Channel state information compression using neural networks
By performing intermediate compression operations between neural network blocks, the balance between complexity and performance is improved, enhancing communication quality in wireless systems.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-10-03
- Publication Date
- 2026-04-09
AI Technical Summary
Neural networks introduce additional complexity and reduce communication quality in wireless communications systems by failing to balance computational complexity and performance in CSI compression, leading to increased processing overhead.
Implement intermediate compression operations between neural network blocks, including reshaping and linear operations, to progressively compress CSI data, reducing the output sequence and improving communication quality.
This approach enhances performance and reduces processing complexity while maintaining communication quality by distributing compression operations across neural network blocks.
Smart Images

Figure CN2024123183_09042026_PF_FP_ABST
Abstract
Description
CHANNEL STATE INFORMATION COMPRESSION USING NEURAL NETWORKS
[0001] FIELD OF TECHNOLOGY
[0002] The following relates to wireless communications, including channel state information (CSI) compression using neural networks.BACKGROUND
[0003] Wireless communications systems are widely deployed to provide various types of communication content such as voice, video, packet data, messaging, broadcast, and so on. These systems may be capable of supporting communication with multiple users by sharing the available system resources (e.g., time, frequency, and power) . Examples of such multiple-access systems include fourth generation (4G) systems such as Long Term Evolution (LTE) systems, LTE-Advanced (LTE-A) systems, or LTE-A Pro systems, and fifth generation (5G) systems which may be referred to as New Radio (NR) systems. These systems may employ technologies such as code division multiple access (CDMA) , time division multiple access (TDMA) , frequency division multiple access (FDMA) , orthogonal FDMA (OFDMA) , or discrete Fourier transform spread orthogonal frequency division multiplexing (DFT-S-OFDM) . A wireless multiple-access communications system may include one or more base stations, each supporting wireless communication for communication devices, which may be known as user equipment (UE) .SUMMARY
[0004] The systems, methods, and devices of this disclosure each have several innovative aspects, no single one of which is solely responsible for the desirable attributes disclosed herein.
[0005] Some devices (e.g., a user equipment (UE) , a network entity) of a wireless communications system may utilize neural networks (e.g., a trained artificial intelligence (AI) model or machine learning (ML) model) to improve performance of some operations. For example, a UE may improve operations associated with compressing information prior to transmitting the information to other devices. In some examples, the UE may use a neural network to compress channel state information (CSI) and may transmit the CSI to a network entity after the compression. However, neural networks may introduce additional complexity to the UE and the network entity. Moreover, some neural network techniques (e.g., for data compression) may not achieve a sufficient balance between the computational complexity and the performance of the neural network, thus negating some benefits of using the neural network, increasing processing overhead, and reducing communication quality in the wireless communications system.
[0006] In accordance with one or more aspects described herein, a UE may support one or more techniques (e.g., CSI compression techniques) that improve the balance between the complexity of a neural network and the performance of the neural network. For example, a UE may be configured to perform one or more intermediate compression operations on a set of data after one or more blocks (e.g., transformer blocks) of the neural network. That is, a neural network may include multiple neural network blocks, and various compression blocks may be distributed between each of the neural network blocks. In some examples, a compression block may include various operations such as one or more reshaping operations, one or more linear operations, one or more appending operations to compress an input sequence. Each compression block may reduce an output sequence of a first neural network block and provide the reduced sequence as an input to a subsequent neural network block. As such, an input sequence (e.g., of CSI data) may be progressively compressed along a processing chain that uses the neural network, which may support enhanced performance, reduced processing complexity, and improved communication quality in a wireless communications system.
[0007] A method for wireless communications by a UE is described. The method may include obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI, performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks, performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks, and transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0008] A UE for wireless communications is described. The UE may include one or more memories storing processor executable code, and one or more processors coupled with the one or more memories. The one or more processors may individually or collectively be operable to execute the code to cause the UE to obtain, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI, perform a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks, perform a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks, and transmit a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0009] Another UE for wireless communications is described. The UE may include means for obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI, means for performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks, means for performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks, and means for transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0010] A non-transitory computer-readable medium storing code for wireless communications is described. The code may include instructions executable by one or more processors to obtain, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI, perform a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks, perform a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks, and transmit a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0011] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for obtaining the CSI prior to obtaining the first sequence, where the CSI includes a set of data that may be associated with a first dimension and performing one or more operations to restructure the set of data, where the set of data may be associated with a second dimension that may be compatible with the neural network after performing the one or more operations.
[0012] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for performing, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, where the first sequence may be associated with a first quantity of dimensions, and where the set of data may be associated with a second quantity of dimensions that may be less than the first quantity of dimensions after performing the one or more operations.
[0013] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for performing a third procedure using the neural network to obtain a fourth sequence including a fourth quantity of elements that may be less than the third quantity of elements, the third procedure including applying one or more third neural network blocks to the third sequence and compressing a third output of the one or more third neural network blocks, where the compressed CSI may be further associated with the fourth sequence.
[0014] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the one or more first neural network blocks may be applied to the first sequence in accordance with a first quantity of repetitions; and the one or more second neural network blocks may be applied to the second sequence in accordance with a second quantity of repetitions.
[0015] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the second quantity of repetitions may be equal to the first quantity of repetitions.
[0016] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the second quantity of repetitions may be different than the first quantity of repetitions.
[0017] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for compressing the first output includes performing a first reshaping operation on the first output that decreases a size of the first output in a first dimension and increases a size of the first output in a second dimension; and performing, after performing the first reshaping operation, a first linear operation that decreases the size of the first output in the second dimension by combining one or more first elements of the first output with one or more second elements of the first output and compressing the second output includes: performing a second reshaping operation on the second output that decreases a size of the second output in a third dimension and increases a size of the second output in a fourth dimension; and performing, after performing the second reshaping operation, a second linear operation that decreases the size of the second output in the fourth dimension by combining one or more third elements of the second output with one or more fourth elements of the second output.
[0018] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, performing the first reshaping operation may include operations, features, means, or instructions for appending, prior to performing the first linear operation, a first element to the first output based on the size of the first output in the second dimension including an odd quantity of elements.
[0019] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the first element may be associated with a parameter that may be learnable by the neural network.
[0020] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for receiving, from the network entity, one or more control messages including one or more first parameters that configure the UE to obtain the CSI and one or more second parameters that configure the UE to transmit the message, where obtaining the first sequence may be in response to receiving the one or more control messages.
[0021] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for receiving, from the network entity, an indication of one or more first parameters associated with a reference neural network, where one or more second parameters of the neural network at the UE may be based on the one or more first parameters, and where the CSI may be compressed in accordance with the one or more second parameters.
[0022] Some examples of the method, UEs, and non-transitory computer-readable medium described herein may further include operations, features, means, or instructions for receiving, from the network entity, an indication of one or more parameters for the neural network at the UE, where the CSI may be compressed in accordance with the one or more parameters.
[0023] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the one or more first neural network blocks and the one or more second neural network blocks may be in accordance with a transformer neural network structure.
[0024] In some examples of the method, UEs, and non-transitory computer-readable medium described herein, the first procedure at the one or more first neural network blocks and the second procedure at the one or more second neural network blocks include one or more identical operations.
[0025] Details of one or more implementations of the subject matter described in this disclosure are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages will become apparent from the description, the drawings, and the claims. Note that the relative dimensions of the following figures may not be drawn to scale.BRIEF DESCRIPTION OF THE DRAWINGS
[0026] FIG. 1 shows an example of a wireless communications system that supports channel state information (CSI) compression using neural networks in accordance with one or more aspects of the present disclosure.
[0027] FIG. 2 shows an example of an artificial neural network (ANN) that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0028] FIG. 3 shows an example of a wireless communications system that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0029] FIGs. 4 and 5 show examples of neural network architectures that support CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0030] FIG. 6 shows an example of a process flow that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0031] FIGs. 7 and 8 show block diagrams of devices that support CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0032] FIG. 9 shows a block diagram of a communications manager that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0033] FIG. 10 shows a diagram of a system including a device that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0034] FIGs. 11 through 13 show flowcharts illustrating methods that support CSI compression using neural networks in accordance with one or more aspects of the present disclosure.DETAILED DESCRIPTION
[0035] Some devices (e.g., a user equipment (UE) , a network entity) of a wireless communications system may utilize neural networks (e.g., a trained artificial intelligence (AI) model or machine learning (ML) model) to improve performance of various operations. For example, a UE may improve operations associated with compressing information, such as channel state information (CSI) (e.g., CSI feedback, channel state feedback (CFE) , a selected precoding matrix, or other data) prior to transmitting the information to other devices (e.g., over-the-air to a network entity) . The network entity may receive the compressed CSI and may also use a neural network to reconstruct the original CSI obtained by the UE. However, neural networks may introduce additional processing complexity by increasing processing operations at the UE or the network entity. The increased processing overhead may be proportional to a quantity of sequence elements (e.g., a quantity of input tokens, input vectors) that are input to the neural network. Although reducing the quantity of input sequence elements may reduce the complexity, the reduction of an input sequence reduction may also reduce performance of the neural network. Moreover, some neural network techniques (e.g., for data compression) may not achieve a sufficient balance between the computational complexity and the performance of the neural network, thus negating some benefits of using the neural network, increasing processing overhead, and reducing communication quality in the wireless communications system.
[0036] In accordance with one or more aspects of the present disclosure, a UE may support one or more techniques (e.g., CSI compression techniques) that improve a balance between the processing complexity and the performance of the neural network. For example, a UE may be configured to perform one or more intermediate (e.g., incremental) compression operations on a set of data (e.g., corresponding to CSI) after one or more blocks (e.g., transformer blocks) of the neural network (e.g., as opposed to compressing an entire sequence before applying the neural network blocks) . That is, a neural network may include multiple neural network blocks, and various compression blocks may be distributed (e.g., interspersed) between each of the neural network blocks. In some examples, a compression block may include various operations such as one or more reshaping operations, one or more linear layer operations, and other operations to compress an input sequence. Each compression block may reduce an output sequence of a first neural network block and provide the reduced sequence as an input to a subsequent neural network block. As such, an input sequence (e.g., of CSI data) may be progressively compressed along a processing chain, which may support enhanced performance, reduced processing complexity, and improved communication quality in a wireless communications system.
[0037] Certain aspects and techniques as described herein may be implemented, at least in part, using an AI program, such as a program that includes a ML or artificial neural network (ANN) model. An example ML model may include mathematical representations or define computing capabilities for making inferences from input data based on patterns or relationships identified in the input data. As used herein, the term “inferences” can include one or more of decisions, predictions, determinations, or values, which may represent outputs of the ML model. The computing capabilities may be defined in terms of certain parameters of the ML model, such as weights and biases. Weights may indicate relationships between certain input data and certain outputs of the ML model, and biases are offsets which may indicate a starting point for outputs of the ML model. An example ML model operating on input data may start at an initial output based on the biases and then update its output based on a combination of the input data and the weights.
[0038] In some aspects, an ML model may be configured to provide computing capabilities for wireless communications. Such an ML model may be configured with weights and biases to perform enhance data compression techniques that support improving accuracy and efficiency of reporting channel conditions. Thus, during operation of a device, the ML model may receive input data (such as channel quality measurements, a precoder matrix, rank information) and make inferences (such as a compressed CSI report) based on the weights and biases.
[0039] ML models may be deployed in one or more devices (e.g., network entities and UEs) and may be configured to enhance various aspects of a wireless communication system. For example, an ML model may be trained to identify patterns or relationships in data corresponding to a network, a device, an air interface, or the like. An ML model may support operational decisions relating to one or more aspects associated with wireless communications devices, networks, or services. For example, an ML model may be utilized for supporting or improving aspects such as signal coding / decoding, network routing, energy conservation, transceiver circuitry controls, frequency synchronization, timing synchronization channel state estimation, channel equalization, channel state feedback, modulation, demodulation, device positioning, beamforming, load balancing, operations and management functions, security, etc.
[0040] ML models may be characterized in terms of types of learning that generate specific types of learned models that perform specific types of tasks. For example, different types of machine learning include supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, etc. ML models may be used to perform different tasks such as classification or regression, where classification refers to determining one or more discrete output values from a set of predefined output values, and regression refers to determining continuous values which are not bounded by predefined output values. Some example ML models configured for performing such tasks include ANNs such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) , transformers, diffusion models, regression analysis models (such as statistical models) , large language models (LLMs) , decision tree learning (such as predictive models) , support vector networks (SVMs) , and probabilistic graphical models (such as a Bayesian network) , etc.
[0041] The description herein illustrates, by way of some examples, how one or more tasks or problems in wireless communications may benefit from the application of one or more ML models (e.g., such as by improved compression of channel state feedback) . To facilitate the discussion, an ML model configured using an ANN is used, but it should be understood, that other types of ML models may be used instead of an ANN. Hence, unless expressly recited, subject matter regarding an ML model is not necessarily intended to be limited to an ANN solution. Further, it should be understood that, unless otherwise specifically stated, terms such “AI / ML model, ” “ML model, ” “trained ML model, ” “ANN, ” “model, ” “algorithm, ” or the like are intended to be interchangeable.
[0042] Aspects of the disclosure are initially described in the context of wireless communications systems. Aspects of the disclosure are further illustrated by and described with reference to ANNs, neural network architectures, process flows, apparatus diagrams, system diagrams, and flowcharts that relate to CSI compression using neural networks.
[0043] FIG. 1 shows an example of a wireless communications system 100 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The wireless communications system 100 may include one or more devices, such as one or more network devices (e.g., network entities 105) , one or more UEs 115, and a core network 130. In some examples, the wireless communications system 100 may be a Long Term Evolution (LTE) network, an LTE-Advanced (LTE-A) network, an LTE-A Pro network, a New Radio (NR) network, or a network operating in accordance with other systems and radio technologies, including future systems and radio technologies not explicitly mentioned herein.
[0044] The network entities 105 may be dispersed throughout a geographic area to form the wireless communications system 100 and may include devices in different forms or having different capabilities. In various examples, a network entity 105 may be referred to as a network element, a mobility element, a radio access network (RAN) node, or network equipment, among other nomenclature. In some examples, network entities 105 and UEs 115 may wirelessly communicate via communication link (s) 125 (e.g., a radio frequency (RF) access link) . For example, a network entity 105 may support a coverage area 110 (e.g., a geographic coverage area) over which the UEs 115 and the network entity 105 may establish the communication link (s) 125. The coverage area 110 may be an example of a geographic area over which a network entity 105 and a UE 115 may support the communication of signals according to one or more radio access technologies (RATs) .
[0045] The UEs 115 may be dispersed throughout a coverage area 110 of the wireless communications system 100, and each UE 115 may be stationary, or mobile, or both at different times. The UEs 115 may be devices in different forms or having different capabilities. Some example UEs 115 are illustrated in FIG. 1. The UEs 115 described herein may be capable of supporting communications with various types of devices in the wireless communications system 100 (e.g., other wireless communication devices, including UEs 115 or network entities 105) , as shown in FIG. 1.
[0046] As described herein, a node of the wireless communications system 100, which may be referred to as a network node, or a wireless node, may be a network entity 105 (e.g., any network entity described herein) , a UE 115 (e.g., any UE described herein) , a network controller, an apparatus, a device, a computing system, one or more components, or another suitable processing entity configured to perform any of the techniques described herein. For example, a node may be a UE 115. As another example, a node may be a network entity 105. As another example, a first node may be configured to communicate with a second node or a third node. In one aspect of this example, the first node may be a UE 115, the second node may be a network entity 105, and the third node may be a UE 115. In another aspect of this example, the first node may be a UE 115, the second node may be a network entity 105, and the third node may be a network entity 105. In yet other aspects of this example, the first, second, and third nodes may be different relative to these examples. Similarly, reference to a UE 115, network entity 105, apparatus, device, computing system, or the like may include disclosure of the UE 115, network entity 105, apparatus, device, computing system, or the like being a node. For example, disclosure that a UE 115 is configured to receive information from a network entity 105 also discloses that a first node is configured to receive information from a second node.
