CSI feedback acquisition based on machine learning

Abstract

There is provided an apparatus and a corresponding method for machine learning based CSI feedback acquisition, the apparatus configured to perform operations comprising: indicating to a network node of a cellular communication network an ability to support machine learning ML based channel state information CSI feedback acquisition; receiving, from the network node, an indication of obtaining using ML-based CSI feedback; acquiring a CSI-RS (Channel State Information-Reference Signal) mode and a CSI feedback size threshold value, wherein the CSI-RS mode and the CSI feedback size threshold value are jointly determined; receiving one or more pilot signals for at least one channel according to the CSI-RS mode; inputting a result of at least one measurement channel on the one or more pilot signals into the ML model and acquiring CSI feedback as an output from the ML model; and sending the CSI feedback to the network node according to the CSI feedback size threshold.

Description

Technical Field

[0001] Various example embodiments relate generally to cellular communications. More specifically, various examples relate to machine learning (ML) based channel state information (CSI) feedback acquisition. Background Art

[0002] CSI feedback acquisition, in particular type IICSI feedback acquisition, may cause problems, for example, in terms of increased overhead. ML has been proposed to perform at least some tasks related to CSI feedback acquisition. However, there seems to be room for improvement in using machine learning-based processes to improve CSI feedback acquisition. Summary of the invention

[0003] According to some aspects, the subject matter of the independent claims is provided. Some further aspects are defined in the dependent claims. Embodiments not belonging to the scope of the claims are to be construed as examples that aid in understanding the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] In the following, the invention will be described in more detail with reference to embodiments and the accompanying drawings, in which:

[0005] Figure 1 A network is presented in which one or more embodiments are applicable;

[0006] Figure 2 The diagram shows the two-step CSI feedback process;

[0007] Figure 3 and Figure 4 illustrates a flow chart according to some embodiments;

[0008] Figure 5 illustrates a signaling diagram according to some embodiments;

[0009] Figure 6 and Figure 7 Some examples are illustrated; and

[0010] Figure 8 and Fig. 9 An apparatus according to some embodiments is illustrated. DETAILED DESCRIPTION

[0011] The following embodiments are exemplary. Although the specification may mention "an", "one" or "some" embodiments at several positions in the text, this does not necessarily mean that each mention refers to the same embodiment (multiple), or that a particular feature is only applicable to a single embodiment. The individual features of different embodiments may also be combined to provide other embodiments. For the purpose of this disclosure, phrases "at least one of A or B", "at least one of A and B" and "A and / or B" refer to (A), (B) or (A and B). For the purpose of this disclosure, phrases "A or B" and "A and / or B" refer to (A), (B) or (A and B). For the purpose of this disclosure, phrases "A, B and / or C" refer to (A), (B), (C), (A and B), (A and C), (B and C) or (A, B and C). As used herein, (multiple) A may refer to one or more A, and similarly, (multiple) B may also refer to one or more B.

[0012] It should be understood that although the terms "first" and "second" etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, without departing from the scope of the example embodiments, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.

[0013] The described embodiments may be implemented in a radio system, such as a radio system including at least one of the following radio access technologies (RATs): Worldwide Interoperability for Microwave Access (WiMAX), Global System for Mobile Communications (GSM, 2G), GSM EDGE Radio Access Network (GERAN), General Packet Radio Service (GRPS), Universal Mobile Telecommunications System (UMTS, 3G) based on basic Wideband Code Division Multiple Access (W-CDMA), High Speed ​​Packet Access (HSPA), Long Term Evolution (LTE), Advanced LTE, and Enhanced LTE (eLTE). The term "eLTE" herein means LTE evolution connected to a 5G core. LTE is also known as Evolved UMTS Terrestrial Radio Access (EUTRA), or Evolved UMTS Terrestrial Radio Access Network (EUTRAN). The term "resource" may refer to a radio resource, such as a physical resource block (PRB), a radio frame, a subframe, a time slot, a subband, a frequency region, a subcarrier, a beam, etc. The term "transmission" and / or "reception" may refer to wireless transmission and / or reception via a wireless propagation channel on a radio resource.

[0014] However, the embodiments are not limited to the systems / RATs given as examples, and those skilled in the art may apply the solution to other communication systems / networks with the necessary attributes. Some examples of suitable communication networks include 5G networks and / or 6G networks. The 3GPP solution for 5G is called New Radio (NR). 6G is considered to be a further development of 5G. NR may enable the use of multiple-input multiple-output (MIMO) multi-antenna transmission technology, with more base stations or nodes (the so-called small cell concept) than the current network deployment of LTE, including macro sites that work in conjunction with smaller local area network access nodes and may also adopt various radio technologies to achieve better coverage and enhanced data rates. 5G will likely consist of a variety of radio access technologies / radio access networks (RAT / RAN), each optimized for specific use cases and / or spectrum. 5G mobile communications may have a wider range of use cases and related applications, including video streaming, augmented reality, different data sharing methods, and various forms of machine-type applications, including vehicle safety, different sensors, and real-time control. It is expected that 5G will have a variety of radio interfaces, namely below 6GHz, cmWave, and mmWave, and can be integrated with existing traditional radio access technologies (such as LTE).

[0015] The current architecture in LTE networks is distributed in the radio and centralized in the core network. Low-latency applications and services in 5G may require content to be close to the radio, resulting in local bursts and multi-access edge computing (MEC). 5G enables analysis and knowledge generation at the data source. This approach requires the use of resources that may not be continuously connected to the network, such as laptops, smartphones, tablets, and sensors. MEC provides a distributed computing environment for application and service hosting. It also has the ability to store and process content near cellular subscribers to achieve faster response times. Edge computing covers a wide range of technologies such as wireless sensor networks, mobile data collection, mobile signature analysis, collaborative distributed peer-to-peer networking and processing, and can also be classified as local cloud / fog computing and grid / mesh computing, dew computing, mobile edge computing, cloudlet, distributed data storage and retrieval, autonomous self-healing networks, remote cloud services, augmented and virtual reality, data caching, Internet of Things (massive connectivity and / or latency critical), critical communications (autonomous driving cars, traffic safety, real-time analysis, time-critical control, healthcare applications). Edge cloud can be introduced into RAN by leveraging network function virtualization (NVF) and software defined networking (SDN). Using edge cloud can mean that access node operations are performed at least partially in a server, host or node that is operably coupled to a remote radio head or base station including radio components. Network slicing allows multiple virtual networks to be created on a shared physical infrastructure. The virtual networks are then customized to meet the specific needs of applications, services, devices, customers or operators.

[0016] In radio communications, node operations may be performed at least in part in a central / centralized unit CU (e.g., a server, host, or node), which is operably coupled to a distributed unit DU (e.g., a radio head / node). Node operations may also be distributed among multiple servers, nodes, or hosts. It should also be understood that the distribution of work between core network operations and base station operations may vary depending on the implementation. The network architecture of 5G and beyond may be based on, but not necessarily on, the so-called CU-DU split. One gNB-CU controls several gNB-DUs. In 5G, the term "gNB" may correspond to an eNB in ​​LTE. (One or more) gNBs may communicate with one or more UEs. A gNB-CU (central node) may control multiple spatially separated gNB-DUs to at least act as a transmit / receive (Tx / Rx) node. However, in some embodiments, the gNB-DU (also referred to as DU) may include, for example, a radio link control (RLC), a medium access control (MAC) layer, and a physical (PHY) layer, while the gNB-CU (also referred to as CU) may include layers above the RLC layer, such as a packet data convergence protocol (PDCP) layer, a radio resource control (RRC), and an Internet Protocol (IP) layer. Other functional splits are also possible. It is assumed that those skilled in the art are familiar with the OSI model and the functions within each layer.

[0017] In one embodiment, a server or CU can generate a virtual network through which the server communicates with the radio node. In general, a virtual network can involve the process of combining hardware and software network resources and network functions into a single software-based management entity (virtual network). Such a virtual network can provide a flexible distribution of operations between the server and the radio head / node. In practice, any digital signal processing task can be performed in the CU or DU, and the boundary of the responsibility transfer between the CU and the DU can be selected according to the implementation.

[0018] Some other possible technological advancements are software defined networking (SDN), big data, and all-IP, to name a few non-limiting examples. For example, network slicing can be a form of virtual network architecture that uses the same principles behind software defined networking (SDN) and network function virtualization (NFV) in fixed networks. SDN and NFV can provide greater network flexibility by allowing traditional network architectures to be segmented into virtual elements that can be linked (also through software). Network slicing allows multiple virtual networks to be created on a shared physical infrastructure. The virtual networks are then customized to meet the specific needs of applications, services, devices, customers, or operators.

