a variety of various operating systems or combinations of operating systems.

[0119] The memory 210 may be configured to include a number of physical drive units, such as redundant array of independent disks (RAID), flash memory, USB flash drive, external hard disk drive, thumb drive, pen drive, key drive, high-density digital versatile disc (HD-DVD) optical disc drive, internal hard disk drive, Blu-Ray optical disc drive, holographic digital data storage (HDDS) optical disc drive, external mini-dual in-line memory module (DIMM), synchronous dynamic random access memory (SDRAM), external micro-DIMM SDRAM, smartcard memory such as tamper resistant module in the form of a universal integrated circuit card (UICC) including one or more subscriber identity modules (SIMs), such as a USIM and / or ISIM, other memory, or any combination thereof. The UICC may for example be an embedded UICC (eUICC), integrated UICC (iUICC) or a removable UICC commonly known as ‘SIM card.’ The memory 210 may allow the UE 200 to access instructions, application programs and the like, stored on transitory or non-transitory memory media, to off-load data, or to upload data. An article of manufacture, such as one utilizing a communication system may be tangibly embodied as or in the memory 210, which may be or comprise a device-readable storage medium.

[0120] The processing circuitry 202 may be configured to communicate with an access network or other network using the communication interface 212. The communication interface 212 may comprise one or more communication subsystems and may include or be communicatively coupled to an antenna 222. The communication interface 212 may include one or more transceivers used to communicate, such as by communicating with one or more remote transceivers of another device capable of wireless communication (e.g., another UE or a network node in an access network). Each transceiver may include a transmitter 218 and / or a receiver 220 appropriate to provide network communications (e.g., optical, electrical, frequency allocations, and so forth). Moreover, the transmitter 218 and receiver 220 may be coupled to one or more antennas (e.g., antenna 222) and may share circuit components, software or firmware, or alternatively be implemented separately.P113381WO01 PCT APPLICATION 27 of 51

[0121] In the illustrated embodiment, communication functions of the communication interface 212 may include cellular communication, Wi-Fi communication, LPWAN communication, data communication, voice communication, multimedia communication, short-range communications such as Bluetooth, near-field communication, location-based communication such as the use of the global positioning system (GPS) to determine a location, another like communication function, or any combination thereof. Communications may be implemented in according to one or more communication protocols and / or standards, such as IEEE 802.11, Code Division Multiplexing Access (CDMA), Wideband Code Division Multiple Access (WCDMA), GSM, LTE, New Radio (NR), UMTS, WiMax, Ethernet, transmission control protocol / intemet protocol (TCP / IP), synchronous optical networking (SONET), Asynchronous Transfer Mode (ATM), QUIC, Hypertext Transfer Protocol (HTTP), and so forth.

[0122] Regardless of the type of sensor, a UE may provide an output of data captured by its sensors, through its communication interface 212, via a wireless connection to a network node. Data captured by sensors of a UE can be communicated through a wireless connection to a network node via another UE. The output may be periodic (e.g., once every 15 minutes if it reports the sensed temperature), random (e.g., to even out the load from reporting from several sensors), in response to a triggering event (e.g., when moisture is detected an alert is sent), in response to a request (e.g., a user initiated request), or a continuous stream (e.g., a live video feed of a patient).

[0123] As another example, a UE comprises an actuator, a motor, or a switch, related to a communication interface configured to receive wireless input from a network node via a wireless connection. In response to the received wireless input the states of the actuator, the motor, or the switch may change. For example, the UE may comprise a motor that adjusts the control surfaces or rotors of a drone in flight according to the received input or to a robotic arm performing a medical procedure according to the received input.

[0124] A UE, when in the form of an Internet of Things (loT) device, may be a device for use in one or more application domains, these domains comprising, but not limited to, city wearable technology, extended industrial application and healthcare. Non-limiting examples of such an loT device are a device which is or which is embedded in: a connected refrigerator or freezer, a TV, a connected lighting device, an electricity meter, a robot vacuum cleaner, a voice controlled smart speaker, a home security camera, a motion detector, a thermostat, a smokeP113381WO01 PCT APPLICATION 28 of 51detector, a door / window sensor, a flood / moisture sensor, an electrical door lock, a connected doorbell, an air conditioning system like a heat pump, an autonomous vehicle, a surveillance system, a weather monitoring device, a vehicle parking monitoring device, an electric vehicle charging station, a smart watch, a fitness tracker, a head-mounted display for Augmented Reality (AR) or Virtual Reality (VR), a wearable for tactile augmentation or sensory enhancement, a water sprinkler, an animal- or item-tracking device, a sensor for monitoring a plant or animal, an industrial robot, an Unmanned Aerial Vehicle (UAV), and any kind of medical device, like a heart rate monitor or a remote controlled surgical robot. A UE in the form of an loT device comprises circuitry and / or software in dependence of the intended application of the loT device in addition to other components as described in relation to the UE 200 shown in FIGURE 7.

