Model framework
The model framework with a main module and sub-modules for specific cells/sites enhances CSI feedback efficiency and adaptability, reducing complexity and storage needs while maintaining performance.
Patent Information
- Application Number
- PCT/CN2024/077779
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-20
- Publication Date
- 2025-08-28
AI Technical Summary
Existing communication systems face challenges in achieving efficient CSI feedback across different cells/sites with high complexity and resource consumption, as well as difficulties in adapting models to real-world deployments.
A model framework with a main module parallel connected to sub-modules tailored for specific cells/sites, allowing for optimized performance and reduced storage requirements, utilizing a new signaling framework for real-world deployment.
The proposed framework achieves better performance and parameter efficiency with minimal inference latency, addressing the trade-offs of existing models by enabling adaptable CSI feedback.
Smart Images

Figure CN2024077779_28082025_PF_FP_ABST
Abstract
Description
MODEL FRAMEWORKFIELD
[0001] Example embodiments of the present disclosure generally relate to the field of communications, and in particular, to a terminal device, a network device, methods, apparatuses, and a computer-readable medium for a model framework.BACKGROUND
[0002] A communication network can be seen as a facility that enables communications between two or more communication devices, or provides communication devices access to a data network. A mobile or wireless communication network is one example of a communication network.
[0003] Such communication networks operate in according with standards such as those provided by 3GPP (Third Generation Partnership Project) or ETSI (European Telecommunications Standards Institute) . Examples of standards are the so-called 5G (5th Generation) standards provided by 3GPP.SUMMARY
[0004] In general, example embodiments of the present disclosure provide a solution for a model framework, especially a model adaptation framework for channel state information (CSI) feedback enhancement.
[0005] In a first aspect, there is provided an apparatus. The apparatus comprises at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: train a main module and a plurality of sub-modules parallel connectable to the main module for channel state information (CSI) feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and perform, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0006] In a second aspect, there is provided a method. The method comprises: training a main module and a plurality of sub-modules parallel connectable to the main module for channel state information (CSI) feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and performing, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0007] In a third aspect, there is provided an apparatus. The apparatus comprises: means training a main module and a plurality of sub-modules parallel connectable to the main module for channel state information (CSI) feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and means for performing, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0008] In a fourth aspect, there is provided a non-transitory computer-readable storage medium having instructions stored thereon. The instructions, when executed on at least one processor, cause the at least one processor to perform the method of the second aspect.
[0009] In a fifth aspect, there is provided a computer program comprising instructions, which, when executed by an apparatus, cause the apparatus at least to: train a main module and a plurality of sub-modules parallel connectable to the main module for channel state information (CSI) feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and perform, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0010] In a sixth aspect, there is provided an apparatus. The apparatus comprises: training circuitry configured to train a main module and a plurality of sub-modules parallel connectable to the main module for channel state information (CSI) feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and performing circuitry configured to perform, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0011] In a seventh aspect, there is provided a terminal device. The terminal device comprises at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the terminal device at least to: receive, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; and connect, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0012] In an eighth aspect, there is provided a network device. The network device comprises at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the network device at least to: select, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; and transmit, to the terminal device, an index of the selected sub-module.
[0013] In a ninth aspect, there is provided a method. The method comprises: receiving, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; and connecting, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0014] In a tenth aspect, there is provided a method. The method comprises: selecting, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; and transmitting, to the terminal device, an index of the selected sub-module.
[0015] In an eleventh aspect, there is provided an apparatus. The apparatus comprises: means for receiving, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; and based on the index, means for connecting a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0016] In a twelfth aspect, there is provided an apparatus. The apparatus comprises: means for selecting, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; and means for transmitting, to the terminal device, an index of the selected sub-module.
[0017] In a thirteenth aspect, there is provided a non-transitory computer-readable storage medium having instructions stored thereon. The instructions, when executed on at least one processor, cause the at least one processor to perform the method of the ninth or tenth aspect.
[0018] In a fourteenth aspect, there is provided a computer program comprising instructions, which, when executed by an apparatus, cause the apparatus at least to: receive, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; and connect, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0019] In a fifteenth aspect, there is provided a computer program comprising instructions, which, when executed by an apparatus, cause the apparatus at least to: select, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; and transmit, to the terminal device, an index of the selected sub-module.
[0020] In a sixteenth aspect, there is provided a terminal device. The terminal device comprises: receiving circuitry configured to receive, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; and connecting circuitry configured to connect, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0021] In a seventeenth aspect, there is provided a network device. The network device comprises: selecting circuitry configured to select, based on a site identifier (ID) , a sub-module among one or more sub-modules at a network device to be parallel connected to a main decoder at the network device; and transmitting circuitry configured to transmit, to a terminal device, an index of the selected sub-module.
[0022] It is to be understood that the summary section is not intended to identify key or essential features of embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will become easily comprehensible through the following description.BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Some example embodiments will now be described with reference to the accompanying drawings, in which:
[0024] FIG. 1 illustrates an example network environment in which some example embodiments of the present disclosure may be implemented;
[0025] FIG. 2 illustrates a flowchart of an example method in accordance with some embodiments of the present disclosure;
[0026] FIG. 3 illustrates a signaling chart illustrating an example communication process in accordance with some example embodiments of the present disclosure;
[0027] FIG. 4A illustrates a schematic diagram of an example model adaptation framework in accordance with some embodiments of the present disclosure;
[0028] FIG. 4B illustrates another schematic diagram of an example model adaptation framework in accordance with some embodiments of the present disclosure;
[0029] FIG. 5A illustrates a block diagram of an example of sub modules parallel connectable to the main module in accordance with some embodiments of the present disclosure;
[0030] FIG. 5B illustrates an example of input / output data of a main module and a sub module parallel connected to the main module in accordance with some embodiments of the present disclosure;
[0031] FIG. 6A illustrates a block diagram of an example alternative hybrid training process in accordance with some embodiments of the present disclosure;
[0032] FIG. 6B illustrates another block diagram of an example alternative hybrid training process in accordance with some embodiments of the present disclosure;
[0033] FIG. 7A illustrates a block diagram of an example growable sub modular training process in accordance with some embodiments of the present disclosure;
[0034] FIG. 7B illustrates another block diagram of an example growable sub modular training process in accordance with some embodiments of the present disclosure;
[0035] FIG. 7C illustrates further another block diagram of an example growable sub modular training process in accordance with some embodiments of the present disclosure;
[0036] FIG. 8 illustrates a block diagram for model inference in accordance with some embodiments of the present disclosure;
[0037] FIG. 9 illustrates a signaling chart illustrating another example communication process in accordance with some embodiments of the present disclosure;
[0038] FIG. 10 illustrates a flowchart of another example method implemented at a terminal device in accordance with some embodiments of the present disclosure;
[0039] FIG. 11 illustrates a flowchart of further another example method implemented at a network device in accordance with some embodiments of the present disclosure;
[0040] FIG. 12A illustrates a block diagram of an example of a transformer-enabled neural network (NN) for CSI feedback in accordance with some example embodiments of the present disclosure;
[0041] FIG. 12B illustrates a block diagram of a modified multi-head attention block in accordance with some example embodiments of the present disclosure;
[0042] FIG. 13 illustrates a simplified block diagram of a device that is suitable for implementing some example embodiments of the present disclosure; and
[0043] FIG. 14 illustrates a block diagram of an example of a computer-readable medium in accordance with some example embodiments of the present disclosure.
[0044] Throughout the drawings, the same or similar reference numerals represent the same or similar elements.DETAILED DESCRIPTION
[0045] Principles of the present disclosure will now be described with reference to some example embodiments. It is to be understood that these embodiments are described for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. The disclosure described herein can be implemented in various manners other than the ones described below.
[0046] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this disclosure belongs.
[0047] References in the present disclosure to “one embodiment, ” “an embodiment, ” “an example embodiment, ” and the like indicate that the embodiment described may include a particular feature, structure, or characteristic, but it is not necessary that every embodiment includes 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 affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.
[0048] It shall be understood that although the terms “first” and “second” etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of example embodiments. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.
[0049] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting of example embodiments. As used herein, the singular forms “a” , “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” , “comprising” , “has” , “having” , “includes” and / or “including” , when used herein, specify the presence of stated features, elements, and / or components etc., but do not preclude the presence or addition of one or more other features, elements, components and / or combinations thereof. As used herein, “at least one of the following: <a list of two or more elements>” and “at least one of <a list of two or more elements>” and similar wording, where the list of two or more elements are joined by “and” or “or” , mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.
[0050] As used in this application, the term “circuitry” may refer to one or more or all of the following:
[0051] (a) hardware-only circuit implementations (such as implementations in only analog and / or digital circuitry) and
[0052] (b) combinations of hardware circuits and software, such as (as applicable) :
[0053] (i) a combination of analog and / or digital hardware circuit (s) with software / firmware and
[0054] (ii) any portions of hardware processor (s) with software (including digital signal processor (s) ) , software, and memory (ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions) and
[0055] (c) hardware circuit (s) and or processor (s) , such as a microprocessor (s) or a portion of a microprocessor (s) , that requires software (for example, firmware) for operation, but the software may not be present when it is not needed for operation.
