Channel state information feedback enhancement with separate training

Abstract

The embodiment of the invention relates to equipment, a method and a device for CSI (Channel State Information) feedback enhancement by utilizing separate training, and a computer readable storage medium. The method comprises: at a terminal device, receiving from a network device a training data set associated with channel state information (CSI), a size of a first codebook of the network device, and a first codeword index set associated with the first codebook, the size of the first codebook being associated with the number of vectors in the first codebook; and determining the CSI feedback compression at least by determining, based on the training data set and a first codeword index set associated with the first codebook, that a second codebook of the terminal device has the same size as that of the first codebook.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure generally relate to the field of telecommunications, and in particular, to devices, methods, apparatuses, and computer-readable storage media for channel state information (CSI) feedback enhancement with separate training. BACKGROUND

[0002] Accurate feedback of CSI is extremely important for efficient transmission of frequency division duplex (FDD) multiple-input multiple-output (MIMO), but the delivery of raw CSI to the base station consumes a large amount of uplink resources. With the great achievements of AI-based feature extraction and image compression, CSI feedback enhancement by using artificial intelligence (AI) / machine learning (ML) methods has been discussed and researched. SUMMARY

[0003] Generally, example embodiments of the present disclosure provide solutions for CSI feedback enhancement with separate training.

[0004] In a first aspect, an apparatus is provided. The apparatus includes 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: receive, from a network device, a training dataset associated with a size of a first codebook of the network device and a first codeword index set associated with the first codebook, wherein the size of the first codebook is associated with a number of vectors in the first codebook; and determine CSI feedback compression of the apparatus at least by determining that a second codebook of the apparatus has a same size as the size of the first codebook based on the training dataset and the first codeword index set associated with the first codebook.

[0005] In a second aspect, an apparatus is provided. The apparatus includes 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: determine a latent vector based on a training dataset associated with CSI; determine a first codeword index set based on the latent vector and a first codebook of the apparatus; and transmit, to a terminal device, the training dataset, the first codeword index set, and a size of the first codebook.

[0006] In a third aspect, a method is provided. The method includes: receiving, at a terminal device, from a network device, a training dataset associated with CSI, a size of a first codebook of the network device, and a first codeword index set associated with the first codebook, wherein the size of the first codebook is associated with a number of vectors in the first codebook; and determining CSI feedback compression of the terminal device at least by determining that a second codebook of the terminal device has a same size as the size of the first codebook based on the training dataset and the first codeword index set associated with the first codebook.

[0007] In a fourth aspect, a method is provided. The method comprises: determining, at a network device, a latent vector based on a training dataset associated with a channel state information, CSI; determining a first set of code word indices based on the latent vector and a first codebook of the network device; and transmitting, to a terminal device, the training dataset, the first set of code word indices, and a size of the first codebook.

[0008] In a fifth aspect, an apparatus is provided. The apparatus comprises: means for receiving, from a network device, a training dataset associated with a channel state information, CSI, a size of a first codebook of the network device, and a first set of code word indices associated with the first codebook, wherein the size of the first codebook is associated with a number of vectors in the first codebook; and means for determining, for a CSI feedback compression of the apparatus, at least by determining that a second codebook of the apparatus has a same size as the size of the first codebook based on the training dataset and the first set of code word indices associated with the first codebook.

[0009] In a sixth aspect, an apparatus is provided. The apparatus comprises: means for determining a latent vector based on a training dataset associated with a channel state information, CSI; means for determining a first set of code word indices based on the latent vector and a first codebook of the apparatus; and means for transmitting, to a terminal device, the training dataset, the first set of code word indices, and a size of the first codebook.

[0010] In a seventh aspect, a computer readable medium is provided. The computer readable medium has stored thereon a computer program, which, when executed by at least one processor of an apparatus, causes the apparatus to carry out the method according to the third aspect or the fourth aspect.

[0011] Other features and advantages of embodiments of the present disclosure will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, which illustrate, by way of example, the principles of embodiments of the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0012] Embodiments of the present disclosure are illustrated by way of example, and not by way of limitation, in the following drawings and specification.

[0013] Figure 1 An example environment in which example embodiments of the present disclosure can be implemented is illustrated.

[0014] Figure 2 An example of an overall structure for CSI compression and reconstruction according to some example embodiments of the present disclosure is illustrated;

[0015] Figure 3 A signaling diagram illustrating an example process according to some example embodiments of the present disclosure is illustrated;

[0016] Figure 4An example of a model for network device side training is shown in accordance with some example embodiments of the present disclosure;

[0017] Figure 5 An example of a model for terminal device side training is shown in accordance with some example embodiments of the present disclosure;

[0018] Figure 6 An example of a joint operation example of a UE side encoder, VQ codebook, and NW side encoder, VQ codebook is shown in accordance with some example embodiments of the present disclosure;

[0019] Figure 7 A flowchart of an example method for a separate training method for CSI feedback is shown in accordance with some example embodiments of the present disclosure;

[0020] Figure 8 A flowchart of an example method for a separate training method for CSI feedback is shown in accordance with some example embodiments of the present disclosure;

[0021] Figure 9 A simplified block diagram of a device suitable for implementing example embodiments of the present disclosure is shown; and

[0022] Figure 10 A block diagram of an example computer readable medium in accordance with some embodiments of the present disclosure is shown.

