CSI compression and decompression

Abstract

A method comprises: generating, at a terminal device, channel state information for a channel between the terminal device and a network device; generating compressed channel state information based on the channel state information by using a compression model; and sending the compressed channel state information and the identification of the compression model to the network device.

Description

Technical Field

[0001] Various example embodiments of the present disclosure relate generally to the field of telecommunications, and in particular, to methods, apparatus, and computer-readable storage media for channel state information (CSI) compression and decompression. Background Art

[0002] Several technologies have been proposed to improve communication performance. For example, communication devices can adopt artificial intelligence / machine learning (AI / ML) models to improve communication quality. AI / ML models can be applied to different scenarios, for example, for reporting CSI. To this end, the terminal device can compress the CSI by using the AI / ML model, and the network device can reconstruct the CSI by using other AI / ML models. Summary of the invention

[0003] In a first aspect of the present disclosure, an apparatus is provided. The apparatus includes: at least one processor; and at least one memory, wherein the at least one memory stores instructions, which when executed by the at least one processor, cause the apparatus to at least: generate channel state information for a channel between the apparatus and a network device; generate compressed channel state information based on the channel state information by using a compression model; and send the compressed channel state information and an identifier of the compression model to the network device.

[0004] In a second aspect of the present disclosure, a device is provided. The device includes: at least one processor; and at least one memory, wherein the at least one memory stores instructions, and when the instructions are executed by the at least one processor, the device at least: receives compressed channel state information for a channel between the terminal device and the device, and an identifier of a compression model for generating the compressed channel state information from a terminal device; determines an indicator corresponding to the compression model based on the identifier; and reconstructs the channel state information of the channel based on the compressed channel state information and the indicator by using the decompression model.

[0005] In a third aspect of the present disclosure, a device is provided. The device includes: at least one processor; and at least one memory, wherein the at least one memory stores instructions, and when the instructions are executed by the at least one processor, the device at least: generates channel state information for a channel between the device and a network device; generates compressed channel state information based on the channel state information and an indicator corresponding to a decompression model for reconstructing the channel state information by using a compression model; and sends the compressed channel state information to the network device.

[0006] In a fourth aspect of the present disclosure, a device is provided. The device includes at least one processor; and at least one memory, the at least one memory storing instructions, which, when executed by the at least one processor, causes the device to at least: receive compressed channel state information for a channel between the terminal device and the device from a terminal device; reconstruct the channel state information of the channel based on the compressed channel state information by using a decompression model, and wherein the compressed channel state information is generated based on the channel state information and an indicator corresponding to the decompression model.

[0007] In a fifth aspect of the present disclosure, a method is provided. The method includes: generating, at a terminal device, channel state information for a channel between the terminal device and a network device; generating compressed channel state information based on the channel state information by using a compression model; and sending the compressed channel state information and an identifier of the compression model to the network device.

[0008] In a sixth aspect of the present disclosure, a method is provided. The method includes: receiving, at a network device, compressed channel state information for a channel between the terminal device and the network device, and an identifier of a compression model used to generate the compressed channel state information from a terminal device; determining an indicator corresponding to the compression model based on the identifier; and reconstructing the channel state information of the channel based on the compressed channel state information and the indicator by using a decompression model.

[0009] In a seventh aspect of the present disclosure, a method is provided. The method includes: generating, at a terminal device, channel state information for a channel between the terminal device and a network device; generating compressed channel state information based on the channel state information and an indicator corresponding to a decompression model for reconstructing the channel state information by using a compression model; and sending the compressed channel state information to the network device.

[0010] In an eighth aspect of the present disclosure, a method is provided. The method includes: receiving, at a network device, compressed channel state information for a channel between the terminal device and the network device from a terminal device; reconstructing the channel state information of the channel based on the compressed channel state information by using a decompression model; wherein the compressed channel state information is generated based on the channel state information and an indicator corresponding to the decompression model.

[0011] In a ninth aspect of the present disclosure, an apparatus is provided. The apparatus includes: a component for generating channel state information for a channel between the apparatus and a network device; a component for generating compressed channel state information based on the channel state information by using a compression model; and a component for sending the compressed channel state information and an identification of the compression model to the network device.

[0012] In a tenth aspect of the present disclosure, an apparatus is provided. The apparatus includes: a component for receiving compressed channel state information for a channel between a terminal device and the apparatus, and an identifier of a compression model for generating the compressed channel state information; a component for determining an indicator corresponding to the compression model based on the identifier; and a component for reconstructing the channel state information of the channel based on the compressed channel state information and the indicator by using a decompression model.

[0013] In an eleventh aspect of the present disclosure, an apparatus is provided. The apparatus includes: a component for generating channel state information for a channel between the apparatus and a network device; a component for generating compressed channel state information based on the channel state information and an indicator corresponding to a decompression model for reconstructing the channel state information by using a compression model; and a component for sending the compressed channel state information to the network device.

[0014] In a twelfth aspect of the present disclosure, an apparatus is provided. The apparatus includes: a component for receiving compressed channel state information for a channel between the terminal device and the apparatus from a terminal device; and a component for reconstructing the channel state information of the channel based on the compressed channel state information by using a decompression model, wherein the compressed channel state information is generated based on the channel state information and an indicator corresponding to the decompression model.

[0015] In a thirteenth aspect of the present disclosure, a computer-readable medium is provided, wherein the computer-readable medium includes instructions stored thereon for causing a device to at least execute the method according to the fifth aspect.

[0016] In a fourteenth aspect of the present disclosure, a computer-readable medium is provided, wherein the computer-readable medium includes instructions stored thereon for causing a device to at least execute the method according to the sixth aspect.

[0017] In a fifteenth aspect of the present disclosure, a computer-readable medium is provided, wherein the computer-readable medium includes instructions stored thereon for causing a device to at least execute the method according to the seventh aspect.

[0018] In a sixteenth aspect of the present disclosure, a computer-readable medium is provided, wherein the computer-readable medium includes instructions stored thereon for causing a device to at least execute the method according to the eighth aspect.