[0056] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.
[0057] As used herein, the term “communication network” refers to a network following any suitable communication standards, such as Long Term Evolution (LTE) , LTE-Advanced (LTE-A) , Wideband Code Division Multiple Access (WCDMA) , High-Speed Packet Access (HSPA) , Narrow Band Internet of Things (NB-IoT) , Wireless Fidelity (WiFi) and so on. Furthermore, the communications between a terminal device and a network device in the communication network may be performed according to any suitable generation communication protocols, including, but not limited to, the fourth generation (4G) , 4.5G, the future fifth generation (5G) , IEEE 802.11 communication protocols, and / or any other protocols either currently known or to be developed in the future. Embodiments of the present disclosure may be applied in various communication systems. Given the rapid development in communications, there will of course also be future type communication technologies and systems with which the present disclosure may be embodied. It should not be seen as limiting the scope of the present disclosure to only the aforementioned system.
[0058] As used herein, the term “network device” refers to a node in a communication network via which a terminal device accesses the network and receives services therefrom. The network device may refer to a base station (BS) or an access point (AP) , for example, a node B (NodeB or NB) , an evolved NodeB (eNodeB or eNB) , a NR NB (also referred to as a gNB) , a Remote Radio Unit (RRU) , a radio header (RH) , a remote radio head (RRH) , a WiFi device, a relay, a low power node such as a femto, a pico, and so forth, depending on the applied terminology and technology. In the following description, the terms “network device” , “AP device” , “AP” and “access point” may be used interchangeably.
[0059] The term “terminal device” refers to any end device that may be capable of wireless communication. By way of example rather than limitation, a terminal device may also be referred to as a communication device, user equipment (UE) , a Subscriber Station (SS) , a Portable Subscriber Station, a Mobile Station (MS) , a station (STA) or station device, or an Access Terminal (AT) . The terminal device may include, but not limited to, a mobile phone, a cellular phone, a smart phone, voice over IP (VoIP) phones, wireless local loop phones, a tablet, a wearable terminal device, a personal digital assistant (PDA) , portable computers, desktop computer, image capture terminal devices such as digital cameras, gaming terminal devices, music storage and playback appliances, vehicle-mounted wireless terminal devices, wireless endpoints, mobile stations, laptop-embedded equipment (LEE) , laptop-mounted equipment (LME) , USB dongles, smart devices, wireless customer-premises equipment (CPE) , an Internet of Things (loT) device, a watch or other wearable, a VR (virtual reality) device, an XR (eXtended reality) device, a head-mounted display (HMD) , a vehicle, a drone, a medical device and applications (for example, remote surgery) , an industrial device and applications (for example, a robot and / or other wireless devices operating in an industrial and / or an automated processing chain contexts) , a consumer electronics device, a device operating on commercial and / or industrial wireless networks, and the like. In the following description, the terms “station” , “station device” , “STA” , “terminal device” , “communication device” , “terminal” , “user equipment” and “UE” may be used interchangeably.
[0060] Communication performance is of great importance to communications. However, good communication performance always relates to high complexity. Therefore, trade-off between performance and complexity / overhead is under study. To this end, one of research topic is cell / site-specific models, and Artificial Intelligence / Machine Learning (AI / ML) models are used for CSI feedback enhancement. A cell / site-specific model tailors (or customizes) its parameters and configurations to the specific characteristics and requirements of an individual cell / site. In other words, a cell / site-specific model aligns with the unique conditions of each cell / site.
[0061] To achieve good performance of an AI / ML model across different cells / sites for CSI feedback enhancement, there are three typical approaches, i.e., a specific model trained for specific cells / sites, or a general model trained with mixing datasets, or a model fine-tuned to fit the deployed environment. If a specific model is trained for specific cells / sites, training models for specific cell / site allows for tailored feature extraction, capturing the most relevant information and discarding unnecessary details for improved efficiency. In this case, when the cell / site where a device (for example, a use equipment (UE) or a network device like a gNB) is located changes, the device may switch from a model specific to the cell / site before the location change to a model specific to the cell / site after the location change. However, the device should store multiple models to cover different cells / sites, which is storage consuming and resource intensive. If a general model is trained with mixing datasets from multiple cells / sites, the model can achieve consistent performance across different cells / sites. However, there may be a performance loss compared to the specific model in some cells / sites. If a model is fine-tuned to fit the deployed environment, the adaptation of a single model ensures the model to remain effective in the deployed environment. However, the model update needs real-time adjustment to the neural network (NN) parameters, which is difficult to implement in real-world deployment.
[0062] Since scenario / configuration / site-specific models may provide performance benefits to the general models in some cases, this concept focuses on a new model adaptation framework to cover different scenarios / configurations / sites in CSI feedback enhancement. More specifically, considering the advantages and disadvantages of the above three approaches (i.e., a specific model trained for specific cells / sites, or a general model trained with mixing datasets from multiple cells / sites, or a model fine-tuned to fit the deployed environment) , a main module parallel connectable with sub modules specified for different cells / sites is proposed in the present disclosure. Such design enables optimized performance in various cells / sites and also reduces storage requirements by only storing several small-scale parallel-connectable sub modules instead of the whole models. On the other hand, since there have been no such models designed for CSI feedback enhancement, new signalling framework is also proposed in some embodiments of this disclosure to investigate standard impact of the real-world deployment.
[0063] FIG. 1 illustrates an example communication system 100 in which some embodiments of the present disclosure can be implemented. The communication system 100, which is a part of a communication network, includes a terminal device (UE) 110 and a network device 120. The terminal device 110 may be, for example, an Internet of Things (IoT) device. The network device 120 may be for example a random access network (RAN) device (like an NG-RAN device, also called as gNB) , or a communication module thereof. The network device 120 is associated with a cell 121, and provides communication service to terminal devices (like UE 110) in the cell 121. As illustrated in FIG. 1, the terminal device 110 is in connection with the network device 120.
[0064] In the system 100, a link from the network device 120 to terminal device 110 is referred to as a downlink (DL) , while a link from terminal device 110 to the network device 120 is referred to as an uplink (UL) . In downlink, the network device 120 is a transmitting (TX) device (or a transmitter) and terminal device 110 is a receiving (RX) device (or a receiver) . In uplink, terminal device 110 is a transmitting TX device (or a transmitter) and the network device 120 is a RX device (or a receiver) .
[0065] The communications in the communication system 100 may conform to any suitable standards including, but not limited to, Long Term Evolution (LTE) , LTE-Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA) , Code Division Multiple Access (CDMA) and Global System for Mobile Communications (GSM) and the like. Furthermore, the communications may be performed according to any generation communication protocols either currently known or to be developed in the future. Examples of the communication protocols include, but not limited to, the first generation (1G) , the second generation (2G) , 2.5G, 2.75G, the third generation (3G) , the fourth generation (4G) , 4.5G, the fifth generation (5G) , 5.5G, 5G-Advanced networks, or the sixth generation (6G) communication protocols.
[0066] It is to be understood that the number of devices (including terminal device 110 and the network device 120) and their connection relationships and types shown in FIG. 1 are only for illustrative purposes without suggesting any limitation. The communication system 100 may include any suitable number of devices adapted for implementing embodiments of the present disclosure.
[0067] FIG. 2 illustrates a flowchart of an example method 200 implemented at an apparatus (for example, the terminal device 110 or network device 120 as illustrated in FIG. 1) in accordance with some embodiments of the present disclosure. For the purpose of discussion and for simplicity, the method 200 will be described from the perspective of the network device 120 with reference to FIG. 2. Those skilled in the art know that the method 200 can also be described from the perspective of the terminal device 110.
[0068] As illustrated in FIG. 2, at block 210, the network device 120 trains a main module and a plurality of sub-modules parallel connectable to the main module for CSI feedback. Here, the plurality of sub-modules are associated with a plurality of sites of different wireless environments (for example, multi-path conditions) . The image of “amain module and a plurality of sub-modules parallel connectable to the main module” may refer to FIGS. 4A, 4B or 5A, which will be described in more detail later. The main module may be or comprise an encoder (herein also referred to as “amain encoder” ) in case that the method 200 is described from the perspective of the terminal device 110. Alternatively, the main module may be or comprise a decoder (herein also referred to as “amain decoder” ) in case that the method 200 is described from the perspective of the network device 120. At block 220, the network device 120 performs, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0069] For example, the main module and the plurality of sub-modules may be within at least one linear layer in an AI / ML model, as illustrated in FIGS. 4B and 5A. The linear layer may be a neural network (NN) layer. Alternatively or in addition, number of the plurality of sub-modules may be predetermined corresponding to number of the plurality of sites. Alternatively or in addition, the plurality of sub-modules may be parallel connectable to the main module via respective switches, as illustrated, for example, in FIGS. 4B and 5A. Alternatively or in addition, if a sub-module is parallel connected to the main module, a same input may be provided to the main module and the sub-module to obtain a first output from the main module and a second output from the sub-module, then, for example, the first output and the second output may be added to obtain the overall output, as illustrated in FIG. 5B. In other words, if a sub-module is parallel connected to the main module, a sum of a first output of the main module and a second output of the sub-module may be provided to a next layer of the AI / ML model.