[0125] As yet another specific example, in an loT scenario, a UE may represent a machine or other device that performs monitoring and / or measurements and transmits the results of such monitoring and / or measurements to another UE and / or a network node. The UE may in this case be an M2M device, which may in a 3GPP context be referred to as an MTC device. As one particular example, the UE may implement the 3GPP NB-IoT standard. In other scenarios, a UE may represent a vehicle, such as a car, a bus, a truck, a ship and an airplane, or other equipment that is capable of monitoring and / or reporting on its operational status or other functions associated with its operation.

[0126] In practice, any number of UEs may be used together with respect to a single use case. For example, a first UE might be or be integrated in a drone and provide the drone’s speed information (obtained through a speed sensor) to a second UE that is a remote controller operating the drone. When the user makes changes from the remote controller, the first UE may adjust the throttle on the drone (e.g. by controlling an actuator) to increase or decrease the drone’s speed. The first and / or the second UE can also include more than one of the functionalities described above. For example, a UE might comprise the sensor and the actuator, and handle communication of data for both the speed sensor and the actuators.

[0127] FIGURE 8 shows a network node 300 in accordance with some embodiments. As used herein, network node refers to equipment capable, configured, arranged and / or operable to communicate directly or indirectly with a UE and / or with other network nodes or equipment, in a telecommunication network. Examples of network nodes include, but are not limited to,P113381WO01 PCT APPLICATION 29 of 51access points (APs) (e.g., radio access points), base stations (BSs) (e.g., radio base stations, Node Bs, evolved Node Bs (eNBs) and NRNodeBs (gNBs)).

[0128] Base stations may be categorized based on the amount of coverage they provide (or, stated differently, their transmit power level) and so, depending on the provided amount of coverage, may be referred to as femto base stations, pico base stations, micro base stations, or macro base stations. A base station may be a relay node or a relay donor node controlling a relay. A network node may also include one or more (or all) parts of a distributed radio base station such as centralized digital units and / or remote radio units (RRUs), sometimes referred to as Remote Radio Heads (RRHs). Such remote radio units may or may not be integrated with an antenna as an antenna integrated radio. Parts of a distributed radio base station may also be referred to as nodes in a distributed antenna system (DAS).

[0129] Other examples of network nodes include multiple transmission point (multi-TRP) 5G access nodes, multi-standard radio (MSR) equipment such as MSR BSs, network controllers such as radio network controllers (RNCs) or base station controllers (BSCs), base transceiver stations (BTSs), transmission points, transmission nodes, multi-cell / multicast coordination entities (MCEs), Operation and Maintenance (O&M) nodes, Operations Support System (OSS) nodes, Self-Organizing Network (SON) nodes, positioning nodes (e.g., Evolved Serving Mobile Location Centers (E-SMLCs)), and / or Minimization of Drive Tests (MDTs).

[0130] The network node 300 includes a processing circuitry 302, a memory 304, a communication interface 306, and a power source 308. The network node 300 may be composed of multiple physically separate components (e.g., a NodeB component and a RNC component, or a BTS component and a BSC component, etc.), which may each have their own respective components. In certain scenarios in which the network node 300 comprises multiple separate components (e.g., BTS and BSC components), one or more of the separate components may be shared among several network nodes. For example, a single RNC may control multiple NodeBs. In such a scenario, each unique NodeB and RNC pair, may in some instances be considered a single separate network node. In some embodiments, the network node 300 may be configured to support multiple radio access technologies (RATs). In such embodiments, some components may be duplicated (e.g., separate memory 304 for different RATs) and some components may be reused (e.g., a same antenna 310 may be shared by different RATs). The network node 300 may also include multiple sets of the various illustrated components for different wireless technologies integrated into network node 300, for exampleP113381WO01 PCT APPLICATION 30 of 51GSM, WCDMA, LTE, NR, WiFi, Zigbee, Z-wave, LoRaWAN, Radio Frequency Identification (RFID) or Bluetooth wireless technologies. These wireless technologies may be integrated into the same or different chip or set of chips and other components within network node 300.

[0131] The processing circuitry 302 may comprise a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application-specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or encoded logic operable to provide, either alone or in conjunction with other network node 300 components, such as the memory 304, to provide network node 300 functionality.

[0132] In some embodiments, the processing circuitry 302 includes a system on a chip (SOC).In some embodiments, the processing circuitry 302 includes one or more of radio frequency (RF) transceiver circuitry 312 and baseband processing circuitry 314. In some embodiments, the radio frequency (RF) transceiver circuitry 312 and the baseband processing circuitry 314 may be on separate chips (or sets of chips), boards, or units, such as radio units and digital units. In alternative embodiments, part or all of RF transceiver circuitry 312 and baseband processing circuitry 314 may be on the same chip or set of chips, boards, or units.

