Techniques for automatically encoding channel feedback
The introduction of time domain correlated codebooks and machine learning-assisted CSI feedback in 3GPP networks addresses high Doppler challenges, reducing overhead and improving throughput in high mobility scenarios.
Patent Information
- Application Number
- CN202280095486.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-29
- Publication Date
- 2024-12-06
AI Technical Summary
Existing 3GPP network technologies face challenges in efficiently handling high Doppler situations, leading to increased feedback overhead and inadequate downlink throughput due to outdated pre-decoder designs that fail to account for time-varying channels and channel aging.
Introduce a codebook design that incorporates time domain correlation and predictive pre-decoders, utilizing Doppler domain compression and machine learning techniques to enhance CSI feedback, allowing for reduced signaling overhead and improved downlink throughput in high mobility scenarios.
The proposed solution reduces feedback overhead and extends CSI feedback validity, effectively handling high Doppler conditions and channel aging, thereby enhancing network performance in high-speed wireless environments.
Smart Images

Figure CN119096479A_ABST
Abstract
Description
Background Art
[0001] In order to facilitate communication between a base station and a user equipment (UE) in a 3rd Generation Partnership Project (3GPP) network, a pre-decoder is implemented by the base station for a signal transmitted by the base station. The base station may determine the value of the pre-decoder based on a channel state information (CSI) signal fed back from the UE. Specifically, the UE performs measurements on the signal received from the base station and feeds back information about the measurements for determining the value of the pre-decoder. BRIEF DESCRIPTION OF THE DRAWINGS
[0002] Figure 1 Example antenna structures for a base station according to some embodiments are illustrated.
[0003] Figure 2 An example spatial beam selection representation is illustrated in accordance with some embodiments.
[0004] Figure 3 An example frequency domain (FD) component selection arrangement according to some embodiments herein is illustrated.
[0005] Figure 4 Another example FD component selection arrangement according to some embodiments is illustrated.
[0006] Figure 5 A first portion of an example bitmap generation process according to some embodiments is illustrated.
[0007] Figure 6 A second portion of an example bitmap generation process is illustrated according to some embodiments.
[0008] Figure 7 Example channel state information (CSI) reporting methods according to some embodiments are illustrated.
[0009] Figure 8 An example codebook and power delay profile with a single table according to some embodiments is illustrated.
[0010] Fig. 9 A portion of a codebook having multiple tables according to some embodiments is illustrated.
[0011] Fig.10 Example spatial beam selection according to some embodiments is illustrated.
[0012] Fig.11 Example FD component selection and indication methods according to some embodiments are illustrated.
[0013] Fig.12 An example quantization design according to some embodiments is illustrated.
[0014] Fig.13Example differential encoding methods according to some embodiments are illustrated.
[0015] Fig.14 A schematic diagram illustrating a UE providing feedback to a base station according to some embodiments.
[0016] Fig.15 Example CSI feedback operations according to some embodiments are illustrated.
[0017] Fig.16 Another example CSI feedback operation according to some embodiments is illustrated.
[0018] Fig.17 Another example CSI feedback operation according to some embodiments is illustrated.
[0019] Fig.18 Another example CSI feedback operation according to some embodiments is illustrated.
[0020] Fig.19 Smoothing operations according to some embodiments are illustrated.
[0021] Fig. 20 Another example CSI feedback operation according to some embodiments is illustrated.
[0022] Fig.21 Another example CSI feedback operation according to some embodiments is illustrated.
[0023] Fig. 22 An example procedure for generating a CSI report according to some embodiments is illustrated.
[0024] Fig.23 A flow chart for generating a CSI report according to some embodiments is illustrated.
[0025] Fig.24 An operational flow / algorithm structure according to some embodiments is illustrated.
[0026] Fig.25 An example UE according to some embodiments is illustrated.
[0027] Fig.26 An example next generation Node B (gNB) according to some embodiments is illustrated. DETAILED DESCRIPTION
[0028] The following specific embodiments relate to the accompanying drawings. The same figure numbers may be used to identify the same or similar elements in different drawings. In the following description, for the purpose of illustration and not limitation, specific details, such as specific structures, architectures, interfaces, technologies, etc., are set forth to provide a thorough understanding of various aspects of various embodiments. However, it will be apparent to those skilled in the art who benefit from the present disclosure that various aspects of various embodiments may be practiced in other examples that deviate from these specific details. In some cases, descriptions of well-known devices, circuits, and methods are omitted so that the descriptions of various embodiments are not obscured by unnecessary details. For the purposes of this document, the phrase "A or B" refers to (A), (B), or (A and B).
[0029] The following is a glossary of terms that may be used in this disclosure.
[0030] As used herein, the term "circuit" refers to, is part of, or includes a hardware component such as an electronic circuit, a logic circuit, a processor (shared, dedicated, or group) or a memory (shared, dedicated, or group) configured to provide the described functions, an application specific integrated circuit (ASIC), a field programmable device (FPD) (e.g., a field programmable gate array (FPGA), a programmable logic device (PLD), a complex PLD (CPLD), a high capacity PLD (HCPLD), a structured ASIC, or a programmable system on chip (SoC)), a digital signal processor (DSP), etc. In some embodiments, the circuit may execute one or more software or firmware programs to provide at least some of the described functions. The term "circuit" may also refer to a combination of one or more hardware elements and a program code for performing the functions of the program code (or a combination of circuits used in an electrical or electronic system). In these embodiments, the combination of hardware elements and program code may be referred to as a specific type of circuit.
[0031] As used herein, the term "processor circuit" refers to, is part of, or includes a circuit capable of sequentially and automatically performing a series of arithmetic or logical operations or recording, storing, or transferring digital data. The term "processor circuit" may refer to an application processor, a baseband processor, a central processing unit (CPU), a graphics processing unit, a single-core processor, a dual-core processor, a triple-core processor, a quad-core processor, or any other device capable of executing or otherwise operating computer-executable instructions (such as program code, software modules, and / or functional processes).
[0032] As used herein, the term "interface circuit" refers to a circuit that enables, is a part of, or includes information exchange between two or more components or devices. The term "interface circuit" may refer to one or more hardware interfaces, such as a bus, an I / O interface, a peripheral component interface, a network interface card, etc.
[0033] As used herein, the term "user equipment" or "UE" refers to a device that has radio communication capabilities and can describe a remote user of network resources in a communication network. In addition, the terms "user equipment" or "UE" may be considered synonymous and may be referred to as a client, mobile phone, mobile device, mobile terminal, user terminal, mobile unit, mobile station, mobile user, subscriber, user, remote station, access agent, user agent, receiver, radio equipment, reconfigurable radio equipment, reconfigurable mobile device, etc. In addition, the term "user equipment" or "UE" may include any type of wireless / wired device or any computing device that includes a wireless communication interface.
[0034] As used herein, the term "computer system" refers to any type of interconnected electronic devices, computer devices, or components thereof. Additionally, the term "computer system" or "system" may refer to various components of a computer that are communicatively coupled to one another. Furthermore, the term "computer system" or "system" may refer to multiple computer devices or multiple computing systems that are communicatively coupled to one another and configured to share computing resources or networked resources.
[0035] As used herein, the term "resource" refers to a physical or virtual device, a physical or virtual component within a computing environment, or a physical or virtual component within a specific device, such as a computer device, a mechanical device, a memory space, a processor / CPU time, a processor / CPU usage, a processor and accelerator load, a hardware time or usage, a power supply, an input / output operation, a port or a network socket, a channel / link allocation, throughput, memory usage, storage, a network, a database and an application, a unit of work, etc. "Hardware resources" may refer to computing, storage or network resources provided by physical hardware elements. "Virtualized resources" may refer to computing, storage or network resources provided by a virtualized infrastructure to an application, a device, a system, etc. The term "network resources" or "communication resources" may refer to resources that a computer device / system can access via a communication network. The term "system resources" may refer to any kind of shared entity that provides a service, and may include computing resources or network resources. System resources may be considered as a set of coherent functions, network data objects or services that can be accessed through a server, wherein such system resources reside on a single host or multiple hosts and can be clearly identified.
[0036] As used herein, the term "channel" refers to any tangible or intangible transmission medium for transmitting data or data streams. The term "channel" may be synonymous or equivalent to "communication channel", "data communication channel", "transmission channel", "data transmission channel", "access channel", "data access channel", "link", "data link", "carrier", "radio frequency carrier" or any other similar term representing a path or medium through which data is transmitted. In addition, as used herein, the term "link" refers to a connection between two devices for sending and receiving information.
[0037] As used herein, the terms "instantiate," "instantiate," and the like refer to the creation of an instance. "Instance" also refers to a specific occurrence of an object, which may occur, for example, during the execution of program code.
[0038] The term "connected" may mean that two or more elements at a common communication protocol layer have an established signaling relationship with each other through a communication channel, link, interface, or reference point.
[0039] As used herein, the term "network element" refers to physical or virtualized equipment or infrastructure for providing wired or wireless communication network services. The term "network element" may be considered synonymous with or referred to as a networked computer, networking hardware, network equipment, network node, virtualized network function, etc.
[0040] The term "information element" refers to a structural element that contains one or more fields. The term "field" refers to the individual contents of an information element, or a data element that contains the contents. An information element may include one or more additional information elements.
[0041] In Release 18 (Rel-18) of the Third Generation Partnership Project (3GPP) for Radio Access Networks (RAN), a codebook design utilizing time domain correlation and a predictive pre-decoder for high Doppler situations may be included. A design utilizing a compact representation for Doppler domain extension is disclosed herein in some embodiments. In this way, low feedback overhead may be achieved and downlink throughput in high Doppler situations may be improved. Using the disclosed CSI feedback, CSI for multiple PDSCH opportunities that may be extended into the time domain may be derived or obtained at a gNB. For each PDSCH opportunity, a PMI and a channel quality indicator (CQI), including a wideband CQI and a sub-band CQI, may be derived or obtained at a base station. In addition, by using an oversampling factor (Rd), multiple pre-decoders for different orthogonal frequency division multiplexing (OFDM) symbols in the same physical downlink shared channel (PDSCH) may be used by a base station (such as a next generation Node B (gNB) (including a gNB 2600( Fig.26 ))Export.
[0042] Embodiments disclosed herein may introduce the time domain in the codebook design. For example, a user equipment (UE) may report a selected spatial beam for each Doppler component, may report a selected frequency domain (FD) component for each Doppler component, and / or may report a selected time domain (TD) component. In addition, the UE may report any one or all of the number of selected spatial beams, the number of selected FD components, and / or the number of selected TD components. The selection of non-zero (NZ) linear coefficients (LC) coefficients may be patterned by a bitmap and a component. Alternatively, the selection of NZ LC coefficients may be by multiple bitmaps, or the Doppler shift may be predicted by at least the spatial beam, the delay offset, and / or the UE position.
[0043] To reduce signaling overhead, the component composition patterns and their frequencies of occurrence can be indicated to the base station. Huffman coding schemes can then be used to refer to those patterns without using bitmaps, thereby reducing signaling overhead. The strongest LC coefficients among all spatial beams, FD components, and TD components can be shifted to the origin position relative to the FD components and TD components. The same shift can be applied to the LC coefficients on all tables.
[0044] LC quantization may be by a fixed quantizer (specified in the specification), a parameterized quantizer with parameters that can be configured by the base station and / or reported by the UE. In some embodiments, the fixed quantizer may be predefined (such as defined in the specification). Additionally, in some embodiments, the UE may report a UE-defined quantizer to the gNB.
[0045] To allow for better quantization, a UE-defined quantizer may be provided to the gNB along with Radio Resource Control (RRC) signaling and / or Medium Access Control (MAC) Control Element (CE) and / or Channel State Information (CSI) reporting. Furthermore, multiple versions may be active simultaneously and the UE may refer to the quantizer version in the CSI reporting.
[0046] For two-stage quantization, the LC coefficients may be divided into one or more sets and a reference amplitude may be determined for each set.The time domain dimension for the reported precoding matrix indicator (PMI) is determined by the maximum gap between the CSI feedback and the time when the last precoder may be used.
[0047] Embodiments herein may support configuration of Rd to allow multiple pre-decoders within the same time slot / same PDSCH, thereby taking into account high Doppler conditions. Embodiments herein may support differential coding of sub-band channel quality indicators (CQIs) across time and / or frequency. In some embodiments, Huffman coding may be used to reduce feedback overhead.
[0048] The embodiments of this document may use Doppler domain compression to achieve Rel-18 Type II enhancement. High-speed scenarios and overhead reduction are two driving factors for utilizing the Doppler domain. Further CSI feedback reduction taking into account the Doppler domain may be achieved by the embodiments of this document. For a single CSI feedback report, the base station may determine a pre-decoder for multiple times of PDSCH transmission. The 15th, 16th, and 17th release (Rel-15 / 16 / 17) design recommendations for a pre-decoder for a base station to determine multiple times of PDSCH transmission from a single CSI feedback report based on PMI / CQI / rank indicator (RI) may be outdated soon after the CSI feedback report due to channel aging. Thus, the design goals in Rel-18 may better handle time-varying channels.
[0049] Note that this is also related to CSI feedback complexity. With Rel-15 / 16 / 17 designs, the base station can still work with the current design even if high mobility is encountered at the UE by triggering frequent CSI reports (full Type-II codebook feedback is not supported on the Physical Uplink Control Channel (PUCCH), so aperiodic CSI reporting should be used in this case). With Rel-18 designs, the UE may be able to generate CSI feedback reports with much longer validity time than the legacy CSI feedback from Rel-15 / 16 / 17.
[0050] Frequent triggering of CSI reports with Rel-15 / 16 / 17 may still not be sufficient to handle channel aging due to the minimum time gap between aperiodic (AP) CSI report triggering and CSI reporting on PUCCH or physical uplink shared channel (PUSCH) (Z) and the minimum gap between AP CSI measurement resources and CSI reporting on PUCCH or PUSCH (Z'). Thus, there is also a real need to handle wireless channels with high mobility.
[0051] Another potential benefit may be lower system overhead for measurement resources. Instead of providing aperiodic (AP) channel state information reference signal (CSI-RS) resources for channel measurement resources (CMR) and / or interference measurement resources (IMR) for each AP CSI report, multiple opportunities for AP CSI-RS resources or semi-persistent CSI-RS resources or static CSI-RS resources may be provided for a single CSI report. When employing these opportunities for CSI-RS resources, the UE may be able to build a prediction model of the channel response and / or a prediction model for the pre-decoder.
[0052] Some embodiments describe adaptation of image processing / video processing techniques to facilitate CSI compression and feedback. Aspects describe various approaches for CSI feedback with machine learning, time domain compression for high-speed wireless channels, and machine learning-assisted feedback.
[0053] In a first approach for CSI feedback, the data may be deemed too complex for humans to decipher. This data, with limited pre-processing or filtering, may be provided to a neural network with a prescribed structure to train the neural network. The trained neural network may be used to form inferences, which may be an inherent part of the training process. More complex training may involve various levels / layers in the neural network and refinement of the network architecture.
[0054] In the second method for CSI feedback, more extensive pre-processing or filtering of the data can be provided to trim / convert the raw data to a suitable domain. Such data processing can facilitate subsequent machine learning processing.
[0055] The proposed machine learning (ML) / artificial intelligence (AI) method can be used to handle the feedback design of linear combination coefficients under various formulations of pre-decoder design. In addition, other CSI feedback design aspects can be utilized consistent with the embodiments described herein.
[0056] Figure 1 An example antenna structure 100 for a base station according to some embodiments is illustrated. Conventional antennas may be placed on a base station antenna array. Specifically, the antenna structure 100 may be placed on a base station such as a gNB 2600 ( Fig.26 )) as part of a base station antenna array.
[0057] The antenna structure 100 may include one or more antennas. The antennas may transmit signals at different antenna polarizations. For example, the illustrated antenna structure 100 may transmit signals using a first polarization (which may be referred to as "polarization 0") and a second polarization (which may be referred to as "polarization 1"). Specifically, the antenna structure 100 shows a first antenna 102 having a first polarization (indicated by a solid line) and a second antenna 104 having a second polarization (indicated by a dotted line). The antenna structure 100 may include one or more antennas having a first polarization (indicated by a solid line) and one or more antennas having a second polarization (indicated by a dotted line). In some embodiments, the second polarization may be orthogonal to the first polarization. Although the first polarization and the second polarization are described as being generated by separate antennas, it should be understood that in other embodiments, a single antenna may implement these two polarizations. In addition, more polarizations may be implemented in other embodiments, where these polarizations may be implemented by a single antenna or different antennas.
[0058] One or more signals may be transmitted by an antenna of the antenna structure 100. A signal transmitted by an antenna having a first polarization may be transmitted in the first polarization, and a signal transmitted by an antenna having a second polarization may be transmitted in the second polarization. One or more predecoders may determine the phase and amplitude of a signal transmitted by an antenna. A predecoder may be used to determine the amplitude of a signal transmitted by an antenna and / or which antenna will transmit a signal. In some embodiments, a predecoder may also be used to determine the direction in which a signal will be transmitted, such as in a beamforming implementation. A predecoder may be defined based on CSI feedback received from a UE. For example, a base station may receive CSI feedback from a UE and may determine a predecoder value for a predecoder corresponding to a UE based on the CSI feedback (e.g., by a signal leakage ratio in a specific implementation). The base station may utilize the determined predecoder value for a predecoder for a precoded signal to be transmitted to the UE.
[0059] The base station may determine the pre-decoder value for the UE's pre-decoder based on a formula for defining a codebook. For example, in a conventional implementation, the base station may determine the pre-decoder value for the UE's pre-decoder based on To determine the pre-decoder value for the spatial layer, where p is a polarization index (e.g., for a polarization at +45° (which may be a first polarization), p=0, and for a polarization at -45° (which may be a second polarization), p=1), when the polarization index is 0, there are B0 significant beams for the transmit (Tx) antenna, and when the polarization index is 1, there are B1 significant beams for the Tx antenna. For the polarization index p, b is a beam with a departure angle (θ b,p ,φ b,p ) of the ray (ray (p,b)), A(θ b,p ,φ b,p ) is (θ b,p ,φ b,p ) array response, τ b,p is the relative delay, and a b is the path gain including the amplitude and phase of ray b. Assuming a conventional antenna element arrangement, then (θ b,p ,φ b,p ) can be mapped to (i1,i2,p1,p2), where p1(0≤p1≤O1-1) and p2(0≤p2≤O2-1) are the oversampling factors in the vertical and horizontal domains, respectively, and (i1,i2) is the spatial beam index. b,p are the complex coefficients of the connected spatial beams and are delayed τ according to the reference receiver timing b,p is the relative delay of the light (p, b). The base station may apply the determined precoder value to a signal sent by the base station to the UE via a precoder.