[0070] In order to train the main module and the plurality of sub-modules, the network device 120 may perform a training process based on a plurality of datasets associated with the plurality of sites, and the training process may comprise a plurality of training steps. More specifically, among the plurality of datasets, a dataset may be specific to a site among the plurality of sites. A site may be a location where the network device 120 is located, like a cell (for example, cell 121 as illustrated in FIG. 1) . During the training process, the network device 120 may train a sub-module jointly with the main module independently (separate) from other sub modules. Alternatively or in addition, the network device 120 may train the main module based on a general dataset. Alternatively or in addition, the network device 120 may train the plurality of sub-modules based on the plurality of datasets respectively.
[0071] In a training step among the plurality of training steps, the main module may be parallel connected with a sub-module among the plurality of sub-modules. In this case, the main module and the sub-module may be trained based on a dataset among the plurality of datasets. Here, the dataset may correspond to a site among the plurality of sites and the cite may be associated with the sub-module, as mentioned above.
[0072] In the training process, parameters of the main module may be updated based on an overall loss which is a sum of a plurality of sub-losses determined based on the plurality of datasets in the plurality of training steps. Also, in the training process, parameters of the sub-module may be updated based on a sub-loss among the plurality of sub-losses. Here, the sub-loss may be determined based on the dataset in a specific training step among the plurality of training steps. More specifically, in case there are n sub modules (n is the number of the plurality of sub modules, and also the number of the plurality of sites, and the number of the plurality of datasets) , sub-loss-i (i = 1, 2, …, n) may be determined based on dataset-i in a training step when training sub module-i. Here, it is assumed that sub module-i is associated with cell / site-i, whose dataset is denoted as dataset-i, and the sub loss in the training step when training sub module-i is sub-loss-i.
[0073] Under some circumstances, before the plurality of training steps, the network device 120 may train the main module based on a general dataset. The general dataset may be, for example, a mixture of the plurality of datasets associated with the plurality of sites. In a training step among the plurality of training steps, the main module may be parallel connected with a sub-module among the plurality of sub-modules, and the main module may be maintained unchanged and the sub-module may be trained based on a dataset among the plurality of datasets. Here, the dataset may correspond to a site among the plurality of sites, and the site may be associated with the sub-module. In this case, in a forward pass of the training step, data from the dataset may be fed through (input to) both the main module and the sub module to obtain a loss. In a back propagation of the training step, gradients of the loss with respect to trainable parameters of the sub-module may be computed, while the parameters of the main module may remain unaffected by the gradients calculated for the sub-module.
[0074] In order to perform the inference, the network device 120 may calculate a plurality of statistical discrepancies between in-field data and a plurality of datasets associated with the plurality of sites. Then, the network device 120 may determine a target site among the plurality of sites based on the plurality of statistical discrepancies, and determine, among the plurality of sub-modules, a target sub-module associated with the target site as the selected sub-module to be parallel connected to the main module for the inference. For example, the target site may correspond to a dataset with a lowest statistical discrepancy among the plurality of statistical discrepancies.
[0075] Through the method 200, a model architecture (also referred to as model generalization / adaptation framework) can be obtained. With such a model architecture, better performance and parameter efficiency can be obtained. Besides, no additional inference latency is introduced.
[0076] FIG. 3 illustrates a signaling chart illustrating an example communication process 300 in accordance with some example embodiments of the present disclosure. For the purpose of discussion, the communication process 300 will be described with reference to FIG. 1. The communication process 300 may involve a terminal device (for example, the terminal device 110 as illustrated in FIG. 1) and a network device (for example, the network device 120 as illustrated in FIG. 1) . In the following, the communication process 300 will be described with reference to the terminal device 110 and the network device 120 as illustrated in FIG. 1.
[0077] As illustrated in FIG. 3, at 310, the network device 120 selects, based on a site identifier (ID) , a sub-module (hereafter, also referred to as “first sub-module” ) among one or more sub-modules (hereafter, also referred to as “first one or more sub-modules” ) at the network device 120 to be parallel connected to a main decoder at the network device 120.
[0078] Then, the network device 120 transmits (320) an index 301 of the selected (first) sub-module to the terminal device 110. On the other side of the communication, the terminal device receives (322) , from the network device 120, the index 301 of the first sub-module among the first one or more sub-modules at the network device 120. Here, the first sub-module is to be parallel connected to the main decoder at the network device 120, as mentioned above.
[0079] At block 330, the terminal device 110 connects, based on the index 301, a second sub-module among second one or more sub-modules at the terminal device 110 parallel to a main encoder at the terminal device 110. Here, the second sub-module is associated with the first sub-module.
[0080] In some circumstances, after the terminal device 110 connects to the second sub-module, the terminal device 110 may transmit, to the network device 120, a message confirming that the second sub-module is to be parallel connected to the main encoder at the terminal device 110. On the other side of communication, the network device 120 may receive, from the terminal device 110, the message confirming that the second sub-module at the terminal device 110 corresponding to the index 301 is to be parallel connected to the main encoder of the terminal device 110.
[0081] In some example embodiments, based on detecting a site drift (asite drift may occur when the terminal device 110 moves from a location to another location, for example, from a cell to another cell, from a restaurant to a mall, etc. ) , the terminal device 110 may transmit a request message to the network device 120 for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device 120 corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device 110. Here, “encoder sub-module” means a sub module among the second one or more sub-modules to be parallel connected with the (main) encoder (the main module) at the terminal device 110, and the request message may comprise a first ID list of the second one or more sub-modules at the terminal device 110. The first ID list may comprise one or more IDs.
[0082] On the other side of communication, the network device 120 may receive, from the terminal device 110, the request message for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device 120 corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device 110, and determine, based on the first ID list of the second on or more sub-modules comprised in the request message, whether there is the at least one decoder sub-module. Here, “decoder sub-module” means a sub module among the first one or more sub-modules to be parallel connected with the (main) decoder (the main module) at the network device 120. Based on determining that there is at least one decoder sub-module, the network device 120 may transmit, to the terminal device 110, a response message corresponding to the request message comprising a second ID list of the at least one decoder sub-module. Similar to the first ID list, the second ID list may comprise one or more IDs. Then, at the terminal device 110, the terminal device 110 may receive, from the network device 120, the response message comprising the second ID list of the at least one decoder sub-module.
[0083] Alternatively, in some other example embodiments, based on detecting a site drift, the network device 120 may transmit a request message to the terminal device 110 for determining whether there is at least one encoder sub-module among the second one or more sub-modules in the terminal device 110 corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device 120. Here, the request message may comprise a first ID list of the first one or more sub-modules at the network device 120. On the other side of communication, the terminal device 110 may receive, from the network device 120, a request message for determining whether there is at least one encoder sub-module among the second one or more sub-modules at the terminal device 110 corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device 120. The request message may comprise the first ID list of the first one or more sub-modules at the network device.
[0084] The terminal device 110 may then determine, based on the first ID list of the first one or more sub-modules, whether there is the at least one encoder sub-module. Based on determining that there is the at least one encoder sub-module, the terminal device 110 may transmit, to the network device, a response message corresponding to the request message comprising a second ID list of the at least one encoder sub-module. Then, the network device 120 may receive, from the terminal device 110, the response message corresponding to the request message comprising the second ID list of the at least one encoder sub-module.
[0085] The terminal device 110 may also transmit pre-defined information to the network device 120 for selecting the first sub-module. On the other side of communication, the network device 120 may receive, from the terminal device 110, the pre-defined information for selecting the sub-module. The pre-defined information may comprise channel state information (CSI) . The channel state information (CSI) may comprise original CSI and / or compressed CSI codewords. Based on the pre-defined information, the network device 120 may classify the drifted site to determine the site ID.
[0086] The terminal device 110 may maintain a sub-module pool which comprises the second one or more sub-modules. Sub-modules in the sub-module pool may be associated with different application conditions (for example, multi-path conditions) . Similarly, the network device 120 may maintain a sub-module pool which comprises the (first) one or more sub-modules. Sub-modules in the sub-module pool may also be associated with different application conditions (for example, multi-path conditions) . With the communication process 300, better performance and parameter efficiency can be obtained. Besides, no additional inference latency is introduced.
[0087] Hereinbefore, some examples of the present disclosure are generally described with reference to a flow chart (i.e., FIG. 2) and a high level signaling chart (i.e., FIG. 3) . In the following, some further examples of the present disclosure are described with reference to FIGS. 4A-12.
[0088] FIG. 4A illustrates a schematic diagram 400A of an example model adaptation framework (also referred to as “model architecture” ) in accordance with some embodiments of the present disclosure. In the example model adaptation framework as illustrated in FIG. 4A, a main module is connected in parallel with small-scale sub modules specified for different cells / sites of different wireless environments (for example, multi-path conditions) for CSI feedback enhancement. Considering models should be deployed at both UE (for example, terminal device 110 as illustrated in FIGS. 1 and 3) and gNB (for example, network device 120 as illustrated in FIGS. 1 and 3) for CSI feedback enhancement, new signalings are introduced for sub module alignment in real-world deployment, which will be described in more detail with reference to FIG. 9.