[0133] The memory 304 may comprise any form of volatile or non-volatile computer-readable memory including, without limitation, persistent storage, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), read-only memory (ROM), mass storage media (for example, a hard disk), removable storage media (for example, a flash drive, a Compact Disk (CD) or a Digital Video Disk (DVD)), and / or any other volatile or non-volatile, non-transitory device-readable and / or computer-executable memory devices that store information, data, and / or instructions that may be used by the processing circuitry 302. The memory 304 may store any suitable instructions, data, or information, including a computer program, software, an application including one or more of logic, rules, code, tables, and / or other instructions capable of being executed by the processing circuitry 302 and utilized by the network node 300. The memory 304 may be used to store any calculations made by the processing circuitry 302 and / or any data received via the communication interface 306. In some embodiments, the processing circuitry 302 and memory 304 is integrated.P113381WO01 PCT APPLICATION 31 of51

[0134] The communication interface 306 is used in wired or wireless communication of signaling and / or data between a network node, access network, and / or UE. As illustrated, the communication interface 306 comprises port(s) / terminal(s) 316 to send and receive data, for example to and from a network over a wired connection. The communication interface 306 also includes radio front-end circuitry 318 that may be coupled to, or in certain embodiments a part of, the antenna 310. Radio front-end circuitry 318 comprises filters 320 and amplifiers 322. The radio front-end circuitry 318 may be connected to an antenna 310 and processing circuitry 302. The radio front-end circuitry may be configured to condition signals communicated between antenna 310 and processing circuitry 302. The radio front-end circuitry 318 may receive digital data that is to be sent out to other network nodes or UEs via a wireless connection. The radio front-end circuitry 318 may convert the digital data into a radio signal having the appropriate channel and bandwidth parameters using a combination of filters 320 and / or amplifiers 322. The radio signal may then be transmitted via the antenna 310. Similarly, when receiving data, the antenna 310 may collect radio signals which are then converted into digital data by the radio front-end circuitry 318. The digital data may be passed to the processing circuitry 302. In other embodiments, the communication interface may comprise different components and / or different combinations of components.

[0135] In certain alternative embodiments, the network node 300 does not include separate radio front-end circuitry 318, instead, the processing circuitry 302 includes radio front-end circuitry and is connected to the antenna 310. Similarly, in some embodiments, all or some of the RF transceiver circuitry 312 is part of the communication interface 306. In still other embodiments, the communication interface 306 includes one or more ports or terminals 316, the radio front-end circuitry 318, and the RF transceiver circuitry 312, as part of a radio unit (not shown), and the communication interface 306 communicates with the baseband processing circuitry 314, which is part of a digital unit (not shown).

[0136] The antenna 310 may include one or more antennas, or antenna arrays, configured to send and / or receive wireless signals. The antenna 310 may be coupled to the radio front-end circuitry 318 and may be any type of antenna capable of transmitting and receiving data and / or signals wirelessly. In certain embodiments, the antenna 310 is separate from the network node 300 and connectable to the network node 300 through an interface or port.

[0137] The antenna 310, communication interface 306, and / or the processing circuitry 302 may be configured to perform any receiving operations and / or certain obtaining operationsP113381WO01 PCT APPLICATION 32 of 51described herein as being performed by the network node. Any information, data and / or signals may be received from a UE, another network node and / or any other network equipment. Similarly, the antenna 310, the communication interface 306, and / or the processing circuitry 302 may be configured to perform any transmitting operations described herein as being performed by the network node. Any information, data and / or signals may be transmitted to a UE, another network node and / or any other network equipment.

[0138] The power source 308 provides power to the various components of network node 300 in a form suitable for the respective components (e.g., at a voltage and current level needed for each respective component). The power source 308 may further comprise, or be coupled to, power management circuitry to supply the components of the network node 300 with power for performing the functionality described herein. For example, the network node 300 may be connectable to an external power source (e.g., the power grid, an electricity outlet) via an input circuitry or interface such as an electrical cable, whereby the external power source supplies power to power circuitry of the power source 308. As a further example, the power source 308 may comprise a source of power in the form of a battery or battery pack which is connected to, or integrated in, power circuitry. The battery may provide backup power should the external power source fail.

[0139] Embodiments of the network node 300 may include additional components beyond those shown in FIGURE 8 for providing certain aspects of the network node’s functionality, including any of the functionality described herein and / or any functionality necessary to support the subject matter described herein. For example, the network node 300 may include user interface equipment to allow input of information into the network node 300 and to allow output of information from the network node 300. This may allow a user to perform diagnostic, maintenance, repair, and other administrative functions for the network node 300.

[0140] FIGURE 9 is a block diagram of a host 400, which may be an embodiment of the host 116 of FIGURE 6, in accordance with various aspects described herein. As used herein, the host 400 may be or comprise various combinations hardware and / or software, including a standalone server, a blade server, a cloud-implemented server, a distributed server, a virtual machine, container, or processing resources in a server farm. The host 400 may provide one or more services to one or more UEs.

[0141] The host 400 includes processing circuitry 402 that is operatively coupled via a bus 404 to an input / output interface 406, a network interface 408, a power source 410, and a memoryP113381WO01 PCT APPLICATION 33 of 51412. Other components may be included in other embodiments. Features of these components may be substantially similar to those described with respect to the devices of previous figures, such as Figures 3 and 4, such that the descriptions thereof are generally applicable to the corresponding components of host 400.