[0060] The base station may determine the precoder for the layer of the UE based on the CSI received from the UE. For example, the precoder for the layer may be composed of a size -P×N3 matrix Given, where W1 is the spatial beam selection or spatial beam-based selection, is the bitmap design and quantizer design for connecting the spatial beam and FD components, and is based on the selection of the FD component. P may be equal to 2N1N2, which may be equal to the number of spatial domain (SD) dimensions, N1 being the number of antenna ports in one dimension (e.g., for the vertical domain, N1=2, Figure 2 ), and N2 is the number of antenna ports in another dimension (e.g., for the horizontal domain, N2=4, Figure 2 ). N3 may be equal to the number of FD dimensions. Precoder normalization may be applied, where the precoder normalization may be defined by the precoding matrix for a given rank, and the units of N3 will be normalized to the norm 1 / sqrt(rank), where sqrt(rank) is the square root of the rank indicator.
[0061] SD selection / compression / quantization may be applied. L spatial domain basis vectors common to both polarizations may be selected (which are mapped to both polarizations, so there are a total of 2L spatial beams for both polarizations). Using Compression / quantization in the spatial domain can be applied to select spatial beams associated with significant power, where is an N1N2×1 orthogonal DFT vector (same as Rel.15 Type II).
[0062] FD selection / compression / quantization can be applied. Compression is performed, where is M size - N3×1 orthogonal DFT vectors to select FD components with significant power for the spatial layer. The number of FD components M may be configurable. L and M may be configured by the gNB. In some embodiments, the FD compression unit may be determined by the number of CQI subbands and {PMI subband size = CQI subband size} as a default, and may be determined by {PMI subband size = CQI subband size / R} as a secondary selection. The value of R may be fixed to 2. The FD compression unit parameter R may be high-level configurable. The number of FD compression units M may be determined by Determine, among which The value of M can be configured at a higher level, such as via R and p. SB The value of N3 (the number of CQI subbands) can be obtained by N3=N SB ×R is determined. R∈{1,2} can be configured by a higher level.
[0063] Figure 2 An exemplary spatial beam selection representation 200 according to some embodiments is illustrated. (i1, i2) may be used to select the main direction of a spatial beam. (q1, q2) may be used to fine-tune the direction of a spatial beam. To ensure an orthogonal basis, the same (q1, q2) may be used for all selected spatial beams.
[0064] The spatial beam selection representation 200 represents the spatial beams that can be transmitted by one or more antennas. Specifically, in some embodiments, the spatial beam selection representation 200 can indicate the spatial beams that can be transmitted by the antenna structure 100 ( Figure 1 ) of the antenna. The spatial beams (represented by circles in the illustrated spatial beam selection representation 200) may be grouped into groups having a plurality of spatial beams (e.g., 16 spatial beams) (as indicated by squares around the spatial beam groups in the illustrated spatial beam selection representation 200), where each of the groups may correspond to an antenna or paired antennas having two different polarizations. (i1, i2) may indicate a selected group, and (q1, q2) may indicate a specific spatial beam within the selected group.
[0065] The spatial beam selection representation 200 may include a plurality of groups (e.g., two groups) in a first direction and a plurality of groups (e.g., four groups) in a second direction, thereby producing a matrix of groups, such as a 2×4 arrangement of groups. Each group may have a plurality of spatial beams (e.g., four spatial beams) in a first direction and a plurality of spatial beams (e.g., four spatial beams) in a second direction. For example, the spatial beam selection representation 200 may include a first group 202. The first group 202 may include 16 spatial beams in a 4×4 arrangement. The first group 202 may include orthogonal discrete Fourier transform (DFT) beams 204, as indicated by the filled circles in the spatial beam selection representation 200. The first group 202 may include rotated DFT beams 206, as indicated by the circles with diagonal lines in the spatial beam selection representation 200. The rotated DFT beams 206 may have a rotation factor and Specifically, the rotated DFT beam 206 may be obtained by rotating one spatial beam of the orthogonal DFT beam 204 in the first direction and rotating two spatial beams in the second direction. The unfilled circles of the first group 202 may include oversampled DFT beams. Each of these groups may have the same beam arrangement as the first group 202. Specifically, the arrangement of the orthogonal DFT beams, the rotated DFT beams, and the oversampled DFT beams may be in the same position relative to these groups as the orthogonal DFT beams 204, the rotated DFT beams 206, and the oversampled DFT beams in the first group 202. The positions of the beams in each of these groups being in the same relative position may ensure an orthogonal basis. It may be possible to select different spatial beams for different antenna polarizations.
[0066] FD component selection can be achieved by the embodiments of this article. Figure 3 An exemplary FD component selection arrangement 300 according to some embodiments of the present invention is illustrated. The FD component is the equivalent of a delay tap. As will be appreciated from wireless channel propagation, a power delay profile typically has a large initial tap (for non-line of sight (NLOS), the strongest tap may not be the earliest tap).
[0067] The FD component selection arrangement 300 may include a plurality of configured CQI subbands 302, the number of which may be represented by a symbol N. SB 27. For example, the FD component selection arrangement 300 includes nine configured CQI subbands in the illustrated embodiment. The configured CQI subbands or a portion thereof may be used by a UE, such as UE 2700 (FIG. 27), for transmission of CSI. For example, the UE may transmit CQI on one or more of the CQI subbands.
[0068] The configured CQI subbands 302 may be configured to have multiple pre-decoders per CQI subband, the number of which may be represented by the symbol R. The number of pre-decoder subbands may be determined based on the number of configured CQI subbands 302 and the number of pre-decoders per CQI subband. For example, N3 = R×N SB , where N3 is the number of predecoder subbands. The number of predecoder subbands may be defined as the number of taps or the number of FD components in the time domain. For example, the number of taps or the number of FD components in the time domain may be equal to the number of predecoder subbands. The FD component selection arrangement 300 may include predecoder subbands 304. The number of predecoder subbands 304 may be defined based on the number of configured CQI subbands 302 and the number of predecoders per CQI subband. For example, in the illustrated embodiment, based on the number of configured CQI subbands 302 being 9 configured CQI subbands and the number of predecoders per CQI subband being 2, the predecoder subbands 304 include 18 predecoder subbands.
[0069] The UE may select a plurality of FD components M based on the CQI subband 302. The number of FD components selected by the UE may be determined based on the number of pre-decoders per CQI subband, the number of pre-decoder subbands, and / or the number of configured CQI subbands 302. For example, The UE may be configured with a value p v , where v is the number of spatial layers (RI) used for CSI feedback. In some embodiments, v may be equal to 1, 2, 3, or 4. In the illustrated embodiment, In the illustrated embodiment, the UE may select five FD components. Specifically, in the illustrated embodiment, the UE may select a first FD component 306, a second FD component 308, a third FD component 310, a fourth FD component 312, and a fifth FD component 314 (as illustrated by the FD components illustrated as having diagonal lines) from the pre-decoder subband 304 for CSI feedback. SB =9, R =2, p1 =1 / 2, N3 =18, and M =5. The UE may report the selected FD component to the base station. Specifically, the UE may send one or more signals indicating the selected FD component to the base station.
[0070] Due to signaling overhead considerations, two-stage FD component selection may be used when the number of taps is large. Figure 4 Another exemplary FD component selection arrangement 400 according to some embodiments is illustrated. The FD component selection arrangement 400 illustrates a two-stage FD component selection example.
[0071] The FD component selection arrangement 400 may include a plurality of configured CQI subbands 402, the number of which may be represented by a symbol N SB For example, the FD component selection arrangement 400 in the illustrated embodiment includes 16 taps due to the 16 configured CQI subbands. The configured CQI subbands or a portion thereof may be selected by a UE (such as UE 2500 ( Fig.25 )) is used for CSI feedback. For example, the UE may feedback the CQI of one or more CQI subbands in the CQI subband.
[0072] The configured CQI subbands 402 may be configured to have multiple pre-decoders per CQI subband, the number of which may be represented by the symbol R. The number of pre-decoder subbands may be determined based on the number of configured CQI subbands 402 and the number of pre-decoders per CQI subband. For example, N3 = R×N SB, where N3 is the number of predecoder subbands. The number of predecoder subbands may be defined as the number of taps or the number of FD components in the time domain. For example, the number of taps or the number of FD components in the time domain may be equal to the number of predecoder subbands. The FD component selection arrangement 400 may include predecoder subbands 404. The number of predecoder subbands 404 may be defined based on the number of configured CQI subbands 402 and the number of predecoders per CQI subband. For example, in the illustrated embodiment, based on the number of configured CQI subbands 402 being 16 configured CQI subbands and the number of predecoders per CQI subband being 2, the predecoder subbands 404 include 32 predecoder subbands.
[0073] The UE may determine an intermediate set 406 from which to select FD components, where the intermediate set (IntS) 406 may be a subset of the predecoder subbands 404. IntS 406 may be determined based on the number M of FD components to be selected by the UE. The number of FD components to be selected may be determined based on the number of predecoders per CQI subband, the number of predecoder subbands, and / or the number of configured CQI subbands 402. For example, The UE may be configured with a value p v , where v is the number of spatial layers (RI) used for CSI feedback. In some embodiments, v may be equal to 1, 2, 3, or 4. In the illustrated embodiment, In the illustrated embodiment, the UE may select eight FD components. IntS 406 may include a set of N3′ FD bases, where N3′=2×M. IntS 406 may be obtained by mod(M initial +n,N3) definition, where M initial It can be selected by the UE and reported to the base station in the uplink control information (UCI) part 2. initial can be selected from the set, where the set is M initial ∈{-(N3′-1),-(N3′-2),…,-1,0}.
[0074] FD component selection arrangement 400 shows IntS 406 from pre-decoder subband 404. In the illustrated embodiment, M initialis -4. Therefore, IntS 406 may extend from predecoder subband index 28 to predecoder subband index 11. The UE may select eight FD components from IntS 406. Specifically, in the illustrated embodiment, the UE may select a first FD component 408, a second FD component 410, a third FD component 412, a fourth FD component 414, a fifth FD component 416, a sixth FD component 418, a seventh FD component 420, and an eighth FD component 422 (as illustrated by the FD components illustrated as having diagonal lines) from IntS 406 for CSI feedback. In the illustrated embodiment, N SB =16, R = 2, p1 = 1 / 2, N3 = 32, M = 8, and N3′ = 16. The UE may report the selected FD component to the base station. Specifically, the UE may send one or more signals indicating the selected FD component to the base station.
[0075] Figure 5 The first part of an exemplary bitmap generation process 500 according to some embodiments is illustrated. The bitmap generation process 500 illustrates an exemplary process that can be performed by a UE (such as UE 2700 (FIG. 27)) to generate a bitmap indicated by LC coefficients for reporting CSI to a base station. A UE implementing the process of the bitmap generation process 500 can reduce the data to be sent for the bitmap (compared to a simple bitmap), for example, by being allowed to select different spatial beams for different antenna polarizations.
[0076] The bitmap generation process 500 may include a bitmap 502 of LC coefficients, as determined based on measurements of a UE. Specifically, the bitmap 502 may indicate the value of the LC coefficient of a signal received from a base station measured by the UE. The UE may generate the bitmap 502 based on the determined value of the LC coefficient. Each square of the bitmap 502 may indicate the LC coefficient of the signal measured by the UE. The x-axis of the bitmap 502 is used for the FD component of the LC coefficient, and the y-axis of the bitmap 502 is used for the selected spatial beam, wherein each square in the bitmap corresponds to an index of the FD component and an index of the spatial beam. In the illustrated embodiment, the bitmap 502 includes eight selected spatial beams for two antenna polarizations, and six FD components. The spatial beams of the bitmap 502 may be divided into a first polarization 504 and a second polarization 506. Specifically, the spatial beams corresponding to the top four rows of the bitmap 502 may have a first polarization 504, and the spatial beams corresponding to the bottom four rows of the bitmap 502 may have a second polarization 506.
[0077] In bitmap 502, unfilled boxes indicate that the amplitude of the LC coefficients of the corresponding spatial beams and frequency components is zero. Specifically, the UE may have determined that the amplitude of the LC coefficients corresponding to the unfilled boxes is equal to zero. For example, in the illustrated embodiment, the LC coefficients of the first coefficients 508 have an amplitude of zero. It should be understood that in some embodiments, reference to having an amplitude of zero may not necessarily mean that the amplitude of the LC coefficients is exactly zero, but in these embodiments, the amplitude of the LC coefficients is within a predefined range of zero. In addition, in these embodiments, non-zero amplitudes may refer to a predefined range in which the amplitude of the LC coefficients is greater than zero.
[0078] In the bitmap 502, the filled boxes indicate that the amplitude of the LC coefficients of the corresponding spatial beam and frequency component is non-zero. Specifically, the UE may have determined that the amplitude of the LC coefficients is non-zero. For example, the UE may have determined that the second coefficient 510, the third coefficient 512, the fourth coefficient 514, the fifth coefficient 516, the sixth coefficient 518, the seventh coefficient 520, the eighth coefficient 522, the ninth coefficient 524, the tenth coefficient 526, and the eleventh coefficient 528 have non-zero amplitudes. It can be seen from the bitmap that the second coefficient 510 to the eighth coefficient 522 have the first polarization 504, and the ninth coefficient 524 to the eleventh coefficient 528 have the second polarization 506.
[0079] The UE may determine whether any FD component of the bitmap 502 does not have any non-zero LC coefficient value. For example, the UE may determine that the FD component 530 corresponding to the fourth column in the bitmap 502 does not include any LC coefficient with a non-zero amplitude. Based on the UE's determination that the FD component does not include any LC coefficient with a non-zero amplitude, the UE may remove the FD component from the bitmap 502 to generate a modified bitmap. Specifically, the UE may remove the column corresponding to the FD component from the bitmap 502, thereby making the modified bitmap smaller than the bitmap 502. In the illustrated embodiment, the UE may remove the FD component 530 to generate a modified bitmap 532 without the FD component. As part of removing the FD component 530, the UE may not report the value of the LC coefficient within the FD component 530. In contrast, the UE may include in the report an indication of the FD component that has been removed from the bitmap 502, rather than the value of each LC coefficient in the FD component, which may result in fewer bits and less overhead included in the report. In addition or alternatively, the UE may indicate a subset of the selected 2L spatial beams to reduce the size of the bitmap. The modified bitmap 532 may maintain the remainder of the FD component and the LC coefficient values from the bitmap 502 .
[0080] For understanding, the illustrated bitmap generation process 500 includes a non-zero indication bitmap 534. The non-zero indication bitmap 534 may indicate which components of the modified bitmap 532 have non-zero values and which components of the modified bitmap have zero values. Specifically, the non-zero indication bitmap 534 indicates "1" in the component positions of the LC coefficients having non-zero values, and indicates "0" in the component positions of the LC coefficients having zero values. In the illustrated example, the non-zero indication bitmap 534 has a first coefficient 536, a second coefficient 538, a third coefficient 540, a fourth coefficient 542, a fifth coefficient 544, a sixth coefficient 546, a seventh coefficient 548, an eighth coefficient 550, a ninth coefficient 552, and a tenth coefficient 554 indicating "1" based on the corresponding LC coefficients being non-zero values. The remaining components of the non-zero indication bitmap 534 may indicate "0" based on the corresponding LC coefficients being zero values. The first coefficient 536 of the non-zero indication bitmap 534 corresponds to the second coefficient 510 from the modified bitmap 532, the second coefficient 538 of the non-zero indication bitmap 534 corresponds to the third coefficient 512 of the modified bitmap 532, the third coefficient 540 of the non-zero indication bitmap 534 corresponds to the fourth coefficient 514 of the modified bitmap 532, the fourth coefficient 542 of the non-zero indication bitmap 534 corresponds to the fifth coefficient 516 of the modified bitmap 532, the fifth coefficient 544 of the non-zero indication bitmap 534 corresponds to the sixth coefficient 518 of the modified bitmap 532, The sixth coefficient 546 of the non-zero indication bitmap 534 corresponds to the seventh coefficient 520 of the modified bitmap 532, the seventh coefficient 548 of the non-zero indication bitmap 534 corresponds to the eighth coefficient 522 of the modified bitmap 532, the eighth coefficient 550 of the non-zero indication bitmap 534 corresponds to the ninth coefficient 524 of the modified bitmap 532, the ninth coefficient 552 of the non-zero indication bitmap 534 corresponds to the tenth coefficient 526 of the modified bitmap 532, and the tenth coefficient 554 of the non-zero indication bitmap 534 corresponds to the eleventh coefficient 528 of the modified bitmap 532.
[0081] Figure 6 The second part of the exemplary bitmap generation process 500 according to some embodiments is illustrated. For example, the second part of the exemplary bitmap generation process 500 can proceed from the modified bitmap 532 generated by the first part of the exemplary bitmap generation process 500. For each polarization, the UE can identify a reference, and the UE can use the stronger reference to normalize all LC coefficients.
[0082] The UE may determine the strongest LC coefficient from the LC coefficients included in the modified bitmap 532. Specifically, the UE may determine the LC coefficient with the largest amplitude included in the modified bitmap 532. In the illustrated embodiment, the UE may determine that the fifth coefficient 516 is the strongest LC coefficient based on the fifth coefficient 516 having the largest amplitude of the LC coefficients included in the modified bitmap 532.
[0083] The UE may also determine which polarization includes the strongest LC coefficient. For example, in the illustrated embodiment, the UE may determine whether the strongest LC coefficient has the first polarization 504 or the second polarization 506. In the illustrated embodiment, the UE may determine that the fifth coefficient 516 has the first polarization 504.
[0084] The UE may also determine the strongest LC coefficient based on other polarizations that do not include the strongest LC coefficient of the entire modified bitmap 532. For example, when the UE determines that the strongest LC coefficient in the illustrated embodiment has the first polarization 504, the UE may determine which LC coefficient having the second polarization 506 has the largest amplitude. In the illustrated embodiment, the UE may determine that the tenth coefficient 526 has the strongest LC coefficient among the LC coefficients having the second polarization 506.
[0085] The UE may normalize the non-zero LC coefficients of the modified bitmap 532 based on the strongest LC coefficient of the modified bitmap 532. Specifically, the UE may divide the values of all LC coefficients having non-zero amplitudes by the value of the strongest LC coefficient in the modified bitmap 532. For example, the UE may divide the values of the non-zero LC coefficients of the modified bitmap 532 by the value of the fifth coefficient 516 to normalize the non-zero LC coefficients.
[0086] The UE may perform high resolution amplitude quantization using the LC coefficients within the modified bitmap 532. For example, the UE may take normalized values for the LC coefficients within the modified bitmap 532 and quantize the normalized values to selected digital values. In some embodiments, high resolution amplitude quantization may be performed using four bits. In some embodiments, the alphabet used for high resolution amplitude quantization may be The step size may be -1.5 decibels (dB). Each normalized value for the LC coefficient may be converted to the corresponding nearest value from the alphabet for high resolution amplitude quantization.