[0019] Multiple gNBs (access points / nodes), each including a CU and one or more DUs, can be connected to each other via an Xn interface that the gNBs can negotiate. The gNBs can also be connected to the 5G core network (5GC) via a next generation (NG) interface, which can be the 5G equivalent of the core network of LTE. This 5G CU-DU split architecture can be implemented using a cloud / server, so that the high layers of the CU are located in the cloud and the DU is closer to, or includes, the actual radio and antenna units. There are similar plans for LTE / LTE-A / eLTE. When eLTE and 5G use similar architectures in the same cloud hardware (HW), the next step may be to combine software (SW) so that one common SW controls two radio access networks / technologies (RAN / RAT). This can enable new methods for controlling radio resources of two RANs. In addition, there may be a configuration in which the entire protocol stack is controlled by the same hardware as the CU and is processed by the same radio unit as the CU.

[0020] It should also be understood that the workload distribution between core network operations and base station operations may be different from LTE, or even non-existent. Some other technological advances that may be used are big data and all-IP, which can change the way networks are constructed and managed. 5G (or New Radio NR) networks are designed to support multiple hierarchical structures, where MEC servers can be placed between the core and the base station or nodeB (gNB). It should be understood that MEC can also be applied to 4G networks.

[0021] 5G can also make use of satellite communications to enhance or complement the coverage of 5G services, for example by providing backhaul services. Possible use cases are to provide service continuity for machine-to-machine (M2M) or Internet of Things (IoT) devices or passengers on board, or to ensure service availability for critical communications and future railway / maritime / aeronautical communications. Satellite communications can make use of geostationary orbit (GEO) satellite systems as well as low Earth orbit (LEO) satellite systems, in particular mega-constellations (systems in which hundreds of (nano) satellites are deployed). Each satellite in a mega-constellation can cover several satellite network entities creating ground cells. Ground cells can be created by ground relay nodes or gNBs located on the ground or in satellites.

[0022] These embodiments may also be applied to narrowband (NB) Internet of Things (IoT) systems, which may enable various devices and services to be connected using cellular telecommunication bands. NB-IoT is a narrowband radio technology designed for the Internet of Things (IoT) and is one of the technologies standardized by the Third Generation Partnership Project (3GPP). Other 3GPP IoT technologies that are also suitable for implementing embodiments include machine type communications (MTC) and eMTC (enhanced machine type communications). NB-IoT focuses on low cost, long battery life, and enabling a large number of connected devices. NB-IoT technology is deployed "in-band" in the spectrum allocated to Long Term Evolution (LTE), using resource blocks within a normal LTE carrier, or unused resource blocks within a guard band of an LTE carrier, or "standalone" in a dedicated spectrum.

[0023] These embodiments may also be applied to device-to-device (D2D), machine-to-machine, peer-to-peer (P2P) communications. These embodiments may also be applied to vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), infrastructure-to-vehicle (I2V), or generally to V2X or X2V communications.

[0024] Figure 1 An example of a communication system to which an embodiment of the present invention can be applied is illustrated. The system may include a control node 110 providing one or more cells (such as cell 100), and a control node 112 providing one or more other cells (such as cell 102). For example, each cell may be a macro cell, a micro cell, a femto cell, or a micro cell. From another perspective, a cell may define a coverage area or a service area of ​​a corresponding access node. The control nodes 110, 112 may be an evolved NodeB (eNB) in LTE and LTE-A, an ng-eNB in ​​eLTE, a gNB in ​​5G, or any other device capable of controlling radio communications within a cell and managing radio resources within a cell. The control nodes 110, 112 may be referred to as base stations, network nodes, or access nodes.

[0025] The system may be a cellular communication system consisting of a radio access network of access nodes, each access node controlling one or more corresponding cells. The access node 110 may provide wireless access to other networks (such as the Internet) to user equipment (UE) 120 (one or more UEs). The wireless access may include downlink (DL) communication from a control node to UE 120, and uplink (UL) communication from UE 120 to a control node.

[0026] In addition, although not shown, one or more local access nodes may be arranged so that the cell provided by the local access node overlaps at least partially with the cell of access node 110 and / or 112. The local access node may provide wireless access within a subcell. Examples of subcells may include microcells, picocells, and / or femtocells. Typically, a subcell provides a hotspot within a macrocell. The operation of a local access node may be controlled by an access node that provides a subcell under its control area. In general, the control node of a small cell may also be referred to as a base station, a network node, or an access node.

[0027] There may be one or more UEs 120, 122 in the system. The UE(s) 120, 122 may be served by one or more control nodes 110, 112. The UE(s) 120, 122 may communicate with each other, for example, using a D2D communication interface established between them. For example, D2D communication may refer to a sidelink (SL) communication, such as an NR sidelink communication.

[0028] The term "terminal equipment" or "UE" refers to any terminal device capable of wireless communication, more specifically cellular wireless communication. By way of example and not limitation, a terminal device may also be referred to as a communication device, user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS) or access terminal (AT). The terminal device may include, but is not limited to, a mobile phone, a cellular phone, a smart phone, a voice over IP (VoIP) phone, a wireless local loop phone, a tablet computer, a wearable terminal device, a personal digital assistant (PDA), a portable computer, a desktop computer, an image capture terminal device (such as a digital camera), a game terminal device, a music storage and playback device, a vehicle-mounted wireless terminal device, a wireless endpoint, a mobile station, a laptop embedded device (LEE), a laptop mounted device (LME), a USB dongle, a smart device, a wireless customer premises equipment (CPE), an Internet of Things (loT) device, a watch or other wearable device, a head mounted display (HMD), a vehicle, a drone, medical equipment and applications (e.g., remote surgery), industrial equipment and applications (e.g., robots and / or other wireless devices operating in an industrial and / or automated process chain environment), consumer electronic devices, devices operating on a commercial and / or industrial wireless network, etc. In the following description, the terms "terminal device", "communication device", "terminal", "user equipment" and "UE" may be used interchangeably.

[0029] In the case where there are multiple access nodes in the communication network, the access nodes can be connected to each other through an interface. The LTE specification refers to this interface as the X2 interface. For IEEE 802.11 networks (i.e., wireless local area networks WLAN (WiFi)), similar interfaces can be provided between access points. The interface between an LTE access point and a 5G access point, or between two 5G access points, can be referred to as Xn. Other communication methods between access nodes are also possible. Access nodes 110 and 112 can also be connected to the core network 116 of the cellular communication system via another interface. The LTE specification specifies the core network as an evolved packet core (EPC), and the core network may include a mobility management entity (MME) and a gateway node. The MME may handle the mobility of terminal devices in a tracking area including multiple cells, and handle the signaling connection between the terminal device and the core network. The gateway node may handle data routing in the core network and to / from the terminal device. The 5G specification specifies the core network as the 5G Core (5GC), and the core network may include, for example, the Access and Mobility Management Function (AMF) and the User Plane Function / Gateway (UPF), to name a few. The AMF may handle termination of non-access stratum (NAS) signaling, NAS encryption and integrity protection, registration management, connection management, mobility management, access authentication and authorization, and security context management. For example, the UPF node may support packet routing and forwarding, packet inspection, and QoS processing.

[0030] It is expected that networks (such as 5G Advanced and 6G) ​​will adopt flexible decentralized and / or distributed computing systems and architectures and ubiquitous computing, based on mobile edge computing, artificial intelligence (AI), machine learning (ML), short packet communications and blockchain technology to achieve local spectrum licensing, spectrum sharing, infrastructure sharing and intelligent automation management. The main features of 6G will include intelligent interconnect management and control functions, programmability, integrated sensing and communication, reduced energy footprint, trusted infrastructure, scalability and affordability. In addition to this, 6G also targets new use cases, which cover the integration of positioning and sensing functions into system definitions to unify the user experience of the physical and digital worlds.

[0031] AI / ML capabilities can be used for a variety of tasks in communication networks. Some of the use cases that 3GPP AI / ML research will focus on are:

[0032] - Initial set of use cases, including:

[0033] CSI feedback enhancements, e.g., reduced overhead, improved accuracy, prediction [RAN1]

[0034] Beam management, e.g., beam prediction in the time and / or spatial domain to reduce overhead and latency and achieve enhanced beam selection accuracy [RAN1]

[0035] Enhanced positioning accuracy in different scenarios, including those with severe NLOS conditions [RAN1]

[0036] - Finalize representative sub-use cases for each use case for characterization and baseline performance evaluation by RAN#98

[0037] The AI / ML approaches for the selected sub-use cases need to be diverse enough to support the various requirements for the gNB-UE collaboration level

[0038] For example, in 3GPP discussions, Type II channel state information (CSI) feedback overhead has been identified as an issue. To achieve good performance for Type I ICSI feedback, hundreds of feedback bits may need to be sent, which in turn consumes UL data bandwidth. In the 3GPP RAN1 AI / ML study project, CSI feedback overhead reduction is included as a priority use case.