[0023] In all of the drawings, like or similar reference numerals can refer to like or similar elements throughout the drawings. DETAILED DESCRIPTION

[0024] The principles of the present disclosure will now be described with reference to some example embodiments. It should be understood that these embodiments are described for illustrative purposes only and help the skilled person to understand and implement the present disclosure, without implying any limitation to the scope of the present disclosure. The embodiments described herein can be implemented in various ways other than those described below.

[0025] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0026] Reference throughout this disclosure to "one embodiment", "an embodiment", "example embodiment" or the like means that a described embodiment can include a particular feature, structure, or characteristic, but every embodiment can not necessarily include the particular feature, structure, or characteristic. Furthermore, such phrases are not necessarily referring to the same embodiment. Furthermore, where a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of those skilled in the art to effect such feature, structure, or characteristic in connection with other

[0027] It should be understood that, although the terms“first,”“second,” etc. can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could also be termed a second element, and, similarly, a second element could also 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 associated listed terms.

[0028] As used herein, “at least one of ,” “one or more of ,” and the like, means at least one, or, alternatively, any single one of the listed elements or combination of any two or more of the listed elements.

[0029] As used herein, unless expressly stated to the contrary, performing an act“in response to” (or variations such as“in reaction to,”“in response to an indication that,” etc.) A happens is not intended to mean that the act recited is to be done immediately in response to A, and can include any number of intermediate steps or delays.

[0030] The terminology used herein is for the purpose of describing particular embodiments only 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,”“includes” and / or“including,” when used herein, specify the presence of stated features, elements, and / or components, but do not preclude the presence or addition of one or more other features, elements, components, and / or groups thereof.

[0031] As used in this application, the term“circuitry” can refer to one or more or all of the following: (a) hardware-only circuitry such as comprises only analog and / or digital circuitry, and (b) combinations of hardware circuits and software, such as (as applicable): (i) combinations of analog and / or digital hardware circuit(s) with software / firmware (ii) any portions of hardware processor(s) with software (including digital signal processors); software; and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions and (c) hardware circuit(s) and / or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but software can not be present when it is not needed for operation.

[0032] The 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 that is at least one integrated circuit or a portion thereof, and / or at least one

[0033] As used herein, the term “communication network” refers to a network that follows any appropriate communication standard, such as New Radio (NR), 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), Enhanced Machine Type Communications (eMTC), etc. Further, the communication between terminal devices and network devices in the communication network can be performed according to any suitable generation communication protocol, including but not limited to, first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G) communication protocols, and / or any other protocols that are currently known or that will be developed in the future. Embodiments of the present disclosure can be applied to various communication systems. In view of the rapid development in communications, it is certain that there will also be future types of communication technologies and systems that utilize which the present disclosure can be applied. This should not be regarded as limiting the scope of the present disclosure only to the above-described systems.

[0034] As used herein, the terms “network device,” “radio network device,” and / or “radio access network device” refer to a node in a communication network via which terminal devices access the network and receive services therefrom. The network device can refer to a base station (BS) or an access point (AP), e.g., a NodeB (NodeB or NB), an evolved NodeB (eNodeB or eNB), an NR NB (also known as gNB), a remote radio unit (RRU), a radio head (RH), a remote radio head (RRH), a relay, an integrated access and backhaul (IAB) node, a low power node (such as a femto, pico, etc.), a non-terrestrial network (NTN) or non-ground network device such as a satellite network device, a low earth orbit (LEO) satellite, and a geosynchronous earth orbit (GEO) satellite, a vehicle network device, etc., depending on the terminology applied and the technology. In some example embodiments, a low earth orbit (RAN) split architecture includes a centralized unit (CU) and a distributed unit (DU). In some other example embodiments, part of or the entirety of a radio access network device can be piggybacked on an airborne or spaceborne NTN vehicle.

[0035] The term "terminal device" refers to any terminal device that can have wireless communication capability. By way of example, and without limitation, a terminal device can also be referred to as a communication device, user equipment (UE), a subscriber station (SS), a portable subscriber station, a mobile station (MS), or an access terminal (AT). The terminal device can include, but is not limited to, a mobile phone, a cellular phone, a smartphone, a voice over Internet Protocol (VoIP) phone, a wireless local loop phone, a tablet, a wearable terminal device, a personal digital assistant (PDA), a portable computer, a desktop computer, an image capture terminal device, such as a digital camera, a game terminal device, a music storage and playback appliance, a vehicular wireless terminal device, a wireless endpoint, a mobile station, a laptop-embedded equipment (LEE), a laptop-mounted equipment (LME), a USB dongle, a smart device, a wireless customer-premise equipment (CPE), an Internet of Things (IoT) device, a watch or other wearable, a head-mounted display (HMD), a vehicle, a drone, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in an industrial and / or an automated processing chain environments), consumer electronics, devices operating on a business and / or industrial wireless network, etc. The terminal device can also correspond to a mobile terminal (MT) part of an IAB node (e.g., a relay node). In the following description, the terms "terminal device", "communication device", "terminal", "user equipment", and "user" can be used interchangeably.