[0089] Specifically, as shown in FIG. 4A, UE possesses a main encoder parallel connected with one of the sub modules from the sub module pool at UE side corresponding to a specific cell / site, and gNB possesses a main decoder parallel connected with one of the sub modules from the sub module pool at gNB side corresponding to the specific cell / site. The main encoder (also referred to as “main module” ) together with the sub modules specific for different cells / sites are pre-trained in advance to ensure consistency between training stage and inference stage.
[0090] Specifically, there may be sub-module pair between the UE and gNB. A sub module in the sub-module pair at UE may be used in pair with the other sub module in the sub-module pair at gNB, since the sub modules in a sub-module pair are both specific for a same cell / site. For example, as illustrated in FIG. 4A, the sub module-1 at UE and the sub module-1 at gNB may be a sub-module pair. In other words, when the UE uses the sub module-1 in parallel with the main encoder, the gNB uses the sub module-1 in parallel with main decoder to successfully decode UL transmission from UE to gNB. The UL transmission may be, for example, channel state information (CSI) . In this case, considering the sub module-1 at UE and the sub module-1 at gNB are used in pair, the sub module-1 at UE may also be referred to as corresponding to the sub module-1 at gNB.
[0091] FIG. 4B illustrates another schematic diagram 400B of an example model adaptation framework for encoder / decoder in CSI feedback in accordance with some embodiments of the present disclosure. In the example model adaptation framework as illustrated in FIG. 4B, a main module is connected in parallel with small-scale sub modules specified for different cells / sites for CSI feedback. As described above, the main module at UE may be a (main) encoder, and the main module at gNB may be a (main) decoder, as illustrated in FIG. 4A. The example model adaptation framework may be utilized to enable the proposed model generalization framework in CSI feedback enhancement.
[0092] Specifically, UE possesses a main encoder parallel connected with one of the sub modules from the sub module pool at UE side corresponding to a specific cell / site, and gNB possesses a main decoder parallel connected with one of the sub modules from the sub module pool at gNB side corresponding to the specific cell / site. The main module and the sub modules specific for different cells / sites of different wireless environments (for example, multi-path conditions) may be pre-trained together in advance to ensure consistency between training stage and inference stage.
[0093] As shown in FIG. 4B, the “main module” box represents the NN layers in the main module and the “sub module-i” (i = 1, 2, …, n) boxes represent the NN layers in the sub modules for specific cell / sites. Here, n represents the number of the sub modules, and i represents the index of a specific sub module among the n sub modules. Each cell / site-specific sub module is parallel connectable to the main module with a switch. When a specific cell / site is detected ( / selected) , the corresponding switch can be turned on to activate that specific sub module branch. Such sub modules for specific cell / sites can be deployed at any dense layers in the main module, while the effectiveness may be different across different layers.
[0094] Hereafter an innovative implementation method for obtaining and deploying the model adaptation framework as illustrated in FIG. 4A or 4B will be discussed with reference to FIGS. 5A-8. In the model adaptation framework, the main module and the parallel connectable cell / site-specific sub modules may be deployed at a terminal device (for example, the terminal device 110 as illustrated in FIGS. 1 and 3) and a network device (for example, the network device 120 as illustrated in FIGS. 1 and 3) for model adaptiveness enhancement in CSI feedback.
[0095] FIG. 5A illustrates an example block diagram 500A of an example of sub modules parallel connectable to the main module in accordance with some embodiments of the present disclosure.
[0096] The model adaptation framework (for example, as illustrated in FIG. 4A) may be obtained by joint pre-training the main module and the parallel connectable sub modules within the linear layers in a CSI feedback model. For the CSI feedback encoder / decoder which consists of multiple linear layers, in one or more linear layers, a parallel sub module may be connected to the original NN structure (i.e., the main module shown in FIG. 5A) , the sub module and main module can be fed with same input, and output (s) from the sub module and output (s) from the main module can be numerically added and merged. Then the merged ( / summarized) output (s) may be fed to the next layer.
[0097] Each sub module (for example, sub module-1, sub module-2, …sub module-n as shown in FIG. 5A) may correspond to a specific cell / site. The sub module number may be pre-determined corresponding to the number of the given cells / sites.
[0098] Each sub module may be trained independently ( / separately) from other sub modules, but each sub module may be trained jointly with the main module. Herein by referring to “training” , it means all the NN model parameters (including weights and biases) are trainable. For example, for the 1st cell / site (also referred to as cell / site-1) , parameters in main module and sub module-1 (which corresponds to the 1st cell / site) are all updated according to the loss function calculated with dataset-1. Herein dataset-1 consists of input data and output labels of the 1st cell / site, which are both CSI matrices (subband *port *I / Q) in CSI feedback use case. Here, “I / Q” is short for “In-phase / Quadrature” .
[0099] For example, alternative hybrid training is designed for training the main module and the n (n is the number of the parallel-connectable sub modules) parallel-connectable sub modules from n different cells / sites, here, n is the number of sub modules which are parallel connectable to the main module. Since each sub module is specific to a cell / site, the number of the different cells / sites is also n. Here, by “alternative” training (among sub modules) , it means for each epoch (or, stage) of loss calculation, it consists of n sub-loss from n scenarios calculated on n datasets corresponding to the n cell / sites of different wireless environments (for example, multi-path conditions) , where each dataset is for a cell / site among the n different cells / sites. For each sub-loss-k (k = 1, 2, …, n) , dataset-k corresponds to site / cell-k is selected, and sub-loss-k is calculated via supervised learning on dataset-k. The overall loss is the summation of k sub-losses (i.e., sub-loss-1 for sub module-1, sub-loss-2 for sub module-2, …, sub-loss-n for sub module-n) .
[0100] By hybrid training (between the main module and a sub module) , it means that, the parameters of the main module are updated based on the overall loss (the sum of sub-loss-1, sub-loss-2, …, sub-loss-n) , while the parameters of sub module-k are updated only based on the sub-loss-k on the dataset-k.
[0101] FIG. 5B illustrates example input / output data 500B of a main module and a sub module parallel connected to the main module in accordance with some embodiments of the present disclosure. In the example illustrated in FIG. 5B, it is supposed that the i-th (i may be 1, 2, …n, as shown in FIG. 5A, where n is the number of the sub modules which are parallel connectable to the main module) branch of the n sub modules is activated, which means that the i-th cell / site-specific sub module is parallel connected to the main module.
[0102] As illustrated in FIG. 5B, the same input x is input into (fed to) the main module and sub module-i (the i-th sub module for the i-th cell / site) . The output of the main module is denoted as z0, and the output of the sub module-i is denoted as zi. z0 and zi are of the same shape (i.e., of the same size) and thus can be numerically added together as the output z, namely z=z0+zi. Then the output z is fed into the next layer of the main module.
[0103] The main module and the parallel-connectable sub modules are trained using different datasets. The main module may be trained with the general dataset, e.g. the mixture of different datasets of different cells / sites, while the i-th sub module may be trained with the dataset specific for the i-th cell / site.
[0104] More specifically, the main module and the parallel connectable sub modules can be trained using “alternative hybrid training” (as roughly described above) or another training approach named “growable sub modular training” . The detailed steps of the two training approaches are illustrated below with reference to FIGS. 6A-7C, among which FIGS. 6A-6B illustrates block diagrams for different steps of an example alternative hybrid training process in accordance with some embodiments of the present disclosure, and FIGS. 7A-7C illustrates block diagrams for different steps of an example growable sub modular training process in accordance with some embodiments of the present disclosure.
[0105] FIG. 6A illustrates a block diagram 600A of an example alternative hybrid training process in accordance with some embodiments of the present disclosure. In the training step illustrated in FIG. 6A, the main module and the sub module-1 (the first sub module) are trained using the cell / site-specific dataset-1 with loss function L (e.g., cosine similarity (SGCS) loss, mean squared error (MSE) loss) . Other sub modules do not participate in both forward pass and backward propagation in this training step; in other words, in this step, among the n sub modules, only sub module-1 participates in forward pass and backward propagation.
[0106] FIG. 6B illustrates another block diagram 600B of an example alternative hybrid training process in accordance with some embodiments of the present disclosure. The training step illustrated in FIG. 6B may follow the training step illustrated in FIG. 6A. As illustrated in FIG. 6B, the main module and the sub module-2 (the 2nd sub module) are trained using the cell / site-specific dataset-2 with the same loss function L. Other sub modules do not participate in both forward pass and backward propagation in this training step; in other words, in this step, among the n sub modules, only sub module-2 participates in forward pass and backward propagation.
[0107] In this way, the alternative hybrid training process is conducted iteratively for all the sub modules (i.e., the n sub modules, for example, as illustrated in FIGS. 4B and 5A) . Therefore, the main module will be trained over all different cells / sites and the sub modules are trained for each specific cells / site.