[0087] The UE may perform normal resolution amplitude quantization and / or phase quantization using the LC coefficients with the first polarization 504. Specifically, the UE may separate the LC coefficients into LC coefficients with the first polarization 504 and LC coefficients with the second polarization 506. The UE may perform normal resolution amplitude quantization using the LC coefficients with the first polarization 504. In some embodiments, normal resolution amplitude quantization may be performed using three bits. In some embodiments, the alphabet for normal resolution amplitude quantization may be Wherein the step size may be -3 decibels (dB). Each value for the LC coefficient may be converted to a corresponding nearest value from the alphabet used for normal resolution amplitude quantization. For example, the LC coefficient may be converted from a high resolution amplitude quantization value to a corresponding nearest value from a normal resolution amplitude quantization.
[0088] The UE may perform phase quantization using the LC coefficients with the first polarization 504. Specifically, the UE may perform phase quantization to indicate the phase of the LC coefficients in the first polarization 504. In some embodiments, four bits may be used to perform phase quantization. For example, in some embodiments, the UE may perform phase quantization for 16-phase shift keying (PSK) using the LC coefficients with the first polarization 504. The phase may be based on the FD component of the strongest LC in the polarization. In the illustrated embodiment, the strongest LC for the first polarization 504 may be the fifth coefficient 516, which is located in the first FD component of the modified bitmap 532 and has a phase of zero. Since the phase of the first polarization 504 is zero in the illustrated embodiment, phase quantization will result in a phase value of zero for the first polarization 504. Therefore, since the phase value of the fifth coefficient 516 will be the same before and after phase quantization, phase quantization may be skipped in the illustrated embodiment.
[0089] The UE may perform normalization, normal resolution amplitude quantization, and / or phase quantization for the LC coefficients of the second polarization 506. Specifically, the UE may normalize the LC coefficients of the second polarization 506 using the strongest LC coefficient of the second polarization 506. For example, the UE may divide the value of the LC coefficient with the second polarization 506 by the value of the strongest LC coefficient to normalize the value of the LC coefficient with the second polarization 506. In the illustrated embodiment, the strongest LC coefficient of the second polarization 506 may be the tenth coefficient 526. The UE may normalize the LC coefficient with the second polarization 506 by dividing the value of the LC coefficient by the value of the tenth coefficient 526.
[0090] The UE may perform normal resolution amplitude quantization using the LC coefficients with the second polarization 506. In some embodiments, normal resolution amplitude quantization may be performed using three bits. In some embodiments, the alphabet used for normal resolution amplitude quantization may be Wherein the step size may be -3dB. Each value for the LC coefficient may be converted to the corresponding closest value from the alphabet used for normal resolution amplitude quantization. For example, the LC coefficient may be converted from a high resolution amplitude quantization value to the corresponding closest value from the normal resolution amplitude quantization.
[0091] The UE may perform phase quantization using the LC coefficients with the second polarization 506. Specifically, the UE may perform phase quantization to indicate the phase of the LC coefficients in the second polarization 506. In some embodiments, four bits may be used to perform phase quantization. For example, in some embodiments, the UE may perform phase quantization for 16PSK using the LC coefficients with the first polarization 504. The phase may be based on the FD component of the strongest LC in the polarization. In the illustrated embodiment, the strongest LC for the second polarization 506 may be the tenth coefficient 526, which is located in the fourth FD component of the modified bitmap 532. The UE may perform phase quantization using the phase from the tenth coefficient 526.
[0092] Figure 7 An example CSI reporting method 700 according to some embodiments is illustrated. According to some embodiments, the CSI reporting method 700 may indicate a method for CSI reporting and precoder generation.
[0093] The CSI reporting method 700 may include a first CSI reporting method 702. The first CSI reporting method 702 may have a single pre-decoder for all OFDM symbols in a PDSCH or a time slot. In the first CSI reporting method 702, a UE (such as UE2500 ( Fig.25 )) can perform measurements of multiple CSI measurement resources 704. For example, the UE can perform measurements of four CSI measurement resources 704. Based on the measurements of the CSI measurement resources 704, the UE can use the obtained measurement values of the CSI measurement resources 704 according to the bitmap generation process 500 ( Figure 5 ) or a process associated with schematic diagram 1400 to perform bitmap generation.
[0094] The UE may generate a CSI report 706 based on the measurement of the CSI measurement resource 704. For example, the UE may generate a CSI report 706 having RI, CQI, and PMI based on the measurement of the CSI measurement resource 704. In some embodiments, the RI, CQI, and / or PMI within the CSI report 706 generated by the UE may be predicative, such that the UE may generate values based on a prediction model of a channel response and / or a prediction model of a pre-decoder. Specifically, the UE may generate values of RI, CQI, and / or PMI that are predictive of a channel response and / or a pre-decoder that may occur in a future time period. The UE may provide a base station (such as a gNB 2600( Fig.26 )) sends a CSI report 706. In some implementations, the CSI report 706 may omit the CSI-reference signal (RS) resource indicator (CRI).
[0095] The base station may generate one or more predecoders for multiple OFDM symbols within the PDSCH based on the CSI report 706. In the first CSI reporting method 702, the base station may generate a single predecoder for multiple portions on the PDSCH. In the illustrated embodiment, the base station may generate a single predecoder from the CSI report 706 for each of the three opportunities 708, and generate a total of three predecoders for the three PDSCH opportunities. Each opportunity 708 may include one or more subbands 720, wherein the predecoder may have subband predecoders that may be used for the subbands 720. In the illustrated embodiment of the first CSI reporting method 702, each opportunity 708 includes five subbands 720, wherein a predecoder having five subband predecoders is used for the five subbands. Therefore, the base station may use the predecoder for the signal transmitted within the subband of the opportunity 708. Generating a pre-decoder for multiple opportunities of the PDSCH from a single CSI report rather than having a single CSI report for each opportunity and / or subband may reduce the number of CSI reports sent during operation of the UE and base station, thereby reducing overhead between the UE and base station.
[0096] The CSI reporting method 700 may include a second CSI reporting method 710. The second CSI reporting method 710 may have multiple pre-decoders at a given subband of multiple parts in the PDSCH. In the second CSI reporting method 710, a UE (such as UE 2500 ( Fig.25 )) can perform measurements of multiple CSI measurement resources 712. For example, the UE can perform measurements of four CSI measurement resources 712. Based on the measurements of the CSI measurement resources 712, the UE can perform bitmap generation according to the bitmap generation process 500 or the process associated with the schematic diagram 1400 using the obtained measurement values of the CSI measurement resources 712.
[0097] The UE may generate a CSI report 714 based on the measurement of the CSI measurement resource 712. For example, the UE may generate a CSI report 714 with RI, CQI, and PMI based on the measurement of the CSI measurement resource 712. In some embodiments, the RI, CQI, and / or PMI within the CSI report 714 generated by the UE may be deterministic, so that the UE may generate values based on a prediction model of a channel response and / or a prediction model of a pre-decoder. Specifically, the UE may generate values of RI, CQI, and / or PMI that are predictive of a channel response and / or a pre-decoder that may occur in a future time period. The UE may provide a base station (such as a gNB 2600( Fig.26 )) sends a CSI report 714. In some implementations, the CSI report 714 may omit the CSI-reference signal (RS) resource indicator (CRI).
[0098] The base station may generate one or more predecoders for multiple OFDM symbols within the PDSCH based on the CSI report 714. In the second CSI reporting method 710, the base station may generate multiple predecoders at given subbands of multiple parts in the PDSCH. In the illustrated embodiment, the base station may generate two predecoders from the CSI report 714 for two parts in the second opportunity 716 (e.g., the first part 718 and the second part 719 of the first opportunity). Each of the first opportunity 715, the second opportunity 716, and the third opportunity 717 may include one or more subbands. In the illustrated embodiment of the second CSI reporting method 710, each of the first opportunity 715, the second opportunity 716, and the third opportunity 717 includes two parts, each of which has five subbands, and a total of ten subbands per opportunity. The base station may apply a first predecoder having five subband predecoders to the five subbands at the first part 718, and apply a second predecoder having five subband predecoders to the five subbands at the second part 719. In some embodiments, the base station may apply the predecoder to the first PDSCH opportunity 715 within the reception period of the CSI report and apply the predecoder to the last PDSCH opportunity after the reception period of the CSI report until the base station receives another CSI report. Generating a predecoder for multiple opportunities of the PDSCH from a single CSI report rather than having a single CSI report for each opportunity / part and / or subband may reduce the number of CSI reports sent during operation of the UE and the base station, thereby reducing overhead between the UE and the base station.
[0099] The predecoder may be represented by a formula defining a predecoder for each polarization. For example, the base station may generate a predecoder for each polarization, where the predecoder may be represented by a formula. For example, the predecoder generated by the base station may be represented by , where the top formula may represent a precoder for a first polarization from which a subband precoder may be generated, and the bottom formula may represent a precoder for a second polarization from which a subband precoder may be generated. These formulas may include terms for Doppler shifts that may be present due to a moving UE and / or a moving base station or a changing propagation environment (e.g., a reflector is moving). b,p . (θ b,p ,φ b,p ) can be defined as the departure angle θ b,p and φ b,p The light at A(θ b,p ,φ b,p ) can be the array response of the beam, C b,p can be a complex coefficient linking the spatial beam, relative delay, and Doppler shift of the light, and τ b,p Can be a delay of the ray.
[0100] This formula can be modified to take into account the normalization based on the shift of the strongest LC coefficient and the frequency offset applied to the LC coefficient. and ray index The strongest LC coefficient associated The frequency of the strongest LC coefficient can be used to shift the frequency offset at the LC coefficient into the formula. The formula can be changed to in Can be the strongest LC coefficient, can be a (relative) frequency offset, and Can be a relative delay. There may be a time delay of the strongest LC coefficient and may be used to shift the LC coefficient. It can be used to shift the strongest LC coefficient to the first FD component (or first tap).
[0101] In order to control the feedback overhead, only the larger and the corresponding (θ b,p ,φ b,p ), Δf b,p =f b,p - and It can be fed back from the UE to the base station. Conceptually, it can be seen that if If smaller, omit and the corresponding amount will not make much difference in the resulting predecoder. If feedback overhead is not an issue, then feedback is possible. Since feedback overhead is a concern in many cases, quantization applied to these quantities can be used. The quantization corresponding to these quantities can be expressed as spatial beam selection and quantization Q1,1 and Q 1,2 , delay taps (which may also be referred to as FD components) quantization Q2, LC coefficient quantization Q3, and Doppler component (which may also be referred to as time domain component or TD component) selection / quantization Q4. The formula then becomes Use quantification instead of quantity.
[0102] The methods described herein may address one or more design challenges presented by conventional systems. For example, the embodiments described herein may address challenges using codebook structures and / or spatial beam, FD component and / or TD component selection. "Table" may be used to refer to a Fig. 9 LC coefficients associated with the same quantized Doppler components shown, where five Doppler frequency offsets are shown (alternatively, five "tables" are shown). These challenges can be addressed by table selection and / or component selection as described throughout this disclosure. In some embodiments, the UE can provide a report on the selected Doppler frequency offset or the selected Doppler table. In some embodiments, table-generic reporting can be implemented, where the UE reports the spatial beam selection and / or FD component selection to the base station to address these challenges. In addition, table-specific reporting can be implemented in some embodiments, where the UE reports the spatial beam selection for each table and / or the FD component selection for each table.
[0103] The embodiments described herein can also address the challenge of non-zero LC coefficient selection. For example, this challenge can be addressed by combining bitmaps and TD component signaling (which can provide efficient encoding of Doppler components). In addition, this challenge can be addressed via a bitmap designed for each table.
[0104] The embodiments described herein can also solve the challenge of non-zero LC coefficient quantization. For example, the challenge can be solved by a fixed quantization design, a quantization design with parameters that can be selected by the UE and / or can be configured by the base station, and / or a quantizer defined by the UE with a standardized interface. These methods can provide a two-stage quantization with various options for reference. In addition, the challenge can be solved by quantizer version selection.
[0105] The methods described herein may implement a codebook structure that takes into account the TD component. For example, a base station (such as gNB 2600 ( Fig.26 )) and / or a UE (such as UE 1800 (FIG. 27)) may implement a codebook structure taking into account selection of a TD component for a pre-decoded signal exchanged between the base station and the UE.
[0106] In some embodiments, the codebook structure implemented by the base station and / or the UE may be based on or it can be formulated in another way as W1 can be the spatial beam selection, Can be used for non-zero LC coefficient selection and quantification, W d It can be TD component selection, in It is M d Size-N4×1 orthogonal DFT vectors to select TD components with significant power at the spatial layer, and Can be FD component selection in is M size - N3×1 orthogonal DFT vectors to select FD components with significant power for the table at the spatial layer. In addition, P can be equal to 2N1N2, which can be equal to the number of SD dimensions. N3 can be equal to the number of FD dimensions. N4 can be equal to the number of time domain dimensions (the maximum number of time units between CSI reporting and the prediction precoder for PDSCH in the most recent valid time unit). The dimension of the linear combination coefficient is 2L×M×M d In some cases, values other than 2L may be used for AI applications as discussed herein. At spatial layer n, 0≤l≤2L-1,0≤m≤M-1,0≤f≤M d -1, where l is the spatial beam index, m is the FD component index (delay tap index), and f is the Doppler component index. In the table f, 0≤f≤M d -1. (in matlab matrix convention) is a 2L×M matrix. is the matrix product along the second dimension from X to Y. is the matrix product along the third dimension from X to Y. Therefore, is a matrix of dimension 2L×M×1. is a matrix of dimension 2L×N3×1, and W is a matrix of dimension 2N1N2×N3×1, which is the precoder for the N3 precoded subband at time t.
[0107] In some embodiments, a codebook may implement multiple tables (also referred to as "pages") for the codebook within a codebook design. Figure 8 An exemplary codebook 800 and power delay profile 850 with a single table according to some embodiments are illustrated. Specifically, the codebook 800 may provide a single table for each spatial layer.
[0108] The illustrated codebook 800 may be used for two spatial layers. Specifically, the codebook 800 may include a first table 802 for a first spatial layer (which may be referred to as "spatial layer 0") and a first table 804 for a second spatial layer (which may be referred to as "spatial layer 1"). The first table 802 may include precoding definitions for the first spatial layer, and the first table 804 may include precoding definitions for the second spatial layer. The base station and / or the UE may use the precoding definitions of the first table 802 for precoded signals sent in the first spatial layer and use the precoding definitions of the first table 804 for precoded signals sent in the second spatial layer. For example, for a stationary UE, a stationary gNB, and a stationary propagation environment, Figure 8 The situation shown in may occur due to insignificant Doppler frequency components in the codebook other than the zero frequency offset component after shifting the frequency offset as described above. Since there is only a single table in the CSI feedback for spatial layer 1 and also a single table in the CSI feedback for spatial layer 2, Figure 8 The Doppler frequency index is omitted from the linear combination coefficients shown, and C l,m,n , instead of C l,m,f,n . This can also happen at one spatial layer where only one table exists, but at another spatial layer where multiple tables exist.
[0109] The power delay profile 850 may illustrate the power delay profile at the spatial beam of the spatial layer. Specifically, the power delay profile 850 may include a first power delay profile 852 and a second power delay profile 854. The first power delay profile 852 may correspond to the first spatial layer, and the second power delay profile 854 may correspond to the second spatial layer. It can be seen that the first power delay profile 852 may reach a peak at an earlier time than the second power delay profile 854, so that the first signal corresponding to the spatial beam at the first spatial layer reaches a peak at a different time than the second signal corresponding to the spatial beam at the second spatial layer.
[0110] Fig. 9 900 with a plurality of tables according to some embodiments. Each table may correspond to a Doppler frequency shift. Specifically, the portion of the codebook 900 may be an example of a portion of the codebook 900 corresponding to a single layer. Fig. 9 The linear combination coefficient C for n=1, 0≤l≤8-1, 0≤m≤6-1, -2≤f≤2 is shown in l,m,f,n .
[0111] The portion of codebook 900 may include multiple tables corresponding to a single spatial layer. For example, in the illustrated embodiment, the portion of codebook 900 may include five tables corresponding to a single spatial layer. Specifically, the portion of codebook 900 may include a first table 902, a second table 904, a third table 906, a fourth table 908, and a fifth table 910.
[0112] Each of these tables may correspond to a frequency offset. For example, the third table 906 may correspond to a spatial layer with no frequency offset, which may be represented as a frequency offset of 0·Δf. The second table 904 may correspond to a spatial layer with a frequency offset of positive 1, which may be represented as a frequency offset of 1·Δf. The first table 902 may correspond to a spatial layer with a frequency offset of positive 2, which may be represented as a frequency offset of 2·Δf. The fourth table 908 may correspond to a spatial layer with a frequency offset of negative 1, which may be represented as a frequency offset of -1·Δf. The fifth table 910 may correspond to a spatial layer with a frequency offset of negative 2, which may be represented as a frequency offset of -2·Δf.
[0113] The base station and / or UE may utilize the table with its corresponding frequency offsets to construct one or more precoders for the signals to be transmitted.
[0114] In the table f,0≤f≤M d -1. (in matlab matrix convention) is a 2L×M matrix. To reduce feedback overhead, table-specific spatial beam selection and / or FD component selection can be used. Let 0≤l≤2L-1,0≤m≤M-1, so that is a 2L×M matrix. So one design is Where S f,1 is a 2L×L′ matrix consisting of elements at 0 or 1, with only one element having “1” on each column and at most one element having “1” on each row, 0≤L′≤2L if TD component selection is not indicated separately, and 0 if TD component selection can be signaled together with table-specific spatial beam selection and / or FD component selection <L′≤2L。S f,1 Indicates the presence of L′ spatial beams with non-zero linear combination (LC) coefficients and their positions between 2L spatial beams (which can be selected jointly for all spatial layers or for a single spatial layer), so table-specific spatial beam selection can help reduce feedback overhead. Similarly, table-specific FD component selection can be achieved through S f,2 , conduct, and YesS f,2 ,S f,2The matrix transpose of is an M×M′ matrix consisting of elements at 0 or 1, there is only one element with “1” on each column, and there is at most one element with “1” on each row, if the TD component selection is not indicated separately, then 0≤M′≤M, and if the TD component selection can be signaled together with the table-specific spatial beam selection and / or FD component selection, then 0 <M′≤M。S f,2 Indicating the presence of M' components with non-zero linear combination (LC) coefficients and their location among FD components (which can be selected jointly for all spatial layers or for a single spatial layer), then table-specific FD component selection can help reduce feedback overhead. It is an L′×M′ matrix. It can be seen that a bitmap matrix with a size of 2L×M is not used to indicate the position of the non-zero LC coefficients, but S f,1 and S f,2 The signaling on the spatial layer only requires a bitmap matrix of size L′×M′, thus reducing the feedback overhead. Note that for the tables at the spatial layer, their corresponding S f,1 can be different; and their corresponding S f,2 Can be different.