[0036] As used herein, the term "resource", "transmission resource", "resource block", "physical resource block" (PRB), "uplink resource", or "downlink resource" can refer to any resource used to perform communication, such as a resource in a time domain, a resource in a frequency domain, a resource in a spatial domain, a resource in a code domain, or any other resource that enables communication, etc. In the following, unless explicitly stated, resources in both frequency and time domains will be used as examples of transmission resources for describing some example embodiments of the present disclosure, noting that example embodiments of the present disclosure are equally applicable to other resources in other domains.

[0037] Figure 1 An example communication network 100 in which example embodiments of the present disclosure can be implemented is shown. As shown in FIG. 1, the communication network 100 can include a terminal device 110. In the following, the terminal device 110 can also be referred to as a UE. Figure 1

[0038] ​The communication network 100 can further include a network device 120. In the following, the network device 120 can also be referred to as gNB or eNB, respectively. The terminal device 110 can communicate with the network device 120 within the coverage of a cell 102 managed by the network device 120.

[0039] It should be appreciated that Figure 1 The number of network devices and terminal devices shown in FIG. 1 is given for the purpose of illustration and is not meant to imply any limitation. The communication network 100 can include any suitable number of network devices and terminal devices.

[0040] In some example embodiments, a link from the network device 120 to the terminal device 110 can be referred to as a downlink (DL), and a link from the terminal device 110 to the network device 120 can be referred to as an uplink (UL). In the DL, the network device 120 is a transmitting (TX) device (or transmitter), and the terminal device 110 is a receiving (RX) device (or receiver). In the UL, the terminal device 110 is a TX device (or transmitter), and the network device 120 is an RX device (or receiver).

[0041] The communication in the communication environment 100 can be implemented according to any suitable communication protocol(s), including but not limited to cellular communication protocols of first generation (1G), second generation (2G), third generation (3G), fourth generation (4G), fifth generation (5G), sixth generation (6G), etc., wireless local area network communication protocols such as Institute of Electrical and Electronics Engineers (IEEE) 802.11, and / or any other protocol that is currently known or to be developed in the future. Moreover, the communication can utilize any suitable wireless communication techniques, including but not limited to code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), frequency division duplexing (FDD), time division duplexing (TDD), multiple-input multiple-output (MIMO), orthogonal frequency division multiplexing (OFDM), discrete Fourier transform spread OFDM (DFT-s-OFDM), and / or any other techniques that are currently known or to be developed in the future.

[0042] As mentioned above, CSI feedback enhancement by using AI / ML methods has been discussed and researched.

[0043] As agreed in the 3GPP discussion, perceptual quantization training and non-perceptual quantization training associated with vector quantization (VQ) and scalar quantization (SQ) are important candidate solutions that can be included in future standards. As for the two typical quantization ways, SQ can be simple and easy to implement, while VQ can be superior in performance. The performance gain of VQ is mainly due to the careful design of the codebook, which is usually jointly learned by the encoder and the decoder.

[0044] It has been agreed that the bilateral model use case can be used in CSI compression, and the quantization of CSI feedback and how to use the quantization method are evaluated and studied based on non-perceptual quantization training, perceptual quantization training, quantization methods (including uniform quantization and non-uniform quantization, scalar quantization and vector quantization), and related parameters (e.g., quantization resolution, etc.).

[0045] Regarding the training framework, 3GPP has defined three types of training framework, i.e., Type 1, Type 2, and Type 3. Type 1 and Type 2 training rely on end-to-end gradient propagation from the decoder back to the encoder to update the encoder and decoder simultaneously, while Type 3 training assumes that the UE-side training for the encoder and the NW-side training for the decoder are done in separate training sessions, so the NW-side decoder training will not need to obtain the detailed model parameters of the UE encoder for its own training, and vice versa.

[0046] Among all three (Type 1 / 2 / 3) types of training in the 3GPP discussion, the separate training (also known as Type 3) is the most preferred solution because it preserves the implementation privacy (such as the encoder / decoder model parameters) between the NW and the UE. That is, contrary to Type 1 and Type 2, the encoder on the UE side and the decoder on the NW side are trained in a separate / independent manner and then can work together with each other for seamless encoding and decoding.

[0047] It has been agreed that the evaluation of examples for Type 3, the following cases are considered for evaluation: aligned AI / ML model structure between the NW side and the UE side; unaligned AI / ML model structure between the NW side and the UE side; reporting AI / ML structure for the UE part model and the NW part model, e.g., different backbone networks (e.g., CNN, Transformer, etc.), or the same backbone network but different structure (e.g., number of layers); different size of dataset between the NW side and the UE side; aligned / different quantization / dequantization method between the NW side and the UE side; and whether / how to evaluate the case of input / output type and / or pre / post-processing unalignment between the NW part model and the UE part model.

[0048] However, the “separate training” of the existing UE-side encoder and NW-side decoder is not completely “separate” for the NW and the UE, because they need to share the VQ codebook. However, according to the spirit of separate training, the implementation details of the UE encoder and the NW decoder (including the carefully designed VQ codebook) should be private and non-public. However, in the NW-first (or UE-first) training, the NW (or UE) needs to share the representation of the channel matrix (or the eigenvector) and its corresponding code word Q. Therefore, one entity can easily interpret the entire VQ codebook of its counterpart, and it is essentially in conflict with the spirit of Type 3 separate training.