[0108] FIG. 7A illustrates a block diagram 700A of an example growable sub modular training process in accordance with some embodiments of the present disclosure. During the growable sub modular training process, in the training step as illustrated in FIG. 7A, first, the main module is trained with the general dataset using the loss function L (e.g., SGCS loss, MSE loss) .
[0109] FIG. 7B illustrates another block diagram 700B of an example growable sub modular training process in accordance with some embodiments of the present disclosure. The training step illustrated in FIG. 7B may follow the training step illustrated in FIG. 7A. In the training step as illustrated in FIG. 7B, sub module-1 is trained using the cell / site-specific dataset-1 with the parameters in the main module frozen. During the training process, the forward pass involves feeding the input data from the specialized dataset-1 through both main module and the parallel connected cell / site-specific sub module-1 to obtain the loss. In the backward backpropagation, the gradients of the loss with respect to the trainable parameters of the sub module-1 are computed. Importantly, since the parameters of the main module are frozen during this training step, they remain unaffected by the gradients calculated for the sub-module-1. This separation between the forward pass and backward propagation allows the sub module-1 to adapt to the specific cell / site (here, the first cell / site corresponding to sub module-1) while preserving the knowledge encoded in the main module. Other sub modules do not participate in both forward pass and backward propagation; in other words, in this step, among the n sub modules, only sub module-1 participates in forward pass and backward propagation.
[0110] FIG. 7C illustrates further another block diagram 700C of an example growable sub modular training process in accordance with some embodiments of the present disclosure. The training step illustrated in FIG. 7C may follow the training step illustrated in FIG. 7B. In the training step as illustrated in FIG. 7C, sub module-2 is trained using the cell / site-specific dataset-2 with the parameters in the main module frozen. During the training process, the forward pass involves feeding the input data from the specialized dataset-2 through both main module and the parallel connected cell / site-specific sub module-2 to obtain the loss. In the backward backpropagation, the gradients of the loss with respect to the trainable parameters of the sub module-2 are computed. Importantly, since the parameters of the main module are frozen during this training step, they remain unaffected by the gradients calculated for the sub-module-2. This separation between the forward pass and backward propagation allows the sub module-2 to adapt to the specific cell / site (here, the second cell / site corresponding to sub module-2) while preserving the knowledge encoded in the main module. Other sub modules do not participate in both forward pass and backward propagation; in other words, in this step, among the n sub modules, only sub module-2 participates in forward pass and backward propagation.
[0111] In this way, the growable sub modular training process is conducted iteratively for all the sub modules (i.e., the n sub modules, for example, as illustrated in FIGS. 4B and 5A) . As described above, first the main module is trained with the general dataset, and the sub modules are then trained for each specific cells / site with the main module frozen.
[0112] The main module as well as the sub modules can be trained offline across vendors via Type I training or trained online via Type III training. In case that the sub modules are trained offline by different vendors, vendors should also align the sub-module ID corresponding to each cell / site offline. As described above, each cell / site may correspond to a specific wireless environment (for example, a multi-path condition) .
[0113] Once the model (which comprises the main module and the sub modules) has been prepared, it can be utilized for inference, as will be described in more detail with reference to FIG. 8.
[0114] FIG. 8 illustrates a block diagram 800 for model inference in accordance with some embodiments of the present disclosure. During model inference as illustrated in FIG. 8, first, for example, the cell / site index may be determined by calculating the statistical discrepancy (for example, maximum mean discrepancy (MMD) ) between the in-field CSI data and the pre-stored datasets of each cell / site. By calculating the discrepancy, it means multiple CSI matrices are collected from the application field, then the similarities between the in-field data and all n datasets can be calculated. For example, maximum mean discrepancy (MMD) is an optional metric. The cell / site corresponding to the dataset with the lowest statistical discrepancy may be considered as the current deployment cell / site. Therefore, it is determined that the parallel-connectable sub module indicated by that cell / site index is to be selected to work together with the main module for inference. Then, the determined sub module is activated to work parallel connect to the main module to be used to perform inference.
[0115] In inference stage, first, the discrepancy between in-field collected data to all n pre-trained datasets is calculated to select the most similar dataset (for example, dataset-k, where k = 1, 2, …, n, where n is the number of the sub modules) out of pre-defined n categories (the n categories correspond to the n cells / sites) . The most similar dataset is designated as current cell / site category-indicator to share between the terminal device and network device. For example, if the i-th dataset (i.e., dataset-i, i = 1, 2, …, n) is selected as the most similar dataset and designated as current cell / site category-indicator, then the i-th dataset (i.e., dataset-i) may be used as an indicator indicating that the ith sub module (i.e., sub module-i) may be used in the terminal device and the network device. Second, the sub module-i’s branch in the terminal device and the network device may be activated to make it work in parallel with the main module. For example, the index i may be shared from gNB to UE, which is described in more detail with reference to FIG. 9.
[0116] FIG. 9 illustrates a signaling chart illustrating another example communication process 900 in accordance with some embodiments of the present disclosure. The communication process 900 may be used for cell / site-specific sub-module selection in CSI feedback. The communication process 900 may involve UE 906 and gNB 908. UE 906 may be an example of the terminal device 110 as illustrated in FIGS. 1 and 3, and gNB 908 may be an example of the network device 120 as illustrated in FIGS. 1 and 3.
[0117] As demonstrated in the signaling framework in FIG. 9, there are 4 stages to deploy the proposed scheme in UE 906 and gNB 908. In the first stage, the main encoder / decoder parallel connectable with the respective sub module pool are prepared. The respective sub module pool for the UE 906 or gNB 908 comprises several cell / site-specific sub modules, as illustrated in FIGS. 4A, 4B and 5A. Then, UE 906 and gNB 908 pre-train the main encoder / decoder and the cell / site-specific sub modules and associate each sub module with a specific cell / site. Based on the determined cell / site category-indicator, UE 906 and gNB 908 use encoder / decoder parallel connectable with the sub module of current cell / site for model inference. In the second stage, UE 906 and gNB 908 check if their NNs are from the same encoder-decoder pair. In the third stage, gNB switches to one appropriate sub module parallel connected with the decoder based on the detected cell / site type. In the fourth stage, UE 906 switches to one sub module parallel-connectable with the encoder based on the sub module ID indicated by gNB 908.
[0118] More specifically, at 910, both UE 906 and gNB 908 maintain a pre-trained main encoder / decoder parallel connectable to cell / site-specific sub modules from sub module pool and have pre-defined the application condition for each sub-module. The cell / site-specific sub modules parallel connectable to the main encoder / decoder may be referred to as a sub module pool. This may be a pre-condition for the communication process 900. Here, the main encoder may be the main module (which may also be referred to as “general model” , “main model” , “general module” ) at the UE 906, and the main decoder may be the main module at the gNB 908. It is assumed that the main encoder / decoder and the parallel-connected sub NN modules are pre-trained offline amongst different vendors and the vendors have aligned the application condition for each sub module. Then in deployment, when UE 906 and gNB 908 work, they use a pair of sub modules for a specific cell / site to parallel connect with the encoder / decoder.
[0119] At 915, a cell / site drift occurs and is detected, for example, by the UE 906 or the gNB 908. Once the cell / site drift is detected by either UE 906 or gNB 908, the sub modules which are parallel connectable with the encoder / decoder may not fit the drifted cell / site. Therefore, UE 906 and gNB 908 may need to communicate with each other to switch to another sub-module pair from the sub module pool for CSI feedback enhancement, as described below.
[0120] At 920, it deals with a situation where cell / site drift is detected by UE 906, i.e., a situation where encoder-decoder pair verification is triggered by UE 906. In this case, UE 906 requests the gNB 908 to verify if there are corresponding sub modules in the sub module pools of UE 906 and gNB 908, i.e., UE 906 requests the gNB 908 to verify if there are sub module pair (s) between the sub module pools of UE 906 and gNB 908. If so (i.e., if there is sub module pair (s) between the sub module pools of UE 906 and gNB 908) , the gNB 908 sends the confirmation message to the UE 906. If not (if there is no sub module pair between the sub module pools of UE 906 and gNB 908) , the communication process 900 terminates.
[0121] At 925, it deals with a situation where cell / site drift is detected by gNB 908, i.e., a situation where encoder-decoder pair verification is triggered by gNB 908. In this case, gNB 908 requests the UE 906 to verify if there are corresponding sub modules in the sub module pools of UE 906 and gNB 908, i.e., gNB 908 requests UE 906 to verify if there are sub module pair (s) between the sub module pools of UE 906 and gNB 908. If so (i.e., if there is sub module pair (s) between the sub module pools of UE 906 and gNB 908) , the UE 906 sends the confirmation message and indicates the indices of corresponding sub modules to the UE 906. If not (if there is no sub module pair between the sub module pools of UE 906 and gNB 908) , the communication process 900 terminates.
[0122] At 930, the UE 906 sends the pre-defined requisite information for drifted cell / site classification. For example, the pre-defined requisite information may be one or more CSI codewords and / or one or more original CSI matrices.