[0019] It should be understood that the abstract is not intended to identify the key or essential features of the 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 readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Some example embodiments will now be described with reference to the accompanying drawings, in which:

[0021] Figure 1 An example communication environment is shown in which example embodiments of the present disclosure may be implemented;

[0022] Figure 2 An example architecture of a two-sided ML model for CSI reporting according to some example embodiments of the present disclosure is shown;

[0023] Figure 3 An example of a decompression model shared by multiple compression models for training according to some example embodiments of the present disclosure is shown;

[0024] Figure 4 An example signaling procedure for reporting CSI in the case of first training at the terminal device side according to some example embodiments of the present disclosure is shown;

[0025] Figure 5A An example of a compression model shared by multiple decompression models for training according to some example embodiments of the present disclosure is shown;

[0026] Figure 5B Another example of a compression model shared by multiple decompression models for training according to some example embodiments of the present disclosure is shown;

[0027] Figure 6 An example signaling flow for reporting CSI in the case of first training on the network device side according to some example embodiments of the present disclosure is shown;

[0028] Figure 7 An example converter architecture for CSI compression and decompression according to some example embodiments of the present disclosure is shown;

[0029] Figure 8 Another example converter architecture for CSI compression and decompression according to some example embodiments of the present disclosure is shown;

[0030] Fig. 9 A flowchart of a method implemented at a terminal device according to some example embodiments of the present disclosure is shown;

[0031] Fig.10 A flowchart of a method implemented at a network device according to some example embodiments of the present disclosure is shown;

[0032] Fig.11 A flowchart of a method implemented at a terminal device according to some example embodiments of the present disclosure is shown;

[0033] Fig.12 A flowchart of a method implemented at a network device according to some example embodiments of the present disclosure is shown;

[0034] Fig.13 shows a simplified block diagram of a device suitable for implementing an example embodiment of the present disclosure; and

[0035] Fig.14 A block diagram of an example computer-readable medium is shown according to some example embodiments of the present disclosure.

[0036] Throughout the drawings, the same or similar reference numerals refer to the same or similar elements. DETAILED DESCRIPTION

[0037] The principles of the present disclosure will now be described with reference to some example embodiments. It should be understood that these embodiments are only for illustrative purposes and are helpful for those skilled in the art to understand and implement the present disclosure without implying any limitation on the scope of the present disclosure. The embodiments described herein can be implemented in various ways different from the ways described below.

[0038] 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 skill in the art to which this disclosure belongs.

[0039] References in this disclosure to "one embodiment," "an embodiment," "an example embodiment," etc. indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment necessarily includes the particular feature, structure, or characteristic. Moreover, such phrases do not necessarily refer to the same embodiment. Furthermore, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it is submitted that it is within the knowledge of those skilled in the art to affect such feature, structure, or characteristic in conjunction with other embodiments, whether or not explicitly described.

[0040] It should be understood that although the terms "first", "second", etc. may 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 element, and they do not limit the order of the nouns. For example, without departing from the scope of the exemplary embodiment, the first element may be referred to as the second element, and similarly, the second element may be referred to as the first element. As used herein, the term "and / or" includes any and all combinations of one or more of the listed terms.

[0041] As used herein, “at least one of: ” and “at least one of ” and similar expressions, where a list of two or more elements is connected by “and” or “or”, means at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

[0042] As used herein, unless explicitly stated, performing a step "in response to A" does not indicate that the step is performed immediately after "A" occurs and may include one or more intervening steps.

[0043] The terms used herein are only used for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include", "have", and / or "include" when used herein specify the presence of the features, elements and / or components, etc., but do not exclude the presence or addition of one or more other features, elements, components and / or combinations thereof.

[0044] As used in this application, the term "circuitry" may refer to one or more or all of the following: (a) Hardware circuit implementation only (such as implementation in simulation only) and / or digital circuit. (b) Combinations of hardware circuits and software such as (where applicable): (i) Analog and analog combinations; digital hardware circuits with software / firmware and (ii) software and memory that work together with any portion of a hardware processor (including a digital signal processor) to enable a device (such as a mobile phone or server) to perform various functions, and (c) hardware circuits and / or processors such as a microprocessor or a portion of a microprocessor that require software (e.g., firmware) for operation, but the software may not be present when it is not required for operation.

[0045] This definition of circuitry applies to all uses of this item in this application, including in any claims. As another example, as used in this application, the term circuitry also covers an implementation of only a hardware circuit or processor (or multiple processors) or a portion of a hardware circuit or processor and its accompanying software and / or firmware. For example, if applicable to the particular claim element, the term circuitry also covers a baseband integrated circuit or processor integrated circuit used in a mobile device or similar integrated circuit in a server, cellular network device, or other computing or network device.

[0046] As used herein, the term "communication network" refers to a network that complies with any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), Advanced LTE (LTE-A), Wideband Code Division Multiple Access (WCDMA), High Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. In addition, the communication between the terminal device and the network device in the communication network can be performed according to any suitable generation communication protocol, including but not limited to the first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G), sixth generation (6G) communication protocol and / or any other protocol currently known or developed in the future. The embodiments of the present disclosure can be applied to various communication systems. In view of the rapid development of communication, there are of course communication technologies and systems of future types that can implement the present disclosure. It should not be considered that the scope of the present disclosure is limited to the above-mentioned system.

[0047] 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 from it. A network device may refer to a base station (BS) or an access point (AP), for example, a Node B (Node B or NB), an evolved NodeB (eNode B or eNB), a NR NB (also referred to as a 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, a micro, a non-terrestrial network (NTN) or a non-terrestrial network device (such as a satellite network device), a low-earth orbit (LEO) satellite and a geosynchronous earth orbit (GEO) satellite, an aircraft network device, etc., depending on the terminology and technology applied. In some example embodiments, a radio access network (RAN) split architecture includes a centralized unit (CU) and a distributed unit (DU) at an IAB donor node. An IAB node includes a mobile terminal (IAB-MT) portion that behaves like a UE toward a parent node, and the DU portion of the IAB node behaves like a base station toward a next-hop IAB node.

[0048] The term "terminal device" refers to any terminal device capable of wireless communication. As an example and not limitation, a terminal device may also be referred to as a communication device, a user equipment (UE), a user station (SS), a portable user station, a mobile station (MS) or an access terminal (AT). The terminal device may include, but is not limited to, a mobile phone, a cellular phone, a smart phone, a voice over IP (VoIP) phone, a wireless local loop phone, a tablet computer, 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 device), a vehicle-mounted wireless terminal device, a wireless endpoint, a mobile station, a laptop embedded device (LEE), a laptop device (LME), a USB dongle, a smart device, a wireless customer premises equipment (CPE), an Internet of Things (IoT) device, a watch or other wearable device, a head mounted display (HMD), a vehicle, a drone, medical equipment and applications (e.g., remote surgery), industrial equipment and applications (e.g., robots and / or other wireless devices operating in the context of an industrial and / or automated processing chain), consumer electronic devices, equipment operating on a commercial and / or industrial wireless network, etc. The terminal device may also correspond to the mobile terminal (MT) part of an IAB node (eg, a relay node).In the following description, the terms "terminal device", "communication device", "terminal", "user equipment" and "UE" may be used interchangeably.

[0049] As used herein, the terms "resource", "transmission resource", "resource block", "physical resource block" (PRB), "uplink resource" or "downlink resource" may refer to any resource used to perform communication, for example, communication between a terminal device and a network device, such as resources in the time domain, resources in the frequency domain, resources in the spatial domain, resources in the code domain, or any other combination of time, frequency, space and / or code domain resources to achieve communication. In the following, unless explicitly stated, resources in the frequency domain and the time domain will be used as examples of transmission resources for describing some example embodiments of the present disclosure. Note that the example embodiments of the present disclosure are equally applicable to other resources in other domains.