[0142] The memory 412 may include one or more computer programs including one or more host application programs 414 and data 416, which may include user data, e.g., data generated by a UE for the host 400 or data generated by the host 400 for a UE. Embodiments of the host 400 may utilize only a subset or all of the components shown. The host application programs 414 may be implemented in a container-based architecture and may provide support for video codecs (e.g., Versatile Video Coding (VVC), High Efficiency Video Coding (HEVC), Advanced Video Coding (AVC), MPEG, VP9) and audio codecs (e.g., FLAC, Advanced Audio Coding (AAC), MPEG, G.711), including transcoding for multiple different classes, types, or implementations of UEs (e.g., handsets, desktop computers, wearable display systems, heads-up display systems). The host application programs 414 may also provide for user authentication and licensing checks and may periodically report health, routes, and content availability to a central node, such as a device in or on the edge of a core network. Accordingly, the host 400 may select and / or indicate a different host for over-the-top services for a UE. The host application programs 414 may support various protocols, such as the HTTP Live Streaming (HLS) protocol, Real-Time Messaging Protocol (RTMP), Real-Time Streaming Protocol (RTSP), Dynamic Adaptive Streaming over HTTP (MPEG-DASH), etc.

[0143] FIGURE 10 is a block diagram illustrating a virtualization environment 500 in which functions implemented by some embodiments may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or devices which may include virtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more virtual machines (VMs) implemented in one or more virtual environments 500 hosted by one or more of hardware nodes, such as a hardware computing device that operates as a network node, UE, core network node, or host. Further, in embodiments in which the virtual node does not require radio connectivity (e.g., a core network node or host), then the node may be entirely virtualized.P113381WO01 PCT APPLICATION 34 of 51

[0144] Applications 502 (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) are run in the virtualization environment Q400 to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0145] Hardware 504 includes processing circuitry, memory that stores software and / or instructions executable by hardware processing circuitry, and / or other hardware devices as described herein, such as a network interface, input / output interface, and so forth. Software may be executed by the processing circuitry to instantiate one or more virtualization layers 506 (also referred to as hypervisors or virtual machine monitors (VMMs)), provide VMs 508a and 508b (one or more of which may be generally referred to as VMs 508), and / or perform any of the functions, features and / or benefits described in relation with some embodiments described herein. The virtualization layer 506 may present a virtual operating platform that appears like networking hardware to the VMs 508.

[0146] The VMs 508 comprise virtual processing, virtual memory, virtual networking or interface and virtual storage, and may be run by a corresponding virtualization layer 506. Different embodiments of the instance of a virtual appliance 502 may be implemented on one or more of VMs 508, and the implementations may be made in different ways. Virtualization of the hardware is in some contexts referred to as network function virtualization (NFV). NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which can be located in data centers, and customer premise equipment.

[0147] In the context of NFV, a VM 508 may be a software implementation of a physical machine that runs programs as if they were executing on a physical, non-virtualized machine. Each of the VMs 508, and that part of hardware 504 that executes that VM, be it hardware dedicated to that VM and / or hardware shared by that VM with others of the VMs, forms separate virtual network elements. Still in the context of NFV, a virtual network function is responsible for handling specific network functions that run in one or more VMs 508 on top of the hardware 504 and corresponds to the application 502.

[0148] Hardware 504 may be implemented in a standalone network node with generic or specific components. Hardware 504 may implement some functions via virtualization. Alternatively, hardware 504 may be part of a larger cluster of hardware (e.g. such as in a data center or CPE) where many hardware nodes work together and are managed via managementP113381WO01 PCT APPLICATION 35 of 51and orchestration 510, which, among others, oversees lifecycle management of applications 502. In some embodiments, hardware 504 is coupled to one or more radio units that each include one or more transmitters and one or more receivers that may be coupled to one or more antennas. Radio units may communicate directly with other hardware nodes via one or more appropriate network interfaces and may be used in combination with the virtual components to provide a virtual node with radio capabilities, such as a radio access node or a base station. In some embodiments, some signaling can be provided with the use of a control system 512 which may alternatively be used for communication between hardware nodes and radio units.

[0149] FIGURE 11 shows a communication diagram of a host 602 communicating via a network node 604 with a UE 606 over a partially wireless connection in accordance with some embodiments. Example implementations, in accordance with various embodiments, of the UE (such as a UE 112a of FIGURE 6 and / or UE 200 of FIGURE 7), network node (such as network node 110a of FIGURE 6 and / or network node 300 of FIGURE 8), and host (such as host 116 of FIGURE 6 and / or host 400 of FIGURE 9) discussed in the preceding paragraphs will now be described with reference to FIGURE 11.

[0150] Like host 400, embodiments of host 602 include hardware, such as a communication interface, processing circuitry, and memory. The host 602 also includes software, which is stored in or accessible by the host 602 and executable by the processing circuitry. The software includes a host application that may be operable to provide a service to a remote user, such as the UE 606 connecting via an over-the-top (OTT) connection 650 extending between the UE 606 and host 602. In providing the service to the remote user, a host application may provide user data which is transmitted using the OTT connection 650.