[0123] At 935, based on the received pre-defined requisite information, the gNB 908 classifies the drifted cell / site with the pre-defined mechanism. For example, based on the received pre-defined requisite information, gNB 908 may determine the current cell / site ID of UE 906 using a pre-defined mechanism. Statistical discrepancy calculation may be an example of pre-defined mechanism to assess ( / evaluate) the similarity between the in-field data and the pre-stored cell / site-specific dataset associated with the sub module pair (s) between the sub module pools of UE 906 and gNB 908 (such sub module pair (s) is determined ( / verified) at 920 or 925, as described above) , then the cell / site with the highest similarity may be selected.
[0124] At 940, according to the classified cell / site, the gNB 908 selects the sub modules from the corresponding sub modules to be parallel connected with the main decoder for inference. More specifically, depending on the cell / site type determined by gNB 908, gNB 908 may select the sub module corresponding to the determined cell / site type from the corresponding sub modules to be parallel connected to the main decoder at the gNB 908 for inference.
[0125] At 945, the gNB 908 sends the index of the selected sub module to UE 906. In other words, gNB 908 informs UE 906 about the index of its newly selected sub NN module. At 950, UE 906 switches to the indicated sub module to parallel connect with the main encoder. At 955, UE 906 sends the model switch confirmation message to gNB 408. With the communication process 900, better performance and parameter efficiency can be obtained. Besides, no additional inference latency is introduced.
[0126] FIG. 10 illustrates a flowchart of an example method 1000 implemented at a terminal device (for example, the terminal device 110 as illustrated in FIGS. 1 and 3) in accordance with some other embodiments of the present disclosure. For the purpose of discussion, the method 1000 will be described from the perspective of the terminal device 110 with reference to FIG. 3.
[0127] As illustrated in FIG. 10, at block 1010, the terminal device 110 receives, from a network device (for example, the network device 120 as illustrated in FIGS. 1-2) , an index (for example, index 301 as illustrated in FIG. 3) of a first sub-module among first one or more sub-modules at the network device. Here, the first sub-module is to be parallel connected to a main decoder at the network device. At block 1020, the terminal device 110 connects, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.
[0128] In some example embodiments, the terminal device 110 may further transmit, to the network device, a message confirming that the second sub-module is to be parallel connected to the main encoder at the terminal device.
[0129] In some example embodiments, based on detecting a site drift, the terminal device 110 may transmit a request message to the network device for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device (here, the request message may comprise a first ID list of the second one or more sub-modules at the terminal device) , and receive, from the network device, a response message comprising a second ID list of the at least one decoder sub-module.
[0130] In some example embodiments, the terminal device 110 may receive, from the network device, a request message for determining whether there is at least one encoder sub-module among the second one or more sub-modules at the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device (here, the request message comprises a third ID list of the first one or more sub-modules at the network device) , and determine, based on the third ID list of the first one or more sub-modules, whether there is the at least one encoder sub-module. Further, based on determining that there is the at least one encoder sub-module, the terminal device 110 may transmit, to the network device, a response message comprising a fourth ID list of the at least one encoder sub-module.
[0131] In some example embodiments, the terminal device 110 may transmit pre-defined information to the network device for selecting the first sub-module. For example, the pre-defined information may comprise channel state information (CSI) . The channel state information (CSI) may comprise original CSI and / or compressed CSI codewords.
[0132] In some example embodiments, the terminal device 110 may maintain a sub-module pool which comprises the second one or more sub-modules, and the sub-module pool may be associated with different application conditions.
[0133] FIG. 11 illustrates a flowchart of an example method 1100 implemented at a network device (for example, the network device 120 as illustrated in FIGS. 1 and 3) in accordance with some other embodiments of the present disclosure. For the purpose of discussion, the method 1100 will be described from the perspective of the network device 120 with reference to FIG. 3.
[0134] As illustrated in FIG. 11, at block 1110, the network device 120 selects, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device 120 to be parallel connected to a main decoder at the network device 120. At block 1120, the network device 120 transmits, to the terminal device, an index (for example, the index 301 as illustrated in FIG. 3) of the selected sub-module.
[0135] In some example embodiments, the one or more sub-modules are first one or more sub-modules corresponding to second one or more sub-modules at the terminal device to be parallel connected to a main encoder at the terminal device.
[0136] In some example embodiments, the sub-module may be a first sub-module, and the network device 120 may further receive, from the terminal device, a message confirming that a second sub-module at the terminal device corresponding to the index is to be parallel connected to the main encoder of the terminal device.
[0137] In some example embodiments, based on detecting a site drift, the network device 120 may transmit a request message to the terminal device for determining whether there is at least one encoder sub-module among the second one or more sub-modules in the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device (here, the request message comprises a first ID list of the first one or more sub-modules at the network device) , and receive, from the terminal device, a response message comprising a second ID list of the at least one encoder sub-module.
[0138] In some example embodiments, the network device 120 may receive, from the terminal device, a request message for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device (here, the request message comprises a third ID list of the second one or more sub-modules at the terminal device) , and determine, based on the third ID list of the second on or more sub-modules, whether there is the at least one decoder sub-module. Based on determining that there is at least one decoder sub-module, the network device 120 may transmit, to the terminal device, a response message comprising a fourth ID list of the at least one decoder sub-module.
[0139] In some example embodiments, the network device 120 may receive, from the terminal device, pre-defined information for selecting the sub-module. The channel state information (CSI) may comprise at least one of original CSI or compressed CSI codewords. Based on the pre-defined information, the network device 120 may classify the drifted site to determine the site ID.
[0140] The network device 120 may maintain a sub-module pool which comprises the one or more sub-modules, and sub-modules in the sub-module pool are associated with different application conditions.
[0141] In some embodiments, an apparatus capable of performing the method 200 may comprise means for performing the respective steps of the method 200. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module.
[0142] In some example embodiments, the apparatus comprises: means for training a main module and a plurality of sub-modules parallel connectable to the main module for CSI feedback, wherein the plurality of sub-modules are associated with a plurality of sites of different wireless environments; and means for performing, based on the main module parallel connected with a selected sub-module among the plurality of sub-modules, inference for CSI feedback.
[0143] In some example embodiments, the main module and the plurality of sub-modules may be within at least one linear layer in an artificial intelligence (AI) / machine learning (ML) model. Alternatively or in addition, sub-module numbers of the plurality of sub-modules may be predetermined corresponding to site numbers of the plurality of sites. Alternatively or in addition, the plurality of sub-modules may be parallel connectable to the main module via respective switches. Alternatively or in addition, in the event that a sub-module is parallel connected to the main module, a same input may be provided to the main module and the sub-module. Alternatively or in addition, in the event that a sub-module is parallel connected to the main module, a sum of a first output of the main module and a second output of the sub-module may be provided to a next layer of an AI / ML model.
[0144] In some example embodiments, the means for training may comprise means for performing a training process comprises a plurality of training steps based on a plurality of datasets associated with the plurality of sites.
[0145] In some example embodiments, a sub-module may be trained independently from other sub modules and is trained jointly with the main module. Alternatively or in addition, the main module may be trained based on a general dataset. Alternatively or in addition, the plurality of sub-modules may be trained based on the plurality of datasets respectively.
[0146] In some example embodiments, in a training step among the plurality of training steps, the main module may be parallel connected with a sub-module among the plurality of sub-modules, and in the training step, the main module and the sub-module may be trained based on a dataset among the plurality of datasets, wherein the dataset corresponds to a site, among the plurality of sites, associated with the sub-module.
[0147] In some example embodiments, in the training process, parameters of the main module may be updated based on an overall loss which is a summation of a plurality of sub-losses determined based on the plurality of datasets in the plurality of training steps, and in the training process, parameters of the sub-module may be updated based on a sub-loss among the plurality of sub-losses. Here, the sub-loss is determined based on the dataset in the training step.
[0148] In some example embodiments, prior to the plurality of training steps, the main module may be trained based on a general dataset. In a training step among the plurality of training steps, the main module may be parallel connected with a sub-module among the plurality of sub-modules, and in the training step, the main module may be maintained unchanged and the sub-module is trained based on a dataset among the plurality of datasets. Here, the dataset may correspond to a site, among the plurality of sites, associated with the sub-module.
[0149] In some example embodiments, in a forward pass of the training step, data from the dataset may be fed through both the main module and the sub module to obtain a loss. In a back propagation of the training step, gradients of the loss with respect to trainable parameters of the sub-module may be computed, and the parameters of the main module may remain unaffected by the gradients calculated for the sub-module.
[0150] In some example embodiments, the means for performing may comprise means for calculating a plurality of statistical discrepancies between in-field data and a plurality of datasets associated with the plurality of sites; means for determining a target site among the plurality of sites based on the plurality of statistical discrepancies; and means for determining, among the plurality of sub-modules, a target sub-module associated with the target site as the selected sub-module to be parallel connected to the main module for the inference.
[0151] In some example embodiments, the target site may correspond to a dataset with a lowest statistical discrepancy among the plurality of statistical discrepancies.