[0050] As used herein, the term "AI / ML model" may refer to a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs. In this disclosure, the term "AI / ML model" may be interchangeable with the terms "model", "AI model", and "ML model".

[0051] The terms “two-sided AI / ML model”, “two-sided model” or “two-sided ML model”, etc., as used herein may refer to a pair of AI / ML models that perform joint reasoning, where the joint reasoning includes AI / ML inference whose reasoning is jointly performed across terminal devices and network devices.

[0052] As used herein, unless explicitly indicated otherwise, the term "compression model" may refer to a model used to compress CSI. Similarly, unless explicitly indicated otherwise, the term "decompression model" may refer to a model used to reconstruct CSI from compressed CSI.

[0053] Figure 1 An example communication environment 100 is shown in which example embodiments of the present disclosure may be implemented. In the communication environment 100, a terminal device 110 and a network device 120 may communicate with each other. In some example embodiments, the terminal device 110 may be, for example, a UE. The network device 120 may be, for example, a gNB.

[0054] In some example embodiments, the link from network device 120 to terminal device 110 is referred to as a downlink (DL), and the link from terminal device 110 to network device 120 is referred to as an uplink (UL). In DL, network device 120 is a transmitting (TX) device (or transmitter), and terminal device 110 is a receiving (RX) device (or receiver). In UL, terminal device 110 is a TX device (or transmitter), and network device 120 is an RX device (or receiver).

[0055] Communications in the communication environment 100 may be implemented according to any suitable communication protocol, including but not limited to cellular communication protocols of the 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, etc., and / or any other protocol currently known or developed in the future. In addition, communications may utilize any suitable wireless communication technology, including but not limited to: code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), frequency division duplex (FDD), time division duplex (TDD), multiple input multiple output (MIMO), orthogonal frequency division multiple access (OFDM), discrete Fourier transform spread OFDM (DFT-s-OFDM) and / or any other technology currently known or developed in the future.

[0056] It should be understood that Figure 1 The number of devices and their connections shown in are for illustration purposes only and do not imply any limitation.Communication environment 100 may include any suitable number of devices configured to implement example embodiments of the present disclosure.

[0057] In the communication environment 100, the terminal device 110 can estimate the channel between the terminal device 110 and the network device 120 and report the CSI to the network device 120. The two-sided ML model can be used for CSI reporting. Figure 2, which illustrates an example architecture of a bilateral ML model 200 for CSI reporting according to some example embodiments of the present disclosure.

[0058] Generally, the bilateral ML model 200 may include a first part 201 for the terminal device side (e.g., UE side) and a second part 202 for the network device side (e.g., gNB side). For example, the CSI obtained by measuring the reference signal from the network device 120 may be Figure 2 as shown. The channel values of the CSI may include complex values, but may also be represented by the real part and the imaginary part. Using the transformation, the CSI with real numbers can represent the channel with elements.

[0059] The compression model deployed at the terminal device 110 may be an ML model, and may compress the CSIw into the compressed CSI, which may be represented as a lower-dimensional vector e(w), (d << N). To transmit the compressed CSI over the wireless link, the quantization module 220 (also referred to as a quantizer) may convert the compressed CSI into compressed bits, which may be represented as and may also be referred to as a bit sequence. For example, the quantization module 220 may perform binarization and map the dimension-reduced value e(w) into a vector of compressed binary elements c represented by b bit values.

[0060] Then, the compressed bits c with b bit values may be transmitted from the terminal device 110 to the network device 120. On the network device side, the dequantization module 230 (also referred to as a dequantizer) may convert the compressed bits received from the terminal device 110 into a set of values, which may be represented as q(w). For example, the dequantization module 230 may perform de-binarization and obtain a set of values q(w) from the binary vector c using a codebook. Then, the decompression model 240 may generate the reconstructed CSI based on these values q(w) For example, the decompression model 240 may reconstruct the channel by mapping q(w) to the channel estimate to reconstruct the channel.

[0061] The compression model 210 and the decompression model 240 may be parameterized machine learning models, such as neural networks. The compression model 210 may be implemented as an encoder, and the decompression model 240 may be correspondingly implemented as a decoder. In the present disclosure, the terms "compression model" and "encoder" may be used interchangeably, and the terms "decompression model" and "decoder" may be used interchangeably.

[0062] It should be understood that the architecture as Figure 2 shown is an example without any limitation. In Figure 2, the compression model 210 and the quantization module 220 are shown respectively, and the decompression model 240 and the inverse quantization module 230 are shown respectively. However, this is for illustration only and is not limited in any way. In some example embodiments, the functions of the compression model 210 and the quantization module 220 as described above may be performed by a single model. Alternatively or additionally, in some example embodiments, the functions of the decompression model 240 and the inverse quantization module 230 as described above may be performed by a single model. In the following, the compression model 210 (or encoder) and the optional quantization module 220 may be referred to as a CSI generation part. Therefore, the decompression model 240 (or decoder) and the optional inverse quantization module 230 may be referred to as a CSI reconstruction part.

[0063] For example Figure 2 The two-sided ML model shown can adopt different training schemes. For example, if the type III training of separate training of encoder and decoder with only exchange of data sets is followed, two training schemes can be used: terminal device-first training (also called UE-first training) scheme and network device-first training (also called network-first training) scheme.

[0064] Using a two-sided machine learning model (such as Figure 2 The CSI compression of the model shown in FIG. 1 is shown to have better performance than the classical method. However, training collaboration is required to make the independently trained compression model working from the terminal device side and the decompression model working from the network device side work together.

[0065] Take UE as an example, terminal equipment and gNB as an example. If the UE-first training method is followed, different compression models can be obtained by different training entities after training. For example, UEs from different sources can use different compression models. The differences between these compression models can be caused by different model structures, different training data sets, different training strategies, different UE versions, different UE types, etc. In other words, the gNB must deploy a corresponding decompression model for each compression model. This will lead to the burden of training and storing multiple network part models at the gNB to pair with different UE part models.

[0066] A similar problem occurs if the network-first training approach is followed. If the network-first training approach is followed, different training entities may obtain different decompression models after training. For example, gNBs from different sources may use different decompression models. The differences between these decompression models may be caused by different model structures, different training datasets, different training strategies, etc. In other words, the UE must deploy a corresponding compression model for each decompression model. This will result in the burden of training and storing multiple UE-part models at the UE to pair with different network-part models.

[0067] Maintenance and storage of multiple models can be expensive. Memory usage will be multiplied by the number of compression models for the gNB. The same problem will occur for the UE if different encoders are trained to match multiple decoders.

[0068] According to some exemplary embodiments of the present disclosure, a scheme for CSI compression and decompression is provided. In some exemplary embodiments, terminal device-first training may be adopted. A decompression model shared by multiple compression models may be used, and the reconstruction of CSI by the decompression model may be assisted by using an indicator corresponding to the compression model as an additional input to the decompression model. In this way, the need for a network device to train and maintain multiple decompression models may be avoided, and the performance of a single decompression model at the network device may be improved.

