Channel prediction

Abstract

There is provided an apparatus for channel prediction, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: obtain a selector identifier, where the selector identifier indicates a quality of quantization of a channel associated with a fine tuning configuration for the channel, where the channel comprises a wireless communication channel between the apparatus and a network node; and sending the selector identifier to the network node.

Description

Technical Field

[0001] Various example embodiments relate to an apparatus (such as a terminal device or a base station), a method, a computer program, and a non-transitory computer-readable medium for fine-tuning a channel prediction model of a communication network. Background Art

[0002] Wireless networks (such as fifth-generation 5G) utilize channel estimation in order to achieve their performance and network capabilities in the growing data traffic. Machine learning (ML) can be used for channel estimation and channel state information (CSI) acquisition, which shows the importance of multi-input multi-output MIMO precoding. Starting from pilots sent by a base station (BS), the channel is estimated at a user equipment (UE). 5G has considered two strategies for CSI reporting, called type I and type II, in order to provide feedback to the BS. These two types of CSI reports include strong compression of the CSI estimated at the UE in order to reduce the over-the-air (OTA) overhead. The compression reduces the accuracy of the CSI, which in turn affects the performance of MIMO precoding.

[0003] Channel prediction can be applied in order to support high-speed UEs. Additionally, in the case of a large channel prediction time domain with low normalized mean square error (NMSE), channel prediction can enable a reduction in CSI reporting overhead. The quality of channel prediction may be impaired as the prediction time increases. Summary of the Invention

[0004] An object of the present invention is to provide a fine-tuning for channel prediction with reduced overhead.

[0005] According to some aspects, the subject matter of the independent claims is provided. Some example embodiments are defined in the dependent claims. The scope of protection sought for the various example embodiments is set out in the independent claims. Example embodiments and features (if any) described in this specification that do not fall within the scope of the independent claims should be construed as examples useful for understanding the various example embodiments.

[0006] According to a first aspect, there is provided an apparatus, comprising: at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause the apparatus to at least: obtain a selector identifier, wherein the selector identifier indicates a quantization quality of a channel associated with a fine-tuning configuration for a channel, wherein the channel comprises a wireless communication channel between a network node and the apparatus; and send the selector identifier to the network node.

[0007] According to a second aspect, there is provided an apparatus, comprising: at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause the apparatus to at least: receive a selector identifier from a terminal device, where the selector identifier indicates a quantified quality of a channel associated with a fine-tuning configuration for a channel, where the channel includes a wireless communication channel between the apparatus and the terminal device; and determine a fine-tuning configuration for channel prediction based at least in part on the received selector identifier.

[0008] According to a third aspect, there is provided a method, comprising: obtaining, by a terminal device, a selector identifier that indicates a quantified quality of a channel associated with a fine-tuning configuration for a channel, where the channel includes a wireless communication channel between a network node and the terminal device; and sending, by the terminal device, the selector identifier to the network node.

[0009] According to a fourth aspect, there is provided a method, comprising: receiving, by a network node, a selector identifier that indicates a quantified quality of a channel associated with a fine-tuning configuration for a channel, where the channel includes a wireless communication channel between the network node and the terminal device; and determining, by the network node, a fine-tuning configuration for channel prediction based at least in part on the selector identifier.

[0010] According to a fifth aspect, there is provided a computer program comprising program instructions stored thereon for performing the method according to the third aspect or the fourth aspect.

[0011] According to a sixth aspect, there is provided a non-transitory computer-readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to at least perform the following operations: obtain a selector identifier, where the selector identifier indicates a quantified quality of a channel associated with a fine-tuning configuration for a channel, where the channel includes a wireless communication channel between a network node and the apparatus; and send the selector identifier to the network node.

[0012] According to a seventh aspect, there is provided a non-transitory computer-readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to at least perform the following operations: receive a selector identifier from a terminal device, where the selector identifier indicates a quantified quality of a channel associated with a fine-tuning configuration for a channel, where the channel includes a wireless communication channel between the apparatus and the terminal device; and determine a fine-tuning configuration for channel prediction based at least in part on the received selector identifier.

[0013] According to an eighth aspect, there is provided an apparatus, comprising: means for obtaining a selector identifier indicative of a quantization quality of a channel associated with a fine-tuning configuration for a channel, where the channel comprises a wireless communication channel between a network node and the apparatus; and means for sending the selector identifier to the network node.

[0014] According to a ninth aspect, there is provided an apparatus, comprising: means for receiving a selector identifier indicative of a quantization quality of a channel associated with a fine-tuning configuration for a channel, where the channel comprises a wireless communication channel between the apparatus and a terminal device; and means for determining, at least in part based on the selector identifier, a fine-tuning configuration for channel prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The network architecture of a communication system is illustrated by way of example;

[0016] Figure 2 The signaling in a cellular network is illustrated by way of example;

[0017] Figure 3 Window selection is illustrated by way of example;

[0018] Figure 4 A deep neural network structure is illustrated by way of example;

[0019] Figure 5 The block diagram of an apparatus is illustrated by way of example;

[0020] Figure 6 The flowchart of a method for a network entity is illustrated by way of example, and

[0021] Figure 7 The flowchart of a method for a terminal device is illustrated by way of example. DETAILED DESCRIPTION

[0022] Channel prediction (CP) for an air interface, models, and error prediction model usage are implemented using a selector identifier, which refers to the most likely channel prediction model among a set of possible channel prediction models. The selector identifier indicates the quantified quality of the channel, which is associated with a fine-tuning configuration for the channel. The quantified quality can include the channel transfer function, power level, measured reference signal, or impulse response of the channel, or any channel information available at the physical layer. A terminal device or user equipment (UE) selects and reports the selector identifier of the channel to a network node or base station. The base station is configured to perform a fine-tuning operation at least in part based on the received selector identifier. The fine-tuning operation includes selecting an optimal fit error model based on the reported selector identifier. The base station is configured to store an error model for channel prediction. The UE does not need to run or have any knowledge of the CP model or error model used at the base station.

[0023] The channel can refer to a wireless communication channel or a radio frequency channel between a terminal device and a network node. Generally, the channel transfer function of a wireless communication channel is the continuous-time Fourier transform of the impulse response of the wireless communication channel. The impulse response is the response of the channel to a short transmitted pulse. A reference signal is used to determine the impulse response of the channel. The reference signal can be transmitted continuously. Physical signals use allocated resource elements.

[0024] Figure 1 The network architecture of a communication system is illustrated by way of example. Hereinafter, a radio access architecture based on Long Term Evolution Advanced (LTE-Advanced), LTE-A, or New Radio (NR) (also known as 5th Generation 5G) will be used as an example of an access architecture to which embodiments can be applied to describe different exemplary embodiments, without limiting the embodiments to such an architecture. It will be apparent to those skilled in the art that the embodiments can also be applied to other kinds of communication networks with suitable means by appropriately adjusting parameters and processes. Some examples of other options for suitable systems are Universal Mobile Telecommunications System, UMTS, Radio Access Network, UTRAN or E-UTRAN, Long Term Evolution, LTE, the same as E-UTRA, Wireless Local Area Network, WLAN or WiFi; Worldwide Interoperability for Microwave Access (WiMAX); Personal Communication Service, PCS Wideband Code Division Multiple Access (WCDMA), systems using Ultra-Wideband (UWB) technology, sensor networks, Mobile Ad-hoc Networks (MANET), and systems of Internet Protocol Multimedia Subsystem (IMS), or any combination thereof.

[0025] Figure 1 Examples illustrate a part of an exemplary radio access network. Figure 1Terminals or user equipments 100 and 102 are shown, which are configured to be wirelessly connected to an access node providing a cell (such as a gNB, i.e., a next-generation node B, or an eNB, i.e., an evolved node B, eNodeB 104) on one or more communication channels in the cell. The physical link from the user equipment to the network node is referred to as the uplink (UL) or reverse link, and the physical link from the network node to the user equipment is referred to as the downlink (DL) or forward link. It should be understood that the network node or its functions can be implemented by any entity such as a node, host, server, or access point suitable for this purpose. A communication system typically includes more than one network node. In this case, the network nodes can also be configured to communicate with each other via wired or wireless links designed for this purpose. These links can be used for signaling purposes. A network node is a computing device configured to control the radio resources of the communication system to which it is coupled. A network node can also be referred to as a base station (BS), an access point, or any other type of interface device including a relay station capable of operating in a wireless environment. The network node includes or is coupled to a transceiver. From the transceiver of the network node, a connection is provided to an antenna unit that establishes a two-way radio link to the user equipment. The antenna unit can include multiple antennas or antenna elements. The network node is also connected to the core network 110 (CN) or next-generation core (NGC). Depending on the system, the corresponding party on the CN side can be a serving gateway (S-GW, routing and forwarding user data packets), a packet data network gateway (P-GW) for providing a connection from the user equipment (UE) to an external packet data network, or a mobility management entity (MME), etc. An example of a network node configured to operate as a relay station is an integrated access and backhaul node (IAB). The distributed unit (DU) part of the IAB node performs the BS function of the IAB node, while the backhaul connection is performed by the mobile terminal (MT) part of the IAB node. The UE function can be performed by the IAB MT, and the BS function can be performed by the IAB DU. The network architecture can include a parent node, i.e., an IAB donor, which can have a wired connection to the CN and a wireless connection to the IAB MT.