[0047] In some examples, network entities 105 may communicate with a core network 130, or with one another, or both. For example, network entities 105 may communicate with the core network 130 via backhaul communication link (s) 120 (e.g., in accordance with an S1, N2, N3, or other interface protocol) . In some examples, network entities 105 may communicate with one another via backhaul communication link (s) 120 (e.g., in accordance with an X2, Xn, or other interface protocol) either directly (e.g., directly between network entities 105) or indirectly (e.g., via the core network 130) . In some examples, network entities 105 may communicate with one another via a midhaul communication link 162 (e.g., in accordance with a midhaul interface protocol) or a fronthaul communication link 168 (e.g., in accordance with a fronthaul interface protocol) , or any combination thereof. The backhaul communication link (s) 120, midhaul communication links 162, or fronthaul communication links 168 may be or include one or more wired links (e.g., an electrical link, an optical fiber link) or one or more wireless links (e.g., a radio link, a wireless optical link) , among other examples or various combinations thereof. A UE 115 may communicate with the core network 130 via a communication link 155.
[0048] One or more of the network entities 105 or network equipment described herein may include or may be referred to as a base station 140 (e.g., a base transceiver station, a radio base station, an NR base station, an access point, a radio transceiver, a NodeB, an eNodeB (eNB) , a next-generation NodeB or giga-NodeB (either of which may be referred to as a gNB) , a 5G NB, a next-generation eNB (ng-eNB) , a Home NodeB, a Home eNodeB, or other suitable terminology) . In some examples, a network entity 105 (e.g., a base station 140) may be implemented in an aggregated (e.g., monolithic, standalone) base station architecture, which may be configured to utilize a protocol stack that is physically or logically integrated within one network entity (e.g., a network entity 105 or a single RAN node, such as a base station 140) .
[0049] In some examples, a network entity 105 may be implemented in a disaggregated architecture (e.g., a disaggregated base station architecture, a disaggregated RAN architecture) , which may be configured to utilize a protocol stack that is physically or logically distributed among multiple network entities (e.g., network entities 105) , such as an integrated access and backhaul (IAB) network, an open RAN (O-RAN) (e.g., a network configuration sponsored by the O-RAN Alliance) , or a virtualized RAN (vRAN) (e.g., a cloud RAN (C-RAN) ) . For example, a network entity 105 may include one or more of a central unit (CU) , such as a CU 160, a distributed unit (DU) , such as a DU 165, a radio unit (RU) , such as an RU 170, a RAN Intelligent Controller (RIC) , such as an RIC 175 (e.g., a Near-Real Time RIC (Near-RT RIC) , a Non-Real Time RIC (Non-RT RIC) ) , a Service Management and Orchestration (SMO) system, such as an SMO system 180, or any combination thereof. An RU 170 may also be referred to as a radio head, a smart radio head, a remote radio head (RRH) , a remote radio unit (RRU) , or a transmission reception point (TRP) . One or more components of the network entities 105 in a disaggregated RAN architecture may be co-located, or one or more components of the network entities 105 may be located in distributed locations (e.g., separate physical locations) . In some examples, one or more of the network entities 105 of a disaggregated RAN architecture may be implemented as virtual units (e.g., a virtual CU (VCU) , a virtual DU (VDU) , a virtual RU (VRU) ) .
[0050] The split of functionality between a CU 160, a DU 165, and an RU 170 is flexible and may support different functionalities depending on which functions (e.g., network layer functions, protocol layer functions, baseband functions, RF functions, or any combinations thereof) are performed at a CU 160, a DU 165, or an RU 170. For example, a functional split of a protocol stack may be employed between a CU 160 and a DU 165 such that the CU 160 may support one or more layers of the protocol stack and the DU 165 may support one or more different layers of the protocol stack. In some examples, the CU 160 may host upper protocol layer (e.g., layer 3 (L3) , layer 2 (L2) ) functionality and signaling (e.g., Radio Resource Control (RRC) , service data adaptation protocol (SDAP) , Packet Data Convergence Protocol (PDCP) ) . The CU 160 (e.g., one or more CUs) may be connected to a DU 165 (e.g., one or more DUs) or an RU 170 (e.g., one or more RUs) , or some combination thereof, and the DUs 165, RUs 170, or both may host lower protocol layers, such as layer 1 (L1) (e.g., physical (PHY) layer) or L2 (e.g., radio link control (RLC) layer, medium access control (MAC) layer) functionality and signaling, and may each be at least partially controlled by the CU 160. Additionally, or alternatively, a functional split of the protocol stack may be employed between a DU 165 and an RU 170 such that the DU 165 may support one or more layers of the protocol stack and the RU 170 may support one or more different layers of the protocol stack. The DU 165 may support one or multiple different cells (e.g., via one or multiple different RUs, such as an RU 170) . In some cases, a functional split between a CU 160 and a DU 165 or between a DU 165 and an RU 170 may be within a protocol layer (e.g., some functions for a protocol layer may be performed by one of a CU 160, a DU 165, or an RU 170, while other functions of the protocol layer are performed by a different one of the CU 160, the DU 165, or the RU 170) . A CU 160 may be functionally split further into CU control plane (CU-CP) and CU user plane (CU-UP) functions. A CU 160 may be connected to a DU 165 via a midhaul communication link 162 (e.g., F1, F1-c, F1-u) , and a DU 165 may be connected to an RU 170 via a fronthaul communication link 168 (e.g., open fronthaul (FH) interface) . In some examples, a midhaul communication link 162 or a fronthaul communication link 168 may be implemented in accordance with an interface (e.g., a channel) between layers of a protocol stack supported by respective network entities (e.g., one or more of the network entities 105) that are in communication via such communication links.
[0051] In some wireless communications systems (e.g., the wireless communications system 100) , infrastructure and spectral resources for radio access may support wireless backhaul link capabilities to supplement wired backhaul connections, providing an IAB network architecture (e.g., to a core network 130) . In some cases, in an IAB network, one or more of the network entities 105 (e.g., network entities 105 or IAB node (s) 104) may be partially controlled by each other. The IAB node (s) 104 may be referred to as a donor entity or an IAB donor. A DU 165 or an RU 170 may be partially controlled by a CU 160 associated with a network entity 105 or base station 140 (such as a donor network entity or a donor base station) . The one or more donor entities (e.g., IAB donors) may be in communication with one or more additional devices (e.g., IAB node (s) 104) via supported access and backhaul links (e.g., backhaul communication link (s) 120) . IAB node (s) 104 may include an IAB mobile termination (IAB-MT) controlled (e.g., scheduled) by one or more DUs (e.g., DUs 165) of a coupled IAB donor. An IAB-MT may be equipped with an independent set of antennas for relay of communications with UEs 115 or may share the same antennas (e.g., of an RU 170) of IAB node (s) 104 used for access via the DU 165 of the IAB node (s) 104 (e.g., referred to as virtual IAB-MT (vIAB-MT) ) . In some examples, the IAB node (s) 104 may include one or more DUs (e.g., DUs 165) that support communication links with additional entities (e.g., IAB node (s) 104, UEs 115) within the relay chain or configuration of the access network (e.g., downstream) . In such cases, one or more components of the disaggregated RAN architecture (e.g., the IAB node (s) 104 or components of the IAB node (s) 104) may be configured to operate according to the techniques described herein.
[0052] For instance, an access network (AN) or RAN may include communications between access nodes (e.g., an IAB donor) , IAB node (s) 104, and one or more UEs 115. The IAB donor may facilitate connection between the core network 130 and the AN (e.g., via a wired or wireless connection to the core network 130) . That is, an IAB donor may refer to a RAN node with a wired or wireless connection to the core network 130. The IAB donor may include one or more of a CU 160, a DU 165, and an RU 170, in which case the CU 160 may communicate with the core network 130 via an interface (e.g., a backhaul link) . The IAB donor and IAB node (s) 104 may communicate via an F1 interface according to a protocol that defines signaling messages (e.g., an F1 AP protocol) . Additionally, or alternatively, the CU 160 may communicate with the core network 130 via an interface, which may be an example of a portion of a backhaul link, and may communicate with other CUs (e.g., including a CU 160 associated with an alternative IAB donor) via an Xn-C interface, which may be an example of another portion of a backhaul link.
[0053] IAB node (s) 104 may refer to RAN nodes that provide IAB functionality (e.g., access for UEs 115, wireless self-backhauling capabilities) . A DU 165 may act as a distributed scheduling node towards child nodes associated with the IAB node (s) 104, and the IAB-MT may act as a scheduled node towards parent nodes associated with IAB node (s) 104. That is, an IAB donor may be referred to as a parent node in communication with one or more child nodes (e.g., an IAB donor may relay transmissions for UEs through other IAB node (s) 104) . Additionally, or alternatively, IAB node (s) 104 may also be referred to as parent nodes or child nodes to other IAB node (s) 104, depending on the relay chain or configuration of the AN. The IAB-MT entity of IAB node (s) 104 may provide a Uu interface for a child IAB node (e.g., the IAB node (s) 104) to receive signaling from a parent IAB node (e.g., the IAB node (s) 104) , and a DU interface (e.g., a DU 165) may provide a Uu interface for a parent IAB node to signal to a child IAB node or UE 115.
[0054] For example, IAB node (s) 104 may be referred to as parent nodes that support communications for child IAB nodes, or may be referred to as child IAB nodes associated with IAB donors, or both. An IAB donor may include a CU 160 with a wired or wireless connection (e.g., backhaul communication link (s) 120) to the core network 130 and may act as a parent node to IAB node (s) 104. For example, the DU 165 of an IAB donor may relay transmissions to UEs 115 through IAB node (s) 104, or may directly signal transmissions to a UE 115, or both. The CU 160 of the IAB donor may signal communication link establishment via an F1 interface to IAB node (s) 104, and the IAB node (s) 104 may schedule transmissions (e.g., transmissions to the UEs 115 relayed from the IAB donor) through one or more DUs (e.g., DUs 165) . That is, data may be relayed to and from IAB node (s) 104 via signaling via an NR Uu interface to MT of IAB node (s) 104 (e.g., other IAB node (s) ) . Communications with IAB node (s) 104 may be scheduled by a DU 165 of the IAB donor or of IAB node (s) 104.
[0055] In the case of the techniques described herein applied in the context of a disaggregated RAN architecture, one or more components of the disaggregated RAN architecture may be configured to support CSI compression using neural networks as described herein. For example, some operations described as being performed by a UE 115 or a network entity 105 (e.g., a base station 140) may additionally, or alternatively, be performed by one or more components of the disaggregated RAN architecture (e.g., components such as an IAB node, a DU 165, a CU 160, an RU 170, an RIC 175, an SMO system 180) .
[0056] A UE 115 may include or may be referred to as a mobile device, a wireless device, a remote device, a handheld device, or a subscriber device, or some other suitable terminology, where the “device” may also be referred to as a unit, a station, a terminal, or a client, among other examples. A UE 115 may also include or may be referred to as a personal electronic device such as a cellular phone, a personal digital assistant (PDA) , a tablet computer, a laptop computer, or a personal computer. In some examples, a UE 115 may include or be referred to as a wireless local loop (WLL) station, an Internet of Things (IoT) device, an Internet of Everything (IoE) device, or a machine type communications (MTC) device, among other examples, which may be implemented in various objects such as appliances, vehicles, or meters, among other examples.
[0057] The UEs 115 described herein may be able to communicate with various types of devices, such as UEs 115 that may sometimes operate as relays, as well as the network entities 105 and the network equipment including macro eNBs or gNBs, small cell eNBs or gNBs, or relay base stations, among other examples, as shown in FIG. 1.
[0058] The UEs 115 and the network entities 105 may wirelessly communicate with one another via the communication link (s) 125 (e.g., one or more access links) using resources associated with one or more carriers. The term “carrier” may refer to a set of RF spectrum resources having a defined PHY layer structure for supporting the communication link (s) 125. For example, a carrier used for the communication link (s) 125 may include a portion of an RF spectrum band (e.g., a bandwidth part (BWP) ) that is operated according to one or more PHY layer channels for a given RAT (e.g., LTE, LTE-A, LTE-A Pro, NR) . Each PHY layer channel may carry acquisition signaling (e.g., synchronization signals, system information) , control signaling that coordinates operation for the carrier, user data, or other signaling. The wireless communications system 100 may support communication with a UE 115 using carrier aggregation or multi-carrier operation. A UE 115 may be configured with multiple downlink component carriers and one or more uplink component carriers according to a carrier aggregation configuration. Carrier aggregation may be used with both frequency division duplexing (FDD) and time division duplexing (TDD) component carriers. Communication between a network entity 105 and other devices may refer to communication between the devices and any portion (e.g., entity, sub-entity) of a network entity 105. For example, the terms “transmitting, ” “receiving, ” or “communicating, ” when referring to a network entity 105, may refer to any portion of a network entity 105 (e.g., a base station 140, a CU 160, a DU 165, a RU 170) of a RAN communicating with another device (e.g., directly or via one or more other network entities, such as one or more of the network entities 105) .
[0059] Signal waveforms transmitted via a carrier may be made up of multiple subcarriers (e.g., using multi-carrier modulation (MCM) techniques such as orthogonal frequency division multiplexing (OFDM) or discrete Fourier transform spread OFDM (DFT-S-OFDM) ) . In a system employing MCM techniques, a resource element may refer to resources of one symbol period (e.g., a duration of one modulation symbol) and one subcarrier, in which case the symbol period and subcarrier spacing may be inversely related. The quantity of bits carried by each resource element may depend on the modulation scheme (e.g., the order of the modulation scheme, the coding rate of the modulation scheme, or both) , such that a relatively higher quantity of resource elements (e.g., in a transmission duration) and a relatively higher order of a modulation scheme may correspond to a relatively higher rate of communication. A wireless communications resource may refer to a combination of an RF spectrum resource, a time resource, and a spatial resource (e.g., a spatial layer, a beam) , and the use of multiple spatial resources may increase the data rate or data integrity for communications with a UE 115.
[0060] The time intervals for the network entities 105 or the UEs 115 may be expressed in multiples of a basic time unit which may, for example, refer to a sampling period of Ts=1 / (Δfmax·Nff) seconds, for which Δfmax may represent a supported subcarrier spacing, and Nf may represent a supported discrete Fourier transform (DFT) size. Time intervals of a communications resource may be organized according to radio frames each having a specified duration (e.g., 10 milliseconds (ms) ) . Each radio frame may be identified by a system frame number (SFN) (e.g., ranging from 0 to 1023) .
[0061] Each frame may include multiple consecutively-numbered subframes or slots, and each subframe or slot may have the same duration. In some examples, a frame may be divided (e.g., in the time domain) into subframes, and each subframe may be further divided into a quantity of slots. Alternatively, each frame may include a variable quantity of slots, and the quantity of slots may depend on subcarrier spacing. Each slot may include a quantity of symbol periods (e.g., depending on the length of the cyclic prefix prepended to each symbol period) . In some wireless communications systems, such as the wireless communications system 100, a slot may further be divided into multiple mini-slots associated with one or more symbols. Excluding the cyclic prefix, each symbol period may be associated with one or more (e.g., Nf) sampling periods. The duration of a symbol period may depend on the subcarrier spacing or frequency band of operation.
[0062] A subframe, a slot, a mini-slot, or a symbol may be the smallest scheduling unit (e.g., in the time domain) of the wireless communications system 100 and may be referred to as a transmission time interval (TTI) . In some examples, the TTI duration (e.g., a quantity of symbol periods in a TTI) may be variable. Additionally, or alternatively, the smallest scheduling unit of the wireless communications system 100 may be dynamically selected (e.g., in bursts of shortened TTIs (sTTIs) ) .
[0063] Physical channels may be multiplexed for communication using a carrier according to various techniques. A physical control channel and a physical data channel may be multiplexed for signaling via a downlink carrier, for example, using one or more of time division multiplexing (TDM) techniques, frequency division multiplexing (FDM) techniques, or hybrid TDM-FDM techniques. A control region (e.g., a control resource set (CORESET) ) for a physical control channel may be defined by a set of symbol periods and may extend across the system bandwidth or a subset of the system bandwidth of the carrier. One or more control regions (e.g., CORESETs) may be configured for a set of the UEs 115. For example, one or more of the UEs 115 may monitor or search control regions for control information according to one or more search space sets, and each search space set may include one or multiple control channel candidates in one or more aggregation levels arranged in a cascaded manner. An aggregation level for a control channel candidate may refer to an amount of control channel resources (e.g., control channel elements (CCEs) ) associated with encoded information for a control information format having a given payload size. Search space sets may include common search space sets configured for sending control information to UEs 115 (e.g., one or more UEs) or may include UE-specific search space sets for sending control information to a UE 115 (e.g., a specific UE) .