[0069] In some example embodiments, network device first training may be used. A compression model shared by multiple decompression models may be used, and the compression of CSI by the compression model may be assisted by using an indicator corresponding to the decompression model as an additional input to the compression model. In this way, the need for training and maintaining multiple compression models by the terminal device may be avoided, and the performance of the terminal device on the terminal device may be improved.

[0070] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. Terminal Equipment-First Training Program

[0071] In some example embodiments, a terminal device-first training scheme may be employed. Each training entity for a compression model may train an end-to-end model, including a compression model, a quantization module, a dequantization module, and a decompression model, such as Figure 2 Model 200. For example, each training entity on the UE side can train an end-to-end autoencoder, which includes an encoder, a quantizer, a dequantizer, and a decoder. Therefore, multiple compression models can be trained by different training entities on the terminal device side.

[0072] In the following, for the purpose of discussion without any limitation, it is assumed that Q compression models can be trained by Q training entities. After training the end-to-end model, at least a portion of the training dataset used to train each of the Q compression models can be shared with the training entity for the decompression model. The training sample {w i ,q i (w)} can be shared with the training entity for the decompression model, where i=1, 2, ..., Q.

[0073] Then, based on the training sample {w i ,q i (w)} train a decompression model shared by multiple compression models. Now refer to Figure 3 , to illustrate an example of training a common decompression model 310 for Q compression models according to some example embodiments of the present disclosure. When training the common decompression model 310, the latent vector q of each training sample i (w) is used as input to the common decompression model 310, and there is no compressed CSIw i Can be used as ground truth.

[0074] In addition, an indicator (eg, a flag) corresponding to each compression model may be used as an additional input to the common decompression model 310. In other words, the training sample {w i ,q i (w)} can be used to train a common decompression model 310 in conjunction with the indicator corresponding to the i-th compression model. Figure 3 As shown, the latent vector q i (w) sign f i Together with the latent vector q i (w) are input to the common decompression model 310 together.

[0075] In such an example embodiment, since the input to the common decompression model 310 is a quantized latent vector, the indicator corresponding to the i-th compression model (e.g., flag f i ) can be added to the latent vector q i (w). For example, an element may be added to the latent vector q i (w) indicates the latent vector q i (w) From which type of terminal device.

[0076] In some example embodiments, the training of the ML model can be performed offline. After training, the Q compression models and the common decompression model 310 can be used for real-time operation. Figure 4, to illustrate an example signaling process 400 for reporting CSI in the case of first training of a terminal device according to some example embodiments of the present disclosure. The signaling process 400 involves a terminal device 110 and a network device 120.

[0077] Terminal device 110 may generate 405 CSI for a channel between terminal device 110 and network device 120. In other words, channel estimation may be performed by terminal device 110. For example, terminal device 110 may estimate the channel based on a reference signal from network device 110 and construct CSI.

[0078] The terminal device 110 may then generate (410) compressed CSI based on the CSI generated using the compression model at 405. For example, the estimated CSIw may be input to the compression model to generate compressed CSIe(w).

[0079] Terminal device 110 may send (415) to network device 120 the compressed CSI and an identification of the compression model used by terminal device 110. For example, a combined message including both the compressed CSI and the identification may be sent to network device 120. The identification may be included as part of the CSI report. Since network device 120 may maintain a decompression model common to multiple compression models, terminal device 110 may instruct network device 120 which compression model to use so that the common decompression model of network device 120 reconstructs CSIw with improved performance.

[0080] The identification of the compression model may include any suitable information that may identify which compression model terminal device 110 uses to compress the CSI. In some example embodiments, the identification of the compression model may be based on or related to the type of the compression model. The type of the compression model may correspond to any suitable characteristic of the compression model, such as, but not limited to, a machine learning algorithm employed by the compression model, a neural network structure of the compression model, a training data set used to train the compression model, a training entity of the compression model, a provider of the compression model, or a model parameter of the compression model.

[0081] Alternatively or additionally, the identification of the compression model may be based on or related to the type of the terminal device 110. The type of the terminal device 110 may correspond to any suitable characteristics of the terminal device 110, such as but not limited to the capabilities of the terminal device 110 (e.g., capabilities associated with ML functions), the release version of the terminal device 110, the provider of the terminal device 110, etc.

[0082] In some example embodiments, in order to send the compressed CSI over a wireless link between the terminal device 110 and the network device 120, the compressed CSI may be quantized into compressed bits. For example, the compressed CSIe(w) may be quantized by a quantization module (e.g., a binarization module) into a vector of compressed binary elements c including b bit values. The terminal device 110 may then send the compressed bits and an identifier of the compression model to the network device 120. It should be noted that the identifier may also be quantized into bits for transmission with the compressed bits. For example, the identifier of the terminal device 110 may be binarized into bits, which may be appended to the compressed bits for transmission.

[0083] On the network side, network device 120 may receive (420) from terminal device 110 the compressed CSI and an identification of a compression model used by terminal device 110. For example, network device 120 may receive from terminal device 110 a combined message including the compressed CSI and an identification of the compression model.

[0084] In some example embodiments, if compressed bits and bits corresponding to the identification are received from terminal device 110, network device 120 may dequantize the compressed bits into a set of values, which may be a latent vector q(w). For example, dequantization module 230 may be used to dequantize the compressed bits to obtain a latent vector. Similarly, the bits corresponding to the identification may be dequantized into an identification of the compression model.

[0085] Then, the network device 120 can determine the indicator corresponding to the compression model used by the terminal device 110 according to the identifier. The indicator used here is the same as the indicator used to train the common decompression model. For example, the flag f as described above can be determined. i In some example embodiments, the indicator may be the same as the identifier. Alternatively, in some example embodiments, multiple indicators (eg, identifiers f1 to f Q ) and a mapping between multiple identifiers can be used to determine the indicator to use.

[0086] By using the common decompression model, the network device 120 can reconstruct the CSI of the channel between the terminal device 110 and the network device 120 based on the compressed CSI and the determined indicator. For example, the latent vector q(w) and the determined indicator f i The decoder may be used as an input to the common decompression model 310. The common decompression model 310 may generate the reconstructed CSI. In other words, the trained decoder may be used to obtain an estimate of the channel between the terminal device 110 and the network device 120.

[0087] Some example embodiments for the case of terminal device-first training are described above. For terminal device-first training, each training entity of the terminal device can independently train an autoencoder with an encoder and a decoder on channel samples. After type III training, the Q training entities on the terminal device side can share their channel and quantized encoding {(w,q1(w)),(w,q2(w)),…,(w,q Q (w))}. The training entity on the network device side may need to use a shared data set to train the pattern to decode each quantization code to reconstruct the channel. By signaling the identity of the compression model to the network device, and also by taking the indicator of the compression model into account in the CSI reconstruction, the CSI reconstruction performance can be ensured or even improved without the need to train and maintain different decompression models. Network Equipment-First Training Program

[0088] In some example embodiments, a network device-first training scheme may be employed. Each training entity for the decompression model may train an end-to-end model, including a compression model, a quantization module, a dequantization module, and a decompression model, such as Figure 2 Model 200. For example, each training entity can train an end-to-end autoencoder, which includes an encoder, a quantizer, a dequantizer, and a decoder. Therefore, multiple decompression models can be trained by different training entities on the network device side.