[0049] Accordingly, the present disclosure proposes mechanisms for channel state information (CSI) feedback enhancement with separate training, particularly based on a Type 3 separate training framework for CSI feedback enhancement with bilateral VQ. In this scheme, a network device can determine a codebook size and a first set of codeword indices of the network device by training a model based on a training dataset associated with CSI, and provide the training dataset, the codebook size of the network device, and the first set of codeword indices to a terminal device. The terminal device can train CSI feedback compression based on the received training dataset, the codebook size of the network device, and the first set of codeword indices.

[0050] In this way, the mechanisms of the Type 3 separate training framework for CSI feedback enhancement with bilateral VQ are implemented, and enhanced system performance for CSI feedback can be achieved.

[0051] Example embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0052] Reference is now made to Figure 2 , which shows an example of an overall structure for CSI compression and reconstruction according to some example embodiments of the present disclosure. Typically, a terminal device 110 (or UE-side entity) compresses raw CSI data through an encoder 101 and quantizes it into codewords through a quantizer 102 for over-the-air delivery. Correspondingly, in a network device 120 (or NW-side entity), a dequantizer 103 and a decoder 104 reconstruct the CSI .

[0053] Reference is now made to Figure 3 , which shows a signaling diagram 300 for communication according to some example embodiments of the present disclosure. As Figure 3 indicated in Figure 1 , the signaling diagram 300 involves a terminal device 110 and a network device 120. For the purpose of discussion, reference is made to to describe the signaling diagram 300.

[0054] In a training phase, the network device 120 can collect a training dataset (e.g., a plurality of stored historical CSI data collected from the terminal device 110), and perform (302) a training procedure based on an AI / ML model for reconstructing CSI feedback data. In this phase, a codebook (e.g., a VQ codebook) of the network device, a decoder at the network device, and a hypothetical encoder can be trained together based on the training dataset associated with CSI (which can be designated as in the following).

[0055] With the trained hypothetical encoder, the network device 120 can compute a first set of codeword indices of the network device a respective codeword index of each data, and generating (304) a dataset pair }, wherein is an index (order) of codewords in the codebook.

[0056] Figure 4 An example of a model 400 for network device side training is shown, according to some example embodiments of the present disclosure;

[0057] As shown, the model 400 for network device side training can include a hypothetical encoder 401, a segmenter 402, a codebook 403 of the network device 120 with size B (i.e., having B codewords in the codebook), and a decoder 407.

[0058] For example, CSI data may be taken as the input of the hypothetical encoder 401, and compressed into a latent vector . Since is a long vector, it can be segmented into W short segments at the segmenter 402, each with a uniform segment length S.

[0059] By looking up in the codebook 403 with size B, each vector , , finds its closest codeword in the codebook, e.g., the 3rd, Bth, and B-1th. Then the sequence {3, B, and B-1} can be outputted.

[0060] According to the codeword index, the codeword is generated as , , which can also be considered to be reconstructed from the index, and assembled into a vector .

[0061] By transmitting the reconstructed codeword vector to the decoder, the CSI is reconstructed. As mentioned above, with the learned hypothetical encoder 401, the network device 120 can calculate the codeword and its respective index for each , and create a dataset pair , }, which can also be referred to as a training dataset and a first codeword index set .

[0062] In the training process, in order to jointly learn the encoder, codebook, and decoder in an end-to-end manner, the loss function for optimization is written as: (1) wherein, represents the true CSI and the reconstruction loss between the CSI reconstructed at the decoder output . The reconstruction loss can be measured in the form of normalized mean squared error (NMSE) or cosine similarity. As mentioned above, the segment latent vector is mapped to the closest codebook vector , which serves as the input to the decoder. Thus, the second term, referred to as the quantization loss, optimizes the codebook such that it becomes as close as possible to the encoder output. In this case, the encoder output using the stop-gradient operator is treated as a constant. Specifically, the codebook can be updated by a dictionary learning algorithm with L2 norm error. The last term is the constraint loss, which in turn optimizes the encoder such that the output is constrained to the codebook.

[0063] After the NW-side model training, the network device 10 can share the data set pair , } and the size B of the NW-side VQ codebook to one or more terminal devices (e.g., the terminal device). Referring back to Figure 3 , the network device 120 can provide (306) the terminal device 110 with the training data set , the codebook size B, and the codeword indices associated with the training data set (i.e., the first set of codeword indices).

[0064] Subsequently, the terminal device 110 can perform (308) a training procedure based on the training data set , the codebook size B, and the codeword indices . Specifically, the terminal device 110 can train the UE-side encoder and the UE-side VQ codebook with the same codebook size B based on the shared data set pair , .

[0065] The training of the UE-side encoder and the UE-side VQ codebook can aim to make the second set of codeword indices identical to the first set of codeword indices, the second set of codeword indices being associated with the UE-side VQ codebook generated based on the UE-side encoder and the UE-side VQ codebook, the first set of codeword indices being associated with the NW-side VQ codebook provided from the network device 120 (i.e., ).

[0066] Figure 5 An example of a model for terminal device-side training is shown in accordance with some example embodiments of the present disclosure.

[0067] As shown, the model 500 for terminal device side training can include an encoder 501, a segmenter 502, a UE-side VQ codebook 503, and a hypothetical decoder 504.

[0068] With the dataset shared from the network device 120, , the terminal device 110 can jointly train the encoder 501, the hypothetical decoder 504, and the VQ codebook 503. The VQ codebook size for the model 500 at the terminal device side is required to be the same as the VQ codebook size at the network device 120 (i.e., NW-side training), i.e., the codebook has B codewords.