[0151] The network node 604 includes hardware enabling it to communicate with the host 602 and UE 606. The connection 660 may be direct or pass through a core network (like core network 106 of FIGURE 6) and / or one or more other intermediate networks, such as one or more public, private, or hosted networks. For example, an intermediate network may be a backbone network or the Internet.

[0152] The UE 606 includes hardware and software, which is stored in or accessible by UE 606 and executable by the UE’s processing circuitry. The software includes a client application, such as a web browser or operator-specific “app” that may be operable to provide a service to a human or non-human user via UE 606 with the support of the host 602. In the host 602, an executing host application may communicate with the executing client application via the OTTP113381WO01 PCT APPLICATION 36 of 51connection 650 terminating at the UE 606 and host 602. In providing the service to the user, the UE's client application may receive request data from the host's host application and provide user data in response to the request data. The OTT connection 650 may transfer both the request data and the user data. The UE's client application may interact with the user to generate the user data that it provides to the host application through the OTT connection 650.

[0153] The OTT connection 650 may extend via a connection 660 between the host 602 and the network node 604 and via a wireless connection 670 between the network node 604 and the UE 606 to provide the connection between the host 602 and the UE 606. The connection 660 and wireless connection 670, over which the OTT connection 650 may be provided, have been drawn abstractly to illustrate the communication between the host 602 and the UE 606 via the network node 604, without explicit reference to any intermediary devices and the precise routing of messages via these devices.

[0154] As an example of transmitting data via the OTT connection 650, in step 608, the host 602 provides user data, which may be performed by executing a host application. In some embodiments, the user data is associated with a particular human user interacting with the UE 606. In other embodiments, the user data is associated with a UE 606 that shares data with the host 602 without explicit human interaction. In step 610, the host 602 initiates a transmission carrying the user data towards the UE 606. The host 602 may initiate the transmission responsive to a request transmitted by the UE 606. The request may be caused by human interaction with the UE 606 or by operation of the client application executing on the UE 606. The transmission may pass via the network node 604, in accordance with the teachings of the embodiments described throughout this disclosure. Accordingly, in step 612, the network node 604 transmits to the UE 606 the user data that was carried in the transmission that the host 602 initiated, in accordance with the teachings of the embodiments described throughout this disclosure. In step 614, the UE 606 receives the user data carried in the transmission, which may be performed by a client application executed on the UE 606 associated with the host application executed by the host 602.

[0155] In some examples, the UE 606 executes a client application which provides user data to the host 602. The user data may be provided in reaction or response to the data received from the host 602. Accordingly, in step 616, the UE 606 may provide user data, which may be performed by executing the client application. In providing the user data, the client application may further consider user input received from the user via an input / output interface of the UEP113381WO01 PCT APPLICATION 37 of 51606. Regardless of the specific manner in which the user data was provided, the UE 606 initiates, in step 618, transmission of the user data towards the host 602 via the network node 604. In step 620, in accordance with the teachings of the embodiments described throughout this disclosure, the network node 604 receives user data from the UE 606 and initiates transmission of the received user data towards the host 602. In step 622, the host 602 receives the user data carried in the transmission initiated by the UE 606.

[0156] One or more of the various embodiments improve the performance of OTT services provided to the UE 606 using the OTT connection 650, in which the wireless connection 670 forms the last segment. More precisely, the teachings of these embodiments may improve the delay to directly activate an SCell by RRC and power consumption of user equipment and thereby provide benefits such as reduced user waiting time and extended battery lifetime.

[0157] In an example scenario, factory status information may be collected and analyzed by the host 602. As another example, the host 602 may process audio and video data which may have been retrieved from a UE for use in creating maps. As another example, the host 602 may collect and analyze real-time data to assist in controlling vehicle congestion (e.g., controlling traffic lights). As another example, the host 602 may store surveillance video uploaded by a UE. As another example, the host 602 may store or control access to media content such as video, audio, VR or AR which it can broadcast, multicast or unicast to UEs. As other examples, the host 602 may be used for energy pricing, remote control of non-time critical electrical load to balance power generation needs, location services, presentation services (such as compiling diagrams etc. from data collected from remote devices), or any other function of collecting, retrieving, storing, analyzing and / or transmitting data.

[0158] In some examples, a measurement procedure may be provided for the purpose of monitoring data rate, latency and other factors on which the one or more embodiments improve. There may further be an optional network functionality for reconfiguring the OTT connection 650 between the host 602 and UE 606, in response to variations in the measurement results. The measurement procedure and / or the network functionality for reconfiguring the OTT connection may be implemented in software and hardware of the host 602 and / or UE 606. In some embodiments, sensors (not shown) may be deployed in or in association with other devices through which the OTT connection 650 passes; the sensors may participate in the measurement procedure by supplying values of the monitored quantities exemplified above, or supplying values of other physical quantities from which software may compute or estimateP113381WO01 PCT APPLICATION 38 of 51the monitored quantities. The reconfiguring of the OTT connection 650 may include message format, retransmission settings, preferred routing etc.; the reconfiguring need not directly alter the operation of the network node 604. Such procedures and functionalities may be known and practiced in the art. In certain embodiments, measurements may involve proprietary UE signaling that facilitates measurements of throughput, propagation times, latency and the like, by the host 602. The measurements may be implemented in that software causes messages to be transmitted, in particular empty or ‘dummy’ messages, using the OTT connection 650 while monitoring propagation times, errors, etc.