[0064] A network entity 105 may provide communication coverage via one or more cells, for example a macro cell, a small cell, a hot spot, or other types of cells, or any combination thereof. The term “cell” may refer to a logical communication entity used for communication with a network entity 105 (e.g., using a carrier) and may be associated with an identifier for distinguishing neighboring cells (e.g., a physical cell identifier (PCID) , a virtual cell identifier (VCID) ) . In some examples, a cell also may refer to a coverage area 110 or a portion of a coverage area 110 (e.g., a sector) over which the logical communication entity operates. Such cells may range from smaller areas (e.g., a structure, a subset of structure) to larger areas depending on various factors such as the capabilities of the network entity 105. For example, a cell may be or include a building, a subset of a building, or exterior spaces between or overlapping with coverage areas 110, among other examples.
[0065] A macro cell generally covers a relatively large geographic area (e.g., several kilometers in radius) and may allow unrestricted access by the UEs 115 with service subscriptions with the network provider supporting the macro cell. A small cell may be associated with a network entity 105 operating with lower power (e.g., a base station 140 operating with lower power) relative to a macro cell, and a small cell may operate using the same or different (e.g., licensed, unlicensed) frequency bands as macro cells. Small cells may provide unrestricted access to the UEs 115 with service subscriptions with the network provider or may provide restricted access to the UEs 115 having an association with the small cell (e.g., the UEs 115 in a closed subscriber group (CSG) , the UEs 115 associated with users in a home or office) . A network entity 105 may support one or more cells and may also support communications via the one or more cells using one or multiple component carriers.
[0066] In some examples, a carrier may support multiple cells, and different cells may be configured according to different protocol types (e.g., MTC, narrowband IoT (NB-IoT) , enhanced mobile broadband (eMBB) ) that may provide access for different types of devices.
[0067] In some examples, a network entity 105 (e.g., a base station 140, an RU 170) may be movable and therefore provide communication coverage for a moving coverage area, such as the coverage area 110. In some examples, coverage areas 110 (e.g., different coverage areas) associated with different technologies may overlap, but the coverage areas 110 (e.g., different coverage areas) may be supported by the same network entity (e.g., a network entity 105) . In some other examples, overlapping coverage areas, such as a coverage area 110, associated with different technologies may be supported by different network entities (e.g., the network entities 105) . The wireless communications system 100 may include, for example, a heterogeneous network in which different types of the network entities 105 support communications for coverage areas 110 (e.g., different coverage areas) using the same or different RATs.
[0068] Some UEs 115, such as MTC or IoT devices, may be relatively low cost or low complexity devices and may provide for automated communication between machines (e.g., via Machine-to-Machine (M2M) communication) . M2M communication or MTC may refer to data communication technologies that allow devices to communicate with one another or a network entity 105 (e.g., a base station 140) without human intervention. In some examples, M2M communication or MTC may include communications from devices that integrate sensors or meters to measure or capture information and relay such information to a central server or application program that uses the information or presents the information to humans interacting with the application program. Some UEs 115 may be designed to collect information or enable automated behavior of machines or other devices. Examples of applications for MTC devices include smart metering, inventory monitoring, water level monitoring, equipment monitoring, healthcare monitoring, wildlife monitoring, weather and geological event monitoring, fleet management and tracking, remote security sensing, physical access control, and transaction-based business charging.
[0069] The wireless communications system 100 may be configured to support ultra-reliable communications or low-latency communications, or various combinations thereof. For example, the wireless communications system 100 may be configured to support ultra-reliable low-latency communications (URLLC) . The UEs 115 may be designed to support ultra-reliable, low-latency, or critical functions. Ultra-reliable communications may include private communication or group communication and may be supported by one or more services such as push-to-talk, video, or data. Support for ultra-reliable, low-latency functions may include prioritization of services, and such services may be used for public safety or general commercial applications. The terms ultra-reliable, low-latency, and ultra-reliable low-latency may be used interchangeably herein.
[0070] In some examples, a UE 115 may be configured to support communicating directly with other UEs (e.g., one or more of the UEs 115) via a device-to-device (D2D) communication link, such as a D2D communication link 135 (e.g., in accordance with a peer-to-peer (P2P) , D2D, or sidelink protocol) . In some examples, one or more UEs 115 of a group that are performing D2D communications may be within the coverage area 110 of a network entity 105 (e.g., a base station 140, an RU 170) , which may support aspects of such D2D communications being configured by (e.g., scheduled by) the network entity 105. In some examples, one or more UEs 115 of such a group may be outside the coverage area 110 of a network entity 105 or may be otherwise unable to or not configured to receive transmissions from a network entity 105. In some examples, groups of the UEs 115 communicating via D2D communications may support a one-to-many (1: M) system in which each UE 115 transmits to one or more of the UEs 115 in the group. In some examples, a network entity 105 may facilitate the scheduling of resources for D2D communications. In some other examples, D2D communications may be carried out between the UEs 115 without an involvement of a network entity 105.
[0071] In some systems, a D2D communication link 135 may be an example of a communication channel, such as a sidelink communication channel, between vehicles (e.g., UEs 115) . In some examples, vehicles may communicate using vehicle-to-everything (V2X) communications, vehicle-to-vehicle (V2V) communications, or some combination of these. A vehicle may signal information related to traffic conditions, signal scheduling, weather, safety, emergencies, or any other information relevant to a V2X system. In some examples, vehicles in a V2X system may communicate with roadside infrastructure, such as roadside units, or with the network via one or more network nodes (e.g., network entities 105, base stations 140, RUs 170) using vehicle-to-network (V2N) communications, or with both.
[0072] The core network 130 may provide user authentication, access authorization, tracking, Internet Protocol (IP) connectivity, and other access, routing, or mobility functions. The core network 130 may be an evolved packet core (EPC) or 5G core (5GC) , which may include at least one control plane entity that manages access and mobility (e.g., a mobility management entity (MME) , an access and mobility management function (AMF) ) and at least one user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW) , a Packet Data Network (PDN) gateway (P-GW) , or a user plane function (UPF) ) . The control plane entity may manage non-access stratum (NAS) functions such as mobility, authentication, and bearer management for the UEs 115 served by the network entities 105 (e.g., base stations 140) associated with the core network 130. User IP packets may be transferred through the user plane entity, which may provide IP address allocation as well as other functions. The user plane entity may be connected to IP services 150 for one or more network operators. The IP services 150 may include access to the Internet, Intranet (s) , an IP Multimedia Subsystem (IMS) , or a Packet-Switched Streaming Service.
[0073] The wireless communications system 100 may operate using one or more frequency bands, which may be in the range of 300 megahertz (MHz) to 300 gigahertz (GHz) . Generally, the region from 300 MHz to 3 GHz is known as the ultra-high frequency (UHF) region or decimeter band because the wavelengths range from approximately one decimeter to one meter in length. UHF waves may be blocked or redirected by buildings and environmental features, which may be referred to as clusters, but the waves may penetrate structures sufficiently for a macro cell to provide service to the UEs 115 located indoors. Communications using UHF waves may be associated with smaller antennas and shorter ranges (e.g., less than one hundred kilometers) compared to communications using the smaller frequencies and longer waves of the high frequency (HF) or very high frequency (VHF) portion of the spectrum below 300 MHz.
[0074] The wireless communications system 100 may also operate using a super high frequency (SHF) region, which may be in the range of 3 GHz to 30 GHz, also known as the centimeter band, or using an extremely high frequency (EHF) region of the spectrum (e.g., from 30 GHz to 300 GHz) , also known as the millimeter band. In some examples, the wireless communications system 100 may support millimeter wave (mmW) communications between the UEs 115 and the network entities 105 (e.g., base stations 140, RUs 170) , and EHF antennas of the respective devices may be smaller and more closely spaced than UHF antennas. In some examples, such techniques may facilitate using antenna arrays within a device. The propagation of EHF transmissions, however, may be subject to even greater attenuation and shorter range than SHF or UHF transmissions. The techniques disclosed herein may be employed across transmissions that use one or more different frequency regions, and designated use of bands across these frequency regions may differ by country or regulating body.
[0075] The wireless communications system 100 may utilize both licensed and unlicensed RF spectrum bands. For example, the wireless communications system 100 may employ License Assisted Access (LAA) , LTE-Unlicensed (LTE-U) RAT, or NR technology using an unlicensed band such as the 5 GHz industrial, scientific, and medical (ISM) band. While operating using unlicensed RF spectrum bands, devices such as the network entities 105 and the UEs 115 may employ carrier sensing for collision detection and avoidance. In some examples, operations using unlicensed bands may be based on a carrier aggregation configuration in conjunction with component carriers operating using a licensed band (e.g., LAA) . Operations using unlicensed spectrum may include downlink transmissions, uplink transmissions, P2P transmissions, or D2D transmissions, among other examples.
[0076] A network entity 105 (e.g., a base station 140, an RU 170) or a UE 115 may be equipped with multiple antennas, which may be used to employ techniques such as transmit diversity, receive diversity, multiple-input multiple-output (MIMO) communications, or beamforming. The antennas of a network entity 105 or a UE 115 may be located within one or more antenna arrays or antenna panels, which may support MIMO operations or transmit or receive beamforming. For example, one or more base station antennas or antenna arrays may be co-located at an antenna assembly, such as an antenna tower. In some examples, antennas or antenna arrays associated with a network entity 105 may be located at diverse geographic locations. A network entity 105 may include an antenna array with a set of rows and columns of antenna ports that the network entity 105 may use to support beamforming of communications with a UE 115. Likewise, a UE 115 may include one or more antenna arrays that may support various MIMO or beamforming operations. Additionally, or alternatively, an antenna panel may support RF beamforming for a signal transmitted via an antenna port.
[0077] The network entities 105 or the UEs 115 may use MIMO communications to exploit multipath signal propagation and increase spectral efficiency by transmitting or receiving multiple signals via different spatial layers. Such techniques may be referred to as spatial multiplexing. The multiple signals may, for example, be transmitted by the transmitting device via different antennas or different combinations of antennas. Likewise, the multiple signals may be received by the receiving device via different antennas or different combinations of antennas. Each of the multiple signals may be referred to as a separate spatial stream and may carry information associated with the same data stream (e.g., the same codeword) or different data streams (e.g., different codewords) . Different spatial layers may be associated with different antenna ports used for channel measurement and reporting. MIMO techniques include single-user MIMO (SU-MIMO) , for which multiple spatial layers are transmitted to the same receiving device, and multiple-user MIMO (MU-MIMO) , for which multiple spatial layers are transmitted to multiple devices.
[0078] Beamforming, which may also be referred to as spatial filtering, directional transmission, or directional reception, is a signal processing technique that may be used at a transmitting device or a receiving device (e.g., a network entity 105, a UE 115) to shape or steer an antenna beam (e.g., a transmit beam, a receive beam) along a spatial path between the transmitting device and the receiving device. Beamforming may be achieved by combining the signals communicated via antenna elements of an antenna array such that some signals propagating along particular orientations with respect to an antenna array experience constructive interference while others experience destructive interference. The adjustment of signals communicated via the antenna elements may include a transmitting device or a receiving device applying amplitude offsets, phase offsets, or both to signals carried via the antenna elements associated with the device. The adjustments associated with each of the antenna elements may be defined by a beamforming weight set associated with a particular orientation (e.g., with respect to the antenna array of the transmitting device or receiving device, or with respect to some other orientation) .
[0079] A network entity 105 or a UE 115 may use beam sweeping techniques as part of beamforming operations. For example, a network entity 105 (e.g., a base station 140, an RU 170) may use multiple antennas or antenna arrays (e.g., antenna panels) to conduct beamforming operations for directional communications with a UE 115. Some signals (e.g., synchronization signals, reference signals, beam selection signals, or other control signals) may be transmitted by a network entity 105 multiple times along different directions. For example, the network entity 105 may transmit a signal according to different beamforming weight sets associated with different directions of transmission. Transmissions along different beam directions may be used to identify (e.g., by a transmitting device, such as a network entity 105, or by a receiving device, such as a UE 115) a beam direction for later transmission or reception by the network entity 105.
[0080] Some signals, such as data signals associated with a particular receiving device, may be transmitted by a transmitting device (e.g., a network entity 105 or a UE 115) along a single beam direction (e.g., a direction associated with the receiving device, such as another network entity 105 or UE 115) . In some examples, the beam direction associated with transmissions along a single beam direction may be determined based on a signal that was transmitted along one or more beam directions. For example, a UE 115 may receive one or more of the signals transmitted by the network entity 105 along different directions and may report to the network entity 105 an indication of the signal that the UE 115 received with a highest signal quality or an otherwise acceptable signal quality.
[0081] In some examples, transmissions by a device (e.g., by a network entity 105 or a UE 115) may be performed using multiple beam directions, and the device may use a combination of digital precoding or beamforming to generate a combined beam for transmission (e.g., from a network entity 105 to a UE 115) . The UE 115 may report feedback that indicates precoding weights for one or more beam directions, and the feedback may correspond to a configured set of beams across a system bandwidth or one or more sub-bands. The network entity 105 may transmit a reference signal (e.g., a cell-specific reference signal (CRS) , a CSI reference signal (CSI-RS) ) , which may be precoded or unprecoded. The UE 115 may provide feedback for beam selection, which may be a precoding matrix indicator (PMI) or codebook-based feedback (e.g., a multi-panel type codebook, a linear combination type codebook, a port selection type codebook) . Although these techniques are described with reference to signals transmitted along one or more directions by a network entity 105 (e.g., a base station 140, an RU 170) , a UE 115 may employ similar techniques for transmitting signals multiple times along different directions (e.g., for identifying a beam direction for subsequent transmission or reception by the UE 115) or for transmitting a signal along a single direction (e.g., for transmitting data to a receiving device) .
[0082] A receiving device (e.g., a UE 115) may perform reception operations in accordance with multiple receive configurations (e.g., directional listening) when receiving various signals from a transmitting device (e.g., a network entity 105) , such as synchronization signals, reference signals, beam selection signals, or other control signals. For example, a receiving device may perform reception in accordance with multiple receive directions by receiving via different antenna subarrays, by processing received signals according to different antenna subarrays, by receiving according to different receive beamforming weight sets (e.g., different directional listening weight sets) applied to signals received at multiple antenna elements of an antenna array, or by processing received signals according to different receive beamforming weight sets applied to signals received at multiple antenna elements of an antenna array, any of which may be referred to as “listening” according to different receive configurations or receive directions. In some examples, a receiving device may use a single receive configuration to receive along a single beam direction (e.g., when receiving a data signal) . The single receive configuration may be aligned along a beam direction determined based on listening according to different receive configuration directions (e.g., a beam direction determined to have a highest signal strength, highest signal-to-noise ratio (SNR) , or otherwise acceptable signal quality based on listening according to multiple beam directions) .
[0083] The UEs 115 and the network entities 105 may support retransmissions of data to increase the likelihood that data is received successfully. Hybrid automatic repeat request (HARQ) feedback is one technique for increasing the likelihood that data is received correctly via a communication link (e.g., the communication link (s) 125, a D2D communication link 135) . HARQ may include a combination of error detection (e.g., using a cyclic redundancy check (CRC) ) , forward error correction (FEC) , and retransmission (e.g., automatic repeat request (ARQ) ) . HARQ may improve throughput at the MAC layer in relatively poor radio conditions (e.g., low signal-to-noise conditions) . In some examples, a device may support same-slot HARQ feedback, in which case the device may provide HARQ feedback in a specific slot for data received via a previous symbol in the slot. In some other examples, the device may provide HARQ feedback in a subsequent slot, or according to some other time interval.