[0026] A terminal device, user equipment, or user equipment (UE) generally refers to a portable computing device that includes a wireless mobile communication device operating with or without a subscriber identity module (SIM), including but not limited to the following types of devices: mobile station, mobile phone, smartphone, personal digital assistant (PDA), handset, device using a wireless modem (such as an alarm or measurement device, etc.), laptop and / or touchscreen computer, Internet of Things (IoT) node, tablet, game console, notebook computer, multimedia device, and connected vehicle connectivity module. It should be understood that the user equipment can also be an almost dedicated uplink-only device, an example of which is a camera or video camera that loads images or video clips onto the network. The user equipment can also be a device capable of operating in an Internet of Things (IoT) network, which is a scenario where objects are provided with the ability to transfer data over a network without the need for human-to-human or human-to-computer interaction.

[0027] Additionally, although the device has been depicted as a single entity, different units, processors, and / or memory units can be implemented within these devices to achieve their functions.

[0028] 5G enables the use of multiple-input-multiple-output (MIMO) technology on both the UE and gNB sides, many more base stations or nodes than LTE, the so-called small cell concept, including macro sites operating in cooperation with smaller stations, and various radio technologies are adopted according to service requirements, use cases, and / or available spectrum. 5G mobile communications support a wide range of use cases and related applications, including video streaming, virtual reality, extended reality, augmented reality, different data sharing methods, and various forms of machine type applications, such as (massive) machine type communication, mMTC, including vehicle safety, different sensors, and real-time control. It is expected that 5G will have multiple radio interfaces, namely below 7 GHz, centimeter wave, and millimeter wave, and can also be integrated with existing traditional radio access technologies (such as LTE). The frequency range below 7 GHz can be referred to as FR1, and above 24 GHz or more precisely, 24 - 52.6 GHz, can be referred to as FR2. The integration with LTE can be implemented as a system at least in the early stage, where macro coverage is provided by LTE, and 5G radio interface access comes from small cells by aggregating to LTE. In other words, 5G is planned to support inter-RAT operability (such as LTE-5G) and inter-RI operability, radio interface inter-operability (such as below 7 GHz - centimeter wave, below 7 GHz - centimeter wave - millimeter wave). One of the concepts considered to be used in 5G networks is network slicing, where multiple independent and dedicated virtual sub-networks or network instances can be created within the same infrastructure to run services with different requirements for latency, reliability, throughput, and mobility.

[0029] The communication system may also be able to communicate with other networks, such as the public switched telephone network, the PSTN or the Internet 112, or utilize services provided by them (e.g., via a server). The communication network may also be able to support the use of cloud services. For example, at least a part of the core network operations may be performed as cloud services, as shown by the "cloud" 114 in Figure 1 The communication system may also include a central control entity or the like, which provides facilities for the networks of different operators to cooperate, for example, in spectrum sharing.

[0030] 5G may also utilize satellite communication to enhance or supplement the coverage of 5G services. For example, by providing backhaul. Possible use cases are to provide service continuity for machine-to-machine (M2M) or Internet of Things (IoT) devices or passengers on vehicles, or to ensure the service availability of critical communications and future railway / sea / air communications. Satellite communication may utilize a geostationary earth orbit (GEO) satellite system, but may also utilize a low earth orbit (LEO) satellite system, especially a mega-constellation system in which hundreds of (nano) satellites are deployed. Each satellite 106 in the constellation may cover a network entity of several supporting satellites that create a ground cell. The ground cell may be created by a ground relay node 104 or by a gNB located on the ground or in the satellite.

[0031] As used herein, a channel prediction model may be referred to as channel prediction (CP). An error prediction model may be referred to as an error model. A window may be an indicator window, which may be a selected portion or subset that quantifies the channel quality. A selector identifier may refer to the most likely channel prediction model among a set of possible channel prediction models. The selector identifier indicates the quantified quality of the channel, which may relate to at least one of the following: the channel transfer function of the channel, the measured reference signal of the channel, the power level of the channel, the channel impulse response, or other channel information available at the physical layer. A fine-tuning configuration may refer to a codebook for channel fine-tuning, a window for channel fine-tuning, time and / or frequency resources for channel fine-tuning, a start frame period and / or slot period for channel fine-tuning; a timer for fine-tuning configuration reference signal information, or a duration for channel fine-tuning configuration.

[0032] As used herein, "at least one of the following: " and "at least one of " and similar phrases, where the list of two or more elements is joined by "and" or "or", means at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

[0033] As used herein, the term "non-transitory" is a limitation of the medium itself (i.e., tangible rather than a signal), rather than a limitation on data storage persistence (e.g., RAM vs. ROM).

[0034] A window or a flagged window can be selected at the gNB. The window indicates a part of the CTF in the frequency space, which includes the relationship between the estimated or observed CTF and the channel prediction based on the error model. The window is selected such that the evolution of the CTF (i.e., a part of the CTF) within the window allows the identification of the best selector identifier, which corresponds to the best-fitting error model identifier within the window. The error prediction model can include an ML error model.

[0035] The UE or the terminal device is configured to identify the selector identifier within the window and report the selector identifier to the gNB. This enables the gNB to search for the best-fitting error model within the window. The UE receives information about the window from the gNB, and the window represents a part of the channel between the UE and the gNB. The window includes the quantization values of one or more selected factors of the channel. According to one embodiment, the UE can be configured to measure the CTF evolution in the frequency sub-band called the window at the prediction time t pred and select the selector identifier indicating the quantization power level defined for the window. The selector identifier identifies the power level value within the window that is closest to the power level observed at the UE. The UE can be configured to search for the quantization quality of the window that has the minimum error for the difference signal between the most recently estimated CTF and the CTF from the window, thereby forming an error subspace. The error can include the normalized mean square error (NMSE). Instead of the NMSE of the channel, the cosine similarity can be utilized. The UE is capable of performing the error calculation and selecting the selector identifier based on this calculation without a further error prediction model.

[0036] The quality of the channel or the quantization quality relates to a communication channel that is a radio channel. The quality can relate to RF characteristics, and the RF characteristics relate to the channel, are characteristics of the channel, and / or describe the channel. The quality of the channel can include the power level or the quantization power level. The quality of the channel can include the angular evolution of the CTF within the window, or the real part and / or the imaginary part of the CTF. The channel characteristics can include the angle-time component, where time can refer to the relevant taps of a certain angle beam. The channel quality can include the angle-multipath component, which is the multipath component of a certain angle beam. The channel quality can include the cosine similarity. The cosine similarity provides a measure of the proximity of the predicted eigenvector to the eigenvector of the real-world radio channel. The scheme can be adapted to predict the cosine similarity, which can be used as another possible quantization quality of the channel.

[0037] Figure 2 Signaling in a cellular network is illustrated by way of example. Figure 2Including signaling an evolved channel transfer function (CTF) from a base station to a terminal device, and signaling one or more selector identifiers for marking a window from the terminal device to the base station. Figure 2 Shows power levels as an example of the quality of quantization of a channel. On the vertical axis, the terminal device or UE is set on the left, the base station gNB is set in the middle, and the cloud representing the network entity is set on the right. Time progresses from top to bottom.

[0038] In stage 201, the base station (gNB) is configured to receive a configuration from the cloud. The base station (gNB) can be configured to download a configuration from the cloud or a network entity. The configuration can include a codebook defining the marking window attributes. 5G includes two different types of codebooks: CSI type I and type II reports. The codebook can include a set of precoders or a set of precoding matrices. The precoding matrix can include complex-valued elements configured to transform data. The transformed data can be configured to be mapped to antenna ports. The configuration can include a codebook that includes at least one of the following: a method for model-based CSI reporting and a rule for model change. The codebook entry can include a parameter that includes one or more of the following: a frequency subband, a frequency subband size, a power quantization value, a number of windows in time and frequency, a weighting and definition, a rule for combined reporting for multiple windows, or any combination thereof.

[0039] In stage 202, the base station (gNB) is configured to send a configuration to the terminal device (UE). The configuration (set of codebooks) can be configured and reported as a combination of messages, such as radio resource control (RRC), media access control identifier (MAC ID), and downlink control information (DCI).

[0040] In stage 203, the base station (gNB) is configured to send a channel state information resource element (CSI RS) carrying information about multiple beams and time instances t1, t2......ts to the UE. This shows the beams sent by the base station (gNB) in a specific direction at specific time instances (t1 - ts).

[0041] In stage 204, the UE is configured to estimate and predict CSI. This is based on the received CSI RS.

[0042] In stage 205, the UE is configured to report the CSI time domain as estimated and predicted. The report is sent to the base station (gNB). The report includes CSI, which includes time domain block reports at time instances t1 to ts.

[0043] In stage 206, the UE is configured to infer a channel prediction model.

[0044] In stage 207, the base station (gNB) is configured to infer a channel prediction model and its parameters.