[0115] Fig.10 An exemplary spatial beam selection 1000 according to a certain embodiment is illustrated. Spatial beam selection 1000 may indicate spatial beam selection for one or more antennas. Specifically, spatial beam selection 1000 may determine beams to be used for transmission by one or more antennas, such as an antenna array.
[0116] Spatial beam selection 1000 may provide beam selection for a spatial layer or multiple spatial layers (e.g., all spatial layers), such as indicated by spatial beam selection representations 1031, 1032, 1033, and 1034. In 1000, four spatial beams are selected in common for both antenna polarizations, resulting in a common spatial beam selection for all tables. After the common spatial beam selection, table-specific spatial beam selection is then performed. Alternatively, spatial beam selection for each table may be performed independently without going through the common spatial beam stage, including selecting different rotation factors. Thus, the spatial beam selection representation 1002 for a given table may include spatial beam selection representation 200 ( Figure 2 ). The spatial beam selection representation 1002 may include an orthogonal DFT beam as indicated by a black filled circle, a rotated DF beam as indicated by a diagonal striped filled circle, and / or an oversampled DFT beam as indicated by an unfilled circle. The rotated DFT beam in the illustrated embodiment may be rotated from an orthogonal DFT beam and In another variant, in order to reduce the implementation effort, the same rotation factor can be used for different tables, but the spatial beam selection is performed independently for different tables.
[0117] Spatial beam selection 1000 may provide beam selection for orthogonal DFT beams of spatial beam selection representation 1002. For example, in the illustrated embodiment, four spatial beams (1031, 1032, 1033, and 1034) may be selected for a first polarization (which may be referred to as "polarization 0") and for a second polarization (which may be referred to as "polarization 1"), for a total of eight spatial beams. These eight spatial beams may be selected together for all spatial layers and / or all tables at a spatial layer. Further selection between these eight spatial beams may be made specifically for a particular table so that a smaller bitmap matrix may be used for that table. This further selection may be made through spatial beam selection and / or combined index encoding.
[0118] For example, spatial beam selection 1000 may provide a spatial beam selection representation 1002 for one or more tables. In the illustrated embodiment, spatial beam selection representation 1002 may include beam selection representations for five tables. Specifically, spatial beam selection representation 1002 may include a first table 1004, a second table 1006, a third table 1008, a fourth table 1010, and a fifth table 1012. First table 1004 may correspond to first table 902, second table 1006 may correspond to second table 904, third table 1008 may correspond to third table 906, fourth table 1010 may correspond to fourth table 908, and fifth table 1012 may correspond to fifth table 910. The top four rows in each of these tables may correspond to the first polarization, and the bottom four rows in each of these tables may correspond to the second polarization. The x-axis of each of these tables may correspond to the FD component, and the y-axis may correspond to the spatial beam, wherein each square in these tables corresponds to the LC coefficient of the FD component and the index of the spatial beam.
[0119] The tables of spatial beam selection 1000 may have frequency offsets of the corresponding tables of that portion of codebook 900. For example, first table 1004 may have a frequency offset of positive 2 (which may be represented as 2·Δf). Second table 1006 may have a frequency offset of positive 1 (which may be represented as 1·Δf). Third table 1008 may have a frequency offset of zero (which may be represented as 0·Δf). Fourth table 1010 may have a frequency offset of negative 1 (which may be represented as -1·Δf). Fifth table 1012 may have a frequency offset of negative 2 (which may be represented as -2·Δf). It should be appreciated that shuffling or reindexing these tables in a deterministic manner does not substantially change the codebook design.
[0120] Each of these tables may indicate a component corresponding to a non-zero LC coefficient detected by the UE. Specifically, a square with fill in the spatial beam selection representation 1002 may indicate that the UE has detected a non-zero LC coefficient for the corresponding spatial beam at the corresponding FD component. Among the detected non-zero LC coefficients, the strongest LC coefficient and / or the strongest LC coefficient of each table may be indicated. For example, the first LC coefficient 1014 in the third table 1008 may correspond to the strongest LC coefficient of the non-zero LC coefficients detected from all tables, as indicated by the black fill of the first LC coefficient 1014. The second LC coefficient 1016 may be the strongest LC coefficient in the first table 1004, as indicated by the vertical line fill of the second LC coefficient 1016. The third LC coefficient 1018 may be the strongest LC coefficient in the second table 1006, as indicated by the vertical line fill of the third LC coefficient 1018. The fourth LC coefficient 1020 may be the strongest LC coefficient in the fifth table 1012, as indicated by the vertical line fill of the fourth LC coefficient 1020. The LC coefficients illustrated with dashed filling may be other detected non-zero LC coefficients, and the LC coefficients without filling may be LC coefficients with zero amplitude.
[0121] The spatial beams may be different for each of the tables as indicated by the non-zero LC coefficients indicated within the tables being in different positions relative to the tables. Different spatial beam selection from table to table may reduce signaling overhead. Additionally, spatial beam selection may be different or common for different polarizations on the same table. Different spatial beam selection from table to table may reduce signaling overhead due to different polarizations on the same table. The signaling overhead caused by the bitmap matrix indicating non-zero LC coefficients in the UE may be reduced. For example, the UE may signal the selected beam for each table and / or for each polarization in the table in a single CSI report, thereby allowing beam selection for multiple frequency offsets without having to send a separate CSI report for each of these PDSCH opportunities.
[0122] At a given spatial layer, there may be a TD component selection (selection "table") first, and then the selected spatial beam is indicated on the selected table. For example, in the illustrated embodiment, the TD component selection may select one of the five tables from the spatial beam selection representation 1002. To facilitate codebook description, a two-stage selection may be considered for FD component selection separately and another two-stage selection may be used for spatial beam selection.
[0123] For FD component selection, in stage 1, multiple FD components may be selected for all tables together, and then in stage 2, the selected FD components for each table may be a subset. For example, the FD components for each table may be selected based on non-zero LC coefficients. The FD component selection for the first table 1004 may be represented by a bitmap [101010] (or a combined index instead of a bitmap to reduce signaling overhead. Other tables are processed without repeatedly iterating this alternative way of reducing overhead), the FD component selection for the second table 1006 may be represented by [011010], the FD component selection for the third table 1008 may be represented by [111001], and the FD component selection for the fifth table 1012 may be represented by [001000], where 1 indicates a column (corresponding to an FD component) having one or more non-zero LC coefficients and 0 indicates a column without any non-zero LC coefficients. Since the fourth table 1010 does not include any non-zero LC coefficients, FD component selection may not be performed in the fourth table 1010. The FD components with non-zero LC coefficients may be a subset of the commonly selected FD components for selection.FD components may also be selected independently for different tables without first going through a stage of identifying commonly selected FD components for all tables.
[0124] For spatial beam selection, in stage 1, 2L spatial beams are selected, and then in stage 2, the selected spatial beams of each table are subsets. For example, the TD component of each table can be selected based on non-zero LC coefficients. The spatial beam selection for the first table 1004 can be represented by a bitmap [0011 0100] (or a combined index instead of a bitmap to reduce signaling overhead. Other tables are processed without repeatedly iterating this alternative way to reduce overhead), the spatial beam selection for the second table 1006 can be represented by [11000100], the spatial beam selection for the third table 1008 can be represented by [0111 0100], and the spatial beam selection for the fifth table 1012 can be represented by [0001 0000], where 1 indicates a row with one or more non-zero LC coefficients (corresponding to a spatial beam) and 0 indicates a row without any non-zero LC coefficients. Since the fourth table 1010 does not include any non-zero LC coefficients, spatial beam selection may not be performed for the fourth table 1010. The spatial beams having non-zero LC coefficients may be a subset of the commonly selected spatial beams for selection.
[0125] In some embodiments, the UE may report the spatial beam selection and / or FD component selection to the base station in the CSI report. For example, the UE may report the spatial beam selection indication and / or FD component selection indication indicated above to the base station in order to select the spatial beam and FD component to be sent by the base station. In the case where a table has no non-zero LC coefficients (such as the fourth table 1010), the UE may indicate in the CSI report that the table has no non-zero LC coefficients, for example by indicating a bitmap of TD components having at least one non-zero LC coefficient or a combined index instead of a bitmap.
[0126] Fig.11 An exemplary FD component selection and indication method 1100 according to some embodiments is illustrated. Specifically, the FD component selection and indication method 1100 may be another method for a UE to select an FD component for beam transmission and indicate the FD component to a base station. The FD component selection and indication method 1100 may be considered as a supplement to the 3GPP Release 16 (Rel-16) design for RAN.
[0127] FD component selection and indication method 1100 may include one or more tables, such as in codebook 900 ( Fig. 9 ) and / or the tables illustrated for the spatial beam selection representation 1002. For example, the FD component selection and indication method 1100 includes a first table 1102, a second table 1104, a third table 1106, a fourth table 1108, and a fifth table 1110. The first table 1102 may have a frequency offset of positive 2 (which may be represented as 2·Δf). The second table 1104 may have a frequency offset of positive 1 (which may be represented as 1·Δf). The third table 1106 may have a frequency offset of zero (which may be represented as 0·Δf). The fourth table 1108 may have a frequency offset of negative 1 (which may be represented as -1·Δf). The fifth table 1110 may have a frequency offset of negative 2 (which may be represented as -2·Δf). The top four rows of each of these tables may correspond to a first polarization (which may be referred to as "polarization 0"), and the bottom four rows of each of these tables may correspond to a second polarization (which may be referred to as "polarization 1"). The tables may indicate the FD component via the x-axis and the spatial beam via the y-axis, where each square in the tables corresponds to an index of the FD component and an index of the spatial beam.
[0128] Each of these tables may indicate a non-zero LC coefficient for a corresponding frequency offset. For example, the UE may determine which spatial beam and FD component combinations have non-zero LC coefficients for each of these frequency offsets based on the CSI-RS. The first table 1102 may indicate non-zero LC coefficients with a frequency offset of positive 2, the second table 1104 may indicate non-zero LC coefficients with a frequency offset of positive 1, the third table 1106 may indicate non-zero LC coefficients with a frequency offset of zero, the fourth table 1108 may indicate non-zero LC coefficients with a frequency offset of negative 1, and the fifth table 1110 may indicate non-zero LC coefficients with a frequency offset of negative 2. Specifically, in the illustrated embodiment, the tables may indicate non-zero LC coefficients via filled boxes, while unfilled boxes (e.g., white boxes) indicate LC coefficients with zero amplitude.
[0129] The UE can identify the strongest LC coefficient (e.g., the LC coefficient with the largest amplitude) from all of these tables. In the illustrated embodiment, the first coefficient 1112 within the third table 1106 can be determined to be the strongest LC coefficient in these tables, as indicated by the first coefficient 1112 being shown filled in black. The UE can also determine the polarization with which the strongest LC coefficient is associated. Since the first coefficient 1112 is located within the top four rows of the third table 1106 in the illustrated embodiment, the UE can determine that the strongest LC coefficient is associated with the first polarization.
[0130] The UE can then determine the strongest LC coefficient for the polarization that does not include the strongest LC coefficient of the entire table. For example, since the strongest LC coefficient is associated with the first polarization in the illustrated embodiment, the UE can determine the strongest LC coefficient associated with the second polarization. In the illustrated embodiment, the second coefficient 1114 located within the second table 1104 can be determined as the strongest LC coefficient associated with the second polarization, as indicated by the second coefficient 1114 being filled with a vertical line. It can be seen that in the illustrated embodiment, the second coefficient 1114 is located in the bottom four rows, indicating that the second coefficient 1114 is associated with the second polarization. The remaining non-zero LC coefficients are illustrated in these tables with dashed line filling.
[0131] The UE may combine the first table 1102, the second table 1104, the third table 1106, the fourth table 1108, and the fifth table 1110 into a single table representation 1116. Specifically, the single table representation 1116 may illustrate all non-zero LC coefficients from the first table 1102, the second table 1104, the third table 1106, the fourth table 1108, and the fifth table 1110 in the single table representation 1116, wherein the LC coefficients maintain their spatial beam and FD component relationships from the first table 1102, the second table 1104, the third table 1106, the fourth table 1108, and the fifth table 1110. For example, the single table representation 1116 may have components of non-zero LC coefficients corresponding to each of these tables, as illustrated by the fill in the square. The single table representation 1116 may also indicate the strongest LC coefficients of these tables, as illustrated by the third coefficient 1118 illustrated with black fill. Other non-zero LC coefficients in the single table representation 1116 are illustrated with dashed fill in the illustrated embodiment. In other cases, a non-zero LC coefficient may be exemplified by 1, while an LC coefficient having zero amplitude may be marked or unmarked by 0. If, on tables 1102, 1104, 1106, 1110, a non-zero LC coefficient is converted to one bit having a logic "1" and a zero LC coefficient is converted to one bit having a logic "0", 1116 may be obtained by performing a logical "OR" operation on four matrices having "0" / "1" as their elements corresponding to tables 1102, 1104, 1106, 1110, respectively.
[0132] The UE may perform spatial beam selection and / or FD component selection based on a single table representation 1116. For example, the UE may perform spatial beam selection based on non-zero LC coefficients, and / or perform FD component selection based on non-zero LC coefficients. In the illustrated embodiment, spatial beam selection may be represented by a bitmap [11111100] (or a combined index instead of a bitmap to reduce overhead), where 1 indicates a row with a non-zero LC coefficient (which represents a spatial beam index) and 0 indicates a row without a non-zero LC coefficient. FD component selection may be represented by a bitmap [111011] (or a combined index instead of a bitmap to reduce overhead), where 1 indicates a column with a non-zero LC coefficient (which represents an FD component index) and 0 indicates a column without a non-zero LC coefficient. In some embodiments, the UE may report spatial beam selection and / or FD component selection to the base station in a CSI report. The base station may communicate with the UE using spatial beam selection and / or FD component selection.
[0133] The UE may also filter the tables to remove tables that do not include non-zero LC coefficients. The UE may indicate to the base station in the CSI report the tables that do not include the filtered non-zero LC coefficients, or may not indicate to the base station in the CSI report the tables that do not include the filtered non-zero LC coefficients. For example, the filtered table representation 1120 in the illustrated embodiment may have filtered the fourth table 1108 based on the fourth table that does not include any non-zero LC coefficients. Thus, the filtered table representation 1120 may include the first table 1102, the second table 1104, the third table 1106, and the fifth table 1110 in the illustrated representation. For each non-zero position on the bitmap (represented by the table in the illustrated embodiment), there is at least one LC coefficient at all selected tables of the filtered table representation 1120. Filtering of tables that do not include non-zero LC coefficients may reduce the size of the signal (such as a CSI report) sent by the UE to the base station, which may reduce overhead.
[0134] The UE may also determine which tables in the filtered table representation 1120 have non-zero LC coefficients at a particular component position (particular table). For example, the UE may determine which tables in the filtered table representation 1120 include non-zero LC coefficients at the position corresponding to the strongest LC coefficient (which is the position of the first coefficient 1112 in the illustrated embodiment). In addition, the UE may generate a representation of the table including the non-zero LC coefficient at the position, which may be referred to as Doppler component component signaling. In the illustrated embodiment, the first table 1102 and the third table 1106 include non-zero LC coefficients at the position of the strongest LC coefficient. The second table 1104 and the fifth table 1110 do not include non-zero LC coefficients at the position of the strongest LC coefficient. The representation of the table having non-zero LC coefficients at the position of the strongest LC coefficient may be a bitmap
[1010] , where 1 indicates a table having a non-zero LC coefficient and 0 indicates a table without a non-zero LC coefficient. The representation with the bitmap
[1010] may be sorted from the lowest table value on the left side of the filtered table representation 1120 to the highest table value on the right side. Since the fourth table 1108 is not included in the filtered table representation 1120, the fourth table 1108 can be omitted from the representation. The determination and generation of the representation can be repeated for other locations of the tables. It should also be understood that the bitmap can be replaced with a combined index to reduce overhead.
[0135] In some embodiments, the UE may report Doppler component composition signaling with a spatial beam selection representation and / or an FD component selection representation from a single table representation 1116 in a CSI report. In these embodiments, the base station may determine the beam to be used for communication with the UE based on the Doppler component composition signaling, the spatial beam selection representation, and / or the FD component selection representation. For example, the base station may select the beam to be utilized from the spatial beam selection representation and / or the FD component selection representation. In some embodiments, the base station may select the beam corresponding to the strongest LC coefficient (represented by the third coefficient 1118 in the illustrated embodiment) based on the spatial beam selection representation and / or the FD component selection representation to communicate with the UE. The base station may then utilize the Doppler component composition signaling to determine at which frequency offsets the selected beam provides a non-zero LC coefficient. The base station may then utilize the selected beam at these frequency offsets to communicate with the UE.
[0136] Fig.12 An exemplary quantization design 1200 according to some embodiments is illustrated. For example, the illustrated quantization design 1200 may include a per-polarization design 1202 and a per-table design 1204 across tables before quantization.
[0137] There may be multiple versions of quantizers (parameterized by different parameter sets) that can be used to achieve lower quantization error. In some embodiments, multiple versions of the quantizer may be UE-defined quantizers that may attempt to achieve lower quantization error for a given overhead. The reference quantizer design for quantization may be a design for each polarization across tables, each polarization / each table, each table, and / or for dividing the LC coefficients into groups. Each polarization / each table design may have two references for two polarizations on each selected table and / or may be applied to a selected table. The design for dividing the LC coefficients into groups may have a reference per group. The strongest LC coefficient among the LC coefficients on all tables may be identified and used to normalize (or divide) all LC coefficients on all tables. The strongest LC coefficient in a group may then be identified and its amplitude may be quantized with a high-resolution amplitude quantizer. Each LC coefficient (or each non-zero LC coefficient) in the group is then divided using the quantized amplitude of the strongest LC coefficient in the group, and each divided LC coefficient is then quantized with a normal resolution quantizer. All phases of the non-zero LC coefficients in all tables may be quantized with phase quantization (e.g., with a 16PSK constellation). It is also possible to perform joint amplitude-phase quantization instead of separate amplitude and phase quantization. In a certain embodiment, one group consists of LC coefficients at the same polarization and on the same table as the strongest LC coefficient from all tables, and the other group consists of LC coefficients at a different polarization or a different table than the strongest LC coefficient from all tables.