[0089] In the following, for the purpose of discussion without any limitation, it is assumed that P decompression models can be trained by P training entities on the network side. After training the end-to-end model, at least a portion of the training data set used to train each of the P decompression models can be shared with the training entity for the compression model. The training sample {w for training the jth decompression model by the jth training entity j ,e j (w)} can be shared with the training entities used for the compression model, where j=1, 2, ..., P.

[0090] Then, based on the training sample {w j ,e j (w)} trains a compression model that is shared by multiple decompression models. Now refer to Figure 5A , which shows an example of training a common compression model 510 for P decompression models according to some example embodiments of the present disclosure. When training the common compression model 510, the uncompressed CSIe of each training sample j is used as input to the common compression model 510, and the compressed CSIe j (w) can be used as ground truth.

[0091] In addition, the indicator (e.g., flag) corresponding to each decompression model can be used as an additional input to the common compression model 510. In other words, the training samples of the jth decompression model can be used to train the common compression model 510 in combination with the indicator corresponding to the jth compression model. Figure 5A As shown, the flag of the compressed CSI is input to the common compression model 510 together with the CSI without compression.

[0092] Alternatively, in some example embodiments, the training samples {w j ,q j (w)} can be shared with the training entities for the compression model, where j = 1, 2, ..., P. In such an example embodiment, Figure 5B As shown, a common compression model 510 is trained together with a quantization module 520. As shown, the uncompressed CSI and the marker f of each training sample are j is used as input to the common compression model 510 and then the quantization module. The compressed bits q j (w) can be used as ground truth. It should be understood that if the compressed bit q j (w) is used to train a general compression model, then the quantization module is the same as the quantization module used to train the decompression model.

[0093] In such an example embodiment, the CSI, which may be a channel matrix w, may have a size of the number of transmit antennas times the number of receive antennas times the number of frequency resource elements (e.g., (32 x 34 x 52)). Since the channel values ​​have multiple dimensions, an indicator dimension may be added instead of using a single element. The indicator dimension may be added to one of the dimensions based on the architecture of the model. For example, in the case of a transformer-based encoder, an indicator may be added to the first dimension. In other words, a 4 x 52 shaped indicator number is added to the input.

[0094] After training, the P decompression models and the common compression model 510 can be used for real-time operation. Figure 6 , to illustrate an example signaling process 600 for reporting CSI in case of first training of a network device according to some example embodiments of the present disclosure. The signaling process 600 involves a terminal device 110 and a network device 120.

[0095] Terminal device 110 may generate 605 CSI for a channel between terminal device 110 and network device 120. In other words, channel estimation may be performed by terminal device 110. For example, terminal device 110 may estimate a channel based on a reference signal from network device 110 and construct CSI.

[0096] Then, by using the compression model, the terminal device 110 may generate (610) compressed CSI based on the CSI generated at 605 and an indicator corresponding to the decompression model for reconstructing the CSI. For example, the estimated CSIw and the flag f may be input to the compression model to generate compressed CSIe(w).

[0097] In some example embodiments, the indicator corresponding to the decompression model at network device 120 may be based on or related to the type of the decompression model. The type of the decompression model may correspond to any suitable characteristic of the decompression model, such as, but not limited to, a machine learning algorithm employed by the decompression model, a neural network structure of the decompression model, a training data set used to train the decompression model, a training entity for the decompression model, a provider of the decompression model, or a model parameter of the decompression model.

[0098] Alternatively or additionally, the indicator corresponding to the decompression model at network device 120 may be based on or associated with a type of network device 120. The type of network device 120 may correspond to any suitable characteristic of network device 120, such as, but not limited to, capabilities of network device 120 (e.g., capabilities associated with ML functionality), a provider of network device 120, etc.

[0099] Here the indicator for the decompression model is the same as the indicator for the decompression model in training. It should also be noted that in such example embodiments, terminal device 110 may know which decompression model will be used by network device 120 in any suitable manner. Example embodiments are not limited in this respect.

[0100] Terminal device 110 may send 615 the compressed CSI to network device 120. For example, a CSI report may be sent to network device 120.

[0101] In some example embodiments, in order to transmit the compressed CSI over the wireless link between terminal device 110 and network device 120, the compressed CSI may be quantized into compressed bits. For example, the compressed CSIe(w) may be quantized by a quantization module (e.g., a binarization module) into a vector of compressed binary elements c including b bit values. Then, terminal device 110 may transmit the compressed bits to network device 120.

[0102] On the network side, the network device 120 may receive 620 the compressed CSI from the terminal device 110. For example, the network device 120 may receive a CSI report from the terminal device 110 that includes the compressed CSI.

[0103] In some example embodiments, if compressed bits are received from terminal device 110, network device 120 may dequantize the compressed bits into a set of values, which may be a latent vector q(w). For example, dequantization module 230 may be used to dequantize the compressed bits to obtain a latent vector.

[0104] Then, the network device 120 can reconstruct the CSI of the channel between the terminal device 110 and the network device 120 according to the compressed CSI through the decompression model. For example, the latent vector q(w) can be used as an input to the decompression model deployed at the network device 120. The decompression model can generate the reconstructed CSI. In other words, the trained decoder can be used to obtain an estimate of the channel between the terminal device 110 and the network device 120.

[0105] Some example embodiments for the case of network device-first training are described above. For network device-first training, each training entity of the network device can independently train an autoencoder with an encoder and a decoder on the channel sample w. After type III training, the P training entity on the network device side can share their channels and encoding {(w, e1(w)), (w, e2(w)), …, (w, e2(w)) with the training entity on the terminal device side. P (w))}. The training entity on the terminal device side may need to use a shared dataset to train one (or more) models to encode the channels for each decompression model. By taking the indicator of the decompression model into account in the CSI generation, the performance can be ensured or even improved without the need to train and maintain different compression models.

[0106] It should be noted that the above exemplary embodiments are described for a single network device for terminal device-first training and a single terminal device for network device-first training. However, this is only an example without any limitation. If the communication involves multiple network devices and multiple terminal devices, the proposed solution can be applied. Experimental results of the embodiment

[0107] Figure 7An example converter architecture (which may also be referred to as TF 1) for CSI compression and decompression according to some example embodiments of the present disclosure is shown. As shown, in encoder 700, the real and imaginary components of the complex CSI value are cascaded at block 701. Block 702 includes a fully connected (FC) layer and a leaky rectified linear unit (LReLU) layer with 64 input nodes and 128 output nodes. Before applying attention, position embedding is performed. Converter block 710 is repeated 5 times. In converter block 710, 16 head attentions are applied at block 703. Then, at block 704, layer normalization is performed. Block 705 includes an FC layer with 128 input nodes and 256 output nodes, an LReLu layer, and another FC layer with 128 input nodes and 256 output nodes. At box 706, layer normalization is performed. Block 702 includes an FC layer with 128*13 input nodes and 64 output nodes, and quantization is then performed at block 708.