[0045] In stage 208, the base station (gNB) is configured to infer a set of Ns channel prediction models. Among the Ns channel prediction models, the gNB is configured to infer a reduced set of Ns_min channel prediction models. The reduced set of Ns_min channel prediction models can be inferred based on their correlation. In stage 209, the base station (gNB) is configured to infer a window or a CTF window or a so-called flagged window. The window indicates a part (frequency sub-band) of the CTF. To identify the window, the base station (gNB) is configured to define a set of all possible CTFs for a certain prediction time (t pred ) The set of all CTFs at a specific time or frequency corresponds to the error subspace of the channel prediction. The base station (gNB) may be configured to: apply different extrapolations with various extrapolation functions for the delay and amplitude of the estimated parameters and for the phase of the selected set of relevant multipath components; and calculate the possible impact of these extrapolations on the evolution of the CTF. Each CTF is characterized by a plurality of multipath components. Each multipath component is described by at least amplitude, delay, and phase parameters. The error subspace covers all possible evolutions of the CTF, where the mutual distances between the CTFs observed in the set of evolving CTFs are almost equal. The selection may be based on the identification of the best-fit window including a subset of the CTF or a frequency sub-band, which is selected such that the evolution of the CTF within the window allows the identification of the best-fit error model identifier based on the measured received signal power of the estimated CTF within the window. The subset of the CTF includes the power level values of the CTF within the window. The relationship between the measured received signal power of the observed CTF can be compared with the power level values of the CTFs forming the error subspace within the window. The relationship may be linear, or at least approximately linear if not completely linear. The window indicates a subset of the CTF in frequency and a set of power level values of the CTF. The power level values of the CTF may be quantized. The set of power level values of the CTF may be stored. Each power level value may be associated with one or exactly one power level identifier.

[0046] In stage 210, the base station (gNB) is configured to send the parameters defining the window inferred in the previous stage 209 to the UE. The window includes a part of the channel indicated by the quantized quality of the channel. In Figure 2 it, the window includes a subset of the channel transfer function in the frequency range and a set of power level values of the channel transfer function, which may be sent to the UE. The window may be sent in at least one of the following: radio resource control message, control element of media access control message, and downlink control information message.

[0047] In stage 211, the base station (gNB) is configured to transmit CSI RS carrying multiple beams to the UE at time instance ts+1. The UE can observe the further evolution of the CTF or the radio channel based on the received CSI RS including multiple beams at time instance ts+1.

[0048] In stage 212, the UE is configured to estimate and predict CSI in response to receiving a CSI RS message from the base station (gNB).

[0049] In stage 213, the UE is configured to infer the power level identifier of the received window based on the received window. The UE is configured to observe the evolution of the CTF at the prediction time t pred at the observation window. The UE can be configured to measure or estimate the received power level of the CTF within the window at time t pred The UE is configured to identify the CTF within the window such that the identified CTF has the minimum difference compared to the most recently observed CTF. The CTF of the difference signal with the minimum normalized mean square error (NMSE) can be selected for the window. This selection can be referred to as a selector identifier indicating the quality of the quantization of the channel. In Figure 2 it, the selection is referred to as the power level identifier for the window, which corresponds to the best-fit error model identifier.

[0050] In stage 214, the UE is configured to report the selector identifier to the base station (gNB), which is the power level identifier in Figure 2 it. The selector identifier can be efficiently reported over the air using only a few or several bits.

[0051] In stage 215, the base station (gNB) is configured to select the corresponding error model using the received selector identifier. The best-fit error model can be identified based on the received selector identifier. The best-fit error model can be referred to as a CSI predictor. The error model enables reducing the difference between the predicted evolution of the radio channel (CTF) and the newly estimated evolution. The selected error model can be used as a new channel prediction model for fine-tuning. Additionally, the output of the channel prediction model can be used to apply the error model on the channel prediction model in order to identify the most likely channel prediction subspace, which includes the set of possible evolutions of the CTF of the current state of a given radio channel.

[0052] Figure 3 Window selection is illustrated by way of example. The base station (gNB) is configured to select a window. The window includes a range of quantization quality parameters for channel prediction. The window can be selected relative to the evolving CTF, which is shown as the curve CTF(f,t in the upper figure s), where the x-axis shows the frequency. The CTF can correspond to the channel transfer function over the full frequency band (i.e., the full RF bandwidth). The CTF can be collected for the entire bandwidth. A window or a marked window can define the position of the window selection in terms of frequency, power, and time. For example, the selection of the window can include processing parameters such as frequency subbands, subband widths, and quantization values. The selection of the window can be implemented using artificial intelligence (e.g., neural networks), which will be described in more detail later. To identify the window, an error prediction subspace is defined, which includes all possible CTFs at a certain prediction time t pred in the set. A subset of the CTFs that includes at least an almost or nearly linear relationship between the observed power levels of the received signals associated with the best-fit error model identifier is selected as the window, or can be referred to as the best marked window. The selected window is shown as a rectangle in the upper graph, and its magnification is shown in the lower graph. The window has been quantized into a finite set of selector identifiers, which are shown as points in the lower graph, and which can equivalently cover the prediction error space. The gNB can be configured to apply different extrapolations with various extrapolation functions to the delay, amplitude, and phase of the selected set of relevant multipath components of the estimated parameters; and the gNB can also be configured to calculate the impact of these on the evolution of the CTF. The error prediction subspace covers all possible evolutions of the CTF within the frequency range and includes at least nearly equal mutual distances between the sets of CTFs. In Figure 3 the lower graph, the lowest curve is the estimated CTF at the UE, and the UE will select the lowest point as the selector identifier, which is the point closest to the estimated CTF.

[0053] Previously, the window has been defined by the power level or the quantized power level. Other channel characteristics can be alternatively utilized, such as the evolution of the phase of the CTF within the window, or the real and / or imaginary parts of the CTF within the window. In addition to the frequency subbands, the window can also be defined by quantizing the phase values and / or by the quantized real and / or imaginary signal values of the CTF. In this case, the UE reports the selector identifier. The selector identifier can refer to a quantity of the power level, as shown in Figure 2 phase 214. Alternatively, the quantity can refer to a phase value identifier or a real / imaginary identifier, which has the closest value to the same quantity derived from the estimated CTF. The reported selector identifier can refer to the power, phase, and / or other characteristic parameters of the channel or the CTF. Other characteristic parameters of the channel can be similarly derived from the time-domain channel impulse response, for example, from the oversampled profiled channel impulse response. In this case, the window includes the time domain instead of the frequency domain. The time-domain window can be defined by a certain subrange of the delay value τ of the channel impulse response.

[0054] Reporting CSI and predicted CSI enables support for multi-user, MU, MIMO precoding. The precoding performance can be defined by the cosine similarity between the reported radio channel and the real-world radio channel. For precoding, the similarity between the reported eigenvectors and the real-world eigenvectors may be more relevant than the normalized mean square error (NMSE) between the reported radio channel and the real-world radio channel. In some cases, the cosine similarity or a similar value can be used as a channel characteristic parameter of the window. The cosine similarity can be calculated from the estimation of the received radio channel through simple data preprocessing.

[0055] If the angular time domain is used for channel characteristics, channel prediction (CP) can be performed for each relevant angular beam. Narrow angular beams can reduce the number of multipath components. Thereby, the prediction of the subspace can be simplified. This can further allow for a lower CSI RS period and result in a larger prediction time domain. The time domain space can be related to the most relevant oversampled taps in the angular domain. The radio channel can be decomposed into a set of the most relevant angular-time components. The window can use a combined indicator as a channel characteristic indicator, which can include a weighted sum over the set of angular-time components. In an example embodiment, the angular-time components can include a set of multipath component parameters such as amplitude, phase, delay, angle (e.g., angle of arrival), etc.

[0056] The channel characteristic identifier can be extended to other channel characteristic parameters of the CTF in addition to the power level identifier, which has been described in more detail in Figure 2 The channel characteristic parameters can be directly related to the radio channel or directly related to the channel characteristics (such as cosine similarity) derived from data preprocessing.

[0057] The window or the marked window can be referred to or shown as a matrix. One dimension of the matrix can be the number of layers. For example, the layer can be defined for type II CSI reporting, where L can correspond to four (4) beams with two polarizations. Then, the size of the matrix can correspond to, for example, 2L times the number of windows. The channel characteristic related identifier or the selector identifier can include one or more coefficients of the matrix. The window can include a subset of the CTF within the frequency range of the entries of the matrix including the window. The CTF includes the frequency domain of the radio channel.

[0058] Previously, the CTF has been used as the quality of quantization of the channel associated with the fine-tuning configuration for the channel. Similarly, other channel information can be used as the quality of quantization of the channel. For example, the power level can be replaced by the measured reference signal, the impulse response, or other channel information available at the physical layer.

[0059] The base station is configured to send a predefined fine-tuning configuration to the UE. The fine-tuning configuration may be sent in a Radio Resource Control (RRC) message; in a Control Element (CE) of a Medium Access Control (MAC) message, or in a Downlink Control Information (DCI) message. For example, the fine-tuning configuration may include one or more of the following: a codebook; a window for channel fine-tuning; a duration for channel fine-tuning; time and / or frequency resources for channel fine-tuning; a start frame and / or slot period; or a timer for fine-tuning configuration reference signal information.