[0069] The learning goal in this phase is not only about compressing and reconstructing the real CSI data as accurately as possible to its original form but also to encode the sequence of latent vectors , , with the same codeword indices in its UE-side VQ codebook (e.g., same as {3, B, B-1}) and in the NW-side VQ codebook . That is, after training, the second set of codeword indices generated by the UE-side encoder and the UE-side VQ codebook is the same or very close to the first set of codeword indices provided by the network device 120.

[0070] It should be understood that the latent vectors need to be segmented into the same W segments at the segmenter 502, while the segment length S’ in the UE-side training does not have to be the same as S in the NW-side training.

[0071] In addition, if the terminal device 110 determines that the second set of codeword indices generated at the terminal device 110 is not satisfied, e.g., the difference between the second set of codeword indices and the first set of codeword indices provided by the network device 120 exceeds a threshold level, the terminal device 110 can request additional shared dataset from the network device 120, i.e., the training dataset associated with the CSI and the corresponding set of codeword indices.

[0072] When the network device 120 and the terminal device complete the separate training in sequence, now the UE-side encoder and its VQ codebook can seamlessly work jointly with the NW-side decoder and its VQ codebook.

[0073] Referring to Figure 3 , the terminal device 110 can encode real CSI data and generate (310) a third set of codeword indices based on the UE-side codebook and provide (312) the generated third set of codeword indices to the network device 120.

[0074] The network device 120 can reconstruct (314) the real CSI data based on the third set of codeword indices and its own codebook.

[0075] Figure 6 An example of joint operation of a UE-side encoder, VQ codebook and NW-side decoder, VQ codebook is shown according to some example embodiments of the present disclosure.

[0076] As shown in FIG. 6, at the joint operation stage, which can also be referred to as the inference stage, the CSI data Figure 6 is used as input to the UE-side encoder 601 and compressed into a latent vector .Subsequently segmented into W short segments each with uniform segment length S’ at the segmenter 602. By looking up in the UE-side VQ codebook 603, the terminal device 110 finds the closest sequence of indices , e.g., {3, B, and B-1} (i.e., the third set of codeword indices). The sequence of indices is transmitted over the air from the terminal device 110 to the network device 120. When receiving the sequence

[0077] in the network device 120, the network device 120 first looks up and finds the codewords in the NW-side VQ codebook 604 according to . It should be appreciated that each NW codeword length S can be different from the UE codeword S’. The sequence of codewords is assembled together and a new latent vector is reconstructed, which is subsequently used by the NW-side decoder 605 to reconstruct the original CSI data .

[0078] As mentioned above, a VQ-based UE encoder and NW decoder scheme under the framework of separate training without shared codebook can be implemented. In summary, according to the solution of the present disclosure, the codebook is learned and acquired independently by the terminal device 110 and the network device 120. That is, the network device 120 can not need to send and share its codewords to the terminal device 110 in both training and inference stages. Instead, the network device 120 only shares the size of the codebook with the terminal device 110, i.e., the number or quantity of codewords in the codebook, typically 2^n, such as 128. The terminal device 110 uses the same size of VQ codebook as indicated by the terminal device 120. For the detailed design of the codewords, it depends on the privacy and implementation of the UE.

[0079] It should be appreciated that the network device 120 can reconstruct the real CSI data based on the training data set , the codebook size B, and the sequence of codeword indices​​ The training procedure is provided to the plurality of terminal devices. The plurality of terminal devices 110 can train themselves based on the received training parameters. In the plurality of terminal devices scenario, each terminal device can acquire and maintain a private VQ codebook, where the codewords and codeword lengths do not even need to be identical, but the indices of the encoded must be identical (e.g., for the same CSI , and the plurality of terminal devices should report {3, B, B-1} to the network device 120.

[0080] In this way, data efficiency can be achieved by passing only the codeword indices instead of the codeword values. Furthermore, by preserving the design privacy of the codebook, the valuable intelligent property can be protected.

[0081] The simulation of the solution based on the present disclosure can be discussed as follows. The configurations for dataset generation are given below. Table 1: Dataset configuration

[0082] In this simulation, there are 100000 feature vector samples as CSI data for training and validation. Among them, 80000 samples are used for training, and 20000 samples are used for validation. Each sample includes real numbers, which correspond to a large feature vector connected by 12 subbands, as follows: (2) where is the feature vector for the k th subband channel.

[0083] Each has been processed to the following format: (3) where and are the real and imaginary parts.

[0084] In this simulation, the SGCS is used as a performance metric between the original CSI data and the reconstructed CSI data.

[0085] The simulation refers to the comparison of three training methods, namely (a) joint training with scalar quantization (SQ) (Type 1), (b) joint training with vector quantization (VQ) (Type 1), and (c) separate training with vector quantization (VQ) (Type 3).

[0086] The detailed model structure of the three training methods can be listed as follows. Table 2: Simulation evaluation results​

[0087] For a) and b), joint training with VQ (0.8061) gives about 5% SGCS gain over SQ (0.7569), which is similar to the current observation.

[0088] For b) and c), separate training with VQ (0.7982) introduces about 0.8% negligible SGCS degradation to joint training with VQ (0.8061).