[0159] FIGURE 12 is a flowchart illustrating an example method 1200 in a second node, according to certain embodiments. In particular embodiments, one or more steps of FIGURE 12 may be performed by UE 200 described with respect to FIGURE 7. The method is for two-sided ML model sequential training. The two-sided ML model comprises a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model. The method is performed by the second node.

[0160] The method may begin at step 1212, where the second node (e.g., UE 200) receives a ML model training dataset or ML model training parameter from a network node for training the second node-side ML model.

[0161] In particular embodiments, the two-sided ML model comprises an autoencoder / decoder ML model (e.g., CSI compression), or any other two-sided ML model described in the examples and embodiments above.

[0162] At step 1214, the second node may receive a testing dataset from the network node for testing the second node-side ML model.

[0163] At step 1216, the second node receives one or more performance targets associated with the second node-side ML model from the network node. The one or more performance targets are based on training of the first node-side ML model using the testing dataset.

[0164] The ML model training dataset or ML model training parameter, the ML model testing dataset, and the one or more performance targets are described in more detail with respect to the embodiments and examples described herein, and may be referred to above as first, second, third, and / or fourth sets of information.

[0165] Although steps 1212-1216 are described as separate steps for discussion purposes, in some embodiments, the receiving steps may comprise one, two, or three steps where theP113381WO01 PCT APPLICATION 39 of 51received information may be received in one, two, three, or any suitable number of messages / containers.

[0166] In particular embodiments, the one or more performance targets comprise one or more of a second node-side output performance target and an end-to-end performance target. In particular embodiments, the one or more performance targets comprise one or more pairs of a second node-side output performance target and an end-to-end performance target associated with the paired second node-side output performance target.

[0167] In particular embodiments, the one or more performance targets comprise multiple values representing statistics across samples of the testing dataset. For example, the multiple values representing statistics across samples of the testing dataset may comprise a mean value and one or more values of one or more specified percentiles.

[0168] In particular embodiments, the testing dataset may comprise multiple sub-datasets and each of the sub-datasets is associated with one or more performance targets.

[0169] In particular embodiments, the training dataset is associated with one or more configuration parameters and each of the one or more performance targets may be associated with a particular configuration parameter or parameter combination.

[0170] In particular embodiments, each of the one or more performance targets may be associated with a particular capability and the second node selects the performance targets that align with a capability of the second node.

[0171] In particular embodiments, receiving the ML model training dataset or the ML model testing dataset comprises receiving an identifier or parameter associated with the dataset and obtaining the data for the dataset based on the identifier or parameter.

[0172] In general, some embodiments include association between the performance target with other information shared by a first node to a second node, such as the training dataset, the encoder output associated with the training dataset, and / or the model parameters (and additionally the model structure) of the second node-side model(s) or the first node-side model(s). Some embodiments include the format / definition for the performance target, performance target(s) for multiple configurations, and / or relation between the performance target and performance monitoring.

[0173] At step 1218, the second node trains the second node-side ML model based on the training dataset, the testing dataset, and the one or more performance targets.P113381WO01 PCT APPLICATION 40 of 51

[0174] In particular embodiments, the one or more performance targets serve as guidance for the second node to train its model part, so that the models trained separately at the first node and the second node may be compatible with each other. The one or more performance targets assist the second node (or a trainer-node / server associated with the second node) to assess if the trained model(s), e.g., second node-side model(s), have converged in such a way that it is expected to have a reasonable performance without having to test it against a model running on the first node (or a trainer-node / server associated with the first node),. This helps achieve better isolation between first node-side training and second node-side training.

[0175] In particular embodiments, the first node comprises a base station and the second node comprises a UE. In particular embodiments, the first node comprises a first UE and the second node comprises a second UE.

[0176] In particular embodiments, the second node trains the second node-side ML model according to any of the embodiments and examples described herein.

[0177] Modifications, additions, or omissions may be made to method 1200 of FIGURE 12. Additionally, one or more steps in the method of FIGURE 12 may be performed in parallel or in any suitable order.

[0178] FIGURE 13 is a flowchart illustrating an example method 1300 in a first node, according to certain embodiments. In particular embodiments, one or more steps of FIGURE 13 may be performed by network node 300 described with respect to FIGURE 8 or UE 200 described with respect to FIGURE 7.

[0179] The method may begin at step 1312, where the first node (e.g., network node 300, UE 200) trains a first node-side ML model using a ML model training dataset or ML model parameters, a ML model testing dataset, and a nominal second node side ML model.

[0180] The first node may train the first node -side ML model according to any of the embodiments and examples described herein.

[0181] At step 1314, the first node determines one or more performance targets based on training the first node-side ML model. The performance targets are described in more detail with respect to FIGURE 12 and with respect to the embodiments and examples described herein.