[0108] In decoder 750, block 751 includes a FC layer with 64 input nodes and 128*13 output nodes. Then, converter block 760 can be repeated 5 times. In converter block 760, 16 head attentions are applied at block 753. Then, at block 754, layer normalization is performed. Block 755 includes a FC layer with 128 input nodes and 256 output nodes, an LReLu layer, and another FC layer with 128 input nodes and 256 output nodes. At box 756, layer normalization is performed. Block 757 includes an LReLu layer and a FC layer with 128 input nodes and 256 output nodes, and then at block 758, complex values ​​can be obtained.

[0109] Figure 8 Another example converter architecture (which may also be referred to as TF 2) for CSI compression and decompression according to some example embodiments of the present disclosure is shown. In encoder 800, linear embedding is performed at block 801. Multi-head attention block 810 is repeated 3 times. Specifically, at block 802, multi-head attention is applied. At block 803, addition and batch normalization are performed. Block 804 performs feedforward, including an FC layer, an LReLu layer, and another FC layer. At block 805, addition and batch normalization are performed. Block 806 includes an FC layer and a normalization layer. Quantization is performed at block 807.

[0110] Therefore, in the decoder 850, block 851 includes a FC layer and a normalization layer. Then, the multi-head attention block 860 is repeated 3 times. In block 860, an attention with 8 heads is applied at block 852. At block 853, addition and batch normalization are performed. At block 854, an attention with 8 heads is applied. At block 855, addition and batch normalization are performed. Block 856 performs feedforward. At block 857, addition and batch normalization are performed. At block 858, softmax is performed. Block 859 includes a FC layer and a normalization layer to generate an output as the reconstructed CSI.

[0111] It should be noted that Figure 7 and Figure 8 The structure of the model shown in is for illustrative purposes only and without any limitation. In addition, the size of data and operations is given as an example without any limitation.

[0112] Two independently trained converter models representing UEs are trained end-to-end on a 504k sample training set. A shared decoder representing a network device (5 conversion blocks with 16-head attention) is trained from scratch using code and channel values ​​{(q1(w),w),(q2(w),w)} assumed to be shared by each UE. The training is repeated by adding binary indicators to the latent vector {(q1(w)0,w),(q2(w)1,w)}. It should be understood that the number of samples in the combined set can be maintained as the number of samples in a single training set so that the number of samples is fixed in all experiments. In this case, the number of samples in the combined set is maintained at 504k instead of 2*504, and there are 252k samples for (q1(w)0,w) and 252k samples for (q2(w)1,w). From Table 1, it can be seen that when the encoder from the UE-side model is run together with a common decoder, the square of the cosine similarity is only slightly reduced for both the indicator and no indicator experiments. Table 1: Quasi-cosine similarity results on the test set for the 2UE model

[0113] The experiment was repeated by adding a convolutional neural network (CNN) based architecture. There are different ways to include the classification input for the remaining input of real value. In the experiment, the one-hot encoded element {[1,0,0], [0,1,0], [0,0,1]} was added to the potential vector for use as input. From Table 2, it can be seen that by adding more UE models, the performance of the shared decoder without indicators deteriorates. Table 2: Quasi-cosine similarity results for the test set of three UE models

[0114] The network is represented using 2 independently trained transformer models. A shared encoder (5 transformer blocks with 16-head attention) represents the UE model on the combined two datasets {w,(q1(w)),(w,q2(w))}. The channel of the first eigenvector of W has 52x32 complex values. As a preprocessing step, the real and imaginary parts are formed into a (52x64) dimensional matrix. The binary indicator matrix is ​​repeatedly added on the second dimension (making input 52x65). Similarly, it should be understood that the number of samples in the combined set can be maintained as the number of samples in the single training set so that the number of samples is fixed in all experiments. Table 3: Quasi-cosine similarity results for the test set for the first training of the network

[0115] In Table 3, it is observed that without the indicator, the common encoder cannot achieve good performance on both decoders. However, with the indicator, the performance is close to the base performance. Example methods, devices, and apparatus

[0116] Fig. 9 A flowchart of an example method 900 implemented at a terminal device according to some example embodiments of the present disclosure is shown. For the purpose of discussion, Figure 1 Method 900 is described from the perspective of the terminal device 110.

[0117] At block 910 , the terminal device 110 generates channel state information for a channel between the terminal device 110 and a network device.

[0118] At block 920, the terminal device 110 generates compressed channel state information based on the channel state information by using the compression model; and

[0119] At block 930, the terminal device 110 sends the compressed channel state information and an identification of the compression model to the network device.

[0120] In some example embodiments, the decompression model used to reconstruct the channel state information is common to a plurality of compression models including the compression model.

[0121] In some example embodiments, an end-to-end model including a compression model and another decompression model is trained using a training dataset, and the decompression model is trained using a combination of: an indicator corresponding to the compression model; and at least a portion of the training dataset.

[0122] In some example embodiments, method 900 further includes: quantizing the compressed channel state information into compressed bits; and sending the compressed bits and an identification of the compression model to a network device.

[0123] In some example embodiments, the identification of the compression model is based on at least one of: a type of compression model, or a type of device.

[0124] Fig.10 A flowchart of an example method 1000 implemented at a network device according to some example embodiments of the present disclosure is shown. For the purpose of discussion, Figure 1 Method 1000 is described from the perspective of network device 120.

[0125] At block 1010 , the network device 120 receives, from a terminal device, compressed channel state information for a channel between the terminal device and the network device 120 , and an identification of a compression model used to generate the compressed channel state information.

[0126] At block 1020 , network device 120 determines an indicator corresponding to the compression model based on the identification.

[0127] At block 1030 , the network device 120 reconstructs channel state information of the channel based on the compressed channel state information and the indicator by using the decompression model.

[0128] In some example embodiments, the decompression model is common to a plurality of compression models including the compression model.

[0129] In some example embodiments, the decompression model is trained using a combination of: an indicator corresponding to each compression model of a plurality of compression models, and at least a portion of a training data set used to train the corresponding compression model.

[0130] In some example embodiments, method 1000 further includes: receiving compressed bits representing compressed channel state information from a terminal device; dequantizing the compressed bits into a set of values; and reconstructing the channel state information based on the set of values ​​and the indicator by using a decompression model.

[0131] In some example embodiments, the identification of the compression model is based on at least one of: a type of compression model, or a type of terminal device.

[0132] Fig.11 A flowchart of an example method 1100 implemented at a terminal device according to some example embodiments of the present disclosure is shown. For the purpose of discussion, Figure 1 Method 1100 is described from the perspective of terminal device 110.