[0138] In each polarization design 1202 across the tables, each of the LC coefficients within the polarizations across the tables may be quantized based on the selected LC coefficients from the tables. For example, the illustrated embodiment of each polarization design 1202 across the tables may include a first table 1206, a second table 1208, a third table 1210, a fourth table 1212, and a fifth table 1214 for the codebook. The first table 1206 may have a frequency offset of positive 2 (which may be represented as 2·Δf). The second table 1208 may have a frequency offset of positive 1 (which may be represented as 1·Δf). The third table 1210 may have a frequency offset of zero (which may be represented as 0·Δf). The fourth table 1212 may have a frequency offset of negative 1 (which may be represented as -1·Δf). The fifth table 1214 may have a frequency offset of negative 2 (which may be represented as -2·Δf). The top four rows in each of the tables may correspond to the first polarity, and the bottom four rows in each of the tables may correspond to the second polarity. The x-axis of each of the tables may correspond to the FD component and the y-axis may correspond to the spatial beam, wherein each square in the tables corresponds to an index of the FD component and an index of the spatial beam.
[0139] The UE may determine the LC coefficients in each of the polarizations, utilizing the LC coefficients to perform quantization of the LC coefficients in the tables. For example, the UE may identify from among all the tables the strongest LC coefficient (e.g., the LC coefficient with the largest amplitude) that the UE finds associated with one polarization (the "stronger" polarization) from the tables, such as the first polarization, and identify the strongest LC coefficient associated with another polarization (the "weaker" polarization) from the tables, such as the second polarization. The UE may quantize the LC coefficients associated with the first polarization using the strongest LC coefficient associated with the first polarization, and quantize the LC coefficients associated with the second polarization using the strongest LC coefficient associated with the second polarization.
[0140] In the illustrated embodiment, the UE may identify a first coefficient 1216 associated with a first polarization (as indicated within the top four rows) as the strongest LC coefficient (as indicated by the black fill of the first coefficient 1216). The UE may identify the amplitude of the first coefficient 1216 and utilize the first coefficient 1216 to normalize the LC coefficients associated with the first polarization (e.g., the LC coefficients within the top four rows). For example, the UE may divide the LC coefficients associated with the two polarizations by the first coefficient 1216 to normalize the LC coefficients. A first quantization scheme may be defined based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may use the first quantization scheme to convert each normalized LC coefficient associated with the "stronger" polarization to a corresponding value in the alphabet for the quantization. The UE may then utilize the quantized values to report the values of the LC coefficients in a CSI report sent to the base station.
[0141] In the illustrated embodiment, after normalizing with the strongest coefficients from the two polarizations, the UE may identify the second coefficient 1218 associated with the "weaker" polarization (in this example, as indicated by the second polarization located within the bottom four rows) as the strongest LC coefficient within the "weaker" polarization (as indicated by the vertical line fill of the second coefficient 1218). A second quantization scheme may be defined based on the number of bits to be used for the quantization to define an alphabet for the quantization, and the second quantization scheme may have more bits than the first quantization scheme. The UE may convert the second coefficient to a corresponding value in the alphabet for the quantization using the second quantization scheme. The UE may identify the quantization amplitude of the second coefficient 1218 and divide the LC coefficient associated with the second polarization (e.g., the LC coefficient located within the bottom four rows) using the quantization amplitude of the second coefficient 1218. For example, the UE may divide the amplitude of the LC coefficient associated with the second polarization by the quantization amplitude of the second coefficient 1218 using the first quantization scheme to normalize the amplitude of the LC coefficient. In some embodiments, the number of bits to be used for the quantization of the second polarization may be the same as the number of bits to be used for the quantization of the first polarization. The UE may convert each normalized LC coefficient associated with the second polarization to a corresponding value in the alphabet for that quantization. In some embodiments, the UE may perform phase quantization on the normalized LC coefficients at the two polarizations. The UE may then use the quantized values to report the values of the LC coefficients in a CSI report sent to the base station.
[0142] In each table design 1204, the strongest LC coefficient among all tables is identified, and all coefficients at all tables can be divided by the strongest LC coefficient among all tables. Afterwards, each of the LC coefficients within the table can be quantized based on the selected LC coefficient from the table. For example, the illustrated embodiment of each table design 1204 can include a first table 1220, a second table 1222, a third table 1224, a fourth table 1226, and a fifth table 1228 for the codebook. The first table 1220 can have a frequency offset of positive 2 (which can be represented as 2·Δf). The second table 1222 can have a frequency offset of positive 1 (which can be represented as 1·Δf). The third table 1224 can have a frequency offset of zero (which can be represented as 0·Δf). The fourth table 1226 can have a frequency offset of negative 1 (which can be represented as -1·Δf). The fifth table 1228 can have a frequency offset of negative 2 (which can be represented as -2·Δf). The top four rows in each of the tables may correspond to the first polarity, and the bottom four rows in each of the tables may correspond to the second polarity. The x-axis of each of the tables may correspond to the FD component, and the y-axis may correspond to the spatial beam, where each square in the tables corresponds to an index of the FD component and an index of the spatial beam.
[0143] The UE may determine the LC coefficients in each of the tables and utilize the LC coefficients to perform quantization of the LC coefficients in each of the tables. For example, the UE may identify the strongest LC coefficient (e.g., the LC coefficient with the largest amplitude) from the first table 1220, the strongest LC coefficient from the second table 1222, the strongest LC coefficient from the third table 1224, the strongest LC coefficient from the fourth table 1226, and the strongest LC coefficient from the fifth table 1228. In the event that a table does not have any non-zero LC coefficients (such as the fourth table 1226 in the illustrated embodiment), quantization of the table may not be performed. The UE may first quantize the amplitude of the strongest coefficient at each table using a high-resolution quantization scheme. In the illustrated embodiment, the UE may normalize (or divide) the LC coefficients within the first table 1220 using the quantized amplitude of the strongest LC coefficient from the first table 1220, normalize (or divide) the LC coefficients within the second table 1222 using the quantized amplitude of the strongest LC coefficient from the second table 1222, normalize (or divide) the LC coefficients within the third table 1224 using the quantized amplitude of the strongest LC coefficient from the third table 1224, and normalize (or divide) the LC coefficients within the fifth table 1228 using the quantized amplitude of the strongest LC coefficient from the fifth table 1228.
[0144] In the illustrated embodiment, the UE may identify the first coefficient 1230 of the first table 1220 as the strongest LC coefficient (as indicated by the vertical stripe fill of the first coefficient 1230). The UE may identify the quantized amplitude of the first coefficient 1230 using a high-resolution quantization scheme and divide the LC coefficients within the first table 1220 using the quantized amplitude of the first coefficient 1230. For example, the UE may divide the amplitude of the LC coefficients within the first table 1220 by the quantized amplitude of the first coefficient 1230 to normalize the amplitude of the LC coefficients. The quantized amplitude of the first coefficient 1230 may be obtained by a high-resolution quantization scheme that defines an alphabet for the quantization based on the number of bits to be used for the quantization. The UE may convert each divided LC coefficient within the first table 1220 to a corresponding value in the alphabet for the quantization using a normal quantization scheme. The UE may then use the quantized value to report the value of the LC coefficient in a CSI report sent to the base station.
[0145] In the illustrated embodiment, the UE may identify the second coefficient 1232 of the second table 1222 as the strongest LC coefficient (as indicated by the vertical stripe fill of the second coefficient 1232). The UE may identify the quantized amplitude of the second coefficient 1232 and utilize the quantized amplitude of the second coefficient 1232 to normalize / divide the LC coefficients within the second table 1222. For example, the UE may divide the amplitude of the LC coefficients within the second table 1222 by the quantized amplitude of the second coefficient 1232 to normalize / divide the amplitude of the LC coefficients. The quantized amplitude of the second coefficient 1232 may be obtained by a high-resolution quantization scheme that is based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the second table 1222 to a corresponding value in the alphabet for the quantization using normal resolution amplitude quantization. The UE may then perform phase quantization on the normalized LC coefficients at the two polarizations. The UE may then utilize the quantized values to report the values of the LC coefficients in a CSI report sent to the base station.
[0146] In the illustrated embodiment, the UE may identify the third coefficient 1234 of the third table 1224 as the strongest LC coefficient (as indicated by the black fill of the third coefficient 1234). The UE may identify the quantized amplitude of the third coefficient 1234 and utilize the quantized amplitude of the third coefficient 1234 to normalize / divide the LC coefficients within the third table 1224. For example, the UE may divide the amplitude of the LC coefficient within the third table 1224 by the quantized amplitude of the third coefficient 1234 to normalize the amplitude of the LC coefficient. The quantized amplitude of the third coefficient 1234 may be obtained by a high-resolution quantization scheme that is based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the third table 1224 to a corresponding value in the alphabet for the quantization using a normal resolution quantization scheme. The UE may then utilize the quantized value to report the value of the LC coefficient in a CSI report sent to the base station.
[0147] In the illustrated embodiment, the UE may identify the fourth coefficient 1236 of the fifth table 1228 as the strongest LC coefficient (as indicated by the vertical stripe fill of the fourth coefficient 1236). The UE may identify the quantized amplitude of the fourth coefficient 1236 and utilize the quantized amplitude of the fourth coefficient 1236 to quantize the LC coefficients within the fifth table 1228. For example, the UE may divide the amplitude of the LC coefficients within the fifth table 1228 by the quantized amplitude of the fourth coefficient 1236 to normalize / divide the amplitude of the LC coefficients. The amplitude of the fourth coefficient 1236 may be obtained by a high-resolution quantization scheme that is based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the fifth table 1228 to a corresponding value in the alphabet for the quantization using a normal resolution quantization scheme. The UE may then perform phase quantization on the normalized LC coefficients at the two polarizations. The UE may then utilize the quantized values to report the values of the LC coefficients in a CSI report sent to the base station.
[0148] In each polarization / each table design, each of the LC coefficients within the table and associated with the polarization may be quantized based on a selected LC coefficient from the table and associated with the polarization. In the event that a table and / or a polarization of a table does not include a non-zero LC coefficient, the table and / or the polarization of the table may not be quantized. For clarity, the description of each polarization / each table design refers to a table per table design 1204.
[0149] Referring to the table of each table design 1204 implementation shown, the UE can identify the strongest LC coefficient from the first polarization of the first table 1220 (as indicated by the top four rows) and identify the strongest LC coefficient from the second polarization of the first table 1220 (as indicated by the bottom four rows).
[0150] The UE may quantize the LC coefficients associated with the first polarization within the first table 1220 using the strongest LC coefficient for the first polarization from the first table 1220. In the illustrated embodiment, the UE may identify the first coefficient 1230 as the strongest LC coefficient for the first polarization from the first table 1220, and quantize the LC coefficients associated with the first polarization within the first table 1220 using the amplitude of the first coefficient 1230. For example, the UE may divide the amplitude of the LC coefficient associated with the first polarization of the first table 1220 (as indicated in the top four rows) by the amplitude of the first coefficient 1230 to normalize the amplitude of the LC coefficient. The amplitude of the first coefficient 1230 may be divided into equal parts based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the first table 1220 to a corresponding value in the alphabet for the quantization. The UE may then use the quantized value to report the value of the LC coefficient in a CSI report sent to the base station.
[0151] The UE may quantize the LC coefficients associated with the second polarization in the first table 1220 using the strongest LC coefficient for the second polarization from the first table 1220. In the illustrated embodiment, the UE may identify the fifth coefficient 1238 as the strongest LC coefficient for the second polarization from the first table 1220, and quantize the LC coefficients associated with the second polarization in the first table 1220 using the amplitude of the fifth coefficient 1238. For example, the UE may divide the amplitude of the LC coefficient associated with the second polarization of the first table 1220 (as indicated in the bottom four rows) by the amplitude of the fifth coefficient 1238 to normalize the amplitude of the LC coefficient. The amplitude of the fifth coefficient 1238 may be divided into equal parts based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient in the first table 1220 to a corresponding value in the alphabet for the quantization. The UE may then use the quantized value to report the value of the LC coefficient in a CSI report sent to the base station. The UE may repeat this process for each of these tables in the codebook.
[0152] In the process of dividing the LC coefficients into groups design, the non-zero LC coefficients can be divided into groups and the LC coefficients within a group can be quantized based on the selected LC coefficients within the group. The definition of the groups into which these groups are divided can be predefined. In some embodiments, these groups can be defined based on the spatial beams and / or FD components to which the LC coefficients belong. For clarity, the description of dividing the LC coefficients into groups design involves the table of each table design 1204.
[0153] For an example of dividing the LC coefficients into groups, the first group may be defined as the first three FD components in the table and the second group may be defined as the last three FD components in the table. For example, in the first table 1220, the LC coefficients located in the three leftmost columns may be defined as the first group and the coefficients located in the three rightmost columns may be defined as the second group. The UE may identify the strongest LC component in the first group and quantize the LC components in the first group using the amplitude of the strongest LC component in the first group. In addition, the UE may identify the strongest LC component in the second group and quantize the LC components in the second group using the amplitude of the strongest LC component in the second group.
[0154] Referring to the table of each table design 1204 implementation shown, the UE can identify the strongest LC coefficients from the leftmost three columns of the first table 1220 and can identify the strongest LC coefficients from the rightmost three columns of the first table 1220. For example, the UE can identify the first coefficient 1230 as the strongest LC coefficient from the rightmost three columns of the first table 1220.
[0155] The UE may quantize the LC coefficients within the first group using the strongest LC coefficients from the first group (defined as the three leftmost columns of the first table 1220). In the illustrated embodiment, the UE may identify the first coefficient 1230 as the strongest LC coefficient from the first group within the first table 1220, and quantize the LC coefficients within the first group within the first table 1220 using the amplitude of the first coefficient 1230. For example, the UE may divide the amplitude of the LC coefficients within the first group of the first table 1220 by the amplitude of the first coefficient 1230 to normalize the amplitude of the LC coefficients. The amplitude of the first coefficient 1230 may be divided into equal parts based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the first group of the first table 1220 to a corresponding value in the alphabet for the quantization. The UE may then use the quantized value to report the value of the LC coefficient in a CSI report sent to the base station.
[0156] The UE may quantize the LC coefficients within the second group using the strongest LC coefficients from the second group (defined as the three rightmost columns of the first table 1220). In the illustrated embodiment, the UE may identify the fifth coefficient 1238 as the strongest LC coefficient from the second group within the first table 1220, and quantize the LC coefficients within the second group within the first table 1220 using the amplitude of the fifth coefficient 1238. For example, the UE may divide the amplitude of the LC coefficients within the second group of the first table 1220 by the amplitude of the fifth coefficient 1238 to normalize the amplitude of the LC coefficients. The amplitude of the fifth coefficient 1238 may be divided into equal parts based on the number of bits to be used for the quantization to define an alphabet for the quantization. The UE may convert each normalized LC coefficient within the second group of the first table 1220 to a corresponding value in the alphabet for the quantization. The UE may then use the quantized value to report the value of the LC coefficient in the CSI report sent to the base station. The UE may repeat this process for each of these tables within the codebook.
[0157] Fig.13 An exemplary differential encoding method 1300 according to some embodiments is illustrated. For example, the differential encoding method 1300 illustrates exemplary differential encoding in the time domain and frequency domain of CQI (such as subband CQI). The method described herein for reporting CQI in a CSI report may implement differential encoding in the time domain and / or differential encoding in the frequency domain.
[0158] The design issues addressed herein may include subband CQI feedback for multiple occasions. For example, the feedback overhead of CQI may be a problem. This problem may exist for subband CQI, and when the number of PDSCH opportunities is large, this problem may also exist for wideband CQI. Feedback overhead may be solved based at least in part on implementing differential coding in the time domain and / or frequency domain. In some embodiments, the differential coding may be Huffman coding for ΔCQI. Since the difference between two subband CQIs at two adjacent subbands (which may include one or two or more parts with a time-varying subband predecoder) at the same PDSCH opportunity may be caused by frequency selective fading and / or frequency selective interference, the change should be gradual when there are sudden changes in frequency selective fading and / or frequency selective interference. Therefore, the statistical mode of its difference (ΔCQI) or change tends to be about zero. And the difference between the subband CQIs at the same subband (which may include one part or two or more parts with a time-varying subband predecoder) at different PDSCH opportunities may be caused by time-selective fading and time-selective interference, and the change should be gradual when there is a sudden change in time-selective fading and / or time-selective interference. Therefore, the statistical pattern of its difference (ΔCQI) or change tends to be about zero. According to the unequal probability of its change or difference (ΔCQI) having values such as 0dB, 1dB, -1dB, 2dB, -2dB, entropy coding (generally Huffman coding) can be particularly helpful in reducing the feedback of ΔCQI, for example, code "0" is used for 0dB, code "100" is used for 1dB, code "101" is used for -1dB, etc. In order to allow consistent understanding between the UE and the base station, the Huffman coding dictionary can be signaled by the base station or indicated by the UE. Due to the use of entropy coding, the payload size of CQI (which may be a wideband CQI and / or a subband CQI using ΔCQI) may not be fixed because it is subject to channel / interference fluctuations in the time domain and / or in the frequency domain. Therefore, the CQI UCI may be carried in UCI part 2 in PUCCH and in CSI part 1 or part of CSI part 2 in PUSCH as in Rel-16, and the CQI UCI size may be carried in UCI part 1 in PUCCH and in part of the hybrid automatic repeat request acknowledgement (HARQ-ACK) feedback in PUSCH as in Rel-16.
[0159] The differential coding method 1300 may include a time domain differential coding method. The time domain differential coding method may include encoding multiple subbands 1302. The x-axis of the multiple subbands 1302 is for time, and the y-axis of the multiple subbands 1302 is for subband frequency. The UE may perform differential coding of CQI on the same subband in the time domain. For example, the UE may identify the first subband 1304 and the second subband 1306 from the multiple subbands 1302. In the time domain, the second subband 1306 may be after the first subband 1304. The UE may determine the difference between the first subband 1304 and the second subband 1306, and may perform coding based on the difference. The UE may perform differential coding on each subband within the multiple subbands 1302. The UE may report a reference CQI and a ΔCQI with the coding in a CSI report sent to the base station.
[0160] The differential encoding method 1300 may include a frequency domain-domain differential encoding method. The frequency domain differential encoding method may include encoding multiple subbands 1308. The x-axis of the multiple subbands 1308 is for time, and the y-axis of the multiple subbands 1308 is for subband frequency. The UE may perform differential encoding of adjacent subbands in the frequency domain. For example, the UE may identify a first subband 1310 and a second subband 1312 from the multiple subbands 1308. In the frequency domain, the second subband 1312 may be after the first subband 1310. The UE may determine the difference between the first subband 1310 and the second subband 1312, and may perform encoding based on the difference. The UE may perform differential encoding for each subband within the multiple subbands 1308. The UE may report a CQI with the encoding in a CSI report sent to the base station.