[0152] In some example embodiments, the apparatus may be a terminal device (for example, the terminal device 110 as illustrated in FIGS. 1 and 3) and the main module may comprise an encoder for CSI feedback.
[0153] In some example embodiments, the apparatus may be a network device (for example, the network device 120 as illustrated in FIGS. 1 and 3) and the main module may comprise a decoder for CSI feedback.
[0154] In some embodiments, the apparatus further comprises means for performing other steps in some embodiments of the method 200. In some embodiments, the means comprises at least one processor and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
[0155] In some embodiments, an apparatus capable of performing the method 1000 may comprise means for performing the respective steps of the method 1000. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module.
[0156] In some example embodiments, the apparatus comprises: means for receiving, from a network device, an index of a first sub-module among first one or more sub-modules at the network device (here, the first sub-module is to be parallel connected to a main decoder at the network device) ; and means for connecting, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device (here, the second sub-module is associated with the first sub-module) .
[0157] In some example embodiments, the apparatus may further comprise means for transmitting, to the network device, a message confirming that the second sub-module is to be parallel connected to the main encoder at the terminal device.
[0158] In some example embodiments, based on detecting a site drift, the apparatus may further comprise means for transmitting a request message to the network device for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device (here, the request message may comprise a first ID list of the second one or more sub-modules at the terminal device) , and means for receiving, from the network device, a response message comprising a second ID list of the at least one decoder sub-module.
[0159] In some example embodiments, the apparatus may further comprise means for receiving, from the network device, a request message for determining whether there is at least one encoder sub-module among the second one or more sub-modules at the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device (here, the request message comprises a third ID list of the first one or more sub-modules at the network device) , and means for determining, based on the third ID list of the first one or more sub-modules, whether there is the at least one encoder sub-module. Further, based on determining that there is the at least one encoder sub-module, the apparatus may further comprise means for transmitting, to the network device, a response message comprising a fourth ID list of the at least one encoder sub-module.
[0160] In some example embodiments, the apparatus may further comprise means for transmitting pre-defined information to the network device for selecting the first sub-module. For example, the pre-defined information may comprise channel state information (CSI) . The channel state information (CSI) may comprise original CSI and / or compressed CSI codewords.
[0161] In some example embodiments, the apparatus may further comprise means for maintaining a sub-module pool which comprises the second one or more sub-modules. For example, the sub-module pool may be associated with different application conditions.
[0162] In some embodiments, the apparatus further comprises means for performing other steps in some embodiments of the method 1000. In some embodiments, the means comprises at least one processor and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
[0163] In some embodiments, an apparatus capable of performing the method 1100 may comprise means for performing the respective steps of the method 1100. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module.
[0164] In some example embodiments, the apparatus comprises: means for selecting, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; and means for transmitting, to the terminal device, an index of the selected sub-module.
[0165] In some example embodiments, the one or more sub-modules may be first one or more sub-modules corresponding to second one or more sub-modules at the terminal device to be parallel connected to a main encoder at the terminal device.
[0166] In some example embodiments, the sub-module may be a first sub-module, and the apparatus may further comprise means for receiving, from the terminal device, a message confirming that a second sub-module at the terminal device corresponding to the index is to be parallel connected to the main encoder of the terminal device.
[0167] In some example embodiments, based on detecting a site drift, the apparatus may further comprise means for transmitting a request message to the terminal device for determining whether there is at least one encoder sub-module among the second one or more sub-modules in the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device (here, the request message comprises a first ID list of the first one or more sub-modules at the network device) , and means for receiving, from the terminal device, a response message comprising a second ID list of the at least one encoder sub-module.
[0168] In some example embodiments, the apparatus may further comprise means for receiving, from the terminal device, a request message for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device (here, the request message comprises a third ID list of the second one or more sub-modules at the terminal device) , and means for determine, based on the third ID list of the second on or more sub-modules, whether there is the at least one decoder sub-module. Based on determining that there is at least one decoder sub-module, the apparatus may further comprise means for transmitting, to the terminal device, a response message comprising a fourth ID list of the at least one decoder sub-module.
[0169] In some example embodiments, the apparatus may further comprise means for receiving, from the terminal device, pre-defined information for selecting the sub-module. The CSI may comprise at least one of original CSI or compressed CSI codewords. Based on the pre-defined information, the network device 120 may classify the drifted site to determine the site ID.
[0170] The apparatus may comprise means for maintaining a sub-module pool which comprises the one or more sub-modules. For example, sub-modules in the sub-module pool are associated with different application conditions.
[0171] In some embodiments, the apparatus further comprises means for performing other steps in some embodiments of the method 1100. In some embodiments, the means comprises at least one processor and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
[0172] To check the performance of the proposed scheme in some embodiments of the present disclosure, a simulation was conducted. In the simulation, the proposed scheme in some embodiments of this disclosure is evaluated with two datasets of cells with different gNB antenna configurations; i.e., in the simulation, n = 2. The detailed parameters of the datasets are shown in Table 1 below.
[0173] Table 1 Parameters of simulation datasets
[0174] Herein, the dataset of (8, 8, 2, 1, 1, 2, 8) antenna configuration is denoted as “282” , and the dataset of (8, 4, 2, 1, 1, 4, 4) antenna configuration is denoted as “442” . Four schemes are selected to evaluate the CSI reconstruction accuracy performance. FIG. 12A provides a basic transformer architecture for CSI feedback.
[0175] FIG. 12A illustrates a block diagram of an example of a transformer-enabled neural network (NN) for CSI feedback in accordance with some example embodiments of the present disclosure. The transformer-enabled neural network (NN) for CSI feedback enhancement is used in the simulation. The left side of FIG. 12A represents blocks at UE encoder, ant the right side of FIG. 12A represents blocks at gNB decoder.
[0176] FIG. 12B illustrates a block diagram of a modified multi-head attention block in accordance with some example embodiments of the present disclosure. As shown in FIG. 12B, the i-th sub module (sub module-i) is represented by new pairs of rank decomposition matrices (QAi, QBi, VAi, VBi) . These rank decomposition matrices are added in a parallel manner to the existing weight matrices (Q0 and V0) . Therefore, the new linear matrices for query and value are obtained as Qnew=Q0+QBi. QAi and Vnew=V0+VBi. VAi. In this way, the output variables change to XQ=QnewX (instead of the original Q0X) and XV=VnewX (instead of the original V0X) .
[0177] As mentioned above, four schemes are selected to evaluate the CSI reconstruction accuracy performance. In the following four schemes, both encoder and decoder use the transformer NN with 6 attention layers, where the embedding dimension equals 128.
[0178] In baseline schemes ( ‘specific’s cheme and ‘mixed’s cheme) , the multi-head attention block is adopted in transformer. More specifically, in the specific schemes, the unmodified transformer architecture is applied. The model is trained with specific 442 / 282 dataset comprising 160K data, respectively. In the mixed scheme, the unmodified transformer architecture is applied. This model is trained with the mixed 160K 442 dataset and 160K 282 dataset.
[0179] In proposed schemes, the multi-head attention block is modified by parallel connecting to cell / site-specific sub modules from the sub module pool. In this case, the transformer in which the modified multi-head attention block parallel connects with the sub module pool is applied. The model is trained with the mixed 160K 442 dataset and 160K 282 dataset.
[0180] In the evaluation, all CSI matrices are compressed into 52-bit codewords. Table 2 illustrates the cosine similarity (SGCS) performances when the 442 / 282 dataset is tested with the models trained by different schemes.
[0181] Table 2 Performance comparisons amongst different training schemes
[0182] According to the results shown in Table 2, it can be observed that, on the 442 testing dataset, the proposed scheme mitigates 44.8% (=1-0.0155 / 0.0282) performance gap between the specific scheme and the mixed scheme. And, on the 282 testing dataset, the proposed scheme mitigates 83.8% (=1-0.0056 / 0.0367) performance gap between the specific scheme and the mixed scheme. On both cases, additional 0.39M (=2.65M-2.26M) trainable parameters are introduced. In this case, to cover 2 cells of different antenna configurations, if the specific scheme is adopted, 4.52M=2*2.26M memory space is needed, while if the proposed scheme is adopted, only 2.65M memory space is needed, which is much more parameter efficient.
[0183] From the above information, it can be observed that the proposed model architecture with parallel-connected sub modules has prominent advantages over other schemes. Specifically, better performance than the mixed scheme can be obtained. As shown in Table 2, the proposed schemes mitigate 44%to 84%of the performance gap between the specific scheme and the mixed scheme, which indicates the proposed schemes are capable to provide performance benefits by parallel connecting cell / site-specific modules to the main module. Further, parameter efficiency can also be obtained. As shown in Table 2, the proposed schemes provide 44%to 84%performance benefits by introducing very small size (i.e., 0.39M, less than 9% (= 0.39M / 4.52M) of trainable parameters of each cell / site-specific sub module. At the same time, in contrast to other adapter methods, the proposed model with parallel-connected modules architecture introduces no additional inference latency.