[0133] At block 1110 , the terminal device 110 generates channel state information for a channel between the device and a network device.

[0134] At block 1120 , the terminal device 110 generates compressed channel state information based on the channel state information and an indicator corresponding to the decompression model for reconstructing the channel state information by using the compression model.

[0135] At block 1130, the terminal device 110 sends the compressed channel state information to the network device.

[0136] In some example embodiments, the compression model is common to a plurality of decompression models including the decompression model.

[0137] In some example embodiments, the compression model is trained by a combination of: an indicator corresponding to each decompression model of a plurality of decompression models, and at least a portion of the training data used to train the corresponding decompression model.

[0138] In some example embodiments, the method 1100 further includes: quantizing the compressed channel state information into compressed bits; and transmitting the compressed bits to the network device.

[0139] In some exemplary embodiments, the indicator corresponding to the decompression model is based on: the type of decompression model, or the type of network device.

[0140] Fig.12 A flowchart of an example method 1200 implemented at a network device according to some example embodiments of the present disclosure is shown. For the purpose of discussion, Figure 1 Method 1200 is described from the perspective of network device 120.

[0141] At block 1210 , the network device 120 receives, from a terminal device, compressed channel state information of a channel between the terminal device and the apparatus.

[0142] At block 1220, the network device 120 reconstructs channel state information of the channel based on the compressed channel state information by using the decompression model. The compressed channel state information is generated based on the channel state information and an indicator corresponding to the decompression model.

[0143] In some example embodiments, the compression model used to generate the compressed channel state information is common to a plurality of decompression models including the decompression model.

[0144] In some example embodiments, an end-to-end model including a decompression model and another compression model is trained using a training dataset, and the compression model is trained using a combination of: an indicator corresponding to the decompression model, and at least a portion of the training dataset.

[0145] In some example embodiments, method 1200 further includes: receiving compressed bits representing compressed channel state information from a terminal device; dequantizing the compressed bits into a set of values; and reconstructing the channel state information based on the set of values ​​by using a decompression model.

[0146] In some exemplary embodiments, the indicator corresponding to the decompression model is based on: the type of decompression model, or the type of network device.

[0147] In some example embodiments, the method 900 can be performed (e.g., Figure 1 The apparatus of any one of the terminal devices 110 in the embodiment of the present invention may include a component for performing the corresponding operation of the method 900. The component may be implemented in any suitable form. For example, the component may be implemented in a circuit or a software module. The apparatus may be implemented as or included in Figure 1 In the terminal device 110.

[0148] In some example embodiments, the apparatus includes: a component for generating channel state information for a channel between the apparatus and a network device; a component for generating compressed channel state information based on the channel state information by using a compression model; and a component for sending the compressed channel state information and an identification of the compression model to the network device.

[0149] In some example embodiments, the decompression model used to reconstruct the channel state information is common to a plurality of compression models including the compression model.

[0150] In some example embodiments, an end-to-end model including a compression model and another decompression model is trained using a training dataset, and the decompression model is trained using a combination of: an indicator corresponding to the compression model; and at least a portion of the training dataset.

[0151] In some example embodiments, the apparatus further comprises: means for quantizing the compressed channel state information into compressed bits; and means for sending the compressed bits and an identification of the compression model to the network device.

[0152] In some example embodiments, the identification of the compression model is based on at least one of: a type of compression model, or a type of device.

[0153] In some example embodiments, the apparatus further comprises means for performing other operations in some example embodiments of the method 900 or the terminal device 110. In some example embodiments, the means comprises at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause performance of the apparatus.

[0154] In some example embodiments, method 1000 can be performed (e.g., Figure 1 The apparatus of any method in the network device 120 in the embodiment of the present invention may include a component for performing the corresponding operation of the method 1000. The component may be implemented in any suitable form. For example, the component may be implemented in a circuit or a software module. The apparatus may be implemented as or included in Figure 1 In the network device 120.

[0155] In some example embodiments, the apparatus includes: a component for receiving compressed channel state information for a channel between a terminal device and the apparatus from a terminal device, and an identifier of a compression model for generating the compressed channel state information; a component for determining an indicator corresponding to the compression model based on the identifier; and a component for reconstructing the channel state information of the channel based on the compressed channel state information and the indicator by using a decompression model.

[0156] In some example embodiments, the decompression model is common to a plurality of compression models including the compression model.

[0157] In some example embodiments, the decompression model is trained using a combination of: an indicator corresponding to each compression model of a plurality of compression models, and at least a portion of a training data set used to train the corresponding compression model.

[0158] In some example embodiments, the apparatus further includes: means for receiving compressed bits representing compressed channel state information from a terminal device; means for dequantizing the compressed bits into a set of values; and means for reconstructing the channel state information based on the set of values ​​and the indicator by using a decompression model.

[0159] In some example embodiments, the identification of the compression model is based on at least one of: a type of compression model, or a type of terminal device.

[0160] In some example embodiments, the apparatus further comprises means for performing other operations in some example embodiments of method 1000 or network device 120. In some example embodiments, the means comprises at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause performance of the apparatus.

[0161] In some example embodiments, the method 1100 can be performed (e.g., Figure 1 The apparatus of any one of the terminal devices 110 in the embodiment of the present invention may include a component for performing the corresponding operation of the method 1100. The component may be implemented in any suitable form. For example, the component may be implemented in a circuit or a software module. The apparatus may be implemented as or included in Figure 1 In the terminal device 110.

[0162] In some example embodiments, the apparatus includes: a component for generating channel state information for a channel between the apparatus and a network device; a component for generating compressed channel state information based on the channel state information and an indicator corresponding to a decompression model for reconstructing the channel state information by using a compression model; and a component for sending the compressed channel state information to the network device.

[0163] In some example embodiments, the compression model is common to a plurality of decompression models including the decompression model.

[0164] In some example embodiments, the compression model is trained by a combination of: an indicator corresponding to each decompression model of a plurality of decompression models, and at least a portion of the training data used to train the corresponding decompression model.

[0165] In some example embodiments, the apparatus further comprises: means for quantizing the compressed channel state information into compressed bits; and means for transmitting the compressed bits to the network device.

[0166] In some exemplary embodiments, the indicator corresponding to the decompression model is based on: the type of decompression model, or the type of network device.

[0167] In some example embodiments, the apparatus further comprises means for performing other operations in some example embodiments of the method 1100 or the terminal device 110. In some example embodiments, the means comprises at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause performance of the apparatus.

[0168] In some example embodiments, the method 1200 can be performed (e.g., Figure 1 The apparatus of any method in the network device 120 in the embodiment of the present invention may include a component for performing the corresponding operation of the method 1200. The component may be implemented in any suitable form. For example, the component may be implemented in a circuit or a software module. The apparatus may be implemented as or included in Figure 1 In the network device 120.

[0169] In some example embodiments, the apparatus includes: a component for receiving compressed channel state information for a channel between the terminal device and the apparatus from a terminal device; and a component for reconstructing the channel state information of the channel based on the compressed channel state information by using a decompression model, and wherein the compressed channel state information is generated based on the channel state information and an indicator corresponding to the decompression model.