[0182] At step 1316, the first node transmits a ML model training dataset or ML model parameters for training the second node-side ML model and the one or more performance targets to the second node. In particular embodiments, the method further comprisesP113381WO01 PCT APPLICATION 41 of51transmitting the ML model testing dataset for testing the second node-side ML model to the second node.

[0183] The first node may transmit the information to the second node according to any of the embodiments and examples described herein.

[0184] Modifications, additions, or omissions may be made to method 1300 of FIGURE 13. Additionally, one or more steps in the method of FIGURE 13 may be performed in parallel or in any suitable order.

[0185] Modifications, additions, or omissions may be made to the methods disclosed herein without departing from the scope of the invention. The methods may include more, fewer, or other steps. Additionally, steps may be performed in any suitable order.

[0186] Some example embodiments follow.Group A Embodiments1. A method performed by a user equipment (UE) for two-sided machine learning (ML) model sequential training, the method comprising:receiving a ML model training dataset from a network node for training a UE-side ML model;receiving a ML model testing dataset from the network node for testing the UE-side ML model;receiving one or more performance targets associated with the UE-side ML model from the network node; andtraining the UE-side ML model based on the training dataset, testing dataset, and one or more performance targets.2. The method of the previous embodiment, wherein the training dataset comprises any of the training datasets described herein.3. The method of any one of the previous embodiments, wherein the testing dataset comprises any of the testing datasets described herein.4. The method of any one of the previous embodiments, wherein the performance targets comprise any of the performance targets described herein.P113381WO01 PCT APPLICATION 42 of 515. A method performed by a wireless device, the method comprising:any of the wireless device steps, features, or functions described above, either alone or in combination with other steps, features, or functions described above.6. The method of the previous embodiment, further comprising one or more additional wireless device steps, features or functions described above.Group B Embodiments7. A method performed by a network node for two-sided machine learning (ML) model sequential training, the method comprising:transmitting a ML model training dataset to a user equipment (UE) for training a UE-side ML model;transmitting a ML model testing dataset to the UE for testing the UE-side ML model; andtransmitting one or more performance targets associated with the UE-side ML model to the UE.8. The method of the previous embodiment, wherein the training dataset comprises any of the training datasets described herein.9. The method of any one of the previous two embodiments, wherein the testing dataset comprises any of the testing datasets described herein.10. The method of any one of the previous three embodiments, wherein the performance targets comprise any of the performance targets described herein.11. A method performed by a network node, the method comprising:any of the steps, features, or functions described above with respect to a network node, either alone or in combination with other steps, features, or functions described above.12. The method of the previous embodiment, further comprising one or more additional network node steps, features or functions described above.P113381WO01 PCT APPLICATION 43 of 51Group C Embodiments13. A user equipment comprising :processing circuitry configured to perform any of the steps of any of the Group A embodiments; andpower supply circuitry configured to supply power to the processing circuitry.14. A network node comprising:processing circuitry configured to perform any of the steps of any of the Group B embodiments;power supply circuitry configured to supply power to the processing circuitry.15. A user equipment (UE) comprising:an antenna configured to send and receive wireless signals;radio front-end circuitry connected to the antenna and to processing circuitry, and configured to condition signals communicated between the antenna and the processing circuitry;the processing circuitry being configured to perform any of the steps of any of the Group A embodiments;an input interface connected to the processing circuitry and configured to allow input of information into the UE to be processed by the processing circuitry;an output interface connected to the processing circuitry and configured to output information from the UE that has been processed by the processing circuitry; anda battery connected to the processing circuitry and configured to supply power to the UE.

[0187] The foregoing description sets forth numerous specific details. It is understood, however, that embodiments may be practiced without these specific details. In other instances, well-known circuits, structures and techniques have not been shown in detail in order not to obscure the understanding of this description. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate functionality without undue experimentation.P113381WO01 PCT APPLICATION 44 of 51

[0188] References in the specification to “one embodiment,” “an embodiment,” “an example embodiment,” etc., indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to implement such feature, structure, or characteristic in connection with other embodiments, whether or not explicitly described.

[0189] Although this disclosure has been described in terms of certain embodiments, alterations and permutations of the embodiments will be apparent to those skilled in the art. Accordingly, the above description of the embodiments does not constrain this disclosure. Other changes, substitutions, and alterations are possible without departing from the scope of this disclosure, as defined by the claims below.P113381WO01 PCT APPLICATION 45 of 51CLAIMS:1. A method for two-sided machine learning (ML) model sequential training, the two-sided ML model comprising a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model, the method performed by the second node, the method comprising:receiving (1212) a ML model training dataset or ML model training parameter from a network node for training the second node -side ML model;receiving (1216) one or more performance targets associated with the second node-side ML model from the network node, wherein the one or more performance targets are based on training of the first node-side ML model using the testing dataset; andtraining (1218) the second node-side ML model based on the training dataset, a testing dataset, and the one or more performance targets.2. The method of claim 1, further comprising receiving (1214) the testing dataset from the network node for testing the second node-side ML model.3. The method of any one of claims 1-2, wherein the one or more performance targets comprise one or more of a second node-side output performance target and an end-to-end performance target.4. The method of any one of claims 1-3, wherein the one or more performance targets comprise one or more pairs of a second node-side output performance target and an end-to-end performance target associated with the paired second node-side output performance target.5. The method of any one of claims 1-4, wherein the one or more performance targets comprise multiple values representing statistics across samples of the testing dataset.6. The method of claim 5, wherein the multiple values representing statistics across samples of the testing dataset comprise a mean value and one or more values of one or more specified percentiles.