[0060] According to one embodiment, the UE may measure or observe the received power level of the CTF within a window at a predicted time t pred The window may include the CTF predicted by an error prediction model and its quantization into a finite set of power levels. Each quantized power level corresponds to a selector identifier. Each quantized power level may correspond to exactly one selector identifier. The window may define a plurality of selector identifiers that equivalently cover the error prediction subspace. The window may include the observed CTF observed at the UE. Then, the UE is able to select the quantized selector identifier that is closest to the measured CTF as the selector identifier for the window. Then, the selected selector identifier for the window is sent to the gNB. The gNB is configured to identify the best-fit error model or the best-fit identifier of the error subspace based on the reported selector identifier. Although the UE does not know the CP model and the error model, the CP model and the associated error model are fine-tuned at the gNB.

[0061] A base station (gNB) may be configured to define a set of quantized qualities of a channel. For example, the gNB may pre-define more than one window. Accordingly, more than one window may be defined by more than one subset of CTFs in a frequency (frequency sub-band). Defining more than one quantized quality of a channel may improve the reliability for identifying a selector identifier. Additionally or alternatively, a terminal device may be configured to observe (e.g., over multiple time or frequency slots) multiple quantized qualities of a channel. This may enable tracking the time evolution of the quantized quality of a channel, such as CTF. The base station (gNB) may define rules for selecting a selector identifier at the terminal device. Adding complexity to the rules may provide accuracy for the selection and enable identifying the most promising (correct or best) selector identifier. The rules may include, for example, using at least one of the following: the average power of the quantized quality of the channel (e.g., CTF) within a window; the average power of the quantized quality of the channel (e.g., CTF) at two predefined time instances or frequency subsets; or the weighted sum of the quantized quality of the channel (e.g., CTF) at two or more predefined frequency subsets and / or time instances. A non-linear quantization curve of the selector identifier may result in a more accurate remapping of the reported selector identifier to an error model at the gNB. This may be achieved at the cost of having a larger set of possible quantized qualities or windows. The gNB reports a predefined fine-tuning configuration to the terminal device. In the case where the gNB reports a window to the terminal device, depending on the size of the sub-band of the window, noise reduction may be applied to the power estimation at the terminal device. To achieve a proper selection of one or more windows, AI, ML, or NN may be used. The window selection may be based on frequency sub-bands, sub-band widths, quantization values, etc.

[0062] In one embodiment, the base station may utilize a predefined NN model. The base station may not know the processing at the terminal device. The base station may update the weights of the NN model. A method of transfer learning may be used, and the base station may update the weights together with the terminal device. This may be achieved via OTA. The update of the NN model may include storing NN model parameters for a specific type of terminal device. The stored parameters may be re-used for the same type of terminal device. For example, vendor-specific parameters may be stored and re-used. When a vendor and / or a certain type of device enters the cell, the stored parameters may be retrieved for use.

[0063] UE processing is minimized to observe or measure the quantized quality of a channel and map the measured quantized quality to the closest predicted quantized quality of the channel.

[0064] Fine-tuning of the CP model enables avoiding the need for a complete new CSI report at each reporting instance. The fine-tuning method for the CP model proposed here avoids the need to implement any CP prediction model at the UE side. Additionally, there is no need to exchange or synchronize machine learning (ML) models between the UE and the gNB. The gNB can have higher processing capabilities compared to the processing capabilities of the UE. The processing and implementation of the channel prediction model and the error prediction model are implemented at the gNB.

[0065] At least for the quantized error prediction subspace, the number of bits used for the selector identifier is small compared to a complete CSI report. Accordingly, reporting the selector identifier uses fewer OTA resources. The error model enables generating the error prediction subspace over the full radio frequency (RF) bandwidth instead of one or a few physical resource blocks (PRBs) or subbands. With just a few bits, the channel prediction can be fine-tuned over the full RF bandwidth. Simulations based on real-world channel sounder measurements show that a high prediction time domain within a range of one lambda for an NMSE of -15 dB is enabled in combination with just a few bits, such as 3 - 6 bits.

[0066] There is no need to have exactly the same channel prediction model and error prediction model at the UE and the gNB. The UE and the gNB do not need to exchange the CP model, nor do they need to exchange the error prediction model. There is no need for complete synchronization of the CP model and the error prediction model between the UE and the gNB to properly map the selector identifier reported by the UE to the error subspace identifier at the gNB, which avoids the risk of losing synchronization for a noisy radio channel.

[0067] The appropriate selection of the window enables achieving a fine-tuning performance close to an implementation where both the UE and the gNB process the same CP model with the same error model in parallel. Artificial intelligence (AI) and its branch machine learning (ML) are used in cellular network implementations, including the prediction of CSI. Channel prediction can be handled by using a Kalman filter (KF)-based method. An autoregressive (AR) model can be used to model the dynamics of the channel, such as the channel transfer function (CTF), and the coefficients of the AR process can be calculated using temporal and spectral correlations under the assumptions of the Jakes fading model and the one-sided exponential power delay profile (PDP). Compared to the KF-based method, a recurrent neural network (RNN) can effectively predict large-scale MIMO channels, thus providing the benefit of multi-step-ahead prediction. Window selection is discussed in more detail next.

[0068] The set of NxL estimated CSIs is represented by:

[0069]

[0070] Among them, at time slots n = 1...N, there are 1 = 1...L CSI resource elements. The CSI here is called the estimated complex CTF at the resource elements of CSI, CSI RS. The observed, estimated, predicted, or measured channel matrix (which can be the relevant observed complex channel matrix) has been estimated at the UE side and reported to the base station (gNB). The predicted channel matrix evolves into time slot t i = n...n + P, where the relevant CSI is collected in the evolving channel matrix which can be a real-world estimate. The evolution of the radio channel is unknown, neither at the UE side nor at the base station. However, the base station is able to apply channel prediction. For this purpose, the set S n of the estimated CSI is fed into the CP model, which is configured to output the predicted channel matrix This predicted channel matrix is considered the best possible approximation of the evolving channel matrix corresponding to the measured channel evolution up to time t pred = n + P.

[0071] Over time, even for a medium prediction time t pred with a medium number P, the mismatch between the evolving channel matrix and the predicted channel matrix may increase. Fine-tuning enables the identification of an error model for the set S n of the estimated CSI with respect to the predicted CSI This error model can cover the entire error subspace for a given current state of the radio channel where M is the size of the error subspace, corresponding to the number of possible expected evolutions of the CSI or CTF. The error subspace can be based on the multipath component parameter estimates from S n for the N first time slots plus various spline extrapolation sets for all relevant multipath component parameters up to time slot t i = n...n + P.

[0072] The entire error subspace M' can be reduced to a smaller equally spaced error subspace of size M << M'. Such a subspace reduction enables the saving of the fine-tuning overhead, because the UE can report only the best-fit selector ID m with B = |log 2 M| bits from a smaller set for the best-fit error model. In previous solutions, the UE may have mirrored the CP model for predicting the predicted channel matrix plus the error model To infer the optimal selector ID m so as to infer the same error subspace at the UE side and the base station side The same set Sn of estimated CSI has been used as the input signal at the UE and the base station.

[0073] Now, an optimal adaptation selector identifier (ID) m is provided to be inferred at the UE side based on the estimated CSI from the evolved radio channel, which is described as an evolved channel matrix This enables avoiding implementing any CP model or error model for calculating the predicted channel matrix at the UE side of The base station is configured to infer a window (a so-called marked window), which is a sub-band (such as a subset ) of the evolved (radio) channel matrix , where L_mark is the frequency sub-band, L_mark = {l mark，1 …l mark，K} and K < L. Additionally, the window is defined by a minimum power P min and a maximum power P max such that the window is configured to cover the entire error subspace at the frequency sub-band defined by L_mark with respect to the known predicted CSI The base station is able to identify values based on the known entire error subspace The quality of the channel characteristic parameter (e.g., the power level of the window) is quantified as corresponding to the number of bits B for reporting M = 2 B selector identifiers, such as p m = p min +(p max - p min ) / M. Using the quantified quality of the channel characteristic parameter (e.g., the power level), the UE is configured to select the selector identifier m, where the power pm is within the window and closest to the power of the predicted evolved radio channel, which can be measured, observed, predicted, or estimated at the UE side:

[0074] The UE can select the selector identifier m, which is configured to minimize, for example, the normalized mean square error (NMSE) between the estimated evolved radio channel and the m-th channel matrix of the error subspace can be utilized according to the following equation:

[0075] The base station is configured to identify the best - fitting window for a given current channel state such that the selector identifier m reported in the previous equation (1) is the same as or as close as possible to the optimal selector identifier m presented in the previous equation (2), even if the UE does not know the current error subspace opt and is as close as possible, even though the UE does not know the current error subspace Contrary to the previous solution, in addition to the estimated evolving radio channel it is not necessary for the UE to have full knowledge of the error subspace, enabling the UE to directly calculate the optimal selector identifier m opt .

[0076] As a supplement or alternative to the foregoing, there may be different criteria for identifying the best - fitting window. In some cases, more than one frequency sub - band and multiple time slots may be defined. This can improve the inference quality.