[0161] As described above, the methods described herein may introduce the time domain in the codebook design. At a given spatial layer, the UE may report selected TD components, the UE may report selected spatial beams for each selected TD component (or each table), and / or the UE may report selected FD components for each selected TD component (or each table). Any one or all of the number of selected TD components, the number of selected spatial beams, and / or the number of selected FD components may be reported by the UE.
[0162] As described above, the first method for selecting non-zero LC coefficients may be through bitmaps and component composition patterns. To reduce signaling overhead, a base station (such as gNB 2600 ( Fig.26 )) indicates the component composition patterns and their frequencies of occurrence; then a Huffman coding scheme can be used to relate to those patterns without using bitmaps, thereby reducing signaling overhead. In a second method, the selection of NZ LC coefficients can be through multiple bitmaps. The strongest LC coefficients among all spatial beams, FD components, and TD components can be shifted to the origin position relative to the FD components and TD components. The same shift can be applied to the LC coefficients on all tables.
[0163] Multiple quantizer designs are described herein. For example, the quantizer design may include LC coefficient quantization, which may be by a fixed quantizer (specified in the specification), a parameterized quantizer with parameters configurable by the gNB and / or reported by the UE. The UE may report the UE-defined quantizer to the base station. To allow for better quantization, the UE-defined quantizer may be provided to the base station along with RRC signaling and / or MAC CE and / or CSI reporting. In addition, multiple versions (all UE defined, UE defined + specified, multiple parameterized, etc.) may be active at the same time, and the UE may refer to the quantizer version in the CSI report. For two-stage quantization, the LC coefficients may be divided into one or more sets, and a reference amplitude may be determined for each set.
[0164] The time domain dimension for the reported PMI may be determined by the maximum gap between the CSI feedback and the time when the last pre-decoder may be used. The method described herein may support this configuration to allow multiple pre-decoders within the same time slot / same PDSCH, thereby accounting for high Doppler situations. In addition, the method herein may support differential coding of wideband / sub-band CQI across time and / or frequency, and Huffman coding may be used to reduce feedback overhead.
[0165] Fig.14 1400 is a diagram illustrating a UE 1404 providing CSI feedback to a base station 1408 according to some embodiments. The UE 1404 may generate a plurality of pages 1412. The pages 1402 may represent 3D coding of LC coefficients corresponding to a spatial layer. A separate page may include the dimensions of a spatial beam of an FD component and may correspond to a frequency offset. The pages 1402 may correspond to Fig. 9 Table of codebook 900 described.
[0166] At 1416, the UE 1404 may generate one or more input matrices based on the page 1412. The input matrices may then be provided to one or more neural networks (NNs) 1420 for autoencoder compression.
[0167] The NN of NN 1420 may have an input layer whose dimensions match the dimensions of the input matrix. The NN may include one or more hidden layers and an output layer with M×1 dimensions that outputs an M×1 codeword. Each of the M values of the codeword may be represented by one or more bits, depending on the desired granularity. The number of bits that the NN uses to represent each of the M values may be referred to as the bit width of the NN.
[0168] Each layer of the NN's layers may have a different number of nodes, each node connected to nodes of adjacent layers or nodes of non-adjacent layers. In general, at some layers, nodes may produce outputs that are nonlinear functions of the sum of their inputs, and provide these outputs to nodes of adjacent layers through corresponding connections. Weights may adjust the strength of connections between nodes of adjacent layers. Weights may be initially set based on a training process with training inputs and expected outputs. The training inputs may be provided to the NN, and the difference between the outputs and the expected outputs may be used to adjust the weights. After training, and in operation, the NN may continue to dynamically update the weights.
[0169] In general, the training, operation, and architecture of NN 1420 may be similar to existing neural networks that provide image / video processing or to the network described in “Deep Learning for Massive MIMO CSI Feedback,” Chao-Kai Wen, Wan-Ting Shih, and Shi Jin, IEEE Wireless Communications Letters (Vol. 7, No. 5, October 2018, pp. 748-751, published March 22, 2018).
[0170] In operation, NN 1420 may output a codeword corresponding to each input matrix. The codewords output by NN 1420 may be sent to base station 1408 in a CSI report.
[0171] Base station 1408 may include NNs 1424 that provide complementary operations with respect to NN 1420 to generate one or more output matrices 1428 corresponding to input matrices 1416. Base station 1408 may deconstruct output matrices 1428 to recover pages 1432 corresponding to pages 1412. Base station 1408 may use pages 1432 and the MIMO codebook to identify desired precoder values to use for downlink transmissions.
[0172] Various embodiments of the present disclosure describe how to generate the input matrix 1416 in a manner that facilitates efficient and effective communication of CSI feedback.
[0173] Fig.15 An example CSI feedback operation 1500 according to some embodiments is illustrated. The CSI feedback operation 1500 may include providing a plurality of pages 1512 to a corresponding plurality of NNs 1516. As shown, there may be m pages 1512 and NNs 1516. In this embodiment, each page may correspond to a respective input matrix. Each of the NNs 1516 may have the same weight and may provide inference output as codewords 1-m, which may be collected and sent to the base station in a CSI report.
[0174] Page 1512 may correspond to tables 902, 904, 906, 908, and 910, in which case m may be equal to 5. Collectively, page 1512 may represent 3D coding of CSI coefficients generated based on measurements of CSI measurement resources. An individual page may correspond to one TD component (e.g., one frequency offset). An individual page may have the dimensions of a spatial beam selected by a selected FD component. Selection of the spatial beam and FD component may be performed as described elsewhere herein.
[0175] In some cases, some of the TD components may be stronger than other TD components, for example, the sum of the squares of the linear combination coefficients on TD component 3 may be greater than the sum of the squares of the linear combination coefficients on any other TD component, or the amplitude of the linear combination coefficient with the maximum amplitude on TD component 3 may be greater than the amplitude of the linear combination coefficient with the maximum amplitude on other TD components, or the maximum amplitude of the real / imaginary part of the linear combination coefficient on TD component 3 may be greater than the amplitude of the maximum amplitude of the real / imaginary part of the linear combination coefficient on other TD components. Therefore, the dynamic range of the linear combination coefficients or the real / imaginary part of the linear combination coefficients at each TD component may be different. This feature can be used to further reduce the CSI feedback overhead. This can be done according to one or more of the following alternatives.
[0176] In a first alternative, the amplitude difference may be reflected as the relative amplitude of each component. For example, the reference amplitude value of the TD component at 0 Hz (assuming that the TD component at 0 Hz is the strongest one) may be set to 1, and the reported reference amplitude values of the other TD components may be relative to the reference amplitude value of the TD component at 0 Hz. For example, a reported reference amplitude value of 0.25 for TD component 2 may result in the NN decoder output value of TD component 2 being scaled by 0.25, while the NN decoder output value of TD component 3 is scaled by 1 (if TD component 3 is considered to be the strongest TD component). All inferred outputs from NN 1516 may have the same bit width.
[0177] In a second alternative, the amplitude difference may be reflected at different quantization granularities (e.g., bit widths) for separate NNs. For example, a NN receiving a TD component at 0 Hz may have, for example, eight bits for each of the M values. However, a NN receiving a different TD component may have fewer bits (e.g., six bits) for each of the M values. Note that the number of outputs (i.e., "M") may be different for different TD components. The UE may need to decide the feedback overhead allocation for different pages through the bit width selection and the M selection.
[0178] Fig.16An example CSI feedback operation 1600 according to some embodiments is illustrated. The CSI feedback operation 1600 may include providing a plurality of pages 1612 to a NN 1616. As shown, there may be two pages 1612 and one NN 1616. In other embodiments, another number of pages 1612 may be used. Page 1612 may be similar to page 1512 or table described elsewhere herein. NN 1616 may provide an inferred output as codeword 1, which may be sent to a base station in a CSI report. When more than one page is fed to a NN, the procedure for deciding the allocation of feedback overhead for different pages may be omitted / circumvented.
[0179] To provide multiple pages 1612 to a NN 1616, the UE may first concatenate the pages 1612 into an input matrix 1614. As shown, the pages 1612 may be combined along the FD component dimension.
[0180] The CSI coefficients from page 1612 may be read out into input matrix 1614 according to various order combinations. For example, the coefficients from the selected spatial beam may be read out first, followed by the coefficients from the FD component second, followed by the coefficients from the selected TD component third. For another example, the coefficients from the selected spatial beam may be read out first, followed by the coefficients from the selected TD component second, followed by the coefficients from the selected FD component third. In other embodiments, the coefficients may be read out in other orders.
[0181] To facilitate combining pages 1612 along the FD component dimension, the total number of spatial beams selected for each page can be the same. In some embodiments, the particular spatial beam selected for TD component 1 can be the same spatial beam selected for TD component 2 (e.g., both TD components 1 and 2 can have spatial beams corresponding to indices 1 to 4). In other embodiments, even if the total number of selected spatial beams is the same, the spatial beams selected for TD component 1 can be different from the spatial beams selected for TD component 2. For example, TD component 1 can have spatial beams corresponding to indices 1 to 4, and TD component 2 can have spatial beams corresponding to indices 2 to 5.
[0182] When pages 1612 are combined along the FD component dimension, the total number of FD components selected (as well as the specific FD components selected) may be different for each page.
[0183] Fig.17 An example CSI feedback operation 1700 is illustrated in accordance with some embodiments. Similar to CSI feedback operation 1600, CSI feedback operation 1700 may include providing multiple pages 1712 to NN 1716. However, instead of combining pages 1712 along the FD component dimension, CSI feedback operation 1700 combines pages 1712 along the spatial beam dimension.
[0184] To facilitate combining pages 1712 along the spatial beam dimension, the total number of FD components selected for each page can be the same. In some embodiments, the particular FD components selected for TD component 1 can be the same FD components selected for TD component 2 (e.g., both TD components 1 and 2 can have FD components corresponding to indices 2 to 8). In other embodiments, even if the total number of selected FD components is the same, the FD components selected for TD component 1 can be different from the FD components selected for TD component 2. For example, TD component 1 can have FD components corresponding to indices 2 to 8, and TD component 2 can have FD components corresponding to indices 4 to 10.
[0185] In all other respects (eg, connections, selection of FD components or spatial beams, readout order, etc.), CSI feedback operation 1700 may be similar to CSI feedback operation 1600 .
[0186] In some embodiments, as described above with respect to Fig.16 and Fig.17 The described combination of multiple pages into an input matrix and providing the input matrix to a NN can resolve any differences in signal strength at different Doppler pages through a quantization process provided by the NN.
[0187] Fig.18 An example CSI feedback operation 1800 according to some embodiments is illustrated. The CSI feedback operation 1800 may include providing a plurality of pages 1812 to a NN 1816. As shown, there may be two pages 1812 and one NN 1816. In other embodiments, another number of pages 1812 may be used. Page 1812 may be similar to page 1512 or table described elsewhere herein. NN 1816 may provide an inferred output as codeword 1, which may be sent to the base station in a CSI report.
[0188] CSI feedback operation 1800 can be similar to CSI feedback operation 1600 in that both operations combine pages along the FD component direction and feed the combined pages to one NN. However, with respect to CSI feedback operation 1800, page 1812 can have a different number of selected spatial beams. Page 1812 can have a different number or the same number of selected FD components. In some embodiments, to facilitate combining pages with different numbers of selected spatial beams along the FD component dimension, zero components or padding components can be added to TD components with fewer selected spatial beams (e.g., TD component 2). The zero components can have dimensions that match the selected FD components of TD component 2 and the difference between the selected spatial beams of TD component 1 and the selected spatial beams of TD component 2. Although Fig.18The zero values are shown added below the TD component 2, but in other embodiments, the zero values may be added additionally / alternatively above the TD component 2. In other embodiments, the zero values may be interleaved with the spatial beams of the TD component 2. This may be done as part of an interpolation process that facilitates smoothing of the values, as described with respect to Fig.19 described in further detail.
[0189] In some embodiments, providing zero / fill values to the TD component can be part of the page generation process. For example, a first page corresponding to TD component 2 can be transformed into a second page including TD component 2 and zero values. The second page can then be combined with the page corresponding to TD component 1 into the input matrix 1814.
[0190] In all other respects (eg, connections, selection of FD components or spatial beams, readout order, etc.), CSI feedback operation 1800 may be similar to CSI feedback operation 1600 .
[0191] In some embodiments, in order to reduce signaling overhead, various constraints may be imposed on the selection of spatial beams or FD components at the page level. For example, among all 2×N_1×N_2 spatial beams, the UE may first determine a common set of spatial beams for all pages. In some embodiments, the common set of spatial beams may be determined by including all spatial beams from the total number of spatial beams, the total number of spatial beams including coefficient values exceeding a predetermined threshold for at least one of the selected pages. The coefficient value may correspond to an aggregated coefficient value, an average value, or some other function. In some embodiments, the common set of spatial beams may be determined by first calculating a wideband covariance matrix with multiple CSI measurement opportunities. Those wideband covariance matrices may be summed to obtain a summed covariance matrix. The projection power may be tested with spatial beams, and the strongest spatial beam may be selected for the common set of spatial beams.
[0192] In some embodiments, all 2×N_1×N_2 spatial beams are selected.
[0193] The common set of spatial beams may be applied to both polarizations. Alternatively, one set of common spatial beams may be defined with respect to a first polarization, while another set of common spatial beams may be defined with respect to a second polarization.
[0194] Once the common spatial beam set is determined, the UE may select a spatial beam from the common spatial beam set for each page of the plurality of pages.
[0195] In some embodiments, the number of selected spatial beams at each page (if a page is selected) may be common among all pages. The number of selected spatial beams may be configured by the base station. Alternatively, the number of selected spatial beams may be first determined by the UE and then reported to the base station.
[0196] In other embodiments, the number of selected spatial beams at each page may be different. The UE may report the number of selected spatial beams to the base station. In some cases, the upper or lower limit of the selected spatial beams may be configured by the base station or predefined in, for example, 3GPP TS.
[0197] Regarding constrained FD component selection, the UE may first determine a common FD component set among all N_3 FD components for all pages. In some embodiments, the common FD component set may be determined by including all FD components from the total number of FD components, the total number of FD components including coefficient values exceeding a predetermined threshold for at least one of the selected pages. The coefficient value may correspond to an aggregated coefficient value, an average value, or some other function. In some embodiments, a common FD component group may be determined by first calculating the FD components of a selected set of spatial beams at each page. The power of the FD component at a given spatial beam may be summed over all pages or summed at all spatial beams on all pages. Then, a strong FD component may be selected for the common FD component set.
[0198] In some embodiments, the FD component set may consist of all available FD components (N3).
[0199] The common set of FD components may apply to both polarizations. Alternatively, one common set of FD components may be defined with respect to a first polarization, while another common set of FD components may be defined with respect to a second polarization.
[0200] Once the common FD component set is determined, the UE may select an FD component from the FD component set for each page of the plurality of pages.
[0201] In some embodiments, the number of selected FD components at each page (if a page is selected) may be common among all pages. The number of selected FD components may be configured by the base station. Alternatively, the number of selected FD components may be first determined by the UE and then reported to the base station.
[0202] In other embodiments, the number of selected FD components at each page may be different. The UE may report the number of selected FD components to the base station. In some cases, an upper or lower limit of the selected FD components may be configured by the base station or predefined in, for example, 3GPP TS.
[0203] It may be expected that the inference output provided by the neural network contains mostly non-zero coefficients even after quantization. When this is not the case, further compression can be performed by adjusting the NN structure to obtain a compact representation of the CSI (e.g., 3D matrix for linear combination coefficients).
[0204] In some cases, a neural network trained on a large amount of data from many sites can provide an inferred output with zero for a particular unit. In some embodiments, a bitmap can be used to represent non-zero coefficients. For example, for an inferred output comprising M elements, each element having a bit width B, an M-element bitmap can be used to represent the position of the non-zero coefficient. For example, for M=4, a bitmap [1 00 1] can indicate two values at the B position instead of four values at the B position. Note that for M=4, and each value consists of 6 bits, without considering the case where some values are zero values, the feedback overhead for the M values is 4×6 bits=24 bits. If a bitmap or other signaling scheme (such as a combined index) is used to indicate the position of zero values and / or the position of non-zero values, the M-bit bitmap [1 0 0 1] consumes 4 bits, while only two non-zero values are required in the feedback, and the feedback overhead is 4 bits + (2×6) = 16 bits.
[0205] Fig.19 Smoothing operation 1900 according to some embodiments is illustrated. In some cases, large transitions between coefficient values of one FD component to an adjacent FD component may impair the effectiveness of the automatic encoding operation. For example, page 1904 may include FD component 1904 having a relatively low value, and an adjacent FD component having a relatively high value. In some cases, the high contrast between these adjacent values may complicate processing by the neural network. Therefore, according to some embodiments, the UE may generate page 1916 by inserting FD component 1910 between FD components 1908 and 1912. FD component 1910 may have an interpolated value set between the values of FD components 1908 and 1912. The interpolated value of FD component 1910 may smooth the transition to facilitate processing by the neural network.
[0206] Embodiments herein provide a method for selecting a spatial beam by first determining a spatial beam selection (e.g., W1 in), TD component selection (for example, W d ) or FD component selection (e.g., In ) describe compact CSI feedback. However, other embodiments may omit one or more of these options and instead rely on various aspects of the autoencoder operation to provide a compact representation of the CSI feedback. Fig. 20 and Fig.21 CSI feedback operations are described in which one or more of these selections may be omitted.
[0207] Fig. 20 An example CSI feedback operation 2000 is illustrated in accordance with some embodiments. The CSI feedback operation 2000 may include providing a plurality of pages 2012 to a NN 2016. As shown, there may be two pages 2012 and one NN 2016. In other embodiments, another number of pages 2012 may be used.
[0208] In general, CSI feedback operation 2000 may be similar to CSI feedback operation 1800; however, in this embodiment, page 2012 may have dimensions of a selected spatial beam and frequency domain (e.g., all subbands). In this embodiment, DFT / inverse DFT (IDFT) to generate FD components / delay taps from a frequency domain representation may not be used.
[0209] The CSI feedback operation 2000 is shown as combining pages in the spatial beam direction 2012. Other embodiments may operate in a similar manner except that pages may be combined in the subband direction.
[0210] Fig.21 An example CSI feedback operation 2100 is illustrated in accordance with some embodiments. The CSI feedback operation 2100 may include providing a plurality of pages 2112 to a NN 2116. As shown, there may be two pages 2112 and one NN 2116. In other embodiments, another number of pages 2112 may be used.
[0211] In general, CSI feedback operation 2100 may be similar to CSI feedback operation 2100; however, in this embodiment, page 2112 may have dimensions of Tx antenna ports and subbands. Thus, skipping spatial beam selection, the page height may be 32 in the Tx antenna port dimension, rather than having a page height of 8 after spatial beam selection.