[0084] In some cases, the wireless communications system 100 may support CSI feedback enhancements. For example, for CSI compression (e.g., two-sided models) a trade-off between performance and complexity (e.g., overhead) may be improved. That is, a spatial and frequency compression may be extended to a spatial, temporal, and frequency compression, cell and / or site specific models, CSI compression plus prediction (compared to a non-AI / ML based approach) . Such techniques may address (e.g., and alleviate) issues related to inter-vendor training collaboration. For CSI prediction (e.g., a one-sided model) , there may be opportunity for a performance gain over non-AI / ML based approaches and associated complexity, while addressing other aspects such as cell and / or site specific models may be considered to improve performance gain.
[0085] In some cases, a precoding matrix (e.g., as opposed to raw channel matrix) may serve as an input (e.g., to an ML model) , which may be associated with per-layer processing with common structure across ranks and layers (e.g., corresponding to various techniques for handling a rank that is greater than or equal to 1) . For some cases (e.g., associated with temporal domain aspects) , a same model structure may be used where appropriate (e.g., a structure based on using the precoding matrix as input) with one or more additional layers or operations at the input, the output domain, at the latent domain, or any combination thereof. The precoding matrix may, in some cases, be associated with an eigen vector in spatial-frequency domain, and / or angular, delay domain representation (e.g., an eType-II W2) . Additionally, or alternatively, the precoding matrix may be associated with a spatial-frequency domain for each CSI observation instance, and / or angular, delay, and Doppler domain representation (e.g., such as eType-II W2) . In some cases, a scalability approach over a quantity of transmission ports, CSI feedback payload sizes, and bandwidths may be considered.
[0086] In some cases, to alleviate and / or resolve the issues related to inter-vendor training collaboration of AI / ML-based CSI compression (e.g., using a two-sided model) , various options may be considered. For instance, a fully standardized reference model (e.g., structure and parameters) , a standardized dataset (e.g., dataset may refer to a set of data samples of CSI feedback and associated target CSI) , a standardized reference model structure with a parameter exchange between network-side and UE-side, a standardized data or dataset format with a dataset exchange between network -side and UE-side, a standardized model format with a reference model exchange between network -side and UE-side, some other option, or a combination thereof may be considered. In such cases, various options utilizing different methods of exchanging one or more parameters, a dataset, a reference model, or other information may affect a feasibility and a collaboration complexity (e.g., over-the-air interface, offline delivery, etc. ) .
[0087] Various implementations may support on-device operation and UE side offline engineering. In a first approach, a sharing of parameters, a reference model, and / or a dataset may enable UE-side offline engineering. Such an approach may support a down-selection of one or more options discussed herein based on their respective feasibility and performance. In some examples, the first approach may be based on whether additional information is shared from network-side to UE-side to enable UE-side encoder training, validation, and testing, based on a proprietary information disclosure by the network entity 105, based on overhead, based on a performance impact due to a mismatch between network-side data distribution and UE-side data distribution, and other factors.
[0088] In a second approach, a sharing of one or more network-side encoder parameters to the UE-side for UE-side inference may be directly associated with on-device operation. In some examples, the second approach may be based on overhead, based on whether it is feasible to use a common encoder across UEs 115 and whether it is feasible for network-side to train multiple encoders for different UEs 115, based on a performance impact due to a mismatch between network-side data distribution and UE-side inference data distribution, based on a proprietary information disclosure for the network entity 105 and / or a UE 115, and other factors.
[0089] In a third approach, one or more fully standardized reference model (s) and parameters with a specified CSI generation part and / or a specified CSI reconstruction part may be used. In some examples, the third approach may be based on a statistical channel model or field data (e.g., how field data is collected) for reference model (s) training, based on a performance impact due to a mismatch between the distribution of a dataset used for reference model (s) training, a UE-side data distribution, and a network-side data distribution, based on additional information that is be specified to enable UE-side encoder training, validation, and testing, and / or network-side decoder training, validation, and testing. In some examples, a combination of the first approach, the second approach, and the third approach may be used.
[0090] In some cases, a reference encoder may be specified, and network-side and / or UE-side may train their CSI generation part and CSI-reconstruction part separately such that they are respectively compatible to the reference encoder. Additionally, or alternatively, a reference decoder may be specified, and network-side and / or UE-side may train their CSI generation part and CSI-reconstruction part separately such that they are respectively compatible to the reference decoder. In some cases, both a reference encoder and a reference decoder may be specified, and network-side and / or UE-side may train their CSI generation part and CSI-reconstruction part separately such that they are compatible to the reference decoder and the reference encoder. In some cases, UE-side data (e.g., in UE-side data distribution) and network-side data (e.g., in network-side data distribution) may be associated with field data. In some cases, one or more issues identified in one approach may be applicable or may not be applicable for other approaches. In some cases, a down selection of one or more approaches described herein may be supported. In some cases, a data distribution mismatch may consider a use of synthetic data and / or field data.
[0091] In some examples, the wireless communications system 100 may support various cases for temporal compression. For example, for an evaluation of temporal domain aspects of an AI / ML-based CSI compression (e.g., using a two-sided model) may be based on Table 1.
[0092] Table 1
[0093] In some cases, for the UE 115, the past CSI information may include past model inputs and / or any information derived therefrom. For the network, the past CSI information may include past CSI feedback instances and / or any information derived therefrom. In some cases (e.g., case 3 and case 4) , the UE may perform prediction as a separate step or jointly with compression. Similarly, the network may perform prediction as a separate step or jointly with reconstruction. In some examples, “Target CSI slot (s) ” may refers to one or more slot (s) to which the CSI feedback in the report corresponds. In some cases, “Present slot” may refers to the slot of the most recent CSI-RS measurement used to generate a CSI report. In some cases, “Future slot (s) ” may include at least one slot after the present slot and may include the present slot as well. In some cases, a down-selection of one or more of such examples may be supported.
[0094] Some devices (e.g., a UE 115, a network entity 105) of a wireless communications system 100 may utilize neural networks (e.g., a trained AI model or ML model) to improve performance of some operations. For example, a UE 115 may improve operations associated with compressing information (e.g., data) prior to transmitting the information to other devices (e.g., via link (s) 125 to a network entity 105) . In some examples, the UE 115 may use a neural network to compress CSI (e.g., CSI feedback, CFE, a selected precoding matrix, CSI measurements) and may transmit the CSI to a network entity 105 after the compression. The network entity 105 may receive the compressed CSI and may also use a neural network to reconstruct the original CSI obtained by the UE 115. However, neural networks may introduce additional complexity to the UE 115 and the network entity 105. For example, processing operations at the UE 115 (e.g., floating point operations, other arithmetic operations) may increase when the UE 115 uses the neural network. Reducing a quantity of input sequence elements may reduce the complexity and may also reduce performance of the neural network. Thus, some neural network techniques (e.g., for data compression) may not achieve a balance between the computational complexity and the performance of the neural network, thus negating some benefits of using the neural network, increasing processing overhead, and reducing communication quality in the wireless communications system 100.
[0095] In accordance with one or more aspects described herein, a UE 115 may support one or more techniques (e.g., CSI compression techniques using an improved neural network architecture) that improves the balance between the complexity of a neural network and the performance of the neural network. For example, a UE 115 may be configured to perform one or more intermediate (e.g., incremental) compression operations on a set of data (e.g., corresponding to CSI) after one or more blocks (e.g., transformer blocks) of the neural network (e.g., instead of compressing the entire sequence before applying the neural network blocks, which may create a processing bottleneck) . That is, a neural network may include multiple neural network blocks, and various compression blocks may be distributed (e.g., interspersed) between each of the neural network blocks. In some examples, a compression block may include various operations such as one or more reshaping operations, one or more linear operations, one or more appending operations to compress an input sequence. Each compression block may reduce an output sequence of a first neural network block and provide the reduced sequence as an input to a subsequent neural network block. As such, an input sequence (e.g., of CSI data) may be progressively compressed along a processing chain that uses the neural network, which may support enhanced performance, reduced processing complexity, and improved communication quality in a wireless communications system 100.
[0096] FIG. 2 is an illustrative block diagram of an example ML model represented by an ANN 200 (e.g., a neural network, an AI model) that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure.
[0097] ANN 200 may receive input data 206 which may include one or more bits of data 202, pre-processed data output from pre-processor 204 (optional) , or some combination thereof. Here, data 202 may include training data, verification data, application-related data, or the like, based, for example, on the stage of deployment of ANN 200. Pre-processor 204 may be included within ANN 200 in some other implementations. Pre-processor 204 may, for example, process all or a portion of data 202 which may result in some of data 202 being changed, replaced, deleted, etc. In some implementations, pre-processor 204 may add additional data to data 202. In some implementations, the pre-processor 204 may be a ML model, such as an ANN.
[0098] ANN 200 includes at least one first layer 208 of artificial neurons 210 to process input data 206 and provide resulting first layer data via connections or “edges” such as edges 212 to at least a portion of at least one second layer 214. Second layer 214 processes data received via edges 212 and provides second layer output data via edges 216 to at least a portion of at least one third layer 218. Third layer 218 processes data received via edges 216 and provides third layer output data via edges 220 to at least a portion of a final layer 222 including one or more neurons to provide output data 224. All or part of output data 224 may be further processed in some manner by (optional) post-processor 226. Thus, in certain examples, ANN 200 may provide output data 228 that is based on output data 224, post-processed data output from post-processor 226, or some combination thereof.
[0099] Post-processor 226 may be included within ANN 200 in some other implementations. Post-processor 226 may, for example, process all or a portion of output data 224 which may result in output data 228 being different, at least in part, to output data 224, as result of data being changed, replaced, deleted, etc. In some implementations, post-processor 226 may be configured to add additional data to output data 224. In this example, second layer 214 and third layer 218 represent intermediate or hidden layers that may be arranged in a hierarchical or other like structure. Although not explicitly shown, there may be one or more further intermediate layers between the second layer 214 and the third layer 218. In some implementations, the post-processor 226 may be a ML model, such as an ANN.
[0100] The structure and training of artificial neurons 210 in the various layers may be tailored to specific requirements of an application. Within a given layer such as first layer 208, second layer 214, or third layer 218 of ANN 200, some or all of the neurons may be configured to process information provided to the layer and output corresponding transformed information from the layer. For example, transformed information from a layer may represent a weighted sum of the input information associated with or otherwise based on a non-linear activation function or other activation function used to “activate” artificial neurons of a next layer. Artificial neurons in such a layer may be activated by or be responsive to parameters such as the previously described weights and biases of ANN 200. The weights and biases of ANN 200 may be adjusted during a training process or during operation of ANN 200. The weights of the various artificial neurons may control a strength of connections between layers or artificial neurons, while the biases may control a direction of connections between the layers or artificial neurons. An activation function may select or determine whether an artificial neuron transmits its output to the next layer or not in response to its received data.
[0101] Different activation functions may be used to model different types of non-linear relationships. By introducing non-linearity into an ML model, an activation function allows the configuration for the ML model to change in response to identifying or detecting complex patterns and relationships in the input data 206. Some non-exhaustive example activation functions include a sigmoid based activation function, a hyperbolic tangent (tanh) based activation function, a convolutional activation function, up-sampling, pooling, and a rectified linear unit (ReLU) based activation function.
[0102] Training of an ML model, such as ANN 200, may be conducted using training data. Training data may include one or more datasets which ANN 200 may use to identify patterns or relationships. Training data may represent various types of information, including written, visual, audio, environmental context, operational properties, etc. During training, the parameters (such as the weights and biases) of artificial neurons 210 may be changed, such as to minimize or otherwise reduce a loss function or a cost function. A training process may be repeated multiple times to fine-tune ANN 200 with each iteration.
[0103] Various ANN model structures are available for consideration. For example, in a feedforward ANN structure, each artificial neuron 210 in layer 214 receives information from the previous layer (such as, one or more artificial neurons 210 in layer 208) and produces information for the next layer (such as, one or more artificial neurons 210 in layer 218) . In a convolutional ANN structure, some layers may be organized into filters that extract features from data, such as the training data or the input data. In a recurrent ANN structure, some layers may have connections that allow for processing of data across time, such as for processing information having a temporal structure, such as time series data forecasting.
[0104] In an autoencoder ANN structure, compact representations of data may be processed and the model trained to predict or potentially reconstruct original data from a reduced set of features. An autoencoder ANN structure may be useful for tasks related to dimensionality reduction and data compression.
[0105] A transformer ANN structure makes use of attention mechanisms that may enable the model to process input sequences in a parallel and efficient manner. An attention mechanism allows the model to focus on different parts of the input sequence at different times. Attention mechanisms may be implemented using a series of layers known as attention layers to compute weighted sums of input features based on a similarity between different elements of the input sequence. A transformer ANN structure may include a series of feedforward ANN layers whose configurations may change in response to identifying non-linear relationships between the input and output sequences, which may also be referred to as a process of “learning” by the ANN layers. The output of a transformer ANN structure may be obtained by applying a linear transformation to the output of a final attention layer. A transformer ANN structure may be of particular use for tasks that involve sequence modeling, or other like processing.
[0106] ANN 200 or other ML models may be implemented in various types of processing circuits along with memory and applicable instructions therein. For example, general-purpose hardware circuits, such as, such as one or more central processing units (CPUs) , one or more graphics processing units (GPUs) , or suitable combinations thereof, may be employed to implement a model. In some implementations, one or more tensor processing units (TPUs) , neural processing units (NPUs) , or other special-purpose processors, field-programmable gate arrays (FPGAs) , application-specific integrated circuits (ASICs) , or the like may also be employed. In some implementations, the ML model may be implemented by a NPU or a TPU embedded in a system on chip (SoC) along with other components, such as one or more CPUs, GPUs, etc. A SoC includes several components manufactured on a shared semiconductor substrate. The NPU or TPU may be controlled by the one or more CPUs by configuring the ML model implemented by the NPU or TPU with weights and biases, providing certain training data to the ML model to configure the ML model, or providing input data to the ML model to obtain related inferences. The one or more CPUs may also receive the inferences and be configured to perform certain actions based on the inferences produced by the ML model. The actions performed by the one or more CPUs may include sending commands to other components of the SoC or components external to the SoC to perform certain actions. For example, the CPU may send commands to a RF transceiver based on the outputs or inferences obtained from an ML model to cause the RF transceiver to operate on a wireless network in accordance with the ML model.
[0107] In example aspects, an ML model may be trained prior to, or at some point following, operation of the ML model, such as ANN 200, on input data. When training the ML model, information in the form of applicable training data may be gathered or otherwise created for use in training an ANN accordingly. For example, training data may be gathered or otherwise created regarding information associated with received / transmitted signal strengths, interference, and resource usage data, as well as any other relevant data that might be useful for training a model to address one or more problems or issues in a communication system. In certain instances, all or part of the training data may originate in a UE or other device in a wireless communication system, or one or more network entities, or aggregated from multiple sources (such as a UE and a network entity / entities, one or more other UEs, the Internet, or the like) . In another example, training data may be generated or collected online, offline, or both online and offline by a UE, network entity, or other device (s) , and all or part of such training data may be transferred or shared (in real or near-real time) , such as through store and forward functions or the like.
[0108] Once an ANN has been configured by setting parameters, including weights and biases, from training data, the ANN’s performance may be evaluated. In some scenarios, evaluation / verification tests may use a validation dataset, which may include data not in the training data, to compare the model’s performance to baseline or other benchmark information. The ANN configuration may be further refined, for example, by changing its architecture, re-training it on the data, or using different optimization techniques, etc.
[0109] As part of a training process, parameters affecting the functioning of the artificial neurons and layers may be adjusted. For example, backpropagation techniques may be used to train an ANN by iteratively adjusting weights or biases of certain artificial neurons associated with errors between a predicted output of the model and a desired output that may be known or otherwise deemed acceptable. Backpropagation may include a forward pass, a loss function, a backward pass, and a parameter update that may be performed in training iteration. The process may be repeated for a certain number of iterations for each set of training data until the weights of the artificial neurons / layers are adequately tuned.