[0077] AI or ML algorithms can be used to identify the best - fitting window. AI, ML, or neural networks can identify the best - fitting window from a set of predefined windows. The window can be defined using the codebook entry c i is defined as a codebook (C), for example, c i =[L_mark, p min , p max , B, w,...], where window (i) is defined by the minimum power p min and the maximum power p max such that the window is configured to cover the entire error subspace at the frequency sub - band L_mark In the case where the set L_mark covers an extended set of frequency sub - bands and time slots, B corresponds to the number of bits, and w may correspond to a weighted cost function or a combination rule. AI, ML, or neural networks can assist in identifying the best - fitting window ci for a specific channel state.

[0078] Figure 4 A deep - neural - network structure is shown by way of example. The DNN structure includes an input layer 410, hidden layers 420, 430, and an output layer 340. The implementation may include using a deep neural network (DNN) to project the codeword - lookup problem as a classification task. The DNN consists of fully - connected layers. The predicted evolving radio channel and the error subspace can be used as an input signal. Each neuron at the output layer corresponds to a codeword in the codebook C. Since multiple windows can be selected within the considered bandwidth, the classifier can be trained for multi-label classification, where more than one codeword is accepted for each input. To achieve this behavior, a sigmoid activation function can be used at the output layer. Additionally, the classifier can be trained with a binary cross-entropy loss function. Each output neuron has a value between [0, 1], where the one closest to 1 is the corresponding codeword. Codewords with probabilities higher than a predefined threshold can be considered the selected windows. Once the window is defined, the quantization level is also defined as the number of bits that are part of the codeword. Uniform quantization can be exploited. Thus, the channel prediction can be corrected.

[0079] In an alternative, a cost function can be exploited such that it minimizes the evolving (radio) channel matrix (CTF) with the difference in channel characteristic parameters (e.g., the power level of the received signal) between the radio channels selected from the error subspace identified by the selector identifier m reported The selector identifier m reported should be as close as possible to the optimal selector identifier m opt , which means that the key performance indicator (KPI) is the confusion matrix of the selector identifier m reported with respect to the optimal selector identifier m opt .

[0080] Training data for AI, ML, or neural networks can be measured from real-world channel sounders or generated from the channel models defined in 3GPP in order to obtain the predicted evolving (radio) channel matrix The predicted evolving radio channel can be used to infer the predicted channel matrix or channel predictor for a given error model and the error subspace Additionally, a method for defining a codebook C including windows is implemented, from which the best window is selected.

[0081] Another alternative is to design the window selection problem as a segmentation task for a generative model. This method avoids the explicit construction of the codebook C, as the segmentation mask has the same dimension as the input signal and can select windows of variable size. For the segmentation task, a conditional generative adversarial network (cGAN) can be trained. In an example cGAN, the segmentation mask defines the window, the number of quantization bits is fixed, and a uniform quantizer is used to derive the selector identifier. For training, corresponding input segmentation mask pairs are constructed. The cGAN includes a generator neural network (NN) and a discriminator neural network (NN). The input to the generator NN is the concatenation of the signals and the output is the real segmentation mask Ms approximation. The input to the discriminator NN is the output from the generator NN and the true segmentation mask M s concatenated. The discriminator NN is a classifier that measures the similarity between its input signals. During training, the generator NN and the discriminator NN compete in a min-max game such that over time, the generator NN learns to generate images similar to the segmentation masks. The loss function for training the cGAN has two terms: a cross-entropy term and an L 2 term, as follows:

[0082] where G and D are the generator NN and the discriminator NN respectively.

[0083] Having a single window has a positive impact on the overhead of the report. Having more than one window at different frequency subbands enables tracking of channel variations at more than one point and thus it is more likely to identify the best selector identifier m to be reported by the UE. For example, assuming B corresponds to 3 bits and the quantization level Q for each window is 8, then the UE has to report LxB = 3x3 = 9 for L = 3 windows instead of reporting 3 bits for a single window. The number of reported bits can be reduced by defining a suitable codebook C marker (CTF) for a given error subspace such that for example for L = 3 windows, less than 9 bits are reported per reporting instance. This can be achieved by exploiting the special structure of the error subspace which may not cover all options and not all options are equally likely. Another option is to use the first two or three bits for the first window. In windows two and three, only a certain subset of options from the known error subspace are possible. This allows adjusting the meaning of the following bits in windows two and three such that these bits only cover the possible residual subspace. This can generate a set of conditional codebooks which change depending on the first bit. When the error subspace changes with changing channel conditions, the base station can update the conditional codebook periodically.

[0084] A full definition of the currently suitable codebook may lead to a relatively large DL overhead for the physical downlink control channel (PDDCH). To avoid the overhead, a set of codebooks can be predefined such that the base station can select the best-fitting codebook identifier from the set of codebooks downwards. The set of codebooks can be predefined and thus known by default at the UE or configured in an RRC message. The base station can report the codebook identifier in the PDCCH instead of the full codebook configuration. This can significantly reduce the associated DL overhead. The latest (used) codebook can be activated at defined time instances according to a predefined method.

[0085] A set of codebooks can be configured and reported as a combination of RRC, MAC ID, and DCI messages. The RRC message can be used to adapt the codebook set to the current radio conditions, i.e., the current scenario. The MAC CE message can reduce the set of codebooks to track the current long-term changes in the radio channel conditions. The DCI message can be a specific downselection of codebook C shorterm for the current short-term channel conditions and the associated error subspace. This can be used by the UE for the codebook index c i in the next UL report. The base station is configured to select the best-fit error subspace index based on the received codebook C shorterm with the codebook index c i .

[0086] According to one embodiment, the reported selector identifier m reported from a certain UE is tracked over multiple sequential reporting instances. With each reported selector identifier m(t), the base station's inference of the best-fit error model becomes more likely. In another embodiment, the frequencies of the subbands L_mark,1, L_mark,2, L_mark,3 of the window can change over time in a predefined manner. For example, the window can be configured to shift sequentially left or right by a few physical resource blocks (PRBs).

[0087] The window can be defined as a subset of the subcarriers of an orthogonal frequency division multiplexing (OFDM) symbol, such as a certain frequency subband of the channel transfer function (CTF 90 ), where 90 indicates time slot n = 90. Alternatively, the window can be defined with respect to a profiled time-domain channel impulse response rather than the channel transfer function.

[0088] The provided fine-tuning can be applied to channel prediction or implementation, where CSI is used as an input and / or output. The provided fine-tuning can be used for each CTF associated with one of the beams of the precoder in the air interface, including CSI compression.

[0089] Figure 5 An apparatus according to an embodiment is illustrated by way of example. The apparatus is capable of determining and transmitting a selector identifier indicating the quantized quality of the channel between the apparatus and a network node. The illustrated device 500 can include, for example, a terminal device such as Figure 1100 and 102. The device 500 includes a processor 510, which may include, for example, a single-core or multi-core processor. A single-core processor includes one processing core, and a multi-core processor includes more than one processing core. The processor 510 generally may include a control device. The processor 510 may include more than one processor. The processor 510 may be a control device. The processing core may include, for example, a Cortex-A8 processing core manufactured by ARM (ARM Holdings) or a Steamroller processing core designed by Advanced Micro Devices Corporation. The processor 510 may include at least one Qualcomm Snapdragon and / or Intel Atom processor. The processor 510 may include at least one application-specific integrated circuit (ASIC). The processor 510 may include at least one field-programmable gate array (FPGA). The processor 510 may be a component for executing method steps in the device 500. The processor 510 may be configured to perform actions at least in part by computer instructions.

[0090] The processor may include circuitry or be configured as one or more circuitry that are configured to perform stages of a method according to example embodiments described herein. As used in this application, the term "circuitry" may refer to one or more or all of the following: (a) only hardware circuit implementations, such as implementations only in analog and / or digital circuitry, and (b) combinations of hardware circuits and software, such as, where applicable: (i) combinations of analog and / or digital hardware circuits with software / firmware, and (ii) any part of a hardware processor with software (including a digital signal processor, software, and memory that work together to cause a device, such as a mobile phone or server, to perform various functions), and (c) hardware circuits and / or processors that require software (e.g., firmware) for operation, such as a microprocessor or part of a microprocessor, but where the software may be absent when not needed for operation.

[0091] This definition of circuitry applies to all uses of the term in this application (including any claims). As another example, as used in this application, the term circuitry also encompasses implementations of only hardware circuits or processors (or multiple processors) or a part of a hardware circuit or processor and its (or their) attendant software and / or firmware. For example and if applicable to a particular claim element, the term circuitry also encompasses a baseband integrated circuit or a processor integrated circuit for a mobile device, or a similar integrated circuit in a server, a cellular network device, or other computing or network device.

[0092] Device 500 may include a memory 520. The memory 520 may include a random access memory and / or a permanent memory. The memory 520 may include at least one RAM chip. The memory 520 may include, for example, solid state, magnetic, optical, and / or holographic memory. The memory 520 may be at least partially accessible by a processor 510. The memory 520 may be at least partially included in the processor 510. The memory 520 may be a component for storing information. The memory 520 may include computer instructions that the processor 510 is configured to execute. When computer instructions configured to cause the processor 510 to perform certain actions are stored in the memory 520 and the device 500 is generally configured to operate under the guidance of the processor 510 using the computer instructions from the memory 520, the processor 510 and / or at least one of its processing cores may be considered to be configured to perform the certain actions. The memory 520 may be at least partially external to the device 500 but accessible by the device 500.