[0170] In some example embodiments, the compression model used to generate the compressed channel state information is common to a plurality of decompression models including the decompression model.

[0171] In some example embodiments, an end-to-end model including a decompression model and another compression model is trained using a training dataset, and the compression model is trained using a combination of: an indicator corresponding to the decompression model, and at least a portion of the training dataset.

[0172] In some example embodiments, the apparatus further includes: a component for receiving compressed bits representing compressed channel state information from a terminal device; a component for dequantizing the compressed bits into a set of values; and a component for reconstructing the channel state information based on the set of values ​​by using a decompression model.

[0173] In some exemplary embodiments, the indicator corresponding to the decompression model is based on: the type of decompression model, or the type of network device.

[0174] In some example embodiments, the apparatus further comprises means for performing other operations in some example embodiments of method 1200 or network device 120. In some example embodiments, the means comprises at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause performance of the apparatus.

[0175] Fig.13 1 is a simplified block diagram of a device 1300 suitable for implementing an example embodiment of the present disclosure. The device 1300 may be used to implement a communication device, such as Figure 1 The terminal device 110 or the network device 120 is shown. As shown in the figure, 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.

[0176] The communication module 1340 is used for two-way communication. The communication module 1340 has one or more communication interfaces to facilitate communication with one or more other modules or devices. The communication interface may represent any interface necessary for communicating with other network elements. In some example embodiments, the communication module 1340 may include at least one antenna.

[0177] As non-limiting examples, processor 1310 may be of any type suitable for a local technology network, and may include one or more of the following: a general purpose computer, a special purpose computer, a microprocessor, a digital signal processor (DSP), and a processor based on a multi-core processor architecture. Device 1300 may have multiple processors, such as application specific integrated circuit chips, which are time-slaved to a clock that synchronizes a master processor.

[0178] The memory 1320 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) 1324, electrically programmable read-only memory (EPROM), flash memory, hard disk, compact disk (CD), digital video disk (DVD), optical disk, laser disk, and other magnetic storage and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 1322 and other volatile memories that do not persist during a power outage.

[0179] The computer program 1330 includes computer executable instructions executed by the associated processor 1310. The instructions of the program 1330 may include instructions for performing the operations / actions of some example embodiments of the present disclosure. The program 1330 may be stored in a memory, such as ROM 1324. The processor 1310 may perform any suitable actions and processes by loading the program 1330 into the RAM 1322.

[0180] The exemplary embodiments of the present disclosure may be implemented by means of a program 1330, so that the device 1300 may execute the Figures 9 to 12 Any process of the present disclosure discussed. The exemplary embodiments of the present disclosure may also be implemented by hardware or a combination of software and hardware.

[0181] In some example embodiments, program 1330 may be tangibly embodied in a computer-readable medium that may be included in device 1300 (such as in memory 1320) or other storage device accessible by device 1300. Device 1300 may load program 1330 from the computer-readable medium to RAM 1322 for execution. In some example embodiments, the computer-readable medium may include any type of non-transitory storage medium, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. As used herein, the term "non-transitory" is a limitation of the medium itself (i.e., tangible rather than a signal), not a limitation on data storage persistence (e.g., RAM versus ROM).

[0182] Fig.14An example of a computer readable medium 1400 is shown which may be in the form of a CD, DVD, or other optical storage disk. The computer readable medium 1400 has a program 1330 stored thereon.

[0183] Generally, various embodiments of the present disclosure may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects may be implemented in hardware, and other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of the present disclosure are shown and described as block diagrams, flow charts, or using some other graphical representations, it should be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general hardware or controllers or other computing devices, or some combination thereof in non-limiting examples.

[0184] Some example embodiments of the present disclosure also provide at least one computer program product tangibly stored on a computer-readable medium such as a non-transitory computer-readable medium. The computer program product includes computer-executable instructions, such as those included in a program module, executed in a device on a target physical or virtual processor to perform any of the methods above. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The functions of the program modules can be combined or split between program modules as needed in various embodiments. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0185] The program code for executing the method of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, as an independent software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0186] In the context of the present disclosure, computer program codes or related data may be carried by any suitable carrier to enable a device, apparatus or processor to perform various processes and operations as described above. Examples of carriers include signals, computer readable media, etc.

[0187] The computer readable medium may be a computer readable signal medium or a computer readable storage medium. The computer readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device or apparatus, or any suitable combination of the foregoing. More specific examples of computer readable storage media would include an electrical connection with one or more wires, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0188] In addition, although operations are depicted in a particular order, this should not be understood as requiring such operations to be performed in the particular order shown or in a sequential order, or performing all of the operations shown to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these details should not be interpreted as limitations on the scope of the present disclosure, but should be interpreted as descriptions of features that may be specific to a particular embodiment. Unless explicitly stated, certain features described in the context of a separate embodiment may also be implemented in combination in a single embodiment. On the contrary, unless explicitly stated, the various features described in the context of a single embodiment may also be implemented in multiple embodiments individually or in any suitable sub-combination.

[0189] Although the disclosure has been described in language specific to structural features and / or methodological acts, it should be understood that the 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 device comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: generating channel state information for a channel between the apparatus and a network device; generating compressed channel state information based on the channel state information by using a compression model; as well as The compressed channel state information and an identifier of the compression model are sent to a network device.

2. The apparatus according to claim 1, wherein a decompression model used to reconstruct the channel state information is common to a plurality of compression models including the compression model.

3. The apparatus of claim 2, wherein the end-to-end model comprising the compression model and another decompression model is trained using a training dataset, and the decompression model is trained using a combination of: an indicator corresponding to the compression model; and at least a portion of the training data set.

4. The apparatus of claim 1 , wherein the instructions, when executed by the at least one processor, further cause the apparatus to: quantizing the compressed channel state information into compressed bits; and The compressed bits and the identification of the compression model are sent to the network device.

5. The apparatus of claim 1 , wherein the identification of the compression model is based on at least one of: the type of compression model, or The type of device.

6. A device comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: generating channel state information for a channel between the apparatus and a network device; generating compressed channel state information based on the channel state information and an indicator corresponding to a decompression model for reconstructing the channel state information by using a compression model; The compressed channel state information is sent to the network device.

7. The apparatus according to claim 6, wherein the compression model is common to a plurality of decompression models including the decompression model.

8. The apparatus of claim 7, wherein the compression model is trained by a combination of: an indicator corresponding to each of the plurality of decompression models, and At least a portion of the training data used to train the corresponding decompression model.

9. The apparatus of claim 6, wherein the instructions, when executed by the at least one processor, further cause the apparatus to: quantizing the compressed channel state information into compressed bits; and The compressed bits are sent to the network device.

10. The apparatus of claim 6, wherein the indicator corresponding to the decompression model is based on: the type of the decompression model, or The type of the network device.

Images