[0089] For c), the UE encoder differs from the NW decoder in two aspects: NN structure (i.e., 3-layer FC in UE vs. 4-layer FC in NW) and codeword length W (6 in UE and 8 in NW).

[0090] Figure 7 A flowchart of an example method 700 for a separate training scheme for CSI feedback according to some example embodiments of the present disclosure is shown. The method 700 can be implemented at the terminal device 110 as shown in Figure 1 For discussion purposes, the method 700 will be described with reference to the terminal device 110 described above. Figure 1

[0091] At 710, the terminal device 110 receives, from a network device, a training dataset associated with CSI, a size of a first codebook of the network device, and a first set of codeword indices associated with the first codebook, wherein the size of the first codebook is associated with a number of vectors in the first codebook.

[0092] At 720, the terminal device 110 determines, at least by determining that a second codebook of the terminal device has a same size as the size of the first codebook based on the training dataset and the first set of codeword indices associated with the first codebook, CSI feedback compression of the terminal device.

[0093] In some example embodiments, the terminal device can generate, at an encoder of the terminal device, a plurality of segmented latent vectors by compressing the training dataset associated with the CSI; and train the encoder of the terminal device for performing the CSI feedback compression with the second codebook of the terminal device such that a second set of codeword indices associated with the second codebook is the same as the first set of codeword indices associated with the first codebook, the second set of codeword indices being determined by the second codebook and the plurality of segmented latent vectors

[0094] In some example embodiments, the terminal device can determine, from the second codebook of the terminal device, a set of codewords closest to the plurality of segmented latent vectors; and determine the second set of codeword indices based on the determined set of codewords.

[0095] ​In some example embodiments, the terminal device can obtain CSI data associated with the CSI; generate, based on the determined CSI feedback compression and the second codebook, a third set of code word indices associated with the second codebook from the CSI data; and transmit the third set of code word indices to the network device.

[0096] In some embodiments, in accordance with a determination that a difference between the second set of code word indices and the first set of code word indices exceeds a threshold level, the terminal device can request another set of shared data from the network device.

[0097] Figure 8 A flowchart of an example method 800 for a separate training scheme for CSI feedback in accordance with some example embodiments of the present disclosure is shown. The method 800 can be implemented at the network device 120 as shown in Figure 1 For purposes of discussion, the method 800 will be described with reference to the Figure 1 described above.

[0098] At 810, the network device 120 determines a latent vector based on a set of training data associated with the CSI.

[0099] At 820, the network device 120 determines a first set of code word indices based on the latent vector and a first codebook of the network device 120.

[0100] At 830, the network device transmits the set of training data, the first set of code word indices, and a size of the first codebook to the terminal device.

[0101] In some example embodiments, the network device 120 can generate a set of segmented latent vectors from the latent vector; determine a plurality of code words corresponding to the segmented latent vectors from the first codebook; and determine the first set of code word indices based on the plurality of code words.

[0102] In some example embodiments, the network device 120 can: receive, from the terminal device, a second set of code word indices associated with a codebook, wherein the second set of code word indices is generated by performing CSI feedback compression on the CSI data at the terminal device; and reconstruct the CSI data associated with the CSI based on the received second set of code word indices and the codebook of the network device.

[0103] In some example embodiments, the network device 120 can: receive, from the terminal device, a third set of code word indices associated with a second codebook of the terminal device, wherein the second codebook has a same size as the size of the first codebook and the third set of code word indices is generated based on the second codebook by performing CSI feedback compression on the CSI data at the terminal device; and reconstruct the set of data associated with the CSI based on the received third set of code word indices and the first codebook of the network device.

[0104] In some example embodiments, the network device 120 can generate a set of code word vectors based on the received third set of code word indices and the first codebook of the network device; and reconstruct the CSI data associated with the CSI by decoding the generated set of code word vectors.

[0105] In some embodiments, the network device 120 can receive a request for another shared data set from the terminal device.

[0106] In some example embodiments, an apparatus (e.g., implemented at the terminal device 110) capable of performing the method 700 can include means for performing the respective steps of the method 700. The means can be implemented in any suitable form. For example, the means can be implemented in circuitry or a software module.

[0107] In some example embodiments, the apparatus includes means for receiving, from the network device, a training data set associated with the CSI, a size of a first codebook of the network device, and a first set of code word indices associated with the first codebook, wherein the size of the first codebook is associated with a number of vectors in the first codebook; and means for determining CSI feedback compression of the apparatus at least by determining that a second codebook of the apparatus has a same size as the size of the first codebook based on the training data set and the first set of code word indices associated with the first codebook.

[0108] In some example embodiments, the apparatus includes means for generating a plurality of segmented latent vectors by compressing, at an encoder of the apparatus, a training data set associated with the CSI; and means for training the encoder of the apparatus for performing CSI feedback compression with a second codebook of the apparatus such that a second set of code word indices associated with the second codebook is the same as a first set of code word indices associated with the first codebook, the second set of code word indices being determined by the second codebook and the plurality of segmented latent vectors.

[0109] In some example embodiments, the apparatus includes means for determining, from the second codebook of the apparatus, a set of code words closest to the plurality of segmented latent vectors; and means for determining the second set of code word indices based on the determined set of code words.