[0039] For the downlink CSI feedback process, the network node may perform CSI-RS transmission that the UE may measure and provide feedback to the network node based on the CSI calculation (e.g., based on type IICSI). Based on the CSI feedback, the network node may perform downlink data scheduling for the UE. The UE reports CSI parameters to the access network node (network node, such as a gNB) as feedback. In general, the CSI may include parameters for multiple-input multiple-output (MIMO) scenarios, such as a channel quality indicator (CQI), a precoding matrix indicator (PMI), and / or a rank indicator (RI). After receiving the (multiple) CSI parameters, the gNB schedules downlink data transmission accordingly (such as modulation scheme, code rate, number of transmission layers, and MIMO precoding).

[0040] To improve CSI feedback, the feedback bits can be reduced while maintaining the same CSI feedback performance, or the feedback bits can be maintained while improving the CSI feedback performance. Both methods improve the trade-off between CSI feedback overhead and CSI feedback performance. CSI feedback performance can be measured by the mean square error (MSE) of channel recovery on the gNB side. As understood by those skilled in the art, channel recovery can represent how well the channel is recovered on the gNB side.

[0041] Figure 2 represents an example of a multi-step CSI feedback process that can be used in modern cellular networks. Figure 2As shown in , the CSI feedback process can be modeled as two steps: channel estimation 208 and CSI encoder 204, both of which can involve high complexity calculations, especially for the CSI encoder module, and when type IICSI feedback is configured to be reported. Type IICSI feedback requires the calculation of the eigendecomposition of the multi-antenna channel, and the eigendecomposition is known for being highly complex and resource consuming, although several commercial libraries have been working to improve efficiency. Box 202 shows channel decompression, which in principle can be understood as the inverse step of box 204. Box 206 represents CSI-RS pattern determination and can therefore be interpreted as a channel compression step.

[0042] In addition to the complexity issue, the two-step modeling of the CSI feedback process also violates the data processing theorem and deviates from the optimal trade-off between CSI feedback overhead and CSI feedback performance. The channel estimation module is regarded as a step of decompressing the channel on the entire TTI-PRB resource grid from the channel on several pilot REs using 12×14 REs, and the CSI encoder module is regarded as a step of compressing the channel on the entire TTI-PRB resource grid into the CSI feedback bit container. Compared with the one-step direct processing from the channel on the pilot RE to the feedback bit, this two-step decompression and compression process loses information.

[0043] In summary, the traditional two-step CSI feedback process can be improved in consideration of both CSI feedback performance (eg, improving channel recovery performance under a fixed number of feedback bits) and reduction of computational complexity.

[0044] Therefore, a solution is provided for improving the CSI feedback process by directly quantizing the channel on the pilots using an ML method to derive the CSI feedback bits.

[0045] Figure 3 and Figure 4 FIG. 1 is a flow chart illustrating a proposed solution according to some embodiments. Figure 3 , a method for an apparatus for a radio access network (RAN) such as a cellular communication network is proposed. The apparatus may be a UE, or be included in a UE. The UE may be, for example, UE 120 or UE 122.

[0046] Figure 3The method may include: indicating to a network node of a cellular communication network a capability to support machine learning (ML) based channel state information (CSI) feedback acquisition (box 302); receiving an indication from the network node to use ML based CSI feedback acquisition (box 304); obtaining a CSI reference signal (CSI-RS) pattern and a CSI feedback size threshold (box 306); receiving one or more pilot signals on at least one channel according to the CSI-RS pattern (box 308); inputting the results of at least one measurement channel on the one or more pilot signals into an ML model, and obtaining CSI feedback as an output from the ML model (box 310); and sending CSI feedback to the network node according to the CSI feedback size threshold (box 312).

[0047] The apparatus may measure the received one or more pilot signals to obtain a measurement channel on the one or more pilot signals. The measurement channel may be input into the ML model to obtain CSI feedback. At least one channel is carried on the one or more pilot signals, and the one or more pilot signals are used for channel acquisition. Thus, the result of at least one measurement channel may be obtained using the one or more pilot signals.

[0048] refer to Figure 4 , a method for an apparatus for a radio access network (RAN) such as a cellular communication network is proposed. The apparatus may be a network node, or be included in a network node. The network node may be, for example, the network node 110 or the network node 112.

[0049] Figure 4 The method may include receiving an indication of a capability to support machine learning (ML) based channel state information (CSI) feedback acquisition from a user equipment (e.g., UE 120, 122) of a cellular communication network (block 402). The receiving step may correspond to the indicating step of block 302.

[0050] The method may include: based on the received indication of box 402, determining to configure ML-based CSI feedback acquisition for the user equipment (box 404); sending an indication to the user equipment to use ML-based CSI feedback acquisition (box 406); and receiving CSI feedback obtained according to a CSI reference signal CSI-RS pattern and a CSI feedback size threshold from the user equipment (box 408).

[0051] In order to efficiently send CSI feedback, a UE (e.g., UE 120) may obtain uplink resources for sending CSI feedback. For example, the uplink resources may be obtained from a network node. CSI feedback may be understood to be sent in one or more CSI feedback containers. According to one embodiment, a CSI feedback size threshold indicates and / or configures a CSI feedback container size. That is, the CSI feedback size threshold may define the size of the CSI feedback container. The CSI feedback size threshold may be indicated as a number of bits. Therefore, the CSI feedback container size may indicate the number of bits of the CSI feedback (i.e., the number of CSI feedback bits). For example, the CSI feedback container size may indicate the maximum number of bits that fit into the container (i.e., the CSI feedback bits).

[0052] If the CSI feedback container size is not large enough to send CSI feedback (e.g., more bits are needed), the CSI feedback process may become inefficient. Therefore, according to the proposed solution, the CSI-RS pattern (abbreviated as P) and the CSI feedback size threshold (abbreviated as L) can be jointly determined. Therefore, at least the CSI feedback size threshold has a probability of being of a suitable size (i.e., not too large to waste resources, nor too small to report all CSI feedback bits). The union herein may mean that the CSI-RS pattern and the CSI feedback size threshold are determined by the same device (e.g., UE or network node) and are determined at least at substantially the same time. For example, the CSI-RS pattern and the CSI feedback size threshold may be determined at the same time. Therefore, it can be understood that the CSI-RS pattern and the CSI feedback size threshold are linked to each other, for example, they can be linked to each other so that when determining another parameter, one parameter or the parameter itself can be considered for determination.

[0053] Determining the CSI-RS pattern and the CSI feedback size threshold may be based on an estimate or measurement of radio channel characteristics. The radio channel here may refer to a channel used to send CSI-RS according to the CSI-RS pattern. For example, for a specific CSI-RS pattern with specific channel characteristics (or radio conditions), a certain number of CSI feedback bits may be required. Therefore, the CSI feedback size threshold may be determined so that the above-mentioned specific number of CSI feedback bits may be suitable for being used to send CSI feedback from the UE (e.g., 120) to the network node (e.g., 110) CSI feedback container.

[0054] Now let's look at Figure 5 , Figure 5 A signaling diagram according to some example embodiments is illustrated. According to one embodiment, reference Figure 5 , Figure 3 The indication of block 302 may be performed using radio resource control (RRC) signaling. Thus, the indication may be an RRC indication, or be included in an RRC message. Figure 5 500. For example, a capability report 502 according to RRC signaling may be sent from the UE 120 to the network node 110, wherein the capability report 502 may indicate that the UE 120 supports, or is capable of supporting, ML-based CSI feedback acquisition.

[0055] ML-based CSI feedback acquisition may refer to, for example, direct quantized CSI feedback, where the measurement channel on the CSI-RS is directly input into the ML model to obtain CSI feedback. Therefore, there is no need for, for example, a two-step CSI feedback process. The ML model can be trained to directly provide CSI feedback based on the measurement channel on the CSI-RS. This provides significant benefits by reducing the steps required to obtain CSI feedback (or merging the two steps into one). An example of an ML model structure is a convolutional neural network (CNN). Therefore, for example, the ML model used may be a CNN model.

[0056] In one embodiment, UE 120 directly inputs the result of at least one measurement channel on (multiple) CSI-RS into the ML model. The result of the CSI-RS or at least one measurement channel on the CSI-RS may be directly input into the ML model without decompression.

[0057] Network node 110 may receive the capability report during RRC establishment 500 and determine to configure UE 120 for ML-based CSI feedback acquisition (eg, direct quantized CSI feedback). In step 504, network node 110 may send an indication to UE 120 to use ML-based CSI feedback acquisition.

[0058] Going back a step, according to one embodiment, the capability report 502 indicates whether the UE 120 supports ML-based CSI feedback acquisition (eg, whether the UE 120 supports direct quantized CSI feedback).