[0093] Device 300 may include a transmitter 530. Device 500 may include a receiver 540. The transmitter 530 and the receiver 550 may be configured to send and receive information respectively according to at least one cellular or non-cellular standard. The transmitter 530 may include more than one transmitter. The receiver 540 may include more than one receiver. For example, the transmitter 530 and / or the receiver 540 may be configured to operate according to Global System for Mobile Communications (GSM), Wideband Code Division Multiple Access (WCDMA), 5G, Long Term Evolution (LTE), IS-95, Wireless Local Area Network (WLAN), Ethernet, and / or Worldwide Interoperability for Microwave Access (WiMAX) standards.

[0094] Device 500 may include a Near Field Communication (NFC) transceiver 550. The NFC transceiver 550 may support at least one NFC technology, such as NFC, Bluetooth, Wibree, or similar technologies.

[0095] Device 500 may include a User Interface (UI) 560. The UI 560 may include at least one of a display, a keyboard, a touch screen, a vibrator arranged to signal the user by vibrating the device 500, a speaker, and a microphone. The user may be able to operate the device 500 via the UI 560, for example, to accept an incoming phone call, initiate a phone call or a video call, browse the Internet, manage digital files stored in the memory 520 or accessible via the cloud through the transmitter 530 and the receiver 540 or via the NFC transceiver 550, and / or play games.

[0096] Device 500 may include or be arranged to receive a user identity module 570. The user identity module 570 may include, for example, a subscriber identity module (SIM) card that may be installed in the device 500. The user identity module 570 may include information identifying the subscription of the user of the device 500. The user identity module 570 may include password information that may be used to authenticate the identity of the user of the device 500 and / or facilitate encryption of the transmitted information and charging of the user of the device 500 for communications implemented via the device 500.

[0097] The processor 510 may be equipped with a transmitter that is arranged to output information from the processor 510 to other devices included in the device 500 via electrical leads inside the device 500. Such a transmitter may include a serial bus transmitter that is arranged to output information, for example, to the memory 520 via at least one electrical lead for storage therein. As an alternative to the serial bus, the transmitter may include a parallel bus transmitter. Similarly, the processor 510 may include a receiver that is arranged to receive information in the processor 510 from other devices included in the device 500 via electrical leads inside the device 500. Such a receiver may include a serial bus receiver that is arranged to receive information, for example, from the receiver 440 via at least one electrical lead for processing in the processor 510. As an alternative to the serial bus, the receiver may include a parallel bus receiver.

[0098] The device 500 may include Figure 5 other devices not shown. For example, in the case where the device 500 includes a smart phone, it may include at least one digital camera. Some devices 500 may include a rear camera and a front camera, where the rear camera may be intended for digital photography and the front camera for video telephony. The device 500 may include a fingerprint sensor arranged to at least partially authenticate the user of the device 500. In some example embodiments, the device 500 lacks at least one of the above devices. For example, some devices 500 may lack the NFC transceiver 550.

[0099] The processor 510, the memory 520, the transmitter 530, the receiver 540, the NFC transceiver 550, the YI 560 and / or the user identity module 570 can be interconnected in a variety of different ways via electrical leads inside the device 500. For example, each of the above devices can be connected to a main bus inside the device 500, respectively, to allow the devices to exchange information. However, as will be appreciated by those skilled in the art, this is only an example, and various ways of interconnecting at least two of the aforementioned devices can be selected depending on the embodiment. The device may include a module, for example, a module configured to send and / or a module configured to receive. The device may include: a module configured to determine a selector identifier associated with a window of a CTF of a wireless communication channel between a network node and the device; and a module configured to send the determined selector identifier to the network node.

[0100] A network node or base station such as Figure 1 104) may include at least partially corresponding parts, such as Figure 5 As shown. The network node may include at least one processor, at least one memory, a transmitter, and a receiver. The network node may be capable of determining the CSI based at least in part on a selector identifier received from a terminal device and associated with a window of a CTF of a wireless communication channel. The apparatus may include a module, for example, a module configured to send and / or a module configured to receive. The apparatus may include: a module configured to receive a selector identifier associated with a window of a CTF of a wireless communication channel between the apparatus and a terminal device; and a module configured to determine the CSI based at least in part on the selector identifier.

[0101] Figure 6It is a flowchart of a method for the network side. The stages of the illustrated method may be executed in a base station (e.g., gNB) or in a control device (when installed therein) configured to control its functions. The method includes: at stage 610, receiving, by a network node, a selector identifier that indicates a quantified quality of a channel associated with a fine-tuning configuration for the channel, where the channel includes a wireless communication channel between the network node and a terminal device. The method includes: at stage 620, determining, by the network node, a fine-tuning configuration for channel prediction at least in part based on the selector identifier. The method may further include: predefined fine-tuning configurations and sending the predefined fine-tuning configurations to the terminal device. The method may further include: sending the predefined fine-tuning configurations in at least one of the following: an RRC message, a CE of a MAC message; or a DL CI message. The method may further include: performing a fine-tuning operation based on the predefined fine-tuning configuration and the received selector identifier. The predefined fine-tuning configuration and / or performing the fine-tuning operation may be implemented by means of one or more of AR, ML, or NN. The method may include: using a predefined NN model to predefined the fine-tuning configuration and for updating the weights of the predefined NN model at the device and with the terminal device. The method may further include: storing the parameters of the predefined NN model for a specific type of terminal device and reusing the stored parameters for the same specific type of terminal device. The method may further include: defining rules for selecting the selector identifier based on at least one of the following: the average power of the quantified quality of the channel; the average power of the quantified quality at two predefined time instances or frequency subsets; the weighted sum of the quantified quality of the channel at two or more predefined frequency subsets and / or time instances; or the preprocessed cosine similarity of the quantified quality of the channel. The method may further include: obtaining the quantified quality of the channel within a window, where the window is referred to as a window matrix, the selector identifier includes one or more coefficients of the window matrix, and the window includes a subset of the quantified quality of the channel that includes the entries of the window matrix. The method may further include: selecting a best-fit error model at least in part based on the received selector identifier and applying the best-fit error model as the channel prediction model for the next channel prediction cycle.

[0102] Figure 7It is a flowchart of a method for the UE side. The stages of the shown method can be executed in, for example, a terminal device, a user equipment, an auxiliary device, or a personal computer, or in a control device configured to control its functions (when installed therein). The method includes: at stage 710, obtaining, by the terminal device, a selector identifier that indicates a quantified quality of a channel associated with a fine-tuning configuration for the channel, where the channel includes a wireless communication channel between a network node and the terminal device. The method further includes: at stage 720, sending, by the terminal device, the selector identifier to the network node. The method may further include: receiving, from the network node, a fine-tuning configuration for the channel. The fine-tuning configuration may be received in at least one of the following: an RRC message, a CE of a MAC message; or a DL CI message. The method may further include: obtaining the selector identifier as the selector identifier associated with the quantified quality of the channel that is closest to the observed value of the corresponding quality of the channel. The method may further include: obtaining the selector identifier based on at least one of the following: the average quantified quality of the channel; the average quality of the channel at two predefined time instances or frequency subsets; the weighted sum of the quantified quality at two or more predefined frequency subsets and / or time instances; or the preprocessed cosine similarity of the quantified quality of the channel. The method may include: obtaining the quantified quality of the channel within a window matrix, where the window is referred to as the window matrix, the selector identifier includes one or more coefficients of the window matrix, and the window includes a subset of the quantified quality of the channel that includes the entries of the window matrix. The method may further include: obtaining the quantified quality of the channel during a plurality of time slots.

[0103] In Figure 6 the method of Figure 7In the method, the fine-tuning configuration may include at least one of the following: codebook, window, duration, time and / or frequency resources, start frame and / or slot period; or a timer for reference signal information. The fine-tuning configuration may include a codebook, which includes one or more of the following parameters: frequency subband; frequency subband size; quantized power level value; real part value and / or imaginary part value of the CTF, angular evolution of the CTF, angular multipath component; angular time component, where time refers to the relevant tap of a certain angular beam; observation time; number of windows in the time and frequency range; weighting and definition; or rules reported for combinations of multiple windows. The quantized quality of the channel may include: CTF of the channel; measured reference signal of the channel; power level of the channel; impulse response of the channel; real part and / or imaginary part of the CTF; angular-time component, where time refers to the relevant tap of a certain angular beam; angular multipath component; or cosine similarity. The quantized quality of the channel may be based on the CTF including at least one of the following: power level of the CTF; angular evolution of the CTF; real part and / or imaginary part of the CTF; angular-time component of a certain angular beam of the CTF; angular multipath component of a certain angular beam of the CTF; or preprocessing cosine similarity of the CTF. Each quantized channel quality may be associated with a selector identifier.

[0104] The window may include a subset of the channel transfer function, which at least includes a close relationship between the observed channel characteristic parameters and the set of channel characteristic parameter values of the channel transfer function. The window may include a marked window of the channel transfer function, and the channel characteristic identifier may include the selector identifier of the marked window of the channel transfer function. The marked window may be referred to or shown as a matrix or a window matrix, where the channel characteristic identifier includes the coefficients of the matrix of the marked window, the marked window includes a subset of the channel transfer function in the frequency range, which includes the entries of the matrix of the marked window, and the channel transfer function includes the frequency domain of the radio channel.