[0212] In the frequency dimension, the width of the page can be 18 instead of 5 (see e.g. Figure 3 ). Instead of using DFT / IDFT for generating FT components from frequency domain representation, the pre-decoders of multiple time periods may be represented as images only.
[0213] The CSI feedback operation 2100 is shown as combining pages in the subband dimension 2112. Other embodiments may operate in a similar manner, except that pages may be combined in the Tx antenna port dimension.
[0214] Fig. 22 An example procedure 2200 for generating a CSI report according to some embodiments is illustrated. Aspects of the example procedure 2200 are also related to Fig.23 2300 is associated with the flowchart.
[0215] The example procedure 2200 may be performed with respect to a base station having N_t=2*N_1×N_2 antenna ports and N_r receive antennas at a UE. N_1 may be the number of columns in a rectangular antenna array, and N_2 may be the number of rows in a rectangular antenna array. There may be N_4 observations of the channel response from channel measurements on the CSI-RS or interpolation processing performed by the UE. In other embodiments, other values may be used.
[0216] Various operations are described with respect to the example procedure 2200 and flowchart 2300. However, some implementations may have only a subset of these operations.
[0217] At 2204, the example procedure 2200 may include forming a subband / time specific covariance matrix. The channel responses at more than one tone / PRB in a subband may be used to accumulate the subband covariance matrix, F_2 may be the number of tones / PRBs in the subband, and for each index f_2, 1≤f2≤F2 there may be a channel response. For a given observation at time t, 1≤t≤N4, at subband f, the channel response may be calculated by calculating the channel response matrix The covariance matrix is formed for subband (f) and observation (t) by accumulation, where f2 is the tone / PRB index within subband f, and there are F_2, The dimension of is N_r×N_t. Ignoring the scaling factor, we get Where R f,t is a N_t×N_t matrix, with f=1,...,N3, t=1,...,N4. The wideband covariance matrix (not considering the scaling factor) can then be formed by:
[0218] At 2208, the example procedure 2200 may include performing dimensionality reduction. This may be done by projecting into a lower dimensional space. For example, the top B spatial beams may be identified based on the following metrics: This identification of the top spatial beam may correspond to spatial beam selection 2304 .
[0219] With the top L spatial beams selected, B can be formed as follows: B = [b1 b2 ... b L ]. If there is oversampling in the vertical domain (O1≥1) or the horizontal domain (O2≥1), spatial beam selection can be performed for each oversampling grid (q1, q2), 0≤q1≤O1-1, 0≤q2≤O2-1. The block diagonal matrix can be formed by:
[0220] and Then it can be formed as: in It is a 2L×2L matrix, f=1,…,N_3, t=1,...,N4.
[0221] As mentioned above about Fig. 20 and Fig.21 Instead of projecting R with W1, f,t , the projection operation can be omitted, so that To accommodate both the dimensionality reduction option and the non-dimensionality reduction option, the notation N_B can be used for the number of spatial beams, so N B =2L and is a 2L×2L matrix in the dimension reduction option, and N B =2N1N2 and is a 2N1N2×2N1N2 matrix in the non-dimensionality reduction option. Regarding the non-dimensionality reduction option, the spatial beam selection 2304 may simply select all spatial beams.
[0222] At 2212, the example procedure 2200 may include performing a singular value decomposition (SVD) or an eigenvalue decomposition (EVD) on the subband / time-specific covariance matrix. This operation may correspond to generating a precoder for the subband at 2308.
[0223] Can Perform SVD / EVD, for spatial layer k, 1≤k≤N s , subband specific / time specific pre-decoder As R f,t The singular vectors of k = 1, 2, ..., N_s, where N_s is the number of spatial layers.
[0224] At 2216, the example procedure 2200 can include performing a phase rotation of the singular vectors or eigenvectors. This operation can correspond to the phase rotation at 2312.
[0225] The following processing is performed according to the spatial layer. And at a given spatial layer k, let For each k=1,…,N s , the spatial beam with the largest metric can be obtained by Find and select the row with the highest metric (spatial beam index ), and α can be determined accordingly k,f,t If a different metric is used, such as the L2 metric instead of the L1 metric The procedure may operate in a similar manner.
[0226] And for spatial layer k, time t and subband i, the phase normalization term can be given by:
[0227] The frequency domain pre-decoder matrix for spatial layer k and time t can be given by Given.
[0228] At 2220, the example procedure 2200 may include obtaining a delayed tap representation from the frequency domain representation. This may correspond to converting from the frequency domain representation (using DFT or IDFT) to a delayed tap representation (N_3 point DFT / IDFT) at 2316.
[0229] At a given spatial layer k, for time t, the procedure can be adopted along P k,t The time domain formula for the pre-decoder can be obtained by OK, where v k,j,i,t are the time domain coefficients of spatial layer k, spatial beam j and tap i for time t.
[0230] For a given spatial layer k, k = 1, ..., N s , then let Q k,t ,t=1,…,N4, we can construct a B 3D matrix Y of ×N3 k , and can be along the first dimension Y k Perform N4-point 1-D IDFT (a total of N B × N3 such that N4-point 1-D IDFT) to obtain Z k , which may have dimensions N4×N B ×N3. This can correspond to obtaining a Doppler domain representation where an N_4 point DFT / IDFT is applied to the dimensions (N4×N B ×N3) time-space delay tap matrix, 2320.
[0231] At 2224, the example procedure may include normalizing the coefficients. This may correspond to normalizing at 2324 with the strongest coefficient across all pages.
[0232] For example, the spatial beam on page t=1 Elements at (corresponding to the DC term in the IDFT) and an FD component of 1 (corresponding to zero delay) are guaranteed to have the largest amplitude among all elements, and Used for normalization of coefficients. Can have dimensions N4×N B ×N3.
[0233] At 2228, the example procedure may include selecting a TD component. This may be accomplished by selecting a TD component having a non-zero coefficient and potentially quantizing the coefficient as described elsewhere herein. This may correspond to selecting TD component 2328.
[0234] As described elsewhere herein, techniques from image / video processing may be utilized to facilitate processing of CSI feedback. Specifically, providing a neural network with autoencoder compression with AI / ML processing may facilitate operation. At 2232, the example procedure 2200 may include performing AI / ML processing. This may be done according to one of the following three options.
[0235] In the first NN coding option, the P provided at 2216 or after phase rotation 2312 k,t , 2×N4 image real(P k,t )、imag(P k,t )(P k,t The real and imaginary parts of N4, 1≤t≤N4, may be fed into one or more neural networks providing autoencoder compression to output one or more codewords that may be sent in a CSI report. A similar process may be performed after the dimensionality reduction option or the non-dimensionality reduction option.
[0236] It can also have a phase-amplitude representation, so it can be used for P k,t The amplitude and phase are compressed separately. Note that the actual phase rotation of the singular vectors / eigenvectors may depend on the UE specific implementation, and the phase rotation may be omitted or modified at operation 2216, for example to achieve a smoother transition when connecting two or more images.
[0237] In the second NN encoding option, Q from operation 2220 or after the conversion at 2316 k,t , which captures the delayed tap representation of the pre-decoder for multiple sub-bands and multiple times, the 2×N4 image real(Q k,t )、imag(Q k,t )(Q k,t The real and imaginary parts of N4), 1≤t≤N4 can be fed into one or more neural networks providing autoencoder compression to output one or more codewords that can be sent in the CSI report.
[0238] In the third NN coding option, the 3D matrix obtained from selecting the TD components 2228 and 2328 can be fed to a single neural network (e.g., an autoencoder designed to process 3D data input). In other embodiments, one or more neural networks can be designed to process 2D data. Thus, concatenated pages from different t can be fed to a single neural network. It can be noted that due to the normalization step and feeding the concatenated images to the same network, setting the quantization granularity for each page can be avoided.
[0239] As described herein, various options may be used for feedback overhead reduction. These options may include, for example, spatial beam selection aspects and FD component selection aspects.
[0240] Regarding the spatial beam selection aspect, instead of the common spatial beam selection corresponding to the above-mentioned dimensionality reduction option, or skipping the dimensionality reduction corresponding to the non-dimensionality reduction option, different spatial beams may be performed for different pages (eg, different t).
[0241] Regarding the FD component selection, in Z k (e.g., as described above with respect to operation 2320, which may occur at page t, Z k (t,:,:) (in matlab matrix convention) has non-zero elements only for some taps. Therefore, some embodiments provide that FD component selection can be performed to remove elements on the remaining taps to achieve overhead reduction.
[0242] In some embodiments, due to the small number of subbands N_3, temporal units N_4, or spatial beams N_B, the dimensions of the connected image or the dimensions of a single page are smaller than the designed capacity of the autoencoder. For example, the autoencoder may be designed for 64×32, while the size of the connected image or page may be 32×13. In these cases, padding with zeros or interpolated values may be applied to one, two, or three dimensions (e.g., to ). For a given dimension, zero padding or interpolation may be applied to the end, the beginning, or both of the non-padded portion. To ensure that the base station and the UE have the same understanding of the padding process, a procedure may be specified to determine the amount of padding at the beginning or the amount of padding at the end in one or more dimensions. Once the base station decoder receives the codeword fed back by the UE and recovers the pre-decoder information, the padded portion may be discarded by the gNB according to the specified procedure. The specified procedure may combine the dimension size of the input (D_1, D_2, D_3 for dimension 1 size, dimension 2 size, dimension 3 size, etc.) and the dimension size expected by the neural network (D_1, D_2, D_3 for dimension 1 size, dimension 2 size, dimension 3 size, etc.) ) as input parameter.
[0243] In some embodiments, due to the large number of subbands N_3, time units N_4, or spatial beams N_B, one of the dimensions of the connected image or the dimensions of the page may exceed the capacity of the neural network design. In these embodiments, segmentation may be performed. or The D_1×D_2 or D_1×D_2×D_3 input of the designed neural network can be split into more than one segment, and if Both padding and splitting can be used, for j, j≤1, or j≤2 for 2D NN, or j≤3 for 3D NN. Taking the first dimension as an example, if Padding (with zeros or interpolation or any other method) may be applied first, so the resulting first dimension has size Padding can be applied to the beginning, end, or both sides (aim for almost the same number of padding elements at the beginning as at the end). For example, And r 1,1 = ceil(r1 / 2), and r 1,2 =floor(r1 / 2), then r 1,1 Elements can be filled at the beginning, and r 1,2 Elements can be filled at the end, or r 1,2 Elements can be filled at the beginning, and r 1,1 Elements can be filled at the end. In one example, D1=9, Then r 1,1 =4, r 1,3 =3.
[0244] You can also get sections and allocate approximately the same number of elements in each section, and then apply padding to each section if necessary.
[0245] Furthermore, when the D1 indices in dimension 1 are divided into multiple parts, overlap can be allowed. Overlapping can provide a way to avoid filling with zeros or other values (such as those interpolated from the input to the NN). In one example, D1 = 24, and Then index 1, ..., 16 go to part 1, and index 9, ..., 24 go to part 2. It is possible to support overlap of each part with its neighbors, or use overlap only for some of them, e.g. only the first and second part have overlap, or only the last and second to last part have overlap.
[0246] Fig.24An operational flow / algorithm structure 2400 according to some embodiments is illustrated. The operational flow / algorithm structure 2400 may be implemented by a UE (such as UE 1404 or UE 2500) or a component thereof (eg, baseband processor 2504A).
[0247] At 2402, the operational flow / algorithm structure 2400 may include performing measurements of CSI measurement resources. The base station may provide an indication of the CSI measurement resources for measurement to the UE via a measurement configuration. The base station may then send a CSI reference signal (CSI-RS) on the measurement resource, and the UE may perform measurements such as layer 1 (L1) reference signal received power (RSRP) measurements on the CSI-RS.
[0248] At 2404, the operational flow / algorithm structure 2400 may also include generating a first page and a second page of channel response coefficients. The pages may be generated as described herein. The pages may be generated using a dimensionality reduction operation (e.g., selection of a spatial beam, FD component, or TD component) or a non-dimensionality reduction operation.
[0249] At 2406, the operational flow / algorithm structure 2400 may also include providing the first page and the second page to one or more neural networks to obtain one or more codewords. In some embodiments, the pages may be provided to the respective neural networks. In other embodiments, the pages may be combined / connected to an input matrix provided to a neural network. Combining / connecting the pages to the input matrix may be performed as described elsewhere herein to facilitate automatic encoding operations of the neural network.
[0250] The operational flow / algorithm structure 2400 may also include generating a CSI report having one or more codewords at 2406. The CSI report may be sent to a base station as channel feedback.
[0251] Fig.25 An exemplary UE 2500 according to some embodiments is illustrated. UE 2500 may be any mobile or non-mobile computing device, such as a mobile phone, a computer, a tablet, an industrial wireless sensor (e.g., a microphone, a carbon dioxide sensor, a pressure sensor, a humidity sensor, a thermometer, a motion sensor, an accelerometer, a laser scanner, a fluid level sensor, an inventory sensor, a voltage / current meter, an actuator, etc.), a video surveillance / monitoring device (e.g., a camera, a video camera, etc.), a wearable device (e.g., a smart watch), a loose IoT device. In some embodiments, UE 2500 may be a RedCap UE or an NR-Light UE.
[0252] UE 2500 may include a processor 2504, an RF interface circuit 2508, a memory / storage 2512, a user interface 2516, a sensor 2520, a driver circuit 2522, a power management integrated circuit (PMIC) 2524, antenna structures 2526, and a battery 2528. The components of UE 2500 may be implemented as an integrated circuit (IC), a portion of an integrated circuit, a discrete electronic device or other module, logical component, hardware, software, firmware, or a combination thereof. Fig.25 The block diagram of UE 2500 is intended to show a high-level view of some of the components of UE 2500. However, some of the components shown may be omitted, additional components may be present, and different arrangements of the components shown may occur in other implementations.
[0253] Components of UE 2500 may be coupled to various other components via one or more interconnects 2532, which may represent any type of interface, input / output, bus (local, system, or extension), transmission line, trace, optical connection, etc., which allows various circuit components (on a common or different chip or chipset) to interact with each other.
[0254] The processor 2504 may include processor circuits such as a baseband processor circuit (BB) 2504A, a central processor unit circuit (CPU) 2504B, and a graphics processor unit circuit (GPU) 2504C. The processor 2504 may include any type of circuit or processor circuit that executes or otherwise operates computer-executable instructions (such as program codes, software modules, or functional processes from the memory / storage device 2512) to cause the UE 2500 to perform operations as described herein.
[0255] In some embodiments, the baseband processor circuit 2504A can access the communication protocol stack 2536 in the memory / storage device 2512 to communicate through a 3GPP compatible network. In general, the baseband processor circuit 2504A can access the communication protocol stack to: perform user plane functions at the PHY layer, MAC layer, RLC layer, PDCP layer, SDAP layer, and PDU layer; and perform control plane functions at the PHY layer, MAC layer, RLC layer, PDCP layer, RRC layer, and non-access layer. In some embodiments, the PHY layer operations can be performed in addition / alternatively by components of the RF interface circuit 2508.
[0256] The baseband processor circuit 2504A may generate or process baseband signals or waveforms that carry information in a 3GPP-compatible network. In some embodiments, the waveforms used for NR may be based on cyclic prefix OFDM (CP-OFDM) in the uplink or downlink, and discrete Fourier transform spread OFDM (DFT-S-OFDM) in the uplink.
[0257] The memory / storage 2512 may include one or more non-transitory computer-readable media including instructions (e.g., the communication protocol stack 2536) that may be executed by one or more processors in the processor 2504 to enable the UE 2500 to perform various operations described herein. The memory / storage 2512 includes any type of volatile or non-volatile memory that may be distributed throughout the UE 2500. In some embodiments, some of the memory / storage 2512 may be located on the processor 2504 itself (e.g., L1 cache and L2 cache), while other memory / storage 2512 is located outside the processor 2504 but is accessible via a memory interface. The memory / storage 2512 may include any suitable volatile or non-volatile memory, such as, but not limited to, dynamic random access memory (DRAM), static random access memory (SRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, solid-state memory, or any other type of memory device technology.
[0258] The RF interface circuit 2508 may include a transceiver circuit and a radio frequency front end module (RFEM), which allows the UE 2500 to communicate with other devices through a radio access network. The RF interface circuit 2508 may include various elements arranged in a transmit path or a receive path. These elements may include, for example, switches, mixers, amplifiers, filters, synthesizer circuits, control circuits, etc.
[0259] In the receive path, the RFEM may receive the radiated signal from the air interface via the antenna structure 2526 and proceed to filter and amplify the signal (using a low noise amplifier). The signal may be provided to the transceiver's receiver, which converts the RF signal down to a baseband signal that is provided to the baseband processor of the processor 2504.
[0260] In the transmit path, the transmitter of the transceiver up-converts the baseband signal received from the baseband processor and provides the RF signal to the RFEM. The RFEM may amplify the RF signal through a power amplifier before the signal is radiated across the air interface via the antenna 2526.
[0261] In various embodiments, the RF interface circuit 2508 may be configured to send / receive signals in a manner compatible with NR access technology.
[0262] Antenna 2526 may include antenna elements to convert electrical signals into radio waves to travel through the air and convert received radio waves into electrical signals. These antenna elements may be arranged into one or more antenna panels. Antenna 2526 may have an antenna panel that is omnidirectional, directional, or a combination thereof to achieve beamforming and multiple input / multiple output communications. Antenna 2526 may include a microstrip antenna, a printed antenna manufactured on the surface of one or more printed circuit boards, a patch antenna, a phased array antenna, etc. Antenna 2526 may have one or more panels that are designed for a specific frequency band included in a band in FR1 or FR2.
[0263] In some embodiments, UE 2500 may include beamforming circuit 1700 ( Fig.17 ), where the beamforming circuit 1700 may be used to communicate with the UE 2500. In some embodiments, components and beamforming circuits of the UE 2500 may be shared. For example, the antenna 2526 of the UE may include the panel 1 1704 and the panel 2 1708 of the beamforming circuit 1700.
[0264] The user interface circuit 2516 includes various input / output (I / O) devices designed to enable a user to interact with the UE 2500. The user interface circuit 2516 includes input device circuits and output device circuits. The input device circuit includes any physical or virtual components for accepting input, including in particular one or more physical or virtual buttons (e.g., reset button), physical keyboard, keypad, mouse, touch pad, touch screen, microphone, scanner, headset, etc. The output device circuit includes any physical or virtual components for displaying information or otherwise conveying information (such as sensor readings, actuator positions, or other similar information). The output device circuit may include any number or combination of audio or visual displays, including in particular one or more simple visual outputs / indicators (e.g., binary state indicators, such as light emitting diodes "LEDs") and multi-character visual outputs, or more complex outputs, such as display devices or touch screens (e.g., liquid crystal displays (LCDs), LED displays, quantum dot displays, projectors, etc.), where the output of characters, graphics, multimedia objects, etc. is generated or produced by the operation of the UE 2500.