[0110] In some implementations, one or more devices or services may support processes relating to a ML model’s usage, maintenance, activation, reporting, or the like. In certain instances, all or part of a dataset or model may be shared across multiple devices, to provide or otherwise augment or improve processing. In some examples, signaling mechanisms may be utilized at various nodes of wireless network to signal the capabilities for performing specific functions related to ML model, support for specific ML models, capabilities for gathering, creating, transmitting training data, or other ML related capabilities. ML models in wireless communication systems may, for example, be employed to support decisions or improve performance relating to wireless resource allocation or selection, wireless channel condition estimation, interference mitigation, beam management, positioning accuracy, energy savings, or modulation or coding schemes, etc. In some implementations, model deployment may occur jointly or separately at various network levels, such as, a UE (e.g., a UE 115) , a network entity (e.g., an network entity 105) such as a base station, or a disaggregated network entity such as a CU (e.g., a CU 160) , a DU (e.g., a DU 165) , an RU (e.g., an RU 170) , or the like.
[0111] A UE and / or a network entity may utilize the ANN 200 to improve operations associated with compressing information (e.g., data) prior to transmitting the information to other devices. In some examples, the UE may use the ANN 200 to compress CSI feedback and may transmit the CSI to a network entity after the compression. However, in some cases, the ANN 200 may introduce additional complexity (e.g., increased floating point operations, other arithmetic operations) to the UE. Reducing a quantity of input sequence elements may reduce the complexity but may also reduce performance of the ANN 200. As such, some techniques may not achieve a balance between the computational complexity and the performance of the ANN 200, thus negating some benefits of the ANN 200.
[0112] In accordance with one or more aspects described herein, a UE may support one or more techniques that improves the balance between complexity and performance of an ANN 200. For example, a UE 115 may be configured to perform one or more intermediate (e.g., incremental) compression operations on a set of data (e.g., corresponding to CSI) after one or more blocks (e.g., transformer blocks) of the ANN200. That is, various compression blocks (e.g., token merge blocks, as part on a post-processor 226) may be distributed (e.g., interspersed) between each of the neural network blocks. In some examples, a compression block may include various operations such as one or more reshaping operations, one or more linear operations, one or more appending operations to compress an input sequence. Each compression block may reduce an output sequence of a first neural network block and provide the reduced sequence as an input to a subsequent neural network block. As such, an input sequence (e.g., of CSI data) may be progressively compressed along a processing chain that uses the ANN 200, which may support enhanced performance, reduced processing complexity, and improved communication quality.
[0113] FIG. 3 shows an example of a wireless communications system 300 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The wireless communications system 300 may implement or be implemented by aspects of the wireless communications system 100 or the ANN 200 as described with reference to FIGs. 1 and 2. For example, the wireless communications system 300 may include a UE 115 and a network entity 105, which may be examples of, or include corresponding devices herein. The network entity 105 may communicate with the UE 115 via one or more communication links 305 such as a communication link 305-a and a communication link 305-b. The communication links 305 may be examples of or include downlink communication interfaces, uplink communication interfaces, or other communication interfaces. Although the network entity 105 and the UE 115 are shown as example devices of the wireless communications system 300, the techniques herein may be applied by one or more other devices described herein, including with reference to FIG. 1.
[0114] In some cases (e.g., with an AI and / or ML based air interface) , the UE 115 and the network entity 105 (e.g., a gNB) may use a trained AI or ML model to implement one or more functions, operations, or procedures. For example, the UE 115 may obtain CSI (e.g., CSF, channel measurements or estimates, a precoding matrix, and other information) and may intend to transmit (e.g., output, convey) the CSI to the network entity 105. In such examples, the UE 115 may use a neural network (e.g., an ANN 200) to derive a compressed representation of the CSI (e.g., having a relatively smaller dimensionality) , such as the compressed CSI 310, which the UE 115 may transmit (e.g., feedback) to the network entity 105. Correspondingly, the network entity 105 may use another neural network to reconstruct the target CSI from the compressed CSI 310 (e.g., compressed representation of the original CSI) received from the UE 115.
[0115] As a non-limiting example, CSI may include an indication of a precoding matrix that the UE 115 recommends (e.g., requests, prefers) the network entity 105 to use based on a channel (e.g., channel characteristics) observed by the UE 115. To ensure accuracy of reconstruction at the network entity 105 (e.g., using the neural network) , the UE-side and NW-side ML models may be trained in a collaborative manner such that the compressed CSI 310 created by the UE-side model is interpreted and decoded correctly by the NW-side model. If such conditions are satisfied, then such a pair of models may be described as being compatible to each other.
[0116] In some cases, the UE 115 may utilize a processing flow 302 (e.g., which may include one or more operations performed by a neural network 304) , which may support a joint spatial and frequency compression of data. For instance, the UE 115 may obtain an input 320, which may be organized in one or more dimensions (e.g., two or three dimensions) . In some cases, each dimension of the input may correspond to a spatial domain, a frequency domain, or a temporal domain. As a non-limiting example, the input 320 may have a first dimension (e.g., frequency domain) with a first size (e.g., equal to a quantity of subbands, nSB) , a second dimension (e.g., spatial domain) with a second size (e.g., equal to an integer multiple of a quantity of ports, nPort, such as 2*nPort) . In some examples, the input 320 may have a third dimension (e.g., temporal domain) with a third size (e.g., corresponding to a quantity of time domain samples, nTD) . That is, a dimensionality of the input 320 may be represented as nSB x 2*nPort x nTD (e.g., for 4 domain samples, 32 ports, and 13 subbands, the dimensionality of the input 320 may be 13x64x4) .
[0117] In some cases, a neural network 304 (e.g., a ANN 200, an ML model, a CSI compression model) may expect a sequence of vectors as an input (e.g., for transformer based encoders and decoders) . Accordingly, an input processing block 325 may reorganize the input 320 into a sequence 330 of vectors 332. In some examples, a sequence 330 may correspond to a single time domain sample, and a quantity of vectors 332 in the sequence 330 may be associated with the first dimension of the input 320 (e.g., nSB) . A quantity of elements in each vector 332 may be associated with the second dimension of the input 320 (e.g., 2*nPort) . That is, after the input processing block 325 the data may be organized a sequence 330 having ‘nSB’ vectors 332 each having ‘2*nPort’ elements (e.g., for 32 ports and 13 subbands, the sequence 330 may have 13 vectors 332 each with size 64) .
[0118] The one or more sequences 330 may be input into an embedding layer 335 of the neural network 304. The embedding layer 335 (e.g., a hidden layer) may include one or more linear operations (e.g., matrix multiplication) that restructure input data to be compatible with one or more neural network blocks 340 (e.g., transformer blocks) . That is, the embedding layer 335 may process incoming data such that a dimensionality of the data input to the one or more neural network blocks 340 is compatible with the model (s) associated with the neural network blocks 340. A quantity of neural network blocks 340 supported by the neural network 304 may be an integer quantity, ‘N’ . In some examples, the quantity ‘N’ may be associated with a quantity of repetitions with which a neural network block 340 is applied (e.g., a quantity of processing iterations by the neural network block 340) . The one or more neural network blocks 340 may output the data to a combining layer 345. The combining layer 345 may produce an output 350 by performing one or more linear operations (e.g., matrix multiplications) that map the data to a sequence having a fourth dimension (e.g., a latent dimension, a one-dimensional sequence) . That is, the output 350 may be associated with a dimension that is used to transmit the data via a wireless channel (e.g., over-the-air) to the network entity 105 (e.g., based on applying the combining layer 345) .
[0119] In some cases, a design of the processing flow 302 may not be defined for temporal compression (e.g., along with spatial and frequency compression) . That is, some implementations of the processing flow 302 may support techniques for spatial and frequency compression but may not support (e.g., the design may not consider) temporal compression. Accordingly, a dimensionality that is to be used by the neural network 304 may be undefined, which may result in ambiguity at the input processing block 325 and the embedding layer 335. Moreover, operation of the combining layer 345 may not be defined in cases of temporal compression, thus resulting in ambiguity of a dimensionality at the output 350. Additionally, when considering the temporal compression, balancing a trade-off between processing complexity and performance may depend on how the processing flow 302 is designed to accommodate the temporal dimension.
[0120] In accordance with one or more techniques described herein, the UE 115 may support CSI compression that incorporates temporal compression (e.g., along with spatial and frequency compression techniques) while balancing the trade-off between the complexity and performance of the neural network 304. For example, the UE 115 may be configured to perform one or more intermediate (e.g., incremental) compression operations on a set of data (e.g., corresponding to CSI) after one or more neural network blocks 340 (e.g., instead of compressing the entire sequence with a single combining layer 345) . That is, and various compression blocks may be distributed (e.g., interspersed) between each of the neural network blocks 340. Such techniques may described in greater detail herein, including with reference to FIGs. 4 and 5.
[0121] In some examples, the UE 115 may receive one or more control messages 315 from the network entity 105. The one or more control messages 315 may include parameters that configure the UE 115 for obtaining and reporting CSI in accordance with the neural network 304 (e.g., an RRC configuration of CSI-RS and ML CSF reporting) . For example, the one or more control messages 315 may include an indication of one or more first parameters associated with a reference neural network, and one or more second parameters (e.g., parameters used for inference by the neural network 304) of the neural network 304 may be based on the one or more first parameters of the reference neural network. Additionally, or alternatively, the one or more control messages 315 may include an indication of one or more parameters for the neural network 304 (e.g., parameter used directly for inference by the neural network 304) . As such, an input 320 (e.g., of CSI data) may be progressively compressed along the processing flow 302 that uses the neural network 304, which may support enhanced performance, reduced processing complexity, and improved communication quality in the wireless communications system 300.
[0122] FIG. 4 shows an example of a neural network architecture 400 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. In some examples, aspects of the neural network architecture 400 may implement or be implemented by aspects of the wireless communications system 100, the ANN 200, or the wireless communications system 300. For example, the neural network architecture 400 may be implemented at a UE 115 (e.g., or a network entity 105) and may include one or more neural network blocks 410 which may be examples of or include neural network blocks 340. Although an example quantity of neural network blocks 410 are shown, the neural network architecture 400 may support any quantity of neural network blocks 410.
[0123] In some cases, to accommodate a temporal compression (e.g., along with spatial and frequency compression) a neural network may support various design implementations. For instance, in one implementation, each time domain sample may be processed by the neural network (e.g., the neural network 304, a transformer-encoder) and may be combined in the latent dimension. However, such implementations may be associated with increased processing complexity (e.g., increased floating point operations and increased multiply and accumulate operations) without a significant performance increase. In another implementation, a relatively longer sequence may be formed based on the quantity of time domain samples (e.g., based on nTD*nSB) . However, such implementations may be associated with increased processing complexity but may support improved performance in limited operating environments (e.g., outdoor scenarios with moderate and high payloads) . In another implementation, time domain samples may be combined via and input processing block (e.g., using multi-layer perception (MLP) , input processing block 325) followed by application of the neural network. However, such implementations may be associated with reduced performance in some environments (e.g., outdoor datasets with UE speeds higher than a threshold) and may create a bottleneck for time domain combining. Thus, techniques are desired that achieve an improved trade-off between processing complexity and performance.
[0124] In accordance with aspects of the present disclose, the neural network architecture 400 may provide the improved balance between processing complexity and performance of a neural network (e.g., a ANN 200, a neural network 304) . For example, the neural network architecture 400 may support a formation of a relatively longer input sequence that is based on the time domain and frequence domain (e.g., based on nTD*nSB) . Then, one or more incremental compression blocks (e.g., incremental token merging) may be used in between neural network blocks 410 (e.g., one or more neural network blocks 340, transformer blocks) .
[0125] In some examples, the neural network architecture 400 may support processing of three-dimensional inputs (e.g., an input ‘V’ , input 320) . A first dimension of an input may be associated with the time domain, where a correlation in the time domain may depend on a velocity of a UE 115 and a periodicity of the CSI-RS. A second dimension of the input may be associated with the frequency domain, where a correlation over subbands may increase for line-of-sight (LOS) samples (e.g., when a delay spread is relatively small) and may decrease for non-LOS samples. In some examples, a quantity of subbands (e.g., nSB) may be larger than a quantity of time samples (e.g., nTD) . A third dimension of the input may be associated with the spatial domain, which may be more difficult to compress (e.g., than the temporal and frequency domains, and may be left for processing by the neural network) . In some examples, it may be beneficial if an input processing layer is not associated with a bottleneck effect (e.g., as a majority of processing power may be in the neural network blocks 410) . For example, when there is medium or low correlation (e.g., below a threshold) in the time domain, combining the time domain samples prior to processing by one or more neural network blocks 410 may be inefficient.
[0126] Accordingly, instead of doing a sequence combining operation (e.g., a single combining operation) at an input processing block (e.g., before neural network blocks 410) , the sequence may be incrementally combined after one or more neural network blocks 410. For example, a combination operation may be performed after each set of one or more neural network blocks 410. Each of the one or more neural network blocks 410 may include an integer quantity of blocks, which may be equal to or different than other sets of the neural network blocks 410. For example, the neural network blocks 410-a, the neural network blocks 410-b, and the neural network blocks 410-c may each include a same quantity of blocks. In some examples, each of the respective sets of neural network blocks 410 may be associated with a same architecture (e.g., model architecture) and may be associated with different weights.
[0127] In some examples, a length, L0, of a sequence at an input 405 may be based on (e.g., equal to) a quantity of subbands times a quantity of time domain sample (e.g., nSB. nTD) . The sequence may be progressively combined after each set of neural network blocks 410. That is, after a first set of neural network blocks 410-a, a length, L1, of the sequence may be less than the length, L0 (e.g., L1 = nSB. nTD / 2) . After a second set of neural network blocks 410-b, a length, L2, of the sequence may be less than the length, L1 (e.g., L2 = L1 / 2) . After a third set of neural network blocks 410-c, a length, L3, of the sequence may be less than the length, L2 (e.g., L3 = L2 / 2) , and so on. As a non-limiting example, a total quantity of neural network blocks 410 may be 8, the sequence length at the input 405 may be 32, and a combining operation may occur after every 2 blocks (e.g., the neural network blocks 410-a, 410-b, and 410-c may each have 2 blocks and a combination operation may be performed after each of them) , which may result in sequence lengths L0 = 32, L1 = 16, L2 = 8, and L3 = 4. Thus, an output 415 may be associated with compressed data (e.g., CSI) relative to the input 405. Additionally, the compression of such data may achieve performance gains without significantly increasing processing complexity.
[0128] FIG. 5 shows an example of a neural network architecture 500 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. In some examples, aspects of the neural network architecture 500 (e.g., a transformer ANN structure, a transformer with incremental compression operations 520 or token merging) may implement or be implemented by aspects of the wireless communications system 100, the ANN 200, the wireless communications system 300, or the neural network architecture 400. For example, the neural network architecture 500 may include multiple neural network blocks 515, which may examples of one or more neural network blocks 340 and neural network blocks 410 as described with reference to FIGs. 3 and 4. In some examples, a UE 115, a network entity 105 or both may implement one or more aspects of the neural network architecture 500.
[0129] In some examples, a UE 115 may obtain a sequence (e.g., a sequence of vectors) at an input 505. The sequence may be obtained (e.g., received, measured) in accordance with one or more dimensions (e.g., three dimensions, as a nSB x 2*nPort x nTD precoder, ‘V’ ) . In some examples, the sequence at the input may have been previously processed by an input processing block (e.g., input 320) to reorganize the sequence data (e.g., a dimensionality of the sequence at the incremental compression operations 520 may be (nTD*nSB) x nPorts) . The sequence may be processed with a restructure operation 510, which may include one or more linear operations and one or more positional embedding operations. After the restructure operation 510, the sequence may be associated with a dimensionality that is compatible with the multiple neural network blocks 515 (e.g., the dimensionality may be (nTD*nSB) x d, where ‘d’ is a dimension that is compatible with the one or more neural network blocks 515) .