[0110] In some example embodiments, the apparatus includes means for obtaining CSI data associated with the CSI; means for generating, from the CSI data, a third set of code word indices associated with the second codebook based on the determined CSI feedback compression and the second codebook; and means for transmitting the third set of code word indices to the network device.

[0111] In some example embodiments, the apparatus includes means for requesting, from the network device, another shared data set in accordance with a determination that a difference between the second set of code word indices and the first set of code word indices exceeds a threshold level.

[0112] In some example embodiments, an apparatus capable of performing the method 800 (e.g., implemented at the terminal device 120) can include means for performing the respective steps of the method 800. The means can be implemented in any suitable form. For example, the apparatus can be implemented in circuitry or a software module.

[0113] In some example embodiments, the apparatus includes means for determining a latent vector based on a training data set associated with the CSI, means for determining a first set of code word indices based on the latent vector and a first codebook of the apparatus, and means for transmitting the training data set, the first set of code word indices, and a size of the first codebook to the terminal device.

[0114] In some example embodiments, the apparatus includes means for generating a set of segmented latent vectors from the latent vector, means for determining a plurality of code words from the first codebook, each code word of the plurality of code words corresponding to a segmented latent vector, and means for determining the first set of code word indices based on the plurality of code words.

[0115] In some example embodiments, the apparatus includes means for receiving, from the terminal device, a second set of code word indices associated with a codebook, wherein the second set of code word indices is generated by performing CSI feedback compression on CSI data at the terminal device, and means for reconstructing the CSI data associated with the CSI based on the received second set of code word indices and the codebook of the network device.

[0116] In some example embodiments, the apparatus includes means for receiving, from the terminal device, a third set of code word indices associated with a second codebook of the terminal device, wherein the second codebook has a same size as a size of the first codebook, and the third set of code word indices is generated by performing CSI feedback compression on the CSI data at the terminal device based on the second codebook, and means for reconstructing the data set associated with the CSI based on the received third set of code word indices and the first codebook of the network device.

[0117] In some example embodiments, the apparatus includes means for generating a set of code word vectors based on the received third set of code word indices and the first codebook of the apparatus, and means for reconstructing the CSI data associated with the CSI by decoding the generated set of code word vectors.

[0118] In some example embodiments, the apparatus includes means for receiving, from the terminal device, a request for another shared data set.

[0119] Figure 9 is a simplified block diagram of a device 900 suitable for implementing example embodiments of the present disclosure. The device 900 can be provided to implement a communication device, for example, as described above in relation to FIG. 1. Figure 1The terminal device 100 or network device 120 shown is illustrated. As shown, device 900 includes one or more processors 910, one or more memories 920 coupled to processors 910, and one or more communication modules 940 coupled to processors 910.

[0120] Communication module 940 is used for bidirectional communication. Communication module 940 has one or more communication interfaces to enable communication with one or more other modules or devices. The communication interface can represent any interface required for communication with other network elements. In some example embodiments, communication module 940 may include at least one antenna.

[0121] Processor 910 can be of any type suitable for a local technology network, and as a non-limiting example, can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Device 900 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock that synchronizes with the main processor.

[0122] Memory 920 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 924, electrically programmable read-only memory (EPROM), flash memory, hard disk, optical disc (CD), digital video disc (DVD), laser disc, and other magnetic and / or optical storage devices. Examples of volatile memories include, but are not limited to, random access memory (RAM) 922, and other volatile memories that do not persist during power-off periods.

[0123] Computer program 930 includes computer-executable instructions that are executed by an associated processor 910. The instructions of program 930 may include instructions for performing operations / actions of some example embodiments of this disclosure. Program 930 may be stored in memory (e.g., ROM 924). The processor 910 can perform any suitable actions and processes by loading program 930 into RAM 922.

[0124] The exemplary embodiments of this disclosure can be implemented by program 930, enabling device 900 to perform as described in the reference. Figure 2 to Figure 8 Any process discussed in this disclosure. Exemplary embodiments of this disclosure may also be implemented by hardware or a combination of software and hardware.

[0125] In some example embodiments, program 930 may be tangibly contained in a computer-readable medium, which may be included in device 900 (such as in memory 920) or in other storage devices accessible by device 900. Device 900 may load program 930 from the computer-readable medium into RAM 922 for execution. In some example embodiments, the computer-readable medium may include any type of non-transient storage medium, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. The term "non-transient" as used herein is a limitation on the medium itself (i.e., tangible, not tactile), not a limitation on the persistence of data storage (e.g., RAM and ROM).

[0126] Figure 10 An example of a computer-readable medium 1000, which may be in the form of a CD, DVD, or other optical storage disc, is shown. The computer-readable medium 1000 has a program 930 stored thereon.

[0127] Generally, the various embodiments of this disclosure can be implemented in hardware or special-purpose circuitry, software, logic, or any combination thereof. Some aspects may be implemented in hardware, while others may be implemented in firmware or software, which may be executed by a controller, microprocessor, or other computing device. Although aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or other illustrated representations, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein may be implemented as non-limiting examples in hardware, software, firmware, special-purpose circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0128] Some exemplary embodiments of this disclosure also provide at least one computer program product tangibly stored on a computer-readable medium (such as a non-transitory computer-readable storage medium). The computer program product includes computer-executable instructions (such as those included in a program module) that are executed in a device on a target physical or virtual processor to perform any of the methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. The functionality of the program modules can be combined or split among program modules as needed in various embodiments. The machine-executable instructions for the program modules can be executed within a local or distributed device. In a distributed device, the program modules can reside on both local and remote storage media.