[0265] Sensor 2520 may include a device, module, or subsystem that is designed to detect events or changes in its environment and send information about the detected events (sensor data) to some other device, module, subsystem, etc. Examples of such sensors include, among others: an inertial measurement unit including an accelerometer, gyroscope, or magnetometer; a microelectromechanical system or nanoelectromechanical system including a three-axis accelerometer, three-axis gyroscope, or magnetometer; a liquid level sensor; a flow sensor; a temperature sensor (e.g., a thermistor); a pressure sensor; a barometric pressure sensor; a gravity meter; an altimeter; an image capture device (e.g., a camera or a lensless aperture); a light detection and ranging sensor; a proximity sensor (e.g., an infrared radiation detector, etc.); a depth sensor; an ambient light sensor; an ultrasonic transceiver; a microphone or other similar audio capture device; and the like.
[0266] The driver circuit 2522 may include software and hardware elements for controlling specific devices embedded in, attached to, or otherwise communicatively coupled to the UE 2500. The driver circuit 2522 may include various drivers to allow other components to interact with or control various input / output (I / O) devices that may be present in or connected to the UE 2500. For example, the driver circuit 2522 may include: a display driver for controlling and allowing access to a display device, a touch screen driver for controlling and allowing access to a touch screen interface, a sensor driver for obtaining sensor readings of the sensor circuit 2520 and controlling and allowing access to the sensor circuit 2520, a driver for obtaining an actuator position of an electromechanical component or controlling and allowing access to an electromechanical component, a camera driver for controlling and allowing access to an embedded image capture device, and an audio driver for controlling and allowing access to one or more audio devices.
[0267] The PMIC 2524 may manage the power provided to various components of the UE 2500. Specifically, with respect to the processor 2504, the PMIC 2524 may control power selection, voltage scaling, battery charging, or DC-DC conversion.
[0268] In some embodiments, the PMIC 2524 may control or otherwise be part of various power saving mechanisms of the UE 2500. For example, if the platform UE is in the RRC_Connected state, in which the platform is still connected to the RAN node because it expects to receive traffic soon, then after a period of inactivity, the platform may enter a state known as discontinuous reception mode (DRX). During this state, the UE 2500 may be powered off for short time intervals, thereby saving power. If there is no data traffic activity for an extended period of time, the UE 2500 may transition to the RRC_Idle state, in which it is disconnected from the network and does not perform operations such as channel quality feedback, handover, etc. The UE 2500 enters a very low power state and performs paging, in which it wakes up again periodically to listen to the network, and then powers off again. The UE 2500 may not receive data in this state; in order to receive data, the platform must transition back to the RRC_Connected state. Additional power saving modes may prevent the device from using the network for a time exceeding the paging interval (from a few seconds to a few hours). During this time, the device is completely unable to connect to the network and can be completely powered off. Any data sent during this time will incur significant latency, and it is assumed that latency is acceptable.
[0269] The battery 2528 can power the UE 2500, but in some examples, the UE 2500 can be installed in a fixed location and can have a power source coupled to the power grid. The battery 2528 can be a lithium-ion battery, a metal-air battery such as a zinc-air battery, an aluminum-air battery, a lithium-air battery, etc. In some specific implementations, such as in vehicle-based applications, the battery 2528 can be a typical lead-acid automobile battery.
[0270] Fig.26 An example gNB 2600 is illustrated in accordance with some embodiments. The gNB 2600 may include a processor 2604, RF interface circuitry 2608, core network (CN) interface circuitry 2612, memory / storage circuitry 2616, and antenna structures 2626.
[0271] Components of the gNB 2600 may be coupled to various other components via one or more interconnects 2628.
[0272] The processor 2604, RF interface circuit 2608, memory / storage circuit 2616 (including communication protocol stack 2610), antenna structure 2626 and interconnect 2628 may be similar to those described in detail in the accompanying drawings. Fig.25 Like-named elements are shown and described.
[0273] The CN interface circuit 2612 may provide connectivity for a core network (e.g., a 5th Generation Core Network (5GC) using a 5th Generation Core Network (5GC) compatible network interface protocol such as a Carrier Ethernet protocol or some other suitable protocol). Network connectivity may be provided to / from the gNB 2600 via optical fiber or wireless backhaul. The CN interface circuit 2612 may include one or more dedicated processors or FPGAs for communicating using one or more of the aforementioned protocols. In some implementations, the CN interface circuit 2612 may include multiple controllers for providing connectivity to other networks using the same or different protocols.
[0274] It is understood that the use of personally identifiable information should be subject to privacy policies and practices that are generally recognized to meet or exceed industry or government requirements for maintaining user privacy. Specifically, personally identifiable information data should be managed and processed to minimize the risk of unintentional or unauthorized access or use, and the nature of the authorized use should be clearly stated to users.
[0275] For one or more embodiments, at least one of the components shown in one or more of the foregoing figures may be configured to perform one or more operations, techniques, processes, or methods described in the following example section. For example, the baseband circuit described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more of the examples described below. For another example, the circuits associated with the UE, base station, network element, etc. described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more of the examples shown in the example section below.
[0276] Example
[0277] In the following sections, additional exemplary embodiments are provided.
[0278] Embodiment 1 includes a method, the method comprising: performing measurement of channel state information (CSI) measurement resources; based on the performance of the measurement, generating a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein the first page has a dimension of a first number of spatial beams multiplied by a second number of frequency domain (FD) components, and the second page has a dimension of a third number of spatial beams multiplied by a fourth number of FD components; providing the first page and the second page to one or more neural networks to obtain one or more codewords; and generating a CSI report having the one or more codewords to send to a base station.
[0279] Embodiment 2 includes the method according to embodiment 1 or some other embodiments herein, the method further comprising: providing the first page to a first neural network among the one or more neural networks; and providing the second page to a second neural network among the one or more neural networks, wherein the first neural network outputs a first codeword among the one or more codewords and the second neural network outputs a second codeword among the one or more codewords.
[0280] Embodiment 3 includes the method of embodiment 2 or some other embodiments herein, wherein the first codeword is associated with a first amplitude and the second codeword is associated with a second amplitude, wherein the second amplitude is defined relative to the first amplitude.
[0281] Embodiment 4 includes the method of embodiment 3 or some other embodiments herein, wherein the first codeword is at a first bit width and the second codeword is at a second bit width, wherein the first bit width is greater than the second bit width.
[0282] Embodiment 5 includes the method according to embodiment 1 or some other embodiments herein, the method further comprising: connecting the first page and the second page to generate an input matrix; and providing the input matrix to a neural network in the one or more neural networks.
[0283] Embodiment 6 includes a method according to embodiment 5 or some other embodiments of the present invention, wherein connecting the first page and the second page includes: filling the CSI coefficients from the first page and the second page based on the following order: spatial beam, FD component and TD component; or spatial beam, TD component and FD component.
[0284] Embodiment 7 includes the method of embodiment 5 or some other embodiments herein, wherein the first quantity is equal to the third quantity and the second quantity is equal to the fourth quantity.
[0285] Embodiment 8 includes the method according to embodiment 7 or some other embodiments herein, wherein the first number of spatial beams and the third number of spatial beams include a common plurality of spatial beam indices or include separate plurality of spatial beam indices.
[0286] Embodiment 9 includes the method of embodiment 7 or some other embodiments herein, wherein the second number of FD components and the fourth number of FD components comprise a common plurality of FD component indices or comprise separate plurality of FD component indices.
[0287] Embodiment 10 includes the method according to embodiment 5 or some other embodiments herein, wherein connecting the first page and the second page includes: combining the first page and the second page along a spatial beam dimension; or combining the first page and the second page along an FD component dimension.
[0288] Embodiment 11 includes the method according to embodiment 1 or some other embodiments of the present invention, and the method also includes: determining a common spatial beam set for multiple pages, the multiple pages including the first page and the second page; generating the first page by selecting a spatial beam to be included in the first number of spatial beams from the common spatial beam set; and generating the second page by selecting a spatial beam to be included in the third number of spatial beams from the common spatial beam set.
[0289] Embodiment 12 includes a method according to embodiment 11 or some other embodiments of the present invention, wherein: the common spatial beam set is a first common spatial beam set for a first polarization, and the method also includes determining a second common spatial beam set for a second polarization; or the common spatial beam set is used for both the first polarization and the second polarization.
[0290] Embodiment 13 includes the method according to embodiment 1 or some other embodiments of the present invention, the method further comprising: determining a common FD component set for multiple pages, the multiple pages including the first page and the second page; generating the first page by selecting FD components to be included in the second number of FD components from the common FD component set; and generating the second page by selecting FD components to be included in the fourth number of FD components from the common FD component beam set.
[0291] Embodiment 14 includes a method according to embodiment 13 or some other embodiments herein, wherein: the common FD component set is a first common FD component set for a first polarization, and the method further includes determining a second common FD component set for a second polarization; or the common FD component set is used for both the first polarization and the second polarization.
[0292] Embodiment 15 includes the method according to embodiment 1 or some other embodiments of the present invention, and the method further includes: determining a page configuration for the first quantity, the second quantity, the third quantity or the fourth quantity, wherein the page configuration is received from a base station or provided to the base station.
[0293] Embodiment 16 includes a method, the method comprising: performing measurement of channel state information (CSI) measurement resources; based on the performance of the measurement, generating a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein the first page has a dimension of a first number of spatial beams multiplied by a second number of frequency domain (FD) components, and the second page has a dimension of a third number of spatial beams multiplied by a fourth number of FD components; generating a third page of channel response coefficients for the first TD component by adding coefficients corresponding to one or more spatial beams or one or more FD components to the first page; providing the third page and the second page to one or more neural networks to obtain one or more codewords; and generating a CSI report having the one or more codewords to send to a base station.
[0294] Embodiment 17 includes a method according to embodiment 16 or some other embodiments herein, wherein the first page includes a first FD component associated with a first value and a second FD component associated with a second value, the first FD component being adjacent to the second FD component, and the method further includes: generating the third page to include a third FD component between the first FD component and the second FD component, wherein the third FD component is associated with a third value between the first value and the second value.
[0295] Embodiment 18 includes the method according to embodiment 16 or some other embodiments herein, the method further comprising: generating the third page to include dimensions that match the dimensions of the second page.
[0296] Embodiment 19 includes a method comprising: receiving one or more codewords in a channel station information (CSI) report; providing the one or more codewords to one or more neural networks to obtain a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein the first page has a dimension of a first number of spatial beams multiplied by a second number of frequency domain (FD) components, and the second page has a dimension of a third number of spatial beams multiplied by a fourth number of frequency domain components; and determining a measurement of a CSI measurement resource based on the first page and the second page.
[0297] Embodiment 20 includes the method according to embodiment 19 or some other embodiments herein, wherein the method further comprises: providing a first codeword of the one or more codewords to a first neural network of the one or more neural networks to obtain the first page; and providing a second codeword of the one or more codewords to a second neural network of the one or more neural networks to obtain the second page.
[0298] Embodiment 21 includes the method according to embodiment 19 or some other embodiments herein, wherein the method further comprises: providing a first codeword of the one or more codewords to a first neural network of the one or more neural networks to obtain an output matrix; and deconstructing the output matrix to obtain the first page and the second page.
[0299] Example 22 may include an apparatus comprising means for performing one or more elements of the method described in or related to any of Examples 1 to 21, or any other method or process described herein.
[0300] Embodiment 23 may include one or more non-transitory computer-readable media, which include instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform one or more elements of a method described in or related to any one of Embodiments 1 to 21 or any other method or process described herein.
[0301] Embodiment 24 may include a device comprising logic components, modules, or circuits for performing one or more elements of a method described in or related to any of Embodiments 1 to 21, or any other method or process described herein.
[0302] Embodiment 25 may include methods, techniques or processes described in or related to any one of Embodiments 1 to 21, or parts or components thereof.
[0303] Embodiment 26 may include a device comprising: one or more processors and one or more computer-readable media, wherein the one or more computer-readable media include instructions that, when executed by the one or more processors, cause the one or more processors to perform a method, technique, or process, or a portion thereof, as described or related to any one of Embodiments 1 to 21.
[0304] Embodiment 27 may include a signal as described or related to any one of Embodiments 1 to 21, or a portion or component thereof.
[0305] Embodiment 28 may include a datagram, information element, packet, frame, segment, PDU or message, or a portion or component thereof, as described or associated with any one of Embodiments 1 to 21, or otherwise described in the present disclosure.
[0306] Embodiment 29 may include a signal encoded with data as described or associated with any one of Embodiments 1 to 21, or a portion or component thereof, or otherwise described in this disclosure.
[0307] Embodiment 30 may include a signal encoded with a datagram, IE, packet, frame, segment, PDU or message as described or associated with any one of Embodiments 1 to 21, or a portion or component thereof, or otherwise described in the present disclosure.
[0308] Embodiment 31 may include an electromagnetic signal carrying computer-readable instructions, wherein execution of the computer-readable instructions by one or more processors will cause the one or more processors to perform a method, technique, or process, or portion thereof, described in or related to any one of Embodiments 1 to 21.
[0309] Embodiment 32 may include a computer program comprising instructions, wherein execution of the program by a processing element causes the processing element to perform a method, technique, or process, or a portion thereof, as described or related to any one of Embodiments 1 to 21.
[0310] Embodiment 33 may include signals in a wireless network as shown and described herein.
[0311] Embodiment 34 may include a method of communicating in a wireless network as shown and described herein.
[0312] Embodiment 35 may include a system for providing wireless communications as shown and described herein.
[0313] Embodiment 36 may include an apparatus for providing wireless communications as shown and described herein.
[0314] The foregoing description of one or more specific implementations provides illustration and description, but is not intended to be exhaustive or to limit the scope of the embodiments to the precise forms disclosed.Modifications and variations are possible in light of the above teachings or may be acquired from practice of various embodiments.
[0315] Although the above embodiments have been described in considerable detail, numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to include all such variations and modifications.
Claims
1. A method, comprising: performing measurement of a channel state information (CSI) measurement resource; generating, based on said performing of said measuring, a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein said first page has dimensions of a first number of spatial beams by a second number of frequency domain (FD) components and said second page has dimensions of a third number of spatial beams by a fourth number of FD components; providing the first page and the second page to one or more neural networks to obtain one or more codewords; and A CSI report having the one or more codewords is generated for transmission to a base station.
2. The method according to claim 1, further comprising: providing the first page to a first neural network of the one or more neural networks; as well as The second page is provided to a second neural network of the one or more neural networks, wherein the first neural network outputs a first codeword of the one or more codewords and the second neural network outputs a second codeword of the one or more codewords. 3 . The method of claim 2 , wherein the first codeword is associated with a first amplitude and the second codeword is associated with a second amplitude, wherein the second amplitude is defined relative to the first amplitude. 4 . The method of claim 3 , wherein the first codeword is at a first bit width and the second codeword is at a second bit width, wherein the first bit width is greater than the second bit width.
5. The method according to claim 1, further comprising: concatenating the first page and the second page to generate an input matrix; as well as The input matrix is provided to a neural network of the one or more neural networks.
6. The method of claim 5, wherein connecting the first page and the second page comprises: populating CSI coefficients from the first page and the second page based on the following order: spatial beam, FD component, and TD component; Or spatial beam, TD component and FD component. The method of claim 5 , wherein the first amount is equal to the third amount and the second amount is equal to the fourth amount. 8 . The method of claim 7 , wherein the first number of spatial beams and the third number of spatial beams comprise a common plurality of spatial beam indices or comprise separate plurality of spatial beam indices. 9 . The method of claim 7 , wherein the second number of FD components and the fourth number of FD components include a common plurality of FD component indices or include separate plurality of FD component indices.
10. The method of claim 5, wherein connecting the first page and the second page comprises: combining the first page and the second page along a spatial beam dimension; or The first page and the second page are combined along an FD component dimension.
11. The method according to claim 1, further comprising: determining a common set of spatial beams for a plurality of pages, the plurality of pages including the first page and the second page; generating the first page by selecting a spatial beam to be included in the first number of spatial beams from the common set of spatial beams; as well as The second page is generated by selecting spatial beams to be included in the third number of spatial beams from the common set of spatial beams.
12. The method according to claim 11, wherein: The common spatial beam set is a first common spatial beam set for a first polarization, and the method further comprises determining a second common spatial beam set for a second polarization; or The common set of spatial beams is used for both the first polarization and the second polarization.
13. The method according to claim 1, further comprising: determining a common set of FD components for a plurality of pages, the plurality of pages including the first page and the second page; generating the first page by selecting FD components to be included in the second number of FD components from the common FD component set; as well as The second page is generated by selecting FD components to be included in the fourth number of FD components from the common FD component beam set.
14. The method according to claim 13, wherein: The set of common FD components is a first set of common FD components for a first polarization, and the method further comprises determining a second set of common FD components for a second polarization; or The common set of FD components is used for both the first polarization and the second polarization.
15. The method according to claim 1, further comprising: determining a page configuration for the first number, the second number, the third number, or the fourth number, The page configuration is received from a base station or provided to the base station.
16. A method comprising: performing measurement of a channel state information (CSI) measurement resource; generating, based on said performing of said measuring, a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein said first page has dimensions of a first number of spatial beams by a second number of frequency domain (FD) components and said second page has dimensions of a third number of spatial beams by a fourth number of FD components; generating a third page of channel response coefficients for the first TD component by adding coefficients corresponding to one or more spatial beams or one or more FD components to the first page; providing the third page and the second page to one or more neural networks to obtain one or more codewords; as well as A CSI report having the one or more codewords is generated for transmission to a base station.
17. The method of claim 16, wherein the first page comprises a first FD component associated with a first value and a second FD component associated with a second value, the first FD component being adjacent to the second FD component, and the method further comprises: The third page is generated to include a third FD component between the first FD component and the second FD component, wherein the third FD component is associated with a third value between the first value and the second value.
18. The method according to claim 16, further comprising: The third page is generated to include dimensions matching the dimensions of the second page.
19. A method comprising: receiving one or more codewords in a channel station information (CSI) report; providing the one or more codewords to one or more neural networks to obtain a first page of channel response coefficients for a first time domain (TD) component and a second page of CSI coefficients for a second TD component, wherein the first page has dimensions of a first number of spatial beams by a second number of frequency domain (FD) components and the second page has dimensions of a third number of spatial beams by a fourth number of frequency domain components; and Measurement of CSI measurement resources is determined based on the first page and the second page.
20. The method according to claim 19, further comprising: providing a first codeword of the one or more codewords to a first neural network of the one or more neural networks to obtain the first page; as well as A second codeword of the one or more codewords is provided to a second neural network of the one or more neural networks to obtain the second page.
