Neural network architecture for CSI feedback

Abstract

Methods and apparatus are provided for machine learning (ML)-based channel state information (CSI) feedback from a user equipment (UE) to a wireless network. For one or more neural network (NN) model types, the UE transmits UE capability signaling to the wireless network to report one or more 3D or 4D input data formats to a CSI feedback encoder. The UE processes NN model configuration information from the wireless network for the one or more NN model types. The UE then generates CSI feedback by providing downlink (DL) channel data or a DL pre-decoder formatted according to the one or more 3D or 4D input formats to the NN model corresponding to the NN model configuration information. The UE then transmits the CSI feedback to the wireless network.

Description

Technical Field

[0001] The present application generally relates to wireless communication systems, including channel state information (CSI) feedback. Background Art

[0002] Wireless mobile communication technologies use various standards and protocols to send data between base stations and wireless communication devices. Wireless communication system standards and protocols may include, for example, the 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) (e.g., 4G), 3GPP New Radio (NR) (e.g., 5G), and the IEEE 802.11 standard for wireless local area networks (WLANs) (commonly referred to within industry organizations as WLANs). ).

[0003] As envisioned by 3GPP, different wireless communication system standards and protocols may use various radio access networks (RANs) to communicate between base stations of the RAN (which may also sometimes be referred to as RAN nodes, network nodes, or simply nodes) and wireless communication devices referred to as user equipment (UE). 3GPP RANs may include, for example, Global System for Mobile Communications (GSM), Enhanced Data Rates for GSM Evolution (EDGE) RAN (GERAN), Universal Terrestrial Radio Access Network (UTRAN), Evolved Universal Terrestrial Radio Access Network (E-UTRAN), and / or Next Generation Radio Access Network (NG-RAN).

[0004] Each RAN may use one or more radio access technologies (RATs) to perform communications between base stations and UEs. For example, GERAN implements GSM and / or EDGE RAT, UTRAN implements Universal Mobile Telecommunications System (UMTS) RAT or other 3GPP RAT, E-UTRAN implements LTE RAT (sometimes referred to as LTE), and NG-RAN implements NR RAT (sometimes referred to herein as 5G RAT, 5G NR RAT, or NR). In some deployments, E-UTRAN may also implement NR RAT. In some deployments, NG-RAN may also implement LTE RAT.

[0005] The base stations used by the RAN may correspond to the RAN. An example of an E-UTRAN base station is an Evolved Universal Terrestrial Radio Access Network (E-UTRAN) Node B (also commonly denoted as an evolved Node B, enhanced Node B, eNodeB, or eNB). An example of an NG-RAN base station is a Next Generation Node B (sometimes also referred to as a g-Node B or gNB).

[0006] The RAN provides communication services together with external entities through its connection with the Core Network (CN). For example, E-UTRAN may utilize the Evolved Packet Core (EPC) and NG-RAN may utilize the 5G Core Network (5GC). BRIEF DESCRIPTION OF THE DRAWINGS

[0007] To easily identify the discussion of any particular element or action, the most significant digit(s) in a reference number refers to the figure numeral that first introduces that element.

[0008] Figure 1 An encoder and a decoder in CSI feedback operation according to certain embodiments are illustrated.

[0009] Figure 2 Antenna configurations that may be used, for example, at a base station are illustrated according to certain embodiments.

[0010] Figure 3 Different dimensions of input data for various neural network model types are illustrated according to certain embodiments.

[0011] Figure 4 is a block diagram illustrating an example 3D convolutional neural network autoencoder that may be used in accordance with certain embodiments.

[0012] Figure 5 is a flow chart illustrating a method for machine learning based CSI feedback according to certain embodiments.

[0013] Figure 6 is a flow chart of a method for a UE to provide machine learning based CSI to a wireless network according to one embodiment.

[0014] Figure 7 It is a flowchart of a method for a base station to configure one or more neural network models for CSI feedback to a UE according to one embodiment.

[0015] Figure 8 An example architecture of a wireless communication system according to embodiments disclosed herein is illustrated.

[0016] Fig. 9 A system for performing signaling between a wireless device and a network device according to embodiments disclosed herein is illustrated. DETAILED DESCRIPTION

[0017] Various embodiments are described with respect to UE. However, reference to UE is provided for illustrative purposes only. The example embodiments may be used with any electronic component that can establish a connection with a network and is configured with hardware, software, and / or firmware for exchanging information and data with the network. Therefore, UE as described herein is used to represent any suitable electronic component.