Claims

P113381WO01 PCT APPLICATION 46 of 517. The method of any one of claims 1-6, wherein the testing dataset may comprise multiple sub-datasets and each of the sub-datasets is associated with one or more performance targets.

8. The method of any one of claims 1-7, wherein the training dataset is associated with one or more configuration parameters and each of the one or more performance targets may be associated with a particular configuration parameter or parameter combination.

9. The method of any one of claims 1-8, wherein each of the one or more performance targets may be associated with a particular capability and the second node selects the performance targets that align with a capability of the second node.

10. The method of any one of claims 1-9, wherein receiving the ML model training dataset or the ML model testing dataset comprises receiving an identifier or parameter associated with the dataset and obtaining the data for the dataset based on the identifier or parameter.

11. The method of any one of claims 1-10, wherein the first node comprises a base station and the second node comprises a user equipment (UE).

12. The method of any one of claims 1-10, wherein the first node comprises a first user equipment (UE) and the second node comprises a second UE.

13. The method of any one of claims 1-12, wherein the two-sided ML model comprises an autoencoder / decoder ML model.

14. A second node (200, 300) capable of two-sided machine learning (ML) model sequential training, the two-sided ML model comprising a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model, the second node comprising processing circuitry (202, 302) operable to:P113381WO01 PCT APPLICATION 47 of 51receive a ML model training dataset from a network node for training the second nodeside ML model;receive one or more performance targets associated with the second node-side ML model from the network node, wherein the one or more performance targets are based on training of the first node-side ML model using the testing dataset; andtrain the second node-side ML model based on the training dataset, a testing dataset, and the one or more performance targets.

15. The second network node of claim 14, wherein the processing circuitry is operable to perform the steps of any one of claims 2-13.P113381WO01 PCT APPLICATION 48 of 5116. A method for two-sided machine learning (ML) model sequential training, the two-sided ML model comprising a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model, the method performed by the first node, the method comprising:training (1312) a first node-side ML model using a ML model training dataset or ML model parameters, a ML model testing dataset, and a nominal second node side ML model;determining (1314) one or more performance targets based on training the first nodeside ML model; andtransmitting (1316) a ML model training dataset or ML model parameters for training the second node-side ML model and the one or more performance targets to the second node.

17. The method of claim 16, further comprising transmitting (1316) the ML model testing dataset for testing the second node-side ML model to the second node.

18. The method of any one of claims 16-17, wherein the one or more performance targets comprise one or more of a second node-side output performance target and an end-to-end performance target.

19. The method of any one of claims 16-18, wherein the one or more performance targets comprise one or more pairs of a second node-side output performance target and an end-to-end performance target associated with the paired second node-side output performance target.

20. The method of any one of claims 16-19, wherein the one or more performance targets comprise multiple values representing statistics across samples of the testing dataset.

21. The method of claim 20, wherein the multiple values representing statistics across samples of the testing dataset comprise a mean value and one or more values of one or more specified percentiles.

22. The method of any one of claims 16-21, wherein the testing dataset may comprise multiple sub-datasets and each of the sub-datasets is associated with one or moreP113381WO01 PCT APPLICATION 49 of 51performance targets.

23. The method of any one of claims 16-22, wherein the training dataset is associated with one or more configuration parameters and each of the one or more performance targets may be associated with a particular configuration parameter or parameter combination.

24. The method of any one of claims 16-23, wherein each of the one or more performance targets may be associated with a particular capability.

25. The method of any one of claims 16-24, wherein transmitting the ML model training dataset or the ML model testing dataset comprises transmitting an identifier or parameter associated with the dataset.

26. The method of any one of claims 16-25, wherein the first node comprises a base station and the second node comprises a user equipment (UE).

27. The method of any one of claims 16-25, wherein the first node comprises a first user equipment (UE) and the second node comprises a second UE.

28. The method of any one of claims 16-27, wherein the two-sided ML model comprises an autoencoder / decoder ML model.

29. A first node (200, 300) capable of two-sided machine learning (ML) model sequential training, the two-sided ML model comprising a first node-side ML model and a second node-side ML model where the output of the second node-side ML model is input to the first node-side ML model, the first node comprising processing circuitry (202, 302) operable to:train a first node -side ML model using a ML model training dataset or ML model parameters, a ML model testing dataset, and a nominal second node side ML model;determine one or more performance targets based on training the first node-side ML model; andtransmit a ML model training dataset or ML model parameters for training the secondP113381WO01 PCT APPLICATION 50 of 51node-side ML model and the one or more performance targets to the second node.

30. The first node of claim 29, the processing circuitry further operable to perform the steps of any one of claims 17-28.

Images