[0105] In the previous Figure 6 and Figure 7 the window may be referred to as a marked window and shown as a matrix. The channel characteristic identifier may include the coefficients of the matrix of the marked window. The marked window may include a subset of the CTF in frequency or a frequency subband, and the subset of the CTF in frequency or the frequency subband includes the entries of the matrix of the marked window. The CTF may include the frequency domain of the radio channel.

[0106] In one embodiment, a device (e.g., a terminal device, UE) may include components for performing Figure 7 the flowchart or the above example embodiments and any combination thereof. In another embodiment, a device (e.g., a base station gNB) may include components for performing Figure 6The components of the flowchart, the above example embodiments, and any combination thereof. In another embodiment, the apparatus may include at least one processor and at least one memory including computer program code, wherein the at least one memory and the computer program code are configured to, together with the at least one processor, cause the apparatus to at least perform the above example embodiments and any combination thereof.

[0107] In one embodiment, the computer program may be configured to cause a method as described in connection with Figure 6 or Figure 7 or the above embodiments and any combination thereof. In an example embodiment, a computer program product embodied on a non-transitory computer-readable medium may be configured to control a processor to execute a process including Figure 6 or Figure 7 the flowchart, the above example embodiments, and any combination thereof.

[0108] It should be understood that, as would be recognized by those of ordinary skill in the relevant art, the disclosed example embodiments are not limited to the specific structures, process steps, or materials disclosed herein, but extend to their equivalents. It should also be understood that the terms used herein are for the purpose of describing particular example embodiments only and are not intended to be limiting.

[0109] References to one example embodiment or an example embodiment throughout the specification mean that the particular features, structures, or characteristics described in connection with the example embodiment are included in at least one example embodiment. Thus, the phrases "in one example embodiment" or "in an example embodiment" appearing throughout the specification do not necessarily all refer to the same example embodiment. Precise numerical values are also disclosed when using terms such as about or substantially in reference to numerical values.

[0110] As used herein, for convenience, a plurality of items, structural elements, compositional elements, and / or parameters may be presented in a common list. However, these lists should be interpreted as if each member of the list is individually identified as a separate and unique member. Thus, individual members of such a list should not be construed as de facto equivalents of any other member of the same list merely based on their presentation in a common group without a contrary indication. Additionally, various example embodiments and examples may be referred to herein in connection with alternatives to their various components. It should be understood that such example embodiments, examples, and alternatives should not be construed as de facto equivalents of one another, but rather should be considered as separate and autonomous representations.

[0111] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more example embodiments. In the foregoing description, numerous specific details are provided, such as examples of parameters, machine learning applications, etc., to provide a thorough understanding of the example embodiments of the present disclosure. However, those skilled in the relevant art will recognize that the present disclosure may be practiced without one or more of the specific details or with other methods, components, entities, etc. In other instances, well-known structures, entities, or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0112] While the foregoing examples illustrate the principles of the example embodiments in one or more particular applications, it will be apparent to those of ordinary skill in the art that many modifications may be made in the form, use, and details of the implementations without departing from the principles and concepts of the present disclosure and without the need for creative effort. Accordingly, the present disclosure is not intended to be limited except as defined by the claims set forth below.

[0113] The verbs “comprise” and “comprising” are used in this document as open limitations that neither exclude nor require the presence of unrecited features. Unless otherwise expressly stated, the features recited in the dependent claims may be freely combined with one another. In addition, it should be understood that the use of “a” or “an” (i.e., the singular form) throughout this document does not exclude a plurality.

Claims

1. A device, comprising: at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause the device to at least: - obtain a selector identifier, where the selector identifier indicates a quantified quality of the channel associated with a fine-tuning configuration for the channel, where the channel includes a wireless communication channel between the device and a network node; and - send the selector identifier to the network node.

2. The device according to claim 1, wherein the fine-tuning configuration includes at least one of the following: - a codebook for fine-tuning; - a window for fine-tuning; - a duration for fine-tuning; - time and / or frequency resources for fine-tuning; - a start frame and / or a time slot period for fine-tuning; or - a timer for fine-tuning reference signal information.

3. The device according to claim 1, wherein the fine-tuning configuration includes a codebook, and the codebook includes parameters, the parameters including one or more of the following: frequency sub-bands, frequency sub-band sizes, quantified power level values, real and / or imaginary part values of a channel transfer function; angular evolution of a channel transfer function; angular multipath components; angle-time components, where time refers to the relevant taps of a certain angular beam; observation time; number of windows, weighting and definition within a time and frequency range, or rules for combined reporting for multiple windows.

4. The device according to any one of the preceding claims, wherein the quantified quality of the channel includes at least one of the following: - the channel transfer function of the channel; - the measured reference signal of the channel; - the power level of the channel; - the impulse response of the channel; - the angular evolution of the channel transfer function; - the real and / or imaginary parts of the channel transfer function; - angle-time components, where time refers to the relevant taps of a certain angular beam; - angular multipath components; or - cosine similarity.

5. The device according to any one of the preceding claims, wherein the device is further caused to: receive the fine-tuning configuration for the channel from a network node.

6. The device according to any one of the preceding claims, wherein the device is further caused to: receive the fine-tuning configuration in at least one of the following: a radio resource control message, a control element of a media access control message, or a downlink control information message.

7. The device according to any one of the preceding claims, wherein each quantified quality of the channel is associated with a selector identifier.

8. The device according to any one of the preceding claims, wherein the device is further caused to: obtain the selector identifier as the selector identifier associated with the quantified quality of the channel that is closest to the observed value of the corresponding quality of the channel.

9. The device according to any one of the preceding claims, wherein the device is further caused to: The selector identifier is obtained based on at least one of the following: the average quantized quality of the channel; the average quality of the channel at two predefined time instances or frequency subsets; the weighted sum of the quality of quantization at two or more predefined frequency subsets and / or time instances; or the preprocessed cosine similarity of the quality of quantization of the channel.

10. The apparatus according to any one of the preceding claims, wherein the quality of quantization of the channel is obtained within a window, and the window is referred to as a window matrix, wherein the selector identifier comprises one or more coefficients of the window matrix, the window comprising a subset of the quality of quantization of the channel, the subset comprising the entries of the window matrix.

11. The apparatus according to any one of the preceding claims, wherein the apparatus is further caused to: Obtain the quality of quantization of the channel during a plurality of time slots.

12. An apparatus, comprising: at least one processor; and at least one memory storing instructions which, when executed by the at least one processor, cause the apparatus to at least: - receive a selector identifier from a terminal device, the selector identifier indicating the quality of quantization of the channel associated with a fine-tuning configuration for a channel, wherein the channel comprises a wireless communication channel between the apparatus and the terminal device; and - determine the fine-tuning configuration for channel prediction at least in part based on the received selector identifier.

13. The apparatus according to claim 12, wherein the fine-tuning configuration comprises at least one of the following: - a codebook for fine-tuning; - a window for fine-tuning; - a duration for fine-tuning; - time and / or frequency resources for fine-tuning; - a start frame and / or slot period for fine-tuning; or - a timer for fine-tuning reference signal information.

14. The apparatus according to claim 12, wherein the fine-tuning configuration comprises a codebook, and the codebook comprises parameters, the parameters comprising one or more of the following: a frequency sub-band, a frequency sub-band size, a quantized power level value, a real part value and / or an imaginary part value of a channel transfer function; an angular evolution of a channel transfer function; an angular multipath component; an angular time component, wherein time refers to a relevant tap for a certain angular beam; an observation time; a number of windows within a time and frequency range, a weighted sum definition, or a rule for combined reporting for a plurality of windows.

15. The apparatus according to any one of claims 12 to 14, wherein the quality of quantization of the channel comprises at least one of the following: - a channel transfer function of the channel; - a measured reference signal of the channel; - a power level of the channel; - an impulse response of the channel; - a real part and / or an imaginary part of a channel transfer function; - an angle-time component, wherein time refers to a relevant tap for a certain angular beam; - an angular multipath component; or - a cosine similarity.

16. The apparatus according to any one of claims 12 to 15, wherein each quantized channel quality of the channel is associated with a selector identifier.

17. The apparatus according to any one of claims 12 to 16, wherein the apparatus is further caused to: Pre - define the fine - tuning configuration; and Send the pre - defined fine - tuning configuration to the terminal device.

18. The apparatus according to claim 17, wherein the apparatus is further caused to: Send the pre - defined fine - tuning configuration in at least one of the following: a radio resource control message, a control element of a media access control message, or a downlink control information message.

19. The apparatus according to any one of claims 17 to 18, wherein the apparatus is further caused to: Perform a fine - tuning operation based on the pre - defined fine - tuning configuration and the received selector identifier.

20. The apparatus according to any one of claims 17 to 19, wherein the apparatus is further caused to: Pre - define the fine - tuning configuration and / or perform the fine - tuning operation by means of at least one of the following: artificial intelligence, machine learning, or neural network.

21. The apparatus according to claim 20, wherein the apparatus is further caused to: Use a pre - defined neural network model to pre - define the fine - tuning configuration; and Update the weights of the pre - defined neural network model at the apparatus and with the terminal device.