[0059] In addition, according to one embodiment, the UE 120 may indicate the CSI-RS pattern (P) and the CSI feedback size threshold (L) to the network node 110. That is, the (L, P) pair. The indicated (L, P) pair may be based on historical information, such as the (L, P) pair of the UE 120 from an earlier or previous session. For example, the UE 120 may store one or more (L, P) pairs from (multiple) previous sessions in its memory. The indicated (L, P) pair may be understood as a recommendation or suggestion of the UE 120 for the (L, P) pair of the current session. According to one embodiment, the (L, P) pair is included in or indicated by the capability report 502. However, depending on the implementation, the UE 120 may send the (L, P) pair separately from the indication that the UE 120 is capable of ML-based CSI feedback acquisition. In addition, the (L, P) pair indication may at least be beneficial for saving computing resources, but is not always necessary, because the (L, P) pair may be determined and configured by, for example, the network node 110.

[0060] In one embodiment, UE 120 indicates a plurality of (L, P) pairs. For example, each indicated pair may be a pair used by UE 120 in an earlier or previous session(s).

[0061] In one embodiment, if UE 120 supports ML-based CSI feedback acquisition, the (L, P) pair is indicated by UE 120. In one embodiment, the (L, P) pair indication may be an implicit indication that UE 120 supports ML-based CSI feedback acquisition.

[0062] In one embodiment, if UE 120 does not support ML-based CSI feedback acquisition, UE 120 may not indicate the (L, P) pair. In this case, UE 120 may indicate to network node 110 that it does not support ML-based CSI feedback acquisition. For example, such an indication may indicate that UE 120 supports two-step CSI feedback acquisition.

[0063] In one embodiment, the indication of whether the UE 120 supports ML-based CSI feedback acquisition (e.g., whether the UE 120 supports direct quantization CSI feedback) is a bit indicator including one or more bits. In one example, the bit indicator is a one-bit indicator, where 0 or 1 indicates that the UE 120 supports ML-based CSI feedback acquisition, and the other of 0 and 1 indicates that the UE does not support ML-based CSI feedback acquisition. For example, 1 may indicate support for ML-based CSI feedback acquisition (e.g., direct quantization CSI feedback), and 0 may indicate that ML-based CSI feedback acquisition (e.g., direct quantization CSI feedback) is not supported, or that the UE only supports one or more traditional CSI feedback processes. For example, the UE 120 may determine whether it supports ML-based CSI feedback acquisition, and generate an indicator based on the determination to indicate the result of the determination. Subsequently, the indicator (e.g., 0 / 1, or any other suitable indicator) may be sent to the network node, for example, as shown in step 502.

[0064] In addition, according to one embodiment, UE 120 may indicate a CSI feedback recovery method to network node 110. For example, UE 120 may indicate a CSI feedback recovery method used by UE 120 in an earlier or previous session. The CSI feedback recovery method indication may be understood as a suggestion or recommendation from UE 120 to network node 110. In one embodiment, capability report 502 includes a CSI feedback recovery method indication. However, the CSI feedback recovery method indication may be performed separately depending on the implementation.

[0065] Therefore, depending on the implementation and whether the UE 120 supports ML-based CSI feedback recovery methods, the capability report 502 may indicate whether the UE 120 supports (multiple) ML-based CSI feedback and / or (L,P) pair recommendations and / or CSI feedback recovery method recommendations.

[0066] Based on the indication(s) in step 502, the network node 110 may determine to configure the UE 120 for CSI feedback. Based on the UE 120 indicating that it supports ML-based CSI feedback acquisition, the network node 110 may determine to configure the UE for ML-based CSI feedback acquisition. Otherwise, the network node 110 may configure the UE 120 for, for example, two-step CSI acquisition according to conventional operations and, for example, as defined in 3gpp.

[0067] For example, when UE 120 indicates to network node 110 during RRC establishment 500 (or RRC establishment phase) that it supports ML-based CSI feedback acquisition process, network node 110 may indicate to UE 120 which CSI feedback process is applied, whether it is conventional CSI feedback acquisition or ML-based CSI feedback acquisition (e.g., direct quantized CSI feedback). Such indication may be made by a one-bit indicator: for example, when "1" is indicated, UE 120 applies ML-based CSI feedback acquisition, and when "0" is indicated, UE 120 applies conventional CSI feedback process. In step 504, such indication (one bit or some other type of indicator) may be sent. In some examples, the indication may be understood as part of ML-based CSI feedback acquisition configuration, as shown in step 504.

[0068] Furthermore, in some embodiments, the network node 110 may indicate the (L, P) pairs and / or the CSI recovery method to the UE 120 in step 504 (i.e., during the RRC establishment 500). The CSI recovery method may indicate how the network node 110 recovers the CSI feedback bits and reconstructs the channel for scheduling DL transmissions. For example, the network node 110 may use the indicated CSI recovery method.

[0069] In one embodiment, the network node 110 indicates the CSI recovery method to be used by the network node 110 among a plurality of predefined CSI recovery methods. For example, the CSI recovery methods and corresponding indicators may be defined in (multiple) specifications. Thus, by sending an indicator corresponding to the desired CSI recovery method, the above CSI recovery method may be indicated to the UE 120. For example, an RRC message field may be used for the indication (i.e., RRC indicator). Thus, the UE may determine which CSI recovery method the network node 110 uses.

[0070] In one embodiment, the network node 110 may determine the CSI recovery method and / or (L, P) pair based at least on the recommendation from the UE 120 (see step 502). Additionally or alternatively, determining the CSI recovery method and / or (L, P) pair may be based on historical information about one or more other UEs. For example, the network node 110 may have participated in (multiple) sessions with the one or more other UEs, and may utilize information about the CSI recovery method and / or (L, P) pair from the (multiple) sessions.

[0071] Indication of a CSI recovery method or use of multiple CSI recovery methods may not always be necessary, as the UE 120 may be configured to use a default CSI feedback recovery method (e.g., the same or similar method as used in Type IICSI feedback acquisition). For conventional CSI feedback recovery methods, the proposed direct quantized CSI feedback is effective, as the ML-based approach can adapt the CNN to any predefined recovery method through a training process. However, utilizing (multiple) simplified recovery methods may be beneficial, as they can potentially reduce computational complexity while maintaining channel recovery performance. One such example is applying feedback bits (i.e., CSI feedback bits) directly as CSI.

[0072] In the case where ML-based CSI feedback acquisition is indicated to the UE 120, the network node 110 may indicate the CSI feedback size threshold, and the associated CSI-RS pattern (i.e., (L, P) pair) at the same or different times. The (L, P) pair is jointly determined. In the conventional CSI feedback process, the CSI feedback size and the CSI-RS pattern are configured and determined separately. For example, the number of CSI feedback bits is implicitly configured with the number of feedback beams and the corresponding co-phase quantization bit width. However, for direct quantized CSI feedback, only one number of total feedback bits (i.e., L) may be configured.

[0073] For example, in step 504, the network node 110 may signal L and P together to the UE 120. The (L, P) pair may be determined by the network node 110 based on the selected CSI feedback recovery method and the expected channel recovery performance (which may be represented by, for example, mean square error (MSE)). For example, for one CSI feedback recovery method, several (L, P) pairs may be defined in a joint coding manner to achieve different channel recovery MSEs while saving signaling overhead.

[0074] In one embodiment, the (L, P) pairs are mapped with corresponding indicators (such as indices). For example, in step 504, the network node may indicate the (L, P) pairs to be used by sending the corresponding indicators to the UE 120, for example as RRC information elements. The UE 120 may then determine the (L, P) pairs to be used based on the mapping and the received (L, P) pair indicators. Such mapping may be implemented, for example, by a table that indexes the (L, P) pairs. For example, the table may be defined in one or more specifications, such as (multiple) 3gpp specifications.

[0075] For example, a default configuration of (L, P) pairs may be calculated by the network node 110 using a conventional type IICSI feedback process with a maximum possible CSI feedback container size, and a CSI-RS pattern with a maximum possible number of pilot resource elements (REs) (or CSI-RS). This may guarantee CSI feedback performance at the outset. However, as discussed later, in this case, the (L, P) pair indication by the network node 110 to the UE 120 may be understood as an initial (L, P) pair indication, and the subsequent step(s) in the process may enable the UE 120 to determine a better (L, P) pair, which indeed enables CSI feedback acquisition to work more efficiently (e.g., fewer feedback bits or fewer CSI-RS pilots).

[0076] In addition, in some embodiments, the network node 110 provides one or more parameters of the ML model for determining CSI feedback to the UE 120. For example, (multiple) CNN parameters. For example, the above (multiple) parameters may be sent in step 504. For example, in the case where the UE 120 has multiple ML models that it can use, the (multiple) parameters may configure the ML model and / or indicate which ML model is to be used.