[0130] In some examples, a procedure 502-a may process the sequence, which may include applying one or more neural network blocks 515-a to the sequence (e.g., a neural network block 515-a may be applied ‘M’ times or ‘M1’ times to the sequence) . Each neural network block 515 may include a respective attention operation 516 (e.g., attention operation 516-a) and a respective MLP operation 518 (e.g., MLP operation 518-a) . An attention operation 516 may be associated with determining a relative importance of each element in the sequence, and an MLP operation 518 may be associated with inferring one or more output of the multiple neural network block 515 (e.g., based on a non-linear function applied to a weighted sum at each neuron of the neural network) . A compression operation 520-a (e.g., a first merge of sequence elements) may be applied to an output of the neural network block (s) 515-a, which may compress a size of at least one dimension of the sequence by a first factor (e.g., a first integer factor, a factor of 2) .
[0131] Each compression operation 520 may be associated with one or more respective linear layer operations 522 and one or more respective reshaping operations 524. In some examples, the one or more respective reshaping operations 524 may include reorganizing the sequence (e.g., nS x d, where nS = nTD*nSB) by increasing a size of a first dimension of the sequence and decreasing a size of a second dimension of the sequence. Moreover, the one or more linear layer operations 522 may decrease the size of the first dimension of the sequence. For example, at 525 the sequence may have a dimensionality of nS x d. At 530, after one or more reshaping operations 524, the sequence may have a dimensionality of nS / a (reduced size) x b*d (increased size) (e.g., where ‘a’a nd ‘b’a re respectively equal to an integer greater than 0, such as 2, nS / 2 x 2*d) . At 535, after one or more linear layer operations 522, the sequence may have a dimensionality of nS / ax d (e.g., nS / 2 x d) . In some examples, a reshaping operation 524 may include appending (e.g., adding, padding) an element (e.g., a token) to the sequence (e.g., if the sequence includes an odd quantity of elements prior to merging) . In some examples, the appended element may be associated a learnable parameter. For example, a value of the element may be based on the learnable parameter (e.g., learnable by the neural network) . Additionally, or alternatively, the element may be appended to the sequence based on a learnable parameter.
[0132] After the procedure 502-a, at least one dimension of the sequence may have a reduced size (e.g., may have dimensionality nTD*nSB / 2 x d after the compression operation 520-a) ) . Subsequently, a procedure 502-b may be performed on the sequence output by the procedure 502-a. In some examples, one or more of the operations of the procedure 502-b may be identical with the procedure 502-a. In some examples, the procedure 502-b may include a respective application of one or more neural network blocks 515-b (e.g., including respective attention operations 516-b and respective MLP operation 518-b) . In some examples, the neural network block 515-b may be applied ‘M’ times or ‘M2’ times to the sequence. A compression operation 520-b may be performed on the output of the one or more neural network blocks 515-b, which may further reduce a size of at least one dimension of the sequence (e.g., may have dimensionality nTD*nSB / 4 x d after the compression operation 520-b) .
[0133] Moreover, a procedure 502-c may be performed on the sequence output by the procedure 502-b. In some examples, one or more of the operations of the procedure 502-c may be identical with the procedure 502-a and / or the procedure 502-b. In some examples, the procedure 502-c may include a respective application of one or more neural network blocks 515-c (e.g., including respective attention operations 516-c and respective MLP operation 518-c) . In some examples, the neural network block 515-c may be applied ‘M’ times or ‘M3’ times to the sequence. A compression operation 520-c may be performed on the output of the one or more neural network blocks 515-c, which may further reduce a size of at least one dimension of the sequence (e.g., may have dimensionality nTD*nSB / 8 x d after the compression operation 520-c) . In some examples, each compression operation 520 may be combined with a respective MLP operation 518 (e.g., an MLP output layer) .
[0134] Accordingly, based progressively reducing a quantity of sequence elements, processing complexity may be reduced at each procedure 502 (e.g., a quantity of floating point operations may be reduced) . As a non-limiting example, nTD may equal 4 and nSB may equal 13, thus a sequence length may be 4*13 = 52. After the procedure 502-a, the sequence length may be 52 / 2 = 26 (e.g., 26 x d) , after the procedure 502-b the sequence length may be 26 / 2 = 13 (e.g., 13 x d) , and after the procedure 502-c the sequence length may be 14 / 2 = 7 (e.g., 7 x d, based on appending a token during the compression operation 520-c) . Although the non-limiting example shown in FIG. 5 includes three procedures 502, the one or more techniques described herein may include using any quantity of procedures 502.
[0135] After the procedure 502-c (e.g., a final procedure 502) , one or more combining operations 540 (e.g., such as operations of a combining layer 345) may be performed on the sequence to obtain the output 545 (e.g., a latent vector) . In some examples, the one or more combining operations 540 may include one or more reshaping operations and one or more linear layer operations, which may reduce a quantity of dimensions in the sequence (e.g., and / or a size of one or more dimensions) . For example, the one or more combining operations 540 may produce a sequence that is associated with a dimension that is used for transmitting information via a wireless channel (e.g., a latent dimension) , which may be reduced relatively to the dimensionality of the sequence after the procedure 502-c. Subsequently, a UE 115 may be configured to transmit the sequence at the output 545 (e.g., a compressed sequence, compressed CSI information) to one or more other devices (e.g., a network entity 105) via a wireless communication channel.
[0136] Thus, by applying one or more techniques herein, a UE 115 may support a neural network that enables spatial, frequency, and temporal compression, which may achieve an improved balance between complexity and compression performance. For example, the interspersed compression operations 520 may support relatively reduced processing overhead without significantly reducing performance of the neural network. Accordingly, a wireless communications system may experience improved processing performance, improved system capacity, reduced communication traffic, and other benefits.
[0137] FIG. 6 shows an example of a process flow 600 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The process flow 600 may implement or be implemented to realize aspects of the wireless communications system 100, the ANN 200, the wireless communications system 300, the neural network architecture 400, or the neural network architecture 500. For example, the process flow 600 illustrates communication between a UE 115 and a network entity 105, which may be examples of corresponding devices described herein. Alternative examples of the following may be implemented. For example, some steps may be performed in a different order than described or may not be performed at all. In some implementations, steps may include additional features not mentioned below, or further steps may be added. Further, although the UE 115 and the network entity 105 are shown performing the operations of the process flow 600, some aspects of some operations may also be performed by one or more other wireless communication devices.
[0138] At 605, in some examples, the UE 115 may receive, from the network entity 105, one or more control messages (e.g., control messages 315, RRC configuration of CSI-RS and ML CSF reporting) that include one or more parameters that configure the UE 115 to obtain CSI and one or more second parameters that configure the UE 115 to transmit a message, which may include the CSI (e.g., a CSI report) . In some examples, obtaining a first sequence (e.g., an input sequence to a neural network) may be based on or in response to receiving the one or more control messages.
[0139] At 610, in some examples, the UE 115 may receive, from the network entity 105, an indication of one or more first parameters associated with a reference neural network. For example, one or more second parameters of a neural network at the UE 115 may be based on the one or more first parameters. In some examples, the CSI may be compressed in accordance with the one or more second parameters. Additionally, or alternatively, the UE 115 may receive, from the network entity 105, an indication of one or more parameters for the neural network at the UE 115 (e.g., for direct implementation by the neural network, for direct inference by the UE 115) . In some examples, the CSI may be compressed in accordance with the one or more parameters.
[0140] At 615, in some examples, the UE 115 may obtain the CSI prior to obtaining the first sequence. In some examples, the CSI may include a set of data that is associated with a first dimension (e.g., (nTD*nSB) x nPorts) . For example, the UE 115 may be configured to perform one or more CSI-RS measurements and may obtain data for the first sequence based on the measurement (s) .
[0141] At 620, the UE 115 may obtain the first sequence, which may be associated with CSI (e.g., the obtained CSI) and may include a first quantity of elements. In some examples, a neural network at the UE 115 may be associated with compressing the first sequence (e.g., the CSI) .
[0142] At 625, in some examples, the UE 115 may perform one or more operations to restructure the set of data (e.g., the obtained data, restructure operation 510) , which may be associated with the first sequence. In some examples, the set of data may be associated with a second dimension (e.g., (nTD*nSB) x d) that is compatible with the neural network after performing the one or more operations.
[0143] At 630, the UE 115 may perform a first procedure using the neural network to obtain a second sequence that includes a second quantity of elements (e.g., (nTD*nSB) / 2) that is less than the first quantity of elements. In some examples, the first procedure may include applying one or more first neural network blocks (e.g., neural network blocks 515-a) to the first sequence and compressing (e.g., using compression operations 520-a) a first output of the one or more first neural network blocks. In some examples, the one or more first neural network blocks may be applied to the first sequence in accordance with a first quantity of repetitions (e.g., ‘M’ or ‘M1’ ) .
[0144] In some examples, compressing the first output (e.g., having dimensionality of nS x d) may include performing a first reshaping operation (e.g., reshaping operations 524) on the first output that decreases a size of the first output in a first dimension and increases a size of the first output in a second dimension (e.g., nS / 2 x 2*d) . In some examples, compressing the first output may further include performing, after the first reshaping operation, a first linear operation (e.g., linear layer operations 522) that decreases the size of the first output in the second dimension by combining one or more first elements of the first output with one or more second elements of the first output (e.g., nS / 2 x d) .
[0145] In some examples, the UE 115 may perform a second procedure using the neural network to obtain a third sequence that includes a third quantity of elements (e.g., (nTD*nSB) / 4) that is less than the second quantity of elements. In some examples, the second procedure may include applying one or more second neural network blocks (e.g., neural network blocks 515-b) to the second sequence and compressing (e.g., using compression operations 520-b) a second output of the one or more second neural network blocks. In some examples, the one or more second neural network blocks may be applied to the second sequence in accordance with a second quantity of repetitions (e.g., ‘M’ or ‘M2’ ) . In some examples, the second quantity of repetitions may be equal to the first quantity of repetitions or may be different than the first quantity of repetitions.
[0146] In some examples, compressing the second output may include performing second reshaping operation on the second output that decreases a size of the second output in a third dimension and increases a size of the second output in a fourth dimension. In some examples, compressing the second output may further include performing, after the second reshaping operation, a second linear operation that decreases the size of the second output in the fourth dimension by combining one or more third elements of the second output with one or more fourth elements of the second output.
[0147] In some examples, the UE 115 may perform a third procedure using the neural network to obtain a fourth sequence that includes a fourth quantity of elements (e.g., (nTD*nSB) / 8) that is less than the third quantity of elements. In some examples, the third procedure may include applying one or more third neural network blocks (e.g., neural network blocks 515-c) to the third sequence and compressing (e.g., using compression operations 520-c) a third output of the one or more third neural network blocks. In some examples, compressed CSI (e.g., used for transmitting) may be further associated with the fourth sequence. In some examples, the one or more third neural network blocks may be applied to the third sequence in accordance with a third quantity of repetitions (e.g., ‘M’ or ‘M3’ ) . In some examples, the third quantity of repetitions may be equal to the first quantity of repetitions and / or the second quantity of repetitions, or may be different than the first quantity of repetitions or the second quantity of repetitions.
[0148] In some examples, the compression of the first output, the second output, the third output, or any combination thereof may include appending, prior to performing a respective linear operation, a first element to the first output, the second output, or the third output, based on the size of the first output, the second output, or the third output in the second dimension including an odd quantity of elements. In some examples, the first element (e.g., the appended element) may be associated with a parameter that is learnable by the neural network.
[0149] In some examples, the one or more first neural network blocks, the one or more second neural network blocks, the one or more third neural network blocks, or any combination thereof may be examples of (e.g., or otherwise in accordance with one or more aspects of) a transformer neural network structure. In some examples, the first procedure at the one or more first neural network blocks, the second procedure at the one or more second neural network blocks, the third procedure at the one or more third neural network blocks, or any combination thereof may include one or more identical operations.
[0150] At 635, in some examples, the UE 115 may perform, after performing the second procedure, one or more operations (e.g., in accordance with the one or more combining operations 540) to restructure a set of data associated with a sequence that is output by the procedure (s) of the neural network. In some examples, the first sequence (e.g., before performing the procedures of the neural network) may be associated with a first quantity of dimensions and the set of data may be associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations.
[0151] At 640, the UE 115 may transmit a message (e.g., CSF, a CSI report) to the network entity 105. The message may include compressed CSI (e.g., which may be obtained based on the procedures of the neural network and other operations) . In some examples, the compressed CSI is associated with a sequence that is output by one or more procedures of the neural network, by one or more combining layer operations, or a combination thereof.
[0152] FIG. 7 shows a block diagram 700 of a device 705 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The device 705 may be an example of aspects of a UE 115 as described herein. The device 705 may include a receiver 710, a transmitter 715, and a communications manager 720. The device 705, or one or more components of the device 705 (e.g., the receiver 710, the transmitter 715, the communications manager 720) , may include at least one processor, which may be coupled with at least one memory, to, individually or collectively, support or enable the described techniques. Each of these components may be in communication with one another (e.g., via one or more buses) .
[0153] The receiver 710 may provide a means for receiving information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to CSI compression using neural networks) . Information may be passed on to other components of the device 705. The receiver 710 may utilize a single antenna or a set of multiple antennas.
[0154] The transmitter 715 may provide a means for transmitting signals generated by other components of the device 705. For example, the transmitter 715 may transmit information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to CSI compression using neural networks) . In some examples, the transmitter 715 may be co-located with a receiver 710 in a transceiver module. The transmitter 715 may utilize a single antenna or a set of multiple antennas.
[0155] The communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof may be examples of means for performing various aspects of CSI compression using neural networks as described herein. For example, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof may be capable of performing one or more of the functions described herein.
[0156] In some examples, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof may be implemented in hardware (e.g., in communications management circuitry) . The hardware may include at least one of a processor, a digital signal processor (DSP) , a CPU, an application-specific integrated circuit (ASIC) , a field-programmable gate array (FPGA) or other programmable logic device, a microcontroller, discrete gate or transistor logic, discrete hardware components, or any combination thereof configured as or otherwise supporting, individually or collectively, a means for performing the functions described in the present disclosure. In some examples, at least one processor and at least one memory coupled with the at least one processor may be configured to perform one or more of the functions described herein (e.g., by one or more processors, individually or collectively, executing instructions stored in the at least one memory) .
[0157] Additionally, or alternatively, the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof may be implemented in code (e.g., as communications management software or firmware) executed by at least one processor (e.g., referred to as a processor-executable code) . If implemented in code executed by at least one processor, the functions of the communications manager 720, the receiver 710, the transmitter 715, or various combinations or components thereof may be performed by a general-purpose processor, a DSP, a CPU, an ASIC, an FPGA, a microcontroller, or any combination of these or other programmable logic devices (e.g., configured as or otherwise supporting, individually or collectively, a means for performing the functions described in the present disclosure) .
[0158] In some examples, the communications manager 720 may be configured to perform various operations (e.g., receiving, obtaining, monitoring, outputting, transmitting) using or otherwise in cooperation with the receiver 710, the transmitter 715, or both. For example, the communications manager 720 may receive information from the receiver 710, send information to the transmitter 715, or be integrated in combination with the receiver 710, the transmitter 715, or both to obtain information, output information, or perform various other operations as described herein.
[0159] The communications manager 720 may support wireless communications in accordance with examples as disclosed herein. For example, the communications manager 720 is capable of, configured to, or operable to support a means for obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The communications manager 720 is capable of, configured to, or operable to support a means for performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The communications manager 720 is capable of, configured to, or operable to support a means for performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The communications manager 720 is capable of, configured to, or operable to support a means for transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0160] By including or configuring the communications manager 720 in accordance with examples as described herein, the device 705 (e.g., at least one processor controlling or otherwise coupled with the receiver 710, the transmitter 715, the communications manager 720, or a combination thereof) may support techniques for reduced processing, reduced power consumption, and more efficient utilization of communication resources, among other benefits.
[0161] FIG. 8 shows a block diagram 800 of a device 805 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The device 805 may be an example of aspects of a device 705 or a UE 115 as described herein. The device 805 may include a receiver 810, a transmitter 815, and a communications manager 820. The device 805, or one or more components of the device 805 (e.g., the receiver 810, the transmitter 815, the communications manager 820) , may include at least one processor, which may be coupled with at least one memory, to support the described techniques. Each of these components may be in communication with one another (e.g., via one or more buses) .
[0162] The receiver 810 may provide a means for receiving information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to CSI compression using neural networks) . Information may be passed on to other components of the device 805. The receiver 810 may utilize a single antenna or a set of multiple antennas.