[0184] FIG. 13 illustrates a simplified block diagram of a device 1300 that is suitable for implementing some example embodiments of the present disclosure. The device 1300 may be provided to implement a communication device, for example, the terminal device 110 and the network device 120 as shown in FIGS. 1 and 3. As shown, the device 1300 includes one or more processors 1310, one or more memories 1320 coupled to the processor 1310, and one or more communication modules 1340 coupled to the processor 1310.
[0185] The communication module 1340 is for bidirectional communications. The communication module 1340 has at least one antenna to facilitate communication. The communication interface may represent any interface that is necessary for communication with other network elements.
[0186] The processor 1310 may be of any type suitable for the local network and may include one or more of the following: general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multicore processor architecture, as non-limiting examples. The device 1300 may have multiple processors, such as an application specific integrated circuit chip that is slaved in time to a clock which synchronizes the main processor.
[0187] The memory 1320 may include one or more non-volatile memories and one or more volatile memories. Examples of the non-volatile memories include, but are not limited to, a Read Only Memory (ROM) 1324, an electrically programmable read only memory (EPROM) , a flash memory, a hard disk, a compact disc (CD) , a digital video disk (DVD) , and other magnetic storage and / or optical storage. Examples of the volatile memories include, but are not limited to, a random access memory (RAM) 1322 and other volatile memories that will not last in the power-down duration.
[0188] A computer program 1330 includes computer executable instructions that are executed by the associated processor 1310. The program 1330 may be stored in the ROM 1324. The processor 1310 may perform any suitable actions and processing by loading the program 1330 into the RAM 1322.
[0189] The embodiments of the present disclosure may be implemented by means of the program 1330 so that the device 1300 may perform any process of the disclosure as discussed with reference to FIGS. 2-3 and 9-11. The embodiments of the present disclosure may also be implemented by hardware or by a combination of software and hardware.
[0190] In some example embodiments, the program 1330 may be tangibly contained in a computer-readable medium which may be included in the device 1300 (such as in the memory 1320) or other storage devices that are accessible by the device 1300. The device 1300 may load the program 1330 from the computer-readable medium to the RAM 1322 for execution. The computer-readable medium may include any types of tangible non-volatile storage, such as ROM, EPROM, a flash memory, a hard disk, CD, DVD, and the like.
[0191] FIG. 14 illustrates a block diagram of an example of a computer-readable medium 1400 in accordance with some example embodiments of the present disclosure. The computer-readable medium 1400 has the program 1330 stored thereon. It is noted that although the computer-readable medium 1400 is depicted in form of CD or DVD in FIG. 8, the computer-readable medium 1400 may be in any other form suitable for carry or hold the program 1330.
[0192] Generally, various embodiments of the present disclosure may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device. While various aspects of embodiments of the present disclosure are illustrated and described as block diagrams, flowcharts, or using some other pictorial representations, it is to be understood that the block, apparatus, system, technique or method described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.
[0193] The present disclosure also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as those included in program modules, being executed in a device on a target real or virtual processor, to carry out any of the method 200-300, 900-1100 as described above with reference to FIGS. 2-3 and 9-11. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, or the like that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split between program modules as desired in various embodiments. Machine-executable instructions for program modules may be executed within a local or distributed device. In a distributed device, program modules may be located in both local and remote storage media.
[0194] Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
[0195] In the context of the present disclosure, the computer program codes or related data may be carried by any suitable carrier to enable the device, apparatus or processor to perform various processes and operations as described above. Examples of the carrier include a signal, computer-readable medium, and the like.
[0196] The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable medium may include but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer-readable storage medium would include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM) , a read-only memory (ROM) , an erasable programmable read-only memory (EPROM or Flash memory) , an optical fiber, a portable compact disc read-only memory (CD-ROM) , an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. The term “non-transitory, ” as used herein, is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM vs. ROM) .
[0197] Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are contained in the above discussions, these should not be construed as limitations on the scope of the present disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features that are described in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable sub-combination.
[0198] Although the present disclosure has been described in languages specific to structural features and / or methodological acts, it is to be understood that the present disclosure defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
Claims
1.A terminal device comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the terminal device to:receive, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; andconnect, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.2.The terminal device of claim 1, wherein the terminal device is further caused to:transmit, to the network device, a message confirming that the second sub-module is to be parallel connected to the main encoder at the terminal device.3.The terminal device of claim 1 or 2, wherein the terminal device is further caused to:based on detecting a site drift, transmit a request message to the network device for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device, wherein the request message comprises a first ID list of the second one or more sub-modules at the terminal device; andreceive, from the network device, a response message comprising a second ID list of the at least one decoder sub-module.4.The terminal device of claim 1 or 2, wherein the terminal device is further caused to:receive, from the network device, a request message for determining whether there is at least one encoder sub-module among the second one or more sub-modules at the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device, wherein the request message comprises a third ID list of the first one or more sub-modules at the network device; anddetermine, based on the third ID list of the first one or more sub-modules, whether there is the at least one encoder sub-module.5.The terminal device of claim 4, the terminal device is further caused to:based on determining that there is the at least one encoder sub-module, transmit, to the network device, a response message comprising a fourth ID list of the at least one encoder sub-module.6.The terminal device of any of claims 3-5, wherein the terminal device is further caused to:transmit pre-defined information to the network device for selecting the first sub-module.7.The terminal device of claim 6, wherein the pre-defined information comprises channel state information (CSI) .8.The terminal device of claim 7, wherein the channel state information (CSI) comprises at least one of original CSI or compressed CSI codewords.9.The terminal device of any of claims 1-8, wherein the terminal device is further caused to:maintain a sub-module pool which comprises the second one or more sub-modules.10.The terminal device of claim 9, wherein sub-modules in the sub-module pool are associated with different application conditions.11.A network device comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the network device to:select, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; andtransmit, to the terminal device, an index of the selected sub-module.12.The network device of claim 11, wherein the one or more sub-modules are first one or more sub-modules corresponding to second one or more sub-modules at the terminal device to be parallel connected to a main encoder at the terminal device.13.The network device of claim 11 or 12, wherein the sub-module is a first sub-module, and the network device is further caused to:receive, from the terminal device, a message confirming that a second sub-module at the terminal device corresponding to the index is to be parallel connected to the main encoder of the terminal device.14.The network device of claim 12 or 13, wherein the network device is further caused to:based on detecting a site drift, transmit a request message to the terminal device for determining whether there is at least one encoder sub-module among the second one or more sub-modules in the terminal device corresponding to at least one decoder sub-module among the first one or more sub-modules at the network device, wherein the request message comprises a first ID list of the first one or more sub-modules at the network device; andreceive, from the terminal device, a response message comprising a second ID list of the at least one encoder sub-module.15.The network device of claim 12 or 13, wherein the network device is further caused to:receive, from the terminal device, a request message for determining whether there is at least one decoder sub-module among the first one or more sub-modules at the network device corresponding to at least one encoder sub-module among the second one or more sub-modules at the terminal device, wherein the request message comprises a third ID list of the second one or more sub-modules at the terminal device; anddetermine, based on the third ID list of the second on or more sub-modules, whether there is the at least one decoder sub-module.16.The network device of claim 15, the network device is further caused to:based on determining that there is at least one decoder sub-module, transmit, to the terminal device, a response message comprising a fourth ID list of the at least one decoder sub-module.17.The network device of any of claims 14-16, wherein the network device is further caused to:receive, from the terminal device, pre-defined information for selecting the sub-module.18.The network device of claim 17, wherein the pre-defined information comprises channel state information (CSI) .19.The network device of claim 18, wherein the channel state information (CSI) comprises at least one of original CSI or compressed CSI codewords.20.The network device of any of claims 17-19, wherein the network device is further caused to:classify, based on the pre-defined information, the drifted site to determine the site ID.21.The network device of any of claims 11-20, wherein the network device is further caused to:maintain a sub-module pool which comprises the one or more sub-modules.22.The network device of claim 21, wherein sub-modules in the sub-module pool are associated with different application conditions.23.A method comprising:receiving, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; andconnecting, based on the index, a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.24.A method comprising:selecting, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; andtransmitting, to the terminal device, an index of the selected sub-module.25.An apparatus comprising:means for receiving, from a network device, an index of a first sub-module among first one or more sub-modules at the network device, wherein the first sub-module is to be parallel connected to a main decoder at the network device; andbased on the index, means for connecting a second sub-module among second one or more sub-modules at the terminal device parallel to a main encoder at the terminal device, wherein the second sub-module is associated with the first sub-module.26.An apparatus comprising:means for selecting, based on a site identifier (ID) , a sub-module among one or more sub-modules at the network device to be parallel connected to a main decoder at the network device; andmeans for transmitting, to the terminal device, an index of the selected sub-module.27.A non-transitory computer-readable medium comprising program instructions stored thereon, the instructions, when executed on at least one processor, causing the at least one processor to perform the method of any of claim 23-24.
Citation Information
Patent Citations
Neural network configuration for wireless communication system assistance
CN115136152A
Machine translation using neural network models
US20200034436A1
Channel state information feedback
WO2021108940A1
Method for feeding back channel state information, method for receiving channel state information, and terminal, base station, and computer-readable storage medium
WO2023011472A1