[0077] In block 506, the network node 110 may send (multiple) CSI-RS according to the CSI-RS pattern, and the UE 120 may monitor and receive the CSI-RS according to the CSI-RS pattern. The UE 120 may then provide CSI feedback to the network node 110. As described herein, the CSI feedback may be acquired using ML-based CSI acquisition (e.g., direct quantization CSI feedback). The CSI feedback may be reported to the network node according to constraints set by a CSI feedback size threshold.

[0078] Direct quantized CSI feedback can work with the proposed initial (L, P) pairs obtained, for example, from the network node 110, but by providing an enhanced method to determine the (L, P) pairs, CSI feedback acquisition can be further improved. Figure 5 , one example of such a process performed by UE 120 is shown. However, it should be noted that the (L, P) pair determination may instead be performed at network node 110 or some other network node, and the results may be provided to UE 120.

[0079] Reference now Figure 5 , in block 508, UE 120 performs data-assisted channel estimation. At this point, note that performing, for example, step 508 may not always require performing steps 502-506, although obtaining an initial (L, P) pair from network node 110 prior to performing step 508 may be beneficial.

[0080] In one embodiment, the data-aided channel estimation includes measuring a channel between the UE 120 and the network node 110. The channel may be a channel used to transmit the CSI-RS.

[0081] In one embodiment, data-assisted channel estimation includes measuring, by UE 120, a full channel grid of a channel between the UE and the network node. In simple terms, this may mean that UE 120 measures the entire channel, rather than measuring only the CSI-RS according to the CSI-RS pattern. For example, the measurement of the full channel may be performed in block 506. Thus, in some examples, block 508 may be part of block 506.

[0082] Therefore, the training / testing data in the proposed ML model (e.g., CNN) based method is the UE channel on the full transmission time interval physical resource block (TTI-PRB) resource grid. The full TTI-PRB resource grid can be obtained using data-assisted channel estimation on the UE side or in combination with UL-DL reciprocity on the network node 110 side. Data-assisted channel estimation is a feature for improving channel estimation performance. When a transport block is successfully decoded, the decoded data RE can be regarded as a pilot. Using the decoded data RE and reference signal RE, the channel frequency response of the entire transport block can be obtained, thereby obtaining the channel on the entire TTI PRB grid.

[0083] In block 509, the UE may determine a CSI-RS pattern and a CSI feedback size threshold based on data-assisted channel estimation. That is, data-assisted channel estimation performed by UE 120. Block 509 may include ML model training to determine the CSI-RS pattern and the CSI feedback size threshold. In an example embodiment, the UE determines an optimized or optimal CSI-RS pattern and CSI feedback size threshold based on data-assisted channel estimation. For example, model training may include determining a ground truth of a channel based on data-assisted channel estimation, and then selecting a CSI-RS pattern and a CSI-RS feedback size threshold by comparing the MSEs of different CSI-RS pattern and CSI feedback size threshold pairs. The MSE for a particular (L, P) pair may be obtained by comparing a channel estimate taken using the above pair with the ground truth of the channel.

[0084] Then, let us look at some examples on how to perform the determination of the CSI-RS pattern and the CSI feedback size threshold simultaneously.

[0085] The results of the measured full channel or full channel grid can be used as training data for the ML model to determine the (L, P) pair. For example, the UE 120 can use the ML model and training data for CSI feedback acquisition to determine multiple CSI feedbacks, where each of the multiple CSI feedbacks determined is associated with a corresponding CSI-RS pattern and a CSI feedback size threshold pair. An example of this is shown in Figure 6 is shown in Figure 6 One embodiment is illustrated.

[0086] refer to Figure 6 , the measured channel grid is shown with reference numeral 600. The measured channel grid may be understood as training data. As described above, the measured channel grid may be a measured full channel grid. However, in some examples or implementations, the measured channel grid may refer to a measured channel on resources (e.g., REs) including (multiple) CSI-RS indicated by a CSI-RS pattern, and some other resources. Thus, for example, substantially all resources of a channel may be measured, but not necessarily all.

[0087] In one embodiment, measuring the channel grid includes measuring the entire channel grid.

[0088] A plurality of different CSI-RS patterns (e.g., P1, P2, and P3) may be utilized by UE 120 to obtain different patterns from the measured channel grid. These patterns are shown with reference numerals 602, 604, and 606. That is, when the channel grid 600 has been measured, different REs of the measured channel grid 600 may be obtained and used as input to the ML model 610.

[0089] like Figure 6 As shown in , each CSI-RS pattern may be associated with a corresponding CSI feedback size threshold (L1, L2, L3). As described above, the CSI feedback size threshold and the CSI-RS pattern may form a pair, such as L1, P1; L2, P2; and L3, P3.

[0090] The CSI feedback size threshold and CSI-RS pattern may be determined by the UE 120. This determination may be based on UE implementation, information from the network node 110, and / or reconfiguration. For example, different resources selected based on the patterns 602, 604, 606 may be input into the ML model 610, and CSI feedback may be obtained as an output. Different (L, P) pairs may provide different MSEs. For example, Figure 6As shown in , the (L1, P1) pair can be associated with MSE1, the (L2, P2) pair can be associated with MSE2, and the (L3, P3) pair can be associated with MSE3. That is, using the associated measurement channel (i.e., measurement resource) as an input to the ML model 610 (e.g., CNN), CSI feedback obtained using different (L, P) pairs can provide different CSI feedback, and these CSI feedbacks can be compared with the measurement channel to determine which (L, P) pair used provides the lowest MSE. For example, the MSE can be determined based on comparing the currently determined CSI feedback with the measurement channel 600. For example, the UE 120 can select a measurement channel (characterizing the channel) obtained using data-assisted channel estimation as the ground truth. The UE 120 can use one or more (L, P) pairs to obtain one or more CSI feedback (i.e., (multiple) channel recovery). The UE 120 can then compare the corresponding CSI feedback with the ground truth measurement channel and determine the MSE of the corresponding CSI feedback based on the comparison. Thus, the MSE of one or more (L, P) pairs may be obtained. For example, UE 120 may select the (L, P) pair associated with the lowest MSE among the (multiple) MSEs determined based on comparing the ground truth channel with the channel recovery using the one or more (L, P) pairs.

[0091] Therefore, UE 120 can obtain CSI-RS pattern and CSI feedback size threshold based on using data-assisted channel estimation and ML model training, as shown in steps 508 and 509, and refer to Figure 6 Explanation. That is, UE 120 may compare the determined multiple CSI feedbacks and select a CSI-RS pattern and CSI feedback size threshold pair based on the comparison. That is, UE 120 may, for example, select the pair associated with the lowest estimated MSE. As discussed, the (L, P) pair-specific MSE for each pair may be determined by comparing the CSI feedback with the ground truth of the measured channel.

[0092] For example, several (L, P) pairs may be specified according to the conventional CSI feedback container size and the conventional CSI-RS pattern in the current standard. For each of the (L, P) pairs, a CNN may be constructed. The training and testing of these CNNs may be performed in parallel. For example, a pair (the weighted sum of the number of CSI RS pilot REs and the CSI feedback container size) that meets the MSE requirement and has the lowest overhead may be selected. Note that the CNN used to directly quantize the CSI feedback may be relatively simple, with only a few layers, so that training and testing may be completed quickly. It should also be noted that this trial and selection method may also be applied if a CSI recovery method other than the conventional method is defined and configured.

[0093] The determined (i.e., determined by the UE 120) CSI-RS pattern and the CSI feedback size threshold may be signaled to the network node 110 from the UE 120, or at least indicated to the network node 110 by the UE 120. For example, and as discussed in more detail below, the indication of the CSI-RS pattern and the CSI feedback size threshold may be performed via RRC signaling (such as via RRC reconfiguration).

[0094] Note again that the determination of the CSI-RS pattern and the CSI feedback size threshold may be performed by the network node 110 similarly as explained with reference to the UE 120 determining the CSI-RS pattern and the CSI-RS feedback size threshold.

[0095] In some examples, UE 120 receives from network node 110 a CSI-RS pattern and a CSI feedback size threshold to be used (e.g., in steps 308 and 312) to obtain CSI feedback that will be sent to network node 110. For example, the CSI-RS pattern and the CSI-RS feedback size threshold may be received after and / or based on indicating the determined or selected CSI-RS pattern and the CSI feedback size threshold to network node 110. On the other hand, if the CSI-RS pattern and the CSI feedback size threshold are determined at network node 110 using the method explained with reference to steps 308, 309, such indication by UE 120 may not be necessary.