[0163] The transmitter 815 may provide a means for transmitting signals generated by other components of the device 805. For example, the transmitter 815 may transmit information such as packets, user data, control information, or any combination thereof associated with various information channels (e.g., control channels, data channels, information channels related to CSI compression using neural networks) . In some examples, the transmitter 815 may be co-located with a receiver 810 in a transceiver module. The transmitter 815 may utilize a single antenna or a set of multiple antennas.
[0164] The device 805, or various components thereof, may be an example of means for performing various aspects of CSI compression using neural networks as described herein. For example, the communications manager 820 may include an input sequence component 825, a neural network component 830, an information output component 835, or any combination thereof. The communications manager 820 may be an example of aspects of a communications manager 720 as described herein. In some examples, the communications manager 820, or various components thereof, may be configured to perform various operations (e.g., receiving, obtaining, monitoring, outputting, transmitting) using or otherwise in cooperation with the receiver 810, the transmitter 815, or both. For example, the communications manager 820 may receive information from the receiver 810, send information to the transmitter 815, or be integrated in combination with the receiver 810, the transmitter 815, or both to obtain information, output information, or perform various other operations as described herein.
[0165] The communications manager 820 may support wireless communications in accordance with examples as disclosed herein. The input sequence component 825 is capable of, configured to, or operable to support a means for obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The neural network component 830 is capable of, configured to, or operable to support a means for performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The neural network component 830 is capable of, configured to, or operable to support a means for performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The information output component 835 is capable of, configured to, or operable to support a means for transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0166] FIG. 9 shows a block diagram 900 of a communications manager 920 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The communications manager 920 may be an example of aspects of a communications manager 720, a communications manager 820, or both, as described herein. The communications manager 920, or various components thereof, may be an example of means for performing various aspects of CSI compression using neural networks as described herein. For example, the communications manager 920 may include an input sequence component 925, a neural network component 930, an information output component 935, a CSI obtaining component 940, a data shaping component 945, a data compression component 950, a control message component 955, a neural network parameter component 960, or any combination thereof. Each of these components, or components or subcomponents thereof (e.g., one or more processors, one or more memories) , may communicate, directly or indirectly, with one another (e.g., via one or more buses) .
[0167] The communications manager 920 may support wireless communications in accordance with examples as disclosed herein. The input sequence component 925 is capable of, configured to, or operable to support a means for obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The neural network component 930 is capable of, configured to, or operable to support a means for performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. In some examples, the neural network component 930 is capable of, configured to, or operable to support a means for performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The information output component 935 is capable of, configured to, or operable to support a means for transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0168] In some examples, the CSI obtaining component 940 is capable of, configured to, or operable to support a means for obtaining the CSI prior to obtaining the first sequence, where the CSI includes a set of data that is associated with a first dimension. In some examples, the data shaping component 945 is capable of, configured to, or operable to support a means for performing one or more operations to restructure the set of data, where the set of data is associated with a second dimension that is compatible with the neural network after performing the one or more operations.
[0169] In some examples, the data shaping component 945 is capable of, configured to, or operable to support a means for performing, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, where the first sequence is associated with a first quantity of dimensions, and where the set of data is associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations.
[0170] In some examples, the neural network component 930 is capable of, configured to, or operable to support a means for performing a third procedure using the neural network to obtain a fourth sequence including a fourth quantity of elements that is less than the third quantity of elements, the third procedure including applying one or more third neural network blocks to the third sequence and compressing a third output of the one or more third neural network blocks, where the compressed CSI is further associated with the fourth sequence.
[0171] In some examples, the one or more first neural network blocks are applied to the first sequence in accordance with a first quantity of repetitions; and the one or more second neural network blocks are applied to the second sequence in accordance with a second quantity of repetitions. In some examples, the second quantity of repetitions is equal to the first quantity of repetitions. In some examples, the second quantity of repetitions is different than the first quantity of repetitions.
[0172] In some examples, compressing the first output includes: performing a first reshaping operation on the first output that decreases a size of the first output in a first dimension and increases a size of the first output in a second dimension; and performing, after performing the first reshaping operation, a first linear operation that decreases the size of the first output in the second dimension by combining one or more first elements of the first output with one or more second elements of the first output. In some examples, compressing the second output includes: performing a second reshaping operation on the second output that decreases a size of the second output in a third dimension and increases a size of the second output in a fourth dimension; and performing, after performing the second reshaping operation, a second linear operation that decreases the size of the second output in the fourth dimension by combining one or more third elements of the second output with one or more fourth elements of the second output.
[0173] In some examples, to support performing the first reshaping operation, the data shaping component 945 is capable of, configured to, or operable to support a means for appending, prior to performing the first linear operation, a first element to the first output based on the size of the first output in the second dimension including an odd quantity of elements. In some examples, the first element may be associated with a parameter that is learnable by the neural network
[0174] In some examples, the control message component 955 is capable of, configured to, or operable to support a means for receiving, from the network entity, one or more control messages including one or more first parameters that configure the UE to obtain the CSI and one or more second parameters that configure the UE to transmit the message, where obtaining the first sequence is in response to receiving the one or more control messages.
[0175] In some examples, the neural network parameter component 960 is capable of, configured to, or operable to support a means for receiving, from the network entity, an indication of one or more first parameters associated with a reference neural network, where one or more second parameters of the neural network at the UE are based on the one or more first parameters, and where the CSI is compressed in accordance with the one or more second parameters.
[0176] In some examples, the neural network parameter component 960 is capable of, configured to, or operable to support a means for receiving, from the network entity, an indication of one or more parameters for the neural network at the UE, where the CSI is compressed in accordance with the one or more parameters.
[0177] In some examples, the one or more first neural network blocks and the one or more second neural network blocks are in accordance with a transformer neural network structure.
[0178] In some examples, the first procedure at the one or more first neural network blocks and the second procedure at the one or more second neural network blocks include one or more identical operations (e.g., one or more operations included in the first procedure at the one or more first neural network blocks are identical to one or more operations included in the second procedure at the one or more second neural network blocks) .
[0179] FIG. 10 shows a diagram of a system 1000 including a device 1005 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The device 1005 may be an example of or include components of a device 705, a device 805, or a UE 115 as described herein. The device 1005 may communicate (e.g., wirelessly) with one or more other devices (e.g., network entities 105, UEs 115, or a combination thereof) . The device 1005 may include components for bi-directional voice and data communications including components for transmitting and receiving communications, such as a communications manager 1020, an input / output (I / O) controller, such as an I / O controller 1010, a transceiver 1015, one or more antennas 1025, at least one memory 1030, code 1035, and at least one processor 1040. These components may be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more buses (e.g., a bus 1045) .
[0180] The I / O controller 1010 may manage input and output signals for the device 1005. The I / O controller 1010 may also manage peripherals not integrated into the device 1005. In some cases, the I / O controller 1010 may represent a physical connection or port to an external peripheral. In some cases, the I / O controller 1010 may utilize an operating system such as or another known operating system. Additionally, or alternatively, the I / O controller 1010 may represent or interact with a modem, a keyboard, a mouse, a touchscreen, or a similar device. In some cases, the I / O controller 1010 may be implemented as part of one or more processors, such as the at least one processor 1040. In some cases, a user may interact with the device 1005 via the I / O controller 1010 or via hardware components controlled by the I / O controller 1010.
[0181] In some cases, the device 1005 may include a single antenna. However, in some other cases, the device 1005 may have more than one antenna, which may be capable of concurrently transmitting or receiving multiple wireless transmissions. The transceiver 1015 may communicate bi-directionally via the one or more antennas 1025 using wired or wireless links as described herein. For example, the transceiver 1015 may represent a wireless transceiver and may communicate bi-directionally with another wireless transceiver. The transceiver 1015 may also include a modem to modulate the packets, to provide the modulated packets to one or more antennas 1025 for transmission, and to demodulate packets received from the one or more antennas 1025. The transceiver 1015, or the transceiver 1015 and one or more antennas 1025, may be an example of a transmitter 715, a transmitter 815, a receiver 710, a receiver 810, or any combination thereof or component thereof, as described herein.
[0182] The at least one memory 1030 may include random access memory (RAM) and read-only memory (ROM) . The at least one memory 1030 may store computer-readable, computer-executable, or processor-executable code, such as the code 1035. The code 1035 may include instructions that, when executed by the at least one processor 1040, cause the device 1005 to perform various functions described herein. The code 1035 may be stored in a non-transitory computer-readable medium such as system memory or another type of memory. In some cases, the code 1035 may not be directly executable by the at least one processor 1040 but may cause a computer (e.g., when compiled and executed) to perform functions described herein. In some cases, the at least one memory 1030 may include, among other things, a basic I / O system (BIOS) which may control basic hardware or software operation such as the interaction with peripheral components or devices.
[0183] The at least one processor 1040 may include one or more intelligent hardware devices (e.g., one or more general-purpose processors, one or more DSPs, one or more CPUs, one or more GPUs, one or more neural processing units (NPUs) (also referred to as neural network processors or deep learning processors (DLPs) ) , one or more microcontrollers, one or more ASICs, one or more FPGAs, one or more programmable logic devices, discrete gate or transistor logic, one or more discrete hardware components, or any combination thereof) . In some cases, the at least one processor 1040 may be configured to operate a memory array using a memory controller. In some other cases, a memory controller may be integrated into the at least one processor 1040. The at least one processor 1040 may be configured to execute computer-readable instructions stored in a memory (e.g., the at least one memory 1030) to cause the device 1005 to perform various functions (e.g., functions or tasks supporting CSI compression using neural networks) . For example, the device 1005 or a component of the device 1005 may include at least one processor 1040 and at least one memory 1030 coupled with or to the at least one processor 1040, the at least one processor 1040 and the at least one memory 1030 configured to perform various functions described herein.
[0184] In some examples, the at least one processor 1040 may include multiple processors and the at least one memory 1030 may include multiple memories. One or more of the multiple processors may be coupled with one or more of the multiple memories, which may, individually or collectively, be configured to perform various functions described herein. In some examples, the at least one processor 1040 may be a component of a processing system, which may refer to a system (such as a series) of machines, circuitry (including, for example, one or both of processor circuitry (which may include the at least one processor 1040) and memory circuitry (which may include the at least one memory 1030) ) , or components, that receives or obtains inputs and processes the inputs to produce, generate, or obtain a set of outputs. The processing system may be configured to perform one or more of the functions described herein. For example, the at least one processor 1040 or a processing system including the at least one processor 1040 may be configured to, configurable to, or operable to cause the device 1005 to perform one or more of the functions described herein. Further, as described herein, being “configured to, ” being “configurable to, ” and being “operable to”may be used interchangeably and may be associated with a capability, when executing code 1035 (e.g., processor-executable code) stored in the at least one memory 1030 or otherwise, to perform one or more of the functions described herein.
[0185] The communications manager 1020 may support wireless communications in accordance with examples as disclosed herein. For example, the communications manager 1020 is capable of, configured to, or operable to support a means for obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The communications manager 1020 is capable of, configured to, or operable to support a means for performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The communications manager 1020 is capable of, configured to, or operable to support a means for performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The communications manager 1020 is capable of, configured to, or operable to support a means for transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence.
[0186] By including or configuring the communications manager 1020 in accordance with examples as described herein, the device 1005 may support techniques for improved communication reliability, improved user experience related to reduced processing, reduced power consumption, more efficient utilization of communication resources, improved coordination between devices, longer battery life, and improved utilization of processing capability, among other benefits.
[0187] In some examples, the communications manager 1020 may be configured to perform various operations (e.g., receiving, monitoring, transmitting) using or otherwise in cooperation with the transceiver 1015, the one or more antennas 1025, or any combination thereof. Although the communications manager 1020 is illustrated as a separate component, in some examples, one or more functions described with reference to the communications manager 1020 may be supported by or performed by the at least one processor 1040, the at least one memory 1030, the code 1035, or any combination thereof. For example, the code 1035 may include instructions executable by the at least one processor 1040 to cause the device 1005 to perform various aspects of CSI compression using neural networks as described herein, or the at least one processor 1040 and the at least one memory 1030 may be otherwise configured to, individually or collectively, perform or support such operations.
[0188] FIG. 11 shows a flowchart illustrating a method 1100 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The operations of the method 1100 may be implemented by a UE or its components as described herein. For example, the operations of the method 1100 may be performed by a UE 115 as described with reference to FIGs. 1 through 10. In some examples, a UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally, or alternatively, the UE may perform aspects of the described functions using special-purpose hardware.
[0189] At 1105, the method may include obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The operations of 1105 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1105 may be performed by an input sequence component 925 as described with reference to FIG. 9.
[0190] At 1110, the method may include performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The operations of 1110 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1110 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0191] At 1115, the method may include performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The operations of 1115 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1115 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0192] At 1120, the method may include transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence. The operations of 1120 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1120 may be performed by an information output component 935 as described with reference to FIG. 9.
[0193] FIG. 12 shows a flowchart illustrating a method 1200 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The operations of the method 1200 may be implemented by a UE or its components as described herein. For example, the operations of the method 1200 may be performed by a UE 115 as described with reference to FIGs. 1 through 10. In some examples, a UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally, or alternatively, the UE may perform aspects of the described functions using special-purpose hardware.
[0194] At 1205, in some examples, the method may include obtaining CSI that includes a set of data that is associated with a first dimension. The operations of 1205 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1205 may be performed by a CSI obtaining component 940 as described with reference to FIG. 9.
[0195] At 1210, in some examples, the method may include performing one or more operations to restructure the set of data, where the set of data is associated with a second dimension that is compatible with a neural network at the UE after performing the one or more operations. The operations of 1210 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1210 may be performed by a data shaping component 945 as described with reference to FIG. 9.
[0196] At 1215, the method may include obtaining, at the UE, a first sequence that is associated with the CSI and includes a first quantity of elements, where the neural network at the UE is associated with compressing the CSI. The operations of 1215 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1215 may be performed by an input sequence component 925 as described with reference to FIG. 9.
[0197] At 1220, the method may include performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The operations of 1220 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1220 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0198] At 1225, the method may include performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The operations of 1225 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1225 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0199] At 1230, the method may include transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence. The operations of 1230 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1230 may be performed by an information output component 935 as described with reference to FIG. 9.
[0200] FIG. 13 shows a flowchart illustrating a method 1300 that supports CSI compression using neural networks in accordance with one or more aspects of the present disclosure. The operations of the method 1300 may be implemented by a UE or its components as described herein. For example, the operations of the method 1300 may be performed by a UE 115 as described with reference to FIGs. 1 through 10. In some examples, a UE may execute a set of instructions to control the functional elements of the UE to perform the described functions. Additionally, or alternatively, the UE may perform aspects of the described functions using special-purpose hardware.
[0201] At 1305, the method may include obtaining, at the UE, a first sequence that is associated with CSI and includes a first quantity of elements, where a neural network at the UE is associated with compressing the CSI. The operations of 1305 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1305 may be performed by an input sequence component 925 as described with reference to FIG. 9.
[0202] At 1310, the method may include performing a first procedure using the neural network to obtain a second sequence including a second quantity of elements that is less than the first quantity of elements, the first procedure including applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks. The operations of 1310 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1310 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0203] At 1315, the method may include performing a second procedure using the neural network to obtain a third sequence including a third quantity of elements that is less than the second quantity of elements, the second procedure including applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks. The operations of 1315 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1315 may be performed by a neural network component 930 as described with reference to FIG. 9.
[0204] At 1320, in some examples, the method may include performing, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, where the first sequence is associated with a first quantity of dimensions, and where the set of data is associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations. The operations of 1320 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1320 may be performed by a data shaping component 945 as described with reference to FIG. 9.
[0205] At 1325, the method may include transmitting a message including compressed CSI to a network entity, where the compressed CSI is associated with the third sequence. The operations of 1325 may be performed in accordance with examples as disclosed herein. In some examples, aspects of the operations of 1325 may be performed by an information output component 935 as described with reference to FIG. 9.