22. The apparatus according to claim 20 or 21, wherein the apparatus is further caused to: Store the parameters of the pre - defined neural network model for a specific type of terminal device; Re - use the stored parameters for the same specific type of terminal device.

23. The apparatus according to any one of claims 12 to 22, wherein the apparatus is further caused to: Define a rule for selecting the selector identifier based on at least one of the following: the average power of the quantized quality of the channel; the average power of the quantized quality of the channel at two or more pre - defined time instances or frequency sub - sets; the weighted sum of the quantized quality of the channel at two or more pre - defined frequency sub - sets and / or time instances; or the pre - processed cosine similarity of the quantized quality of the channel.

24. The apparatus according to any one of claims 12 to 23, wherein the apparatus is further caused to obtain the quantized quality of the channel within a window, wherein the window is referred to as a window matrix, wherein the selector identifier includes one or more coefficients of the window matrix, the window includes a subset of the quantized quality of the channel, and the subset includes the entries of the window matrix.

25. The apparatus according to any one of claims 12 to 24, wherein the apparatus is further caused to: Select an optimal - fit error model at least partly based on the received selector identifier; and Apply the optimal - fit error model as a channel prediction model for the next channel prediction period.

26. A method, comprising: - Obtaining, by a terminal device, a selector identifier that indicates the quantized quality of the channel associated with a fine - tuning configuration for a channel, wherein the channel comprises a wireless communication channel between a network node and the terminal device; - Sending, by the terminal device, the selector identifier to the network node.

27. The method according to claim 26, wherein the fine-tuning configuration includes at least one of the following: - A codebook for fine-tuning; - A window for fine-tuning; - A duration for fine-tuning; - Time and / or frequency resources for fine-tuning; - A start frame and / or a time slot period for fine-tuning; or - A timer for fine-tuning reference signal information.

28. The method according to claim 26, wherein the fine-tuning configuration includes a codebook, and the codebook includes parameters, and the parameters include one or more of the following: a frequency sub-band, a frequency sub-band size, a quantized power level value, a real value and / or an imaginary value of a channel transfer function; an angular evolution of a channel transfer function; an angular multipath component; an angle-time component, where time refers to a relevant tap of a certain angular beam; an observation time; a number, a weighting and a definition of a window within a time and frequency range, or a rule for combined reporting for multiple windows.

29. The method according to any one of claims 26 to 28, wherein the quantized quality of the channel includes at least one of the following: - The channel transfer function of the channel; - A measured reference signal of the channel; - The power level of the channel; - The impulse response of the channel; - An angular evolution of a channel transfer function; - A real part and / or an imaginary part of a channel transfer function; - An angle-time component, where time refers to a relevant tap of a certain angular beam; - An angular multipath component; or - A cosine similarity.

30. The method according to any one of claims 26 to 29, further comprises: The terminal device receiving, from the network node, the fine-tuning configuration for the channel.

31. The method according to any one of claims 26 to 30, further comprises: The terminal device receiving, from the network node, the fine-tuning configuration in at least one of the following: a radio resource control message, a control element of a media access control message, a downlink control information message.

32. The method according to any one of claims 26 to 31, wherein each quantized quality of the channel is associated with a selector identifier.

33. The method according to any one of claims 26 to 32, further comprises: The terminal device obtaining the selector identifier as the selector identifier associated with the quantized quality of the channel that is closest to an observed value of the corresponding quality of the channel.

34. The method according to any one of claims 26 to 33, further comprises: The terminal device obtaining the selector identifier based on at least one of the following: an average quantized quality of the channel; An average quality of the channel at two predefined time instances or frequency subsets; A weighted sum of the quantized qualities at two or more predefined frequency subsets and / or time instances; or a preprocessed cosine similarity of the quantized quality of the channel.

35. The method according to any one of claims 26 to 34, comprises: Obtain the quantified quality of the channel within a window by the terminal device, where the window is referred to as a window matrix, and the selector identifier includes one or more coefficients of the window matrix, the window includes a subset of the quantified quality of the channel, and the subset includes the entries of the window matrix.

36. The method according to any one of claims 26 to 35, further comprises: Obtain the quantified quality of the channel during a plurality of time slots.

37. A method, comprises: - Receive, by a network node, a selector identifier from a terminal device, the selector identifier indicating the quantified quality of the channel associated with a fine-tuning configuration for the channel, where the channel includes a wireless communication channel between the network node and the terminal device; and - Determine, by the network node, the fine-tuning configuration for channel prediction at least partially based on the selector identifier.

38. The method according to claim 37, wherein the fine-tuning configuration includes at least one of the following: - A codebook for fine-tuning; - A window for fine-tuning; - A duration for fine-tuning; - Time and / or frequency resources for fine-tuning; - A start frame and / or a time slot period for fine-tuning; or - A timer for fine-tuning reference signal information.

39. The method according to claim 37, wherein the fine-tuning configuration includes a codebook, and the codebook includes parameters, the parameters including one or more of the following: frequency subbands, frequency subband sizes, quantified power level values, real and / or imaginary part values of the channel transfer function; angular evolution of the channel transfer function; angular multipath components; angle-time components, where time refers to the relevant taps for a certain angular beam; observation time; number of windows within a time and frequency range, weighting and definition, or rules for combined reporting for multiple windows.

40. The method according to any one of claims 37 to 39, wherein the quantified quality of the channel includes at least one of the following: - The channel transfer function of the channel; - The measured reference signal of the channel; - The power level of the channel; - The impulse response of the channel; - The real and / or imaginary part of the channel transfer function; - Angle-time components, where time refers to the relevant taps for a certain angular beam; - Angular multipath components; or - Cosine similarity.

41. The method according to any one of claims 37 to 40, wherein each quantified channel quality of the channel is associated with a selector identifier.

42. The method according to any one of claims 37 to 41, further comprises: Predefine the fine-tuning configuration by the network node; Send the predefined fine-tuning configuration from the network node to the terminal device.

43. The method according to claim 42, further comprises: Send the predefined fine-tuning configuration by the network node in at least one of the following: radio resource control message, control element of media access control message, downlink control information message.

44. The method according to any one of claims 42 to 43, further comprises: The fine-tuning operation is performed by the network node based on the predefined fine-tuning configuration and the received selector identifier.

45. The method according to claims 42 to 43, further comprising: predefining the fine-tuning configuration and / or performing the fine-tuning operation by the network node by means of at least one of: artificial intelligence, machine learning, or neural network.

46. The method according to claim 45, further comprising: predefining the fine-tuning configuration by the network node using a predefined neural network model, and for updating weights of the predefined neural network model at the network node and with the terminal device.

47. The method according to claim 45 or 46, further comprising: storing, by the network node, parameters of the predefined neural network model for a specific type of terminal device; reusing, by the network node, the stored parameters for the same type of terminal device.

48. The method according to any one of claims 37 to 47, further comprising: defining, by the network node, rules for selecting a selector identifier based on at least one of: the average power of the quantized quality of the channel; the average power of the quantized quality at two predefined time instances or frequency subsets; the weighted sum of the quantized quality of the channel at two or more predefined frequency subsets and / or time instances; or the preprocessed cosine similarity of the quantized quality of the channel.

49. The method according to any one of claims 37 to 48, further comprising: obtaining, by the network node, the quantized quality of the channel within a window, the window being referred to as a window matrix, wherein the selector identifier includes one or more coefficients of the window matrix, and the window includes a subset of the quantized quality of the channel, the subset including entries of the window matrix.

50. The method according to any one of claims 37 to 49, further comprising: selecting, by the network node, an optimal fitting error model at least partially based on the received selector identifier; and applying, by the network node, the optimal fitting error model as a channel prediction model for a next channel prediction cycle.

51. A non-transitory computer-readable medium comprising program instructions that, when executed by a device, cause the device to perform at least the following operations: - obtaining a selector identifier, wherein the selector identifier indicates the quantized quality of the channel associated with a fine-tuning configuration for the channel, wherein the channel comprises a wireless communication channel between a network node and the device; and - sending the selector identifier to the network node.

52. A non-transitory computer-readable medium comprising program instructions that, when executed by a device, cause the device to perform at least the following operations: - receiving a selector identifier from a terminal device, wherein the selector identifier indicates the quantized quality of the channel associated with a fine-tuning configuration for the channel, wherein the channel comprises a wireless communication channel between the device and the terminal device; and -Determine the fine-tuning configuration for channel prediction at least partially based on the received selector identifier.

53. A computer program comprising program instructions stored thereon for performing the method according to at least one of claims 26 to 36 or 37 to 50.

54. An apparatus, comprising: -means for obtaining a selector identifier indicating the quantified quality of the channel associated with the fine-tuning configuration for the channel, wherein the channel comprises a wireless communication channel between a network node and the apparatus; and -means for sending the selector identifier to the network node.

55. The apparatus according to claim 54, comprising means for performing the method according to any one of claims 26 to 36.

56. An apparatus, comprising: -means for receiving a selector identifier indicating the quantified quality of the channel associated with the fine-tuning configuration for the channel, wherein the channel comprises a wireless communication channel between the apparatus and a terminal device; and -means for determining the fine-tuning configuration for channel prediction at least partially based on the selector identifier.

57. The apparatus according to claim 56, comprising means for performing the method according to any one of claims 37 to 50.

Images