[0096] In the event that the UE 120 indicates the determined and / or selected CSI-RS pattern and CSI feedback size threshold to the network node 110, the network node 110 may utilize the indicated CSI-RS pattern and CSI feedback size threshold when determining the CSI-RS pattern and CSI feedback size threshold to be used by the UE 120 to determine CSI feedback. In one example, the network node 110 may confirm the determined and / or selected CSI-RS pattern and CSI feedback size threshold to the UE 120. Regardless, in at least some examples, the network node 110 may be able to configure the CSI-RS pattern and CSI feedback size threshold to the UE 120 based on its determination. This may be beneficial because both the UE 120 and the network node 110 may be aware of the parameters L and P to be used for CSI feedback using the direct quantization CSI feedback process.

[0097] Reference now Figure 5In one embodiment, the CSI-RS pattern and CSI feedback size threshold determined and / or selected by UE 120 based on data-assisted channel estimation are sent to network node 110 using RRC signaling (step 552). For example, RRC reconfiguration 550 signaling may be used. Network node 110 may receive the RRC signaling of step 552.

[0098] In step 554, the network node 110 may respond by sending the above-mentioned CSI-RS pattern to be used and the CSI feedback size threshold to the UE 120. This may be performed based on and / or in response to the RRC signaling of step 552.

[0099] In one embodiment, step 554 further includes: the network node 110 indicating the CSI feedback recovery method to the UE 120 .

[0100] UE 120 may utilize the information received from the network node in step 554 and perform a CSI feedback process based on the information. For example, this may mean that UE 120 measures the CSI-RS sent by network node 110 according to the CSI-RS pattern, performs direct quantization using, for example, CNN, and reports the result (i.e., CSI feedback) to network node 110 while taking into account the CSI feedback container size. Similarly, network node 110 may perform transmission of CSI-RS according to the above-mentioned CSI-RS pattern and may also receive CSI feedback from UE 120 on a given resource. Network node 110 may also utilize CSI feedback, for example, in scheduling decision(s).

[0101] Simulation results show that under the same CSI-RS pattern and the same number of feedback bits, the performance of direct quantized CSI feedback using CNN can be better than the traditional two-step CSI feedback process. After the CNN is trained, only sequential addition and multiplication operations are required in the inference phase. Therefore, direct quantized CSI feedback can also reduce the complexity of CSI feedback calculation.

[0102] In the current solution, the CSI-RS pattern and CSI feedback bit width are configured independently, just like in the conventional approach, and the intermediate variable ( Figure 2 Recovery Channel ) separates the two modules, channel estimation and CSI encoder, so there is no performance loss in doing so. For direct quantized CSI feedback, there is no such intermediate variable to separate or orthogonalize the two steps of decompression (channel estimation) and compression (CSI encoder). Instead, a one-step process is applied. Therefore, they are highly related, and it is beneficial to determine / configure both together: CSI-RS pattern (number of CSI-RS REs and their distribution in the TTI-PRB resource grid) and CSI feedback container size (total number of feedback bits; also known as CSI feedback size threshold). Based on this observation, it is proposed in this article to jointly configure CSI-RS pattern and CSI feedback container size (i.e., bit width).

[0103] Figure 7 Represents the ground truth CSI feedback E, the traditional CSI feedback E' and the proposed direct quantized CSI feedback This example should be understood as an example showing how direct quantized CSI feedback compares with ground truth CSI feedback and / or traditional CSI feedback.

[0104] If the ground truth CSI feedback E is optimally measured without any error or loss, the ground truth CSI feedback E can be understood to represent the channel.

[0105] The conventional CSI feedback E' may be understood to represent CSI feedback obtained using a two-step CSI feedback process (eg, a type IICSI feedback process).

[0106] CSI Feedback It can be understood as CSI feedback obtained using the methods and processes described in this document. For example, it can refer to directly quantized CSI feedback.

[0107] exist Figure 7 , H (also shown with reference numeral 702) represents the ground truth channel over the entire TTI-PRB resource grid. The ground truth CSI feedback E can be obtained as follows: Starting from the ground truth channel H, eigendecomposition can be used to derive eigenvectors of the channel. These eigenvectors can be used as training / testing labels for the ground truth CSI feedback.

[0108] The traditional CSI feedback E' can be obtained using a two-step process 706. Starting from the ground truth channel H, the channel H is sampled with a pilot RE based on the configured CSI-RS pattern (see reference numeral 704). Then, interpolation is used to recover the sampled channel, and the recovered channel H' (708) can be obtained. This can correspond to the first step in the traditional CSI feedback process, i.e., the channel estimation step. After this, we input the recovered channel H' into the eigendecomposition module 710, and the eigenvectors can be calculated to simulate the CSI feedback of the traditional process. This can correspond to the second step: CSI encoder.

[0109] In direct quantized CSI feedback, instead of using channel estimation and eigendecomposition, CNN 730 replaces these two steps and calculates CSI feedback in one step. And the behavior of CNN should be similar to the straight-through direct quantization function. The detailed training / testing of CNN and the comparison with traditional methods are described as follows.

[0110] An exemplary implementation of a CNN may include training the CNN:

[0111] - Input → Pilot

[0112] -Output→

[0113] -Tag → E

[0114] -Loss → E and MSE between

[0115] Comparison between direct quantized CSI feedback and traditional CSI feedback:

[0116] - Input → Pilot

[0117] -E with The MSE between E and E'

[0118] Benefits of the proposed solution may include: Direct quantized CSI feedback has lower MSE compared to the traditional 2-step CSI feedback process. This seems to be consistent with the data processing theorem; more processing may actually lead to more information loss. It is also important to note that after the CNN is trained, it may only require sequential addition and multiplication operations for inference. The computational complexity of the directional quantized CSI feedback at the inference stage may be lower than the traditional eigendecomposition-based process involving iterative operations.

[0119] like Figure 8As shown in, an embodiment provides a device 10, which includes a control circuit system (CTRL) 12 (such as at least one processor), and at least one memory 14 storing instructions (INSTRUCT.), which, when executed by at least one processor, causes the device to perform at least any one of the above processes. In one example, the at least one processor and the instructions are configured to, together with the at least one memory, cause the device to perform any one of the above processes. The memory can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The memory may include a database for storing data.

[0120] In one embodiment, the apparatus 10 comprises a terminal device of a communication system, such as a user terminal (UT), a computer (PC), a laptop, a tabloid computer, a cellular phone, a mobile phone, a communicator, a smart phone, a PDA, a mobile transportation device (such as a car), a household appliance or any other communication device, generally referred to as a UE in the specification. Alternatively, the apparatus is included in such a terminal device. In addition, the apparatus may be or may include a module (attached to the UE) providing a connection, such as a plug-in unit, a "USB dongle", or any other kind of unit. The unit may be installed inside the UE, or may be attached to the UE via a connector or even wirelessly.

[0121] In one embodiment, the apparatus 10 is a UE 120, or is included in a UE 120. The apparatus may be caused to perform some of the functions of the above-described procedures.

[0122] The device 10 may include a radio interface (TRX) 16, which includes hardware and / or software for implementing a communication connection according to one or more communication protocols. The TRX may provide the device with communication capabilities to access a radio access network. For example, the TRX may enable SL communication.

[0123] The apparatus may further comprise a user interface 18, for example comprising at least one keyboard, microphone, touch display, display, speaker, etc. The user interface may be used by a user to control the apparatus.

[0124] In one embodiment, the control circuit system 12 includes a circuit for performing at least Figure 3 The step 302 of the instruction circuit system 20; for performing at least Figure 3 The receiving circuit system 21 of step 304; for performing at least Figure 3 The acquisition circuit system 22 of step 306; for performing at least Figure 3The execution circuit system 23 of step 308 is used to perform at least Figure 3 The input circuit system 24 of step 310; and for performing at least Figure 3 The transmitting circuit system 25 of step 312.

[0125] like Fig. 9 As shown, one embodiment provides a device 50, which includes a control circuit system (CTRL) 52 (such as at least one processor), and at least one memory 54 storing instructions (INSTRUCT.), which, when executed by at least one processor, causes the device to at least perform any one of the above processes. In one example, the at least one processor and the instructions are configured to cause the device to perform any one of the above processes together with the at least one memory. The memory can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The memory may include a database for storing data.

[0126] In one embodiment, the apparatus 50 may be a network node, or may be included in a network node, such as in a gNB / gNB-CU / gNB-DU of 5G. In one embodiment, the apparatus is a network node 110, or may be included in a network node 110. The apparatus may be enabled to perform some of the functions of the above-described process.

[0127] The device 50 may include a radio interface (TRX) 56, which includes hardware and / or software for implementing a communication connection according to one or more communication protocols. The TRX may provide the device with communication capabilities to access a radio access network. For example, the TRX may enable SL communication.