[0206] The following provides an overview of aspects of the present disclosure:
[0207] Aspect 1: A method for wireless communications by a UE, comprising: obtaining, at the UE, a first sequence that is associated with channel state information and comprises a first quantity of elements, wherein a neural network at the UE is associated with compressing the channel state information; performing a first procedure using the neural network to obtain a second sequence comprising a second quantity of elements that is less than the first quantity of elements, the first procedure comprising applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks; performing a second procedure using the neural network to obtain a third sequence comprising a third quantity of elements that is less than the second quantity of elements, the second procedure comprising applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks; and transmitting a message comprising compressed channel state information to a network entity, wherein the compressed channel state information is associated with the third sequence.
[0208] Aspect 2: The method of aspect 1, further comprising: obtaining the channel state information prior to obtaining the first sequence, wherein the channel state information comprises a set of data that is associated with a first dimension; and performing one or more operations to restructure the set of data, wherein the set of data is associated with a second dimension that is compatible with the neural network after performing the one or more operations.
[0209] Aspect 3: The method of any of aspects 1 through 2, further comprising: performing, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, wherein the first sequence is associated with a first quantity of dimensions, and wherein the set of data is associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations.
[0210] Aspect 4: The method of any of aspects 1 through 3, further comprising: performing a third procedure using the neural network to obtain a fourth sequence comprising a fourth quantity of elements that is less than the third quantity of elements, the third procedure comprising applying one or more third neural network blocks to the third sequence and compressing a third output of the one or more third neural network blocks, wherein the compressed channel state information is further associated with the fourth sequence.
[0211] Aspect 5: The method of any of aspects 1 through 4, wherein the one or more first neural network blocks are applied to the first sequence in accordance with a first quantity of repetitions; and the one or more second neural network blocks are applied to the second sequence in accordance with a second quantity of repetitions.
[0212] Aspect 6: The method of aspect 5, wherein the second quantity of repetitions is equal to the first quantity of repetitions.
[0213] Aspect 7: The method of any of aspects 5 through 6, wherein the second quantity of repetitions is different than the first quantity of repetitions.
[0214] Aspect 8: The method of any of aspects 1 through 7, wherein compressing the first output comprises performing a first reshaping operation on the first output that decreases a size of the first output in a first dimension and increases a size of the first output in a second dimension; and performing, after performing the first reshaping operation, a first linear operation that decreases the size of the first output in the second dimension by combining one or more first elements of the first output with one or more second elements of the first output; and compressing the second output comprises: performing a second reshaping operation on the second output that decreases a size of the second output in a third dimension and increases a size of the second output in a fourth dimension; and performing, after performing the second reshaping operation, a second linear operation that decreases the size of the second output in the fourth dimension by combining one or more third elements of the second output with one or more fourth elements of the second output.
[0215] Aspect 9: The method of aspect 8, wherein performing the first reshaping operation comprises: appending, prior to performing the first linear operation, a first element to the first output based at least in part on the size of the first output in the second dimension comprising an odd quantity of elements.
[0216] Aspect 10: The method of aspect 9, wherein the first element is associated with a parameter that is learnable by the neural network.
[0217] Aspect 11: The method of any of aspects 1 through 10, further comprising: receiving, from the network entity, one or more control messages comprising one or more first parameters that configure the UE to obtain the channel state information and one or more second parameters that configure the UE to transmit the message, wherein obtaining the first sequence is in response to receiving the one or more control messages.
[0218] Aspect 12: The method of any of aspects 1 through 11, further comprising: receiving, from the network entity, an indication of one or more first parameters associated with a reference neural network, wherein one or more second parameters of the neural network at the UE are based at least in part on the one or more first parameters, and wherein the channel state information is compressed in accordance with the one or more second parameters.
[0219] Aspect 13: The method of any of aspects 1 through 11, further comprising: receiving, from the network entity, an indication of one or more parameters for the neural network at the UE, wherein the channel state information is compressed in accordance with the one or more parameters.
[0220] Aspect 14: The method of any of aspects 1 through 13, wherein the one or more first neural network blocks and the one or more second neural network blocks are in accordance with a transformer neural network structure.
[0221] Aspect 15: The method of any of aspects 1 through 14, wherein the first procedure at the one or more first neural network blocks and the second procedure at the one or more second neural network blocks comprise one or more identical operations.
[0222] Aspect 16: A UE for wireless communications, comprising one or more memories storing processor-executable code, and one or more processors coupled with the one or more memories and individually or collectively operable to execute the code to cause the UE to perform a method of any of aspects 1 through 15.
[0223] Aspect 17: A UE for wireless communications, comprising at least one means for performing a method of any of aspects 1 through 15.
[0224] Aspect 18: A non-transitory computer-readable medium storing code for wireless communications, the code comprising instructions executable by one or more processors to perform a method of any of aspects 1 through 15.
[0225] It should be noted that the methods described herein describe possible implementations. The operations and the steps may be rearranged or otherwise modified and other implementations are possible. Further, aspects from two or more of the methods may be combined.
[0226] Although aspects of an LTE, LTE-A, LTE-A Pro, or NR system may be described for purposes of example, and LTE, LTE-A, LTE-A Pro, or NR terminology may be used in much of the description, the techniques described herein are applicable beyond LTE, LTE-A, LTE-A Pro, or NR networks. For example, the described techniques may be applicable to various other wireless communications systems such as Ultra Mobile Broadband (UMB) , Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi) , IEEE 802.16 (WiMAX) , IEEE 802.20, Flash-OFDM, as well as other systems and radio technologies not explicitly mentioned herein.
[0227] Information and signals described herein may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.
[0228] The various illustrative blocks and components described in connection with the disclosure herein may be implemented or performed using a general-purpose processor, a DSP, an ASIC, a CPU, a GPU, a neural processing unit (NPU) , an FPGA or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor but, in the alternative, the processor may be any processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices (e.g., a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration) . Any functions or operations described herein as being capable of being performed by a processor may be performed by multiple processors that, individually or collectively, are capable of performing the described functions or operations.
[0229] The functions described herein may be implemented using hardware, software executed by a processor, firmware, or any combination thereof. If implemented using software executed by a processor, the functions may be stored as or transmitted using one or more instructions or code of a computer-readable medium. Other examples and implementations are within the scope of the disclosure and appended claims. For example, due to the nature of software, functions described herein may be implemented using software executed by a processor, hardware, firmware, hardwiring, or combinations of any of these. Features implementing functions may also be physically located at various positions, including being distributed such that portions of functions are implemented at different physical locations.
[0230] Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one location to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer. By way of example, and not limitation, non-transitory computer-readable media may include RAM, ROM, electrically erasable programmable ROM (EEPROM) , flash memory, compact disk (CD) ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that may be used to carry or store desired program code means in the form of instructions or data structures and that may be accessed by a general-purpose or special-purpose computer or a general-purpose or special-purpose processor. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) , or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of computer-readable medium. Disk and disc, as used herein, include CD, laser disc, optical disc, digital versatile disc (DVD) , floppy disk, and Blu-ray disc. Disks may reproduce data magnetically, and discs may reproduce data optically using lasers. Combinations of the above are also included within the scope of computer-readable media. Any functions or operations described herein as being capable of being performed by a memory may be performed by multiple memories that, individually or collectively, are capable of performing the described functions or operations.
[0231] As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of” or “one or more of” ) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C) . Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on. ”
[0232] As used herein, including in the claims, the article “a” before a noun is open-ended and understood to refer to “at least one” of those nouns or “one or more” of those nouns. Thus, the terms “a, ” “at least one, ” “one or more, ” and “at least one of one or more” may be interchangeable. For example, if a claim recites “a component” that performs one or more functions, each of the individual functions may be performed by a single component or by any combination of multiple components. Thus, the term “a component” having characteristics or performing functions may refer to “at least one of one or more components” having a particular characteristic or performing a particular function. Subsequent reference to a component introduced with the article “a” using the terms “the” or “said” may refer to any or all of the one or more components. For example, a component introduced with the article “a” may be understood to mean “one or more components, ” and referring to “the component” subsequently in the claims may be understood to be equivalent to referring to “at least one of the one or more components. ” Similarly, subsequent reference to a component introduced as “one or more components” using the terms “the” or “said” may refer to any or all of the one or more components. For example, referring to “the one or more components” subsequently in the claims may be understood to be equivalent to referring to “at least one of the one or more components. ”
[0233] The term “determine” or “determining” encompasses a variety of actions and, therefore, “determining” can include calculating, computing, processing, deriving, investigating, looking up (such as via looking up in a table, a database, or another data structure) , ascertaining, and the like. Also, “determining” can include receiving (e.g., receiving information) , accessing (e.g., accessing data stored in memory) , and the like. Also, “determining” can include resolving, obtaining, selecting, choosing, establishing, and other such similar actions.
[0234] In the appended figures, similar components or features may have the same reference label. Further, various components of the same type may be distinguished by following the reference label by a dash and a second label that distinguishes among the similar components. If just the first reference label is used in the specification, the description is applicable to any one of the similar components having the same first reference label irrespective of the second reference label or other subsequent reference label.
[0235] The description set forth herein, in connection with the appended drawings, describes example configurations and does not represent all the examples that may be implemented or that are within the scope of the claims. The term “example” used herein means “serving as an example, instance, or illustration” and not “preferred” or “advantageous over other examples. ” The detailed description includes specific details for the purpose of providing an understanding of the described techniques. These techniques, however, may be practiced without these specific details. In some figures, known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described examples.
[0236] Also, as used herein, the phrase “a set” shall be construed as including the possibility of a set with one member. That is, the phrase “a set” shall be construed in the same manner as “one or more. ”
[0237] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.
Claims
1.A user equipment (UE) , comprising:one or more memories storing processor-executable code; andone or more processors coupled with the one or more memories and individually or collectively operable to execute the code to cause the UE to:obtain, at the UE, a first sequence that is associated with channel state information and comprises a first quantity of elements, wherein a neural network at the UE is associated with compressing the channel state information;perform a first procedure using the neural network to obtain a second sequence comprising a second quantity of elements that is less than the first quantity of elements, the first procedure comprising application of one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks;perform a second procedure using the neural network to obtain a third sequence comprising a third quantity of elements that is less than the second quantity of elements, the second procedure comprising application of one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks; andtransmit a message comprising compressed channel state information to a network entity, wherein the compressed channel state information is associated with the third sequence.2.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:obtain the channel state information prior to obtaining the first sequence, wherein the channel state information comprises a set of data that is associated with a first dimension; andperform one or more operations to restructure the set of data, wherein the set of data is associated with a second dimension that is compatible with the neural network after performing the one or more operations.3.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:perform, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, wherein the first sequence is associated with a first quantity of dimensions, and wherein the set of data is associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations.4.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:perform a third procedure using the neural network to obtain a fourth sequence comprising a fourth quantity of elements that is less than the third quantity of elements, the third procedure comprising application of one or more third neural network blocks to the third sequence and compressing a third output of the one or more third neural network blocks, wherein the compressed channel state information is further associated with the fourth sequence.5.The UE of claim 1, wherein:the application of the one or more first neural network blocks to the first sequence is in accordance with a first quantity of repetitions; andthe application of the one or more second neural network blocks to the second sequence is in accordance with a second quantity of repetitions.6.The UE of claim 5, wherein the second quantity of repetitions is equal to the first quantity of repetitions.7.The UE of claim 5, wherein the second quantity of repetitions is different than the first quantity of repetitions.8.The UE of claim 1, wherein:to compress the first output, the one or more processors are individually or collectively operable to execute the code to cause the UE to:perform a first reshaping operation on the first output that decreases a size of the first output in a first dimension and increases a size of the first output in a second dimension; andperform, after performing the first reshaping operation, a first linear operation that decreases the size of the first output in the second dimension by combining one or more first elements of the first output with one or more second elements of the first output; andto compress the second output, the one or more processors are individually or collectively operable to execute the code to cause the UE to:perform a second reshaping operation on the second output that decreases a size of the second output in a third dimension and increases a size of the second output in a fourth dimension; andperform, after performing the second reshaping operation, a second linear operation that decreases the size of the second output in the fourth dimension by combining one or more third elements of the second output with one or more fourth elements of the second output.9.The UE of claim 8, wherein, to perform the first reshaping operation, the one or more processors are individually or collectively operable to execute the code to cause the UE to:append, prior to performing the first linear operation, a first element to the first output based at least in part on the size of the first output in the second dimension comprising an odd quantity of elements.10.The UE of claim 9, wherein the first element is associated with a parameter that is learnable by the neural network.11.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:receive, from the network entity, one or more control messages comprising one or more first parameters that configure the UE to obtain the channel state information and one or more second parameters that configure the UE to transmit the message, wherein the one or more processors are individually or collectively operable to execute the code to cause the UE to obtain the first sequence in response to receiving the one or more control messages.12.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:receive, from the network entity, an indication of one or more first parameters associated with a reference neural network, wherein one or more second parameters of the neural network at the UE are based at least in part on the one or more first parameters, and wherein the one or more processors are individually or collectively operable to execute the code to cause the UE to compress the channel state information in accordance with the one or more second parameters.13.The UE of claim 1, wherein the one or more processors are individually or collectively further operable to execute the code to cause the UE to:receive, from the network entity, an indication of one or more parameters for the neural network at the UE, wherein the one or more processors are individually or collectively operable to execute the code to cause the UE to compress the channel state information in accordance with the one or more parameters.14.The UE of claim 1, wherein the one or more first neural network blocks and the one or more second neural network blocks are in accordance with a transformer neural network structure.15.The UE of claim 1, wherein:the first procedure at the one or more first neural network blocks and the second procedure at the one or more second neural network blocks comprise one or more identical operations.16.A method for wireless communications by a user equipment (UE) , comprising:obtaining, at the UE, a first sequence that is associated with channel state information and comprises a first quantity of elements, wherein a neural network at the UE is associated with compressing the channel state information;performing a first procedure using the neural network to obtain a second sequence comprising a second quantity of elements that is less than the first quantity of elements, the first procedure comprising applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks;performing a second procedure using the neural network to obtain a third sequence comprising a third quantity of elements that is less than the second quantity of elements, the second procedure comprising applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks; andtransmitting a message comprising compressed channel state information to a network entity, wherein the compressed channel state information is associated with the third sequence.17.The method of claim 16, further comprising:obtaining the channel state information prior to obtaining the first sequence, wherein the channel state information comprises a set of data that is associated with a first dimension; andperforming one or more operations to restructure the set of data, wherein the set of data is associated with a second dimension that is compatible with the neural network after performing the one or more operations.18.The method of claim 16, further comprising:performing, after performing the second procedure, one or more operations to restructure a set of data associated with the third sequence, wherein the first sequence is associated with a first quantity of dimensions, and wherein the set of data is associated with a second quantity of dimensions that is less than the first quantity of dimensions after performing the one or more operations.19.A non-transitory computer-readable medium storing code for wireless communications, the code comprising instructions executable by one or more processors to:obtain, at a user equipment (UE) , a first sequence that is associated with channel state information and comprises a first quantity of elements, wherein a neural network at the UE is associated with compressing the channel state information;perform a first procedure using the neural network to obtain a second sequence comprising a second quantity of elements that is less than the first quantity of elements, the first procedure comprising applying one or more first neural network blocks to the first sequence and compressing a first output of the one or more first neural network blocks;perform a second procedure using the neural network to obtain a third sequence comprising a third quantity of elements that is less than the second quantity of elements, the second procedure comprising applying one or more second neural network blocks to the second sequence and compressing a second output of the one or more second neural network blocks; andtransmit a message comprising compressed channel state information to a network entity, wherein the compressed channel state information is associated with the third sequence.20.The non-transitory computer-readable medium of claim 19, wherein the instructions are further executable by the one or more processors to:obtain the channel state information prior to obtaining the first sequence, wherein the channel state information comprises a set of data that is associated with a first dimension; andperform one or more operations to restructure the set of data, wherein the set of data is associated with a second dimension that is compatible with the neural network after performing the one or more operations.
Citation Information
Patent Citations
Large-scale MIMO channel state information feedback method based on FCFNN
CN109672464A
Channel state information omissions from channel state information reports
WO2024032088A1
CSI compression feedback method based on deep learning
WO2024045689A1