[0018] Downlink channel state information (CSI) may be transmitted from the UE to the base station via a feedback channel (e.g., for frequency division duplex (FDD) operation). The base station may use CSI feedback to, for example, reduce interference and increase throughput of massive multiple-input multiple-output (MIMO) communications. However, such feedback uses excessive overhead. Vector quantization or codebook-based feedback may be used to reduce feedback overhead. However, the amount of feedback generated by these methods scales linearly with the number of transmit antennas, which may be difficult when using hundreds or thousands of centralized or distributed transmit antennas.

[0019] Artificial intelligence (AI) and / or machine learning (ML) may be used for CSI feedback enhancement to reduce overhead, improve accuracy, and / or generate predictions. AI and / or ML may also be used, for example, to perform beam management (e.g., beam prediction in the time / space domain to reduce overhead and latency and improve beam selection accuracy) and / or improve positioning accuracy.

[0020] CSI feedback using AI and / or ML can be formulated as a joint optimization of the encoder and decoder. See, for example, Chao-Kai Wen, Wan-Ting Shih, and Shi Jin, "Deep Learning for Massive MIMO CSI Feedback," IEEE Wireless Communications Letters, Volume 7, Issue 5, October 2018. Since this early paper by Chao-Kai Wen et al., autoencoders and many variants have been considered. Given the latest wave of applications of ML in image processing / video processing, image processing / video processing techniques have been used for CSI compression, which may be a natural choice. In addition, when formulated in a suitable domain, CSI feedback has similarities with image / video streams.

[0021] At high levels, Figure 1An encoder 102 of a UE and a decoder 104 of a base station (e.g., a gNB) in an AI-based CSI feedback operation according to certain embodiments are illustrated. As shown, the encoder 102 receives a downlink (DL) channel H and outputs an AI-based CSI feedback. The encoder 102 learns the transformation from the original transformation matrix to the compressed representation (codeword) through training data. The decoder 104 learns the inverse transformation from the codeword to the original channel. Therefore, the decoder 104 may receive AI-based CSI feedback (codeword) from the encoder 102 and output the reconstructed channel H. End-to-end learning (e.g., using an unsupervised learning algorithm) may be used to train the encoder 102 and the decoder 104. Typically, the normalized mean square error (NMSE) or cosine similarity is an optimization metric. In some designs, the DL channel H may be replaced by a DL precoder. Therefore, the encoder 102 takes the DL precoder as input and generates AI-based CSI feedback, and the decoder 104 takes the AI-based CSI feedback and reconstructs the DL precoder.

[0022] Various types of neural network (NN) encoders / decoders can be trained for different purposes with different tradeoffs in complexity, overhead, and performance. Convolutional neural networks (CNNs) can be used, for example, for CSI feedback for frequency and spatial domain CSI reference signals (CSI-RS) compression. Other examples include the use of transformers or generative adversarial networks (GANs). Depending on the number and rank of receive antennas, channel feedback or precoding matrix indicator (PMI) feedback can be used. For example, for four receive antennas, rank 1 and rank 2 feedback can potentially use AI NNs trained with eigenvectors as input, while rank 3 and rank 4 can potentially use channel state information as input to the trained AI NN. Data preprocessing can be used for the input to the AI ​​model. Preprocessing from the frequency domain to the time domain can be used, and some of the small paths in the small paths can be removed before input to the AI ​​NN. The maximum rank indicates the maximum number of layers per UE, which corresponds to the lack of correlation or interference between the antennas of the UE. For example, rank 1 corresponds to a maximum of one spatial layer for the UE, rank 2 corresponds to a maximum of two spatial layers for the UE, rank 3 corresponds to a maximum of three layers for the UE, and rank 4 corresponds to a maximum of four layers for the UE.

[0023] The input to the encoder 102 may include data without angle domain processing, which may be referred to as spatial domain projection (e.g., from 32 transmit ports to 8 spatial beams). In addition, the input to the encoder 102 may include data without frequency domain (FD) processing (e.g., omitting the conversion from N3 subbands to M taps or FD components). There are four combinations ({with angle domain processing, with frequency domain processing}, {with angle domain processing, without frequency domain processing}, {without angle domain processing, with frequency domain processing}, and {without angle domain processing, without frequency domain processing}) with or without angle domain processing and with or without frequency domain processing. For different combinations, a two-dimensional (2D) matrix is ​​fed to the encoder 102 for a complex-valued NN, or two 2D matrices (the real and imaginary parts of the complex 2D matrix) are fed to the encoder for a real-valued NN.