[0128] The apparatus 50 may further comprise a user interface 58, for example comprising at least one keyboard, microphone, touch display, display, speaker, etc. The user interface may be used by a user to control the apparatus.

[0129] In one embodiment, the control circuit system 52 includes a circuit for performing at least Figure 4 The receiving circuit system 60 of step 402; for performing at least Figure 4 The determining circuit system 62 of step 404; for performing at least Figure 4 The transmitting circuit system 64 of step 406; and for performing at least Figure 4 Receive circuitry 66 of step 408. In one embodiment, receive circuitry 60 and receive circuitry 66 are the same circuitry.

[0130] In one embodiment, the device for performing at least some of the above embodiments comprises at least one processor and at least one memory comprising instructions, which, when executed by the at least one processor, causes the device to perform functions according to any one of the above embodiments. According to one aspect, when the at least one processor executes the instruction, the instruction causes the device to perform functions according to any one of the above embodiments. According to another embodiment, the device for performing at least some of the above embodiments comprises at least one processor and at least one memory comprising instructions, wherein the at least one memory and the instructions perform at least some of the functions according to any one of the above embodiments. Therefore, the at least one processor, the memory and the instructions form a processing unit for performing at least some of the above embodiments. According to another embodiment, the device for performing at least some of the above embodiments comprises a circuit system, which comprises at least one processor and at least one memory comprising instructions. When activated, the circuit system causes the device to perform at least some of the functions according to any one of the above embodiments.

[0131] As used in this application, the term "circuitry" refers to all of the following: (a) pure hardware circuit implementations, such as implementations in analog and / or digital circuitry only, and (b) combinations of circuitry and software (and / or firmware), such as (where applicable): (i) a combination of (multiple) processors, or (ii) portions of (multiple) processors / software, including (multiple) digital signal processors, software, and (multiple) memories, which work together to enable the device to perform various functions, and (c) circuitry, such as (multiple) microprocessors, or portions of (multiple) microprocessors, which also require software and firmware to operate, even if the software or firmware is not physically present. This definition of "circuitry" applies to all uses of the term in this application. As another example, as used in this application, the term "circuitry" would also cover implementations of only a processor (or multiple processors), or a portion of a processor and its (their) accompanying software and / or firmware. For example, and if applicable to the particular element, the term "circuitry" would also cover a baseband integrated circuit or application processor integrated circuit for a mobile phone, or a similar integrated circuit in a server, cellular network device, or another network device.

[0132] In one embodiment, at least some of the above processes may be performed by a device including corresponding components for performing at least some of the above processes. Some example components for performing these processes may include at least one of the following items: a detector, a processor (including dual-core and multi-core processors), a digital signal processor, a controller, a receiver, a transmitter, an encoder, a decoder, a memory, a RAM, a ROM, software, firmware, a display, a user interface, a display circuit system, a user interface circuit system, a user interface software, a display software, a circuit, an antenna, an antenna circuit system, and a circuit system.

[0133] The term "non-transitory" as used herein is a restriction on the medium itself (ie, tangible, not a signal), not on the persistence of data storage (eg, RAM vs. ROM).

[0134] As used herein, the term "component" should be interpreted in the singular, i.e., to refer to a single element; or in the plural, i.e., to refer to a combination of single elements. Thus, the term "components [for performing A, B, C]" should be interpreted to cover devices in which there is only one component for performing A, B, and C; or in which there are separate components for performing A, B, and C; or in which there are partially or completely overlapping components for performing A, B, and C. In addition, the term "components for performing A, components for performing B, components for performing C" will be interpreted to cover devices in which there is only one component for performing A, B, and C; or in which there are separate components for performing A, B, and C; or in which there are partially or completely overlapping components for performing A, B, and C.

[0135] For example, the components of the device can be implemented in hardware and / or software. They may include, for example, at least one processor for executing processor instructions to perform the required functions, at least one memory for storing instructions, or both. Alternatively, they may include, for example, a circuit system designed or configured to implement the required functions, such as implemented in a chipset or chip (such as an integrated circuit). Typically, the device may include, for example, one or more processing components or processors.

[0136] The techniques and methods described herein can be implemented in various ways. For example, these techniques can be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or a combination thereof. For hardware implementations, the (multiple) devices of the embodiments can be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof. For firmware or software, the implementation can be performed by a module (e.g., a process, a function, etc.) of at least one chipset that performs the functions described herein. The software code can be stored in a memory unit and executed by a processor. The memory unit can be implemented in a processor or outside a processor. In the latter case, it can be coupled to the processor communication via various means known in the art. In addition, the components of the system described herein can be rearranged and / or supplemented by other components to facilitate the implementation of the various aspects described therein, and they are not limited to the precise configurations listed in a given figure, as will be understood by those skilled in the art.

[0137] The above-described embodiments can also be executed in the form of a computer process defined by a computer program or a part thereof. The embodiments of the above-described method can be implemented by executing at least a portion of a computer program including corresponding instructions. The computer program can be in source code form, object code form, or some intermediate form, and can be stored in a carrier, which can be any entity or device capable of carrying a program. For example, the computer program can be stored on a computer program distribution medium readable by a computer or processor. The computer program medium can be, for example, but not limited to, a recording medium, a computer memory, a read-only memory, an electric carrier signal, a telecommunication signal, and a software distribution package. The computer program medium can be a non-transient medium. The software coding for executing the illustrated and described embodiments is fully within the scope of those of ordinary skill in the art.

[0138] Example 1. An apparatus comprising:

[0139] at least one processor; and

[0140] At least one memory storing instructions, which, when executed by at least one processor, cause the apparatus to at least perform: indicating to a network node of a cellular communication network a capability of supporting channel state information (CSI) feedback acquisition based on machine learning (ML); receiving an indication of using ML-based CSI feedback acquisition from the network node; acquiring a CSI reference signal (CSI-RS) pattern and a CSI feedback size threshold, wherein the CSI-RS pattern and the CSI-RS feedback size threshold are jointly determined; receiving one or more pilot signals for at least one channel according to the CSI-RS pattern; inputting results of at least one measurement channel on the one or more pilot signals into an ML model, and acquiring CSI feedback as an output from the ML model; and sending CSI feedback to the network node according to the CSI feedback size threshold.

[0141] Example 2. The apparatus of Example 1, wherein the CSI-RS pattern and the CSI feedback size threshold are determined based on data-assisted channel estimation.

[0142] Example 3. The apparatus of Example 2, wherein data-aided channel estimation comprises: measuring a channel between the apparatus and a network node.

[0143] Example 4. The apparatus of Example 3, wherein data-aided channel estimation comprises: measuring a full channel grid of channels.

[0144] Example 5. The apparatus according to any one of the preceding examples is configured to perform: determining a CSI-RS pattern and a CSI feedback size threshold.

[0145] Example 6. An apparatus according to Example 5, wherein the determination is based on historical information.

[0146] Example 7. The apparatus according to Example 5 is configured to perform: measuring a channel between the apparatus and a network node; using the result of the measurement as training data for the ML model; determining a plurality of CSI feedbacks using the ML model and the training data, wherein each of the plurality of CSI feedbacks determined is associated with a corresponding CSI-RS pattern and a CSI feedback size threshold pair; comparing the plurality of CSI feedbacks determined; and selecting a CSI-RS pattern and a CSI feedback size threshold pair based on the comparison.

[0147] Example 8. The apparatus according to Example 5, 6 or 7 is configured to perform: indicating the determined or selected CSI-RS pattern and CSI feedback size threshold to a network node; and receiving from the network node the CSI-RS pattern and CSI feedback size threshold to be used to obtain CSI feedback, which will be sent to the network node.

[0148] Example 9. The apparatus according to any one of the preceding Examples 1 to 4 is configured to perform: receiving a CSI-RS pattern and a CSI feedback size threshold from a network node.

[0149] Example 10. An apparatus according to any of the preceding examples, being caused to perform: receiving an indication of a CSI recovery method to be used from a network node.

[0150] Example 11. The apparatus of any of the preceding examples is caused to perform: sending at least one indicator bit to a network node, the at least one indicator bit indicating a capability of supporting ML-based CSI feedback acquisition.

[0151] Example 12. The apparatus of Example 11, wherein the at least one indicator bit is sent in a radio resource control (RRC) message.

[0152] Example 13. An apparatus according to any of the preceding examples, wherein the results of the at least one measurement channel are directly input into the ML model without decompression.

[0153] Example 14. An apparatus comprising:

[0154] at least one processor; and

[0155] At least one memory storing instructions which, when executed by at least one processor, cause the apparatus to at least perform: receiving an indication of a capability to support machine learning-based channel state information (CSI) feedback acquisition from a user equipment (UE) of a cellular communication network; determining, based on the indication, to configure ML-based CSI feedback acquisition for the user equipment; sending an indication to the user equipment to use ML-based CSI feedback acquisition; and receiving CSI feedback acquired according to a CSI reference signal (CSI-RS) pattern and a CSI feedback size threshold from the user equipment.