[0129] Program code for performing the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that, when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine (as a standalone software package), partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0130] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0131] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. More specific examples of computer-readable storage media include electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0132] Furthermore, although the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order or sequence shown, or that all of the operations shown be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these details should not be construed as limiting the scope of this disclosure, but rather as a description of features that may be specific to particular embodiments. Unless explicitly stated otherwise, certain features described in the context of a single embodiment may also be implemented in combination in a single embodiment. Conversely, unless explicitly stated otherwise, the various features described in the context of a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination.

[0133] Although this disclosure has been described in language specific to structural features and / or methodological actions, it should be understood that this disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as examples of implementing the claims.

Claims

1. An apparatus comprising: At least one processor; as well as At least one memory stores instructions that, when executed by the at least one processor, cause the device to at least: The network device receives a training dataset associated with Channel State Information (CSI), the size of a first codebook of the network device, and a first codeword index set associated with the first codebook, wherein the size of the first codebook is associated with the number of vectors in the first codebook. as well as CSI feedback compression is determined by at least determining that the second codebook of the device has the same size as the first codebook, based on the training dataset and the first codeword index set associated with the first codebook.

2. The apparatus of claim 1, wherein the apparatus is further configured to: Multiple segmented latent vectors are generated by compressing the training dataset associated with the CSI at the encoder of the device; and By training the encoder of the device to perform the CSI feedback compression together with the second codebook of the device, the second codeword index set associated with the second codebook is the same as the first codeword index set associated with the first codebook, the second codeword index set being determined by the second codebook and the plurality of segmented latent vectors.

3. The apparatus of claim 2, wherein the apparatus is further configured to: Determine the set of codewords closest to the plurality of segmented potential vectors from the second codebook of the device; and Based on the determined codeword set, the second codeword index set is determined.

4. The apparatus of claim 1, wherein the apparatus is further configured to: Obtain CSI data; Based on the determined CSI feedback compression and the second codebook, a third codeword index set associated with the second codebook is generated from the CSI data; and The third codeword index set is sent to the network device.

5. The apparatus according to claim 2 or 3, wherein the apparatus is further configured to: If the difference between the second codeword index set and the first codeword index set exceeds a threshold level, another shared dataset is requested from the network device.

6. An apparatus comprising: At least one processor; as well as At least one memory stores instructions that, when executed by the at least one processor, cause the device to at least: Based on the training dataset associated with Channel State Information (CSI), potential vectors are determined; Based on the potential vector and the first codebook of the device, a first codeword index set is determined; as well as The training dataset, the first codeword index set, and the size of the first codebook are sent to the terminal device.

7. The apparatus of claim 6, wherein the apparatus is further configured to: Generate a segmented set of latent vectors from the latent vectors; A plurality of codewords are determined from the first codebook, each of the plurality of codewords corresponding to a segmented latent vector; and Based on the multiple codewords, the first codeword index set is determined.

8. The apparatus of claim 6, wherein the apparatus is further configured to: Receive from the terminal device a third codeword index set associated with a second codebook of the terminal device, wherein the second codebook has the same size as the first codebook, and the third codeword index set is generated by performing CSI feedback compression on CSI data at the terminal device based on the second codebook; and The CSI data is reconstructed based on the received third codeword index set and the first codebook of the device.

9. The apparatus of claim 8, wherein the apparatus is further configured to: Based on the received third codeword index set and the first codebook of the device, a codeword vector set is generated; and The CSI data is reconstructed by decoding the codeword vector set generated.

10. The apparatus of claim 6, wherein the apparatus is further configured to: Receive a request for another shared dataset from the terminal device.

11. A method comprising: At the terminal device, a training dataset associated with Channel State Information (CSI), the size of a first codebook of the network device, and a first codeword index set associated with the first codebook are received from the network device, wherein the size of the first codebook is associated with the number of vectors in the first codebook. as well as CSI feedback compression is determined by at least determining that the second codebook of the terminal device has the same size as the first codebook, based on the training dataset and the first codeword index associated with the first codebook.

12. A method comprising: At the network device, potential vectors are determined based on a training dataset associated with Channel State Information (CSI). Based on the potential vector and the first codebook of the network device, a first codeword index set is determined; as well as The training dataset, the first codeword index set, and the size of the first codebook are sent to the terminal device.

13. An apparatus comprising: A component for receiving from a network device a training dataset associated with Channel State Information (CSI), the size of a first codebook of the network device, and a first codeword index set associated with the first codebook, wherein the size of the first codebook is associated with the number of vectors in the first codebook. as well as A component for determining CSI feedback compression by determining, at least based on the training dataset and the first codeword index set associated with the first codebook, that the second code of the device has the same size as the first codebook.

14. An apparatus comprising: A component used to determine potential vectors based on a training dataset associated with Channel State Information (CSI); A component for determining a first codeword index set based on the potential vector and the first codebook of the device; as well as A component for sending the training dataset, the first codeword index set, and the size of the first codebook to the terminal device.

15. A computer-readable medium comprising instructions that, when executed by a device, cause the device to perform at least the method of claim 11 or the method of claim 12.

Images