[0156] Example 15. The apparatus of Example 14, further caused to indicate a CSI-RS pattern and a CSI feedback size threshold to a user equipment.

[0157] Example 16. The apparatus of Example 15, wherein the CSI-RS pattern and the CSI feedback size threshold are determined based on data-assisted channel estimation.

[0158] Example 17. The apparatus of Example 16, wherein data-assisted channel estimation comprises: measuring a channel between the apparatus and the user equipment.

[0159] Example 18. The apparatus of Example 16, wherein data-aided channel estimation comprises: measuring a full channel grid of channels.

[0160] Example 19. The apparatus according to any one of the aforementioned Examples 14 to 18 is further configured to perform: determining or selecting a CSI-RS mode and a CSI feedback size threshold; and indicating the determined or selected CSI-RS mode and CSI feedback size threshold.

[0161] Example 20. The apparatus according to Example 19 is configured to perform: receiving a CSI-RS mode and a CSI feedback size threshold recommendation from a user equipment; and determining or selecting a CSI-RS mode and a CSI-RS feedback size threshold based at least on the received CSI-RS mode and CSI feedback size threshold recommendation.

[0162] Example 21. An apparatus according to Example 20, wherein determining or selecting the CSI-RS pattern and the CSI feedback size threshold is based on historical information about at least one user equipment.

[0163] Example 22. An apparatus comprising:

[0164] A component for indicating to a network node of a cellular communication network the ability to support machine learning (ML) based channel state information (CSI) feedback acquisition; a component for receiving an indication of using ML based CSI feedback acquisition from a network node; a component for acquiring a CSI reference signal (CSI-RS) pattern and a CSI feedback size threshold, wherein the CSI-RS pattern and the CSI-RS feedback size threshold are jointly determined; a component for receiving one or more pilot signals for at least one channel according to the CSI-RS pattern; a component for inputting the results of at least one measurement channel on the one or more pilot signals into an ML model and acquiring CSI feedback as an output from the ML model; and a component for sending CSI feedback to the network node according to the CSI feedback size threshold.

[0165] Example 23. An apparatus comprising:

[0166] A component for receiving an indication of the ability to support machine learning (ML)-based channel state information (CSI) feedback acquisition from a user equipment (UE) of a cellular communication network; a component for determining, based on the indication, to configure ML-based CSI feedback acquisition for the user equipment; a component for sending an indication of using ML-based CSI feedback acquisition to the user equipment; and a component for receiving CSI feedback acquired according to a CSI reference signal (CSI-RS) pattern and a CSI feedback size threshold from the user equipment.

[0167] Although the present invention has been described above with reference to the examples according to the accompanying drawings, it is obvious that the present invention is not limited thereto, but can be modified in various ways within the scope of the appended claims. Therefore, all words and expressions should be interpreted broadly, and they are intended to illustrate rather than limit the embodiments. It is obvious to those skilled in the art that, with the advancement of technology, the inventive concept can be implemented in various ways. In addition, it is obvious to those skilled in the art that the described embodiments can, but are not required to be combined with other embodiments in various ways.

Claims

1. A device for communication, comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least perform: Indicating to a network node of a cellular communication network a capability to support ML-based channel state information CSI feedback acquisition; receiving, from the network node, an indication to acquire CSI feedback using the ML-based method; Acquire a CSI reference signal CSI-RS pattern and a CSI feedback size threshold, wherein the CSI-RS pattern and the CSI-RS feedback size threshold are jointly determined; receiving one or more pilot signals for at least one channel according to the CSI-RS pattern; inputting the result of at least one measurement channel on the one or more pilot signals into an ML model, and obtaining CSI feedback as an output from the ML model; The CSI feedback is sent to the network node according to the CSI feedback size threshold. 2 . The apparatus of claim 1 , wherein the CSI-RS pattern and the CSI feedback size threshold are determined based on data-assisted channel estimation.

3. The apparatus of claim 2, wherein the data-aided channel estimation comprises: A channel between the device and the network node is measured.

4. The apparatus of claim 3, wherein the data-aided channel estimation comprises: A full channel grid of the channels is measured.

5. The apparatus according to any one of claims 1 to 4, being configured to perform: The CSI-RS pattern and the CSI feedback size threshold are determined. The apparatus of claim 5 , wherein the determination is based on historical information.

7. The apparatus according to claim 5, being configured to perform: measuring a channel between the device and the network node; using the results of the measurements as training data for the ML model; Determine a plurality of CSI feedbacks using the ML model and the training data, wherein each CSI feedback of the determined plurality of CSI feedbacks is associated with a corresponding CSI-RS pattern and a CSI feedback size threshold pair; comparing the determined plurality of CSI feedbacks; as well as Based on the comparison, a CSI-RS pattern and CSI feedback size threshold pair is selected.

8. The apparatus according to claim 5, being configured to perform: indicating the determined or selected CSI-RS pattern and CSI feedback size threshold to the network node; and A CSI-RS pattern and a CSI feedback size threshold to be used to obtain the CSI feedback to be sent to the network node are received from the network node.

9. The apparatus according to any one of claims 1 to 4, being configured to perform: The CSI-RS pattern and the CSI feedback size threshold are received from the network node.

10. The apparatus according to any one of claims 1 to 4, being configured to perform: An indication of a CSI recovery method to be used is received from the network node.

11. The apparatus according to any one of claims 1 to 4, being configured to perform: At least one indicator bit is sent to the network node, the at least one indicator bit indicating the capability of supporting ML-based CSI feedback acquisition.

12. The apparatus of claim 11, wherein the at least one indicator bit is sent in a Radio Resource Control (RRC) message.

13. The apparatus according to any one of claims 1 to 4, wherein the results of the at least one measurement channel are directly input into the ML model without decompression.

14. An apparatus for communication, comprising: at least one processor; as well as at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least perform: receiving, from a user equipment UE of a cellular communication network, an indication of a capability to support channel state information CSI feedback acquisition based on machine learning ML; Based on the indication, determining to configure the ML-based CSI feedback acquisition for the user equipment; sending an indication to the user equipment to obtain CSI feedback using the ML; and A CSI feedback is received from the user equipment according to a CSI reference signal CSI-RS pattern and a CSI feedback size threshold.

15. The apparatus according to claim 14, further configured to: The CSI-RS pattern and the CSI feedback size threshold are indicated to the user equipment.

16. The apparatus of claim 15, wherein the CSI-RS pattern and the CSI feedback size threshold are determined based on data-assisted channel estimation.

17. The apparatus of claim 16, wherein the data-aided channel estimation comprises: A channel between the apparatus and the user equipment is measured.

18. The apparatus of claim 16, wherein the data-aided channel estimation comprises: A full channel grid of the channels is measured.

19. An apparatus according to any one of claims 14 to 18, configured to perform: determining or selecting the CSI-RS pattern and the CSI feedback size threshold; and Indicating the determined or selected CSI-RS pattern and the CSI feedback size threshold.

20. The apparatus according to claim 19, being configured to perform: receiving a CSI-RS pattern and a CSI feedback size threshold recommendation from the user equipment; and The CSI-RS pattern and the CSI-RS feedback size threshold are determined or selected based at least on the received CSI-RS pattern and the CSI feedback size threshold suggestion.

21. The apparatus of claim 20, wherein determining or selecting the CSI-RS pattern and the CSI feedback size threshold is based on historical information about at least one user equipment.

22. A method for communication, comprising: An apparatus of a cellular communication network indicates to a network node of the cellular communication network a capability of supporting channel state information (CSI) feedback acquisition based on machine learning (ML); receiving, from the network node, an indication to acquire CSI feedback using the ML-based method; Acquire a CSI reference signal CSI-RS pattern and a CSI feedback size threshold, wherein the CSI-RS pattern and the CSI-RS feedback size threshold are jointly determined; receiving one or more pilot signals for at least one channel according to the CSI-RS pattern; inputting the result of at least one measurement channel on the one or more pilot signals into an ML model, and obtaining CSI feedback as an output from the ML model; The CSI feedback is sent to the network node according to the CSI feedback size threshold.

23. A method for communication, comprising: Receiving, by an apparatus of a cellular communication network, an indication of a capability to support channel state information (CSI) feedback acquisition based on machine learning (ML) from a user equipment (UE) of the cellular communication network; Based on the indication, determining to configure the ML-based CSI feedback acquisition for the user equipment; sending an indication to the user equipment to obtain CSI feedback using the ML; and A CSI feedback is received from the user equipment according to a CSI reference signal CSI-RS pattern and a CSI feedback size threshold.

Images