[0024] Example use cases include space-frequency domain compression (which uses 3GPP Release (Rel)-16 or Rel-17 Type II codebooks for comparison) and time-space-frequency domain compression or prediction. For time-space-frequency domain compression or prediction, if there are multiple channel matrices or precoding matrices at times t1, t2, t3, t4, ..., the autoencoder can be modified accordingly. For example, if the input data of the complex-valued NN is of dimension N4×N tx × N3 (e.g., N4 = 8 time slots, N tx =32 transmit ports, and N3 =9 subbands), the dimensionality of the decoder or relevant parts of the decoder / encoder can be increased (e.g., instead of 2D convolutions, three-dimensional (3D) convolutions are performed).

[0025] Figure 2 An antenna configuration 200 that may be used, for example, at a base station according to certain embodiments is illustrated. The example antenna configuration 200 includes antennas with a first polarization (shown in solid lines) and antennas with a second polarization (shown in dashed lines) arranged in two rows and four columns. A skilled person will recognize from the disclosure herein that other antenna configurations with different numbers of rows, columns, and / or polarizations may also be used. Typically, the number of antenna ports is determined by the number of rows, the number of columns, and the number of polarizations. For AI and / or ML based methods, the step for spatial basis selection may be omitted in certain designs (e.g., thereby omitting spatial beam projection), and a more refined representation is possible.

[0026] Some designs (e.g., the paper by Wen et al. discussed above) do not utilize 2D antenna structures, such as Figure 2The antenna configuration 200 shown in the figure uses vectorization instead, where a one-dimensional (1D) vector is provided to the input of the ML encoder. However, the embodiments disclosed herein provide a NN architecture for CSI feedback based on antenna structure and configuration parameters to provide space-frequency domain compression, time-space-frequency domain compression, and time-space-frequency domain prediction. The compression implementation scheme (i.e., space-frequency domain compression or time-space-frequency domain compression) provides feedback with lower overhead. The prediction implementation scheme mitigates the degradation caused by channel aging by proactively predicting what the actual channel state of the channel may be when used in the future based on the available observed channel.

[0027] The antenna structure and configuration parameters include the number of antenna columns N1 at the base station, the number of antenna rows N2 at the base station (or alternatively, the number of antenna columns N2 at the base station, the number of antenna rows N1 at the base station), two polarizations of the base station antenna (e.g., -45° and +45°, or horizontal and vertical), and the number of subbands N3 for CSI feedback. In some embodiments, for time-space-frequency domain compression and time-space-frequency domain prediction, the parameters also include the number of opportunities (e.g., time slots) N4 for generating N3 pre-decoders for N3 subbands, respectively. For space-frequency domain compression, for example, N4=1.

[0028] Input data dimensions for different NN model types

[0029] In some embodiments, the antenna structure and configuration parameters of the base station are used as input values ​​to the NN model at the encoder of the UE to generate CSI feedback. Different NN model types are configured to receive and process different dimensions or combinations of antenna structure and configuration parameters.

[0030] For example, Figure 3 Different dimensions of input data for various NN model types according to certain embodiments are illustrated. A first input data option 302 includes a 3D matrix or tensor corresponding to N1 antenna columns in a first dimension, N2 antenna rows in a second dimension, and N3 subbands in a third dimension, or corresponding to (N1 x N2 x N3). Input data arranged as shown for the first input data option 302 can be provided to the NN model for each channel (e.g., polarization). However, for some NN model types, the input data is concatenated along one of the dimensions (e.g., to input data for two polarizations simultaneously). For example, a second input data option 304 includes a 3D matrix or tensor corresponding to 2x N1 antenna columns, N2 antenna rows, and N3 subbands (2 . The third input data option 306 includes a 3D matrix or tensor corresponding to data stacked in the N1 dimension (N1 x N2 x N3). The third input data option 306 includes data corresponding to N1 antenna columns, 2 x N2 antenna rows, and N3 subbands (N1 x 2. The fourth input data option 308 includes a 3D matrix or tensor corresponding to data stacked in the N2 dimension (N2 x N3). The fourth input data option 308 includes data corresponding to N1 antenna columns, N2 antenna rows, and 2 x N3 subbands (N1 xN2 x 2 . N3) corresponds to a 3D matrix or tensor of data stacked in N3 dimensions. Figure 3 Not shown, but a fourth dimension of each of the illustrated input data options (eg, for time-space-frequency domain compression or prediction) may include a number of opportunities N4.

[0031] Space-frequency domain compression (complex-valued NN)

[0032] The complex-valued NN uses a single channel for both in-phase (I) and quadrature (Q) complex values. For space-frequency domain compression using a complex-valued NN, the data input to the first NN model type (referred to herein as space-frequency (SF) option 1 complex or SF-1c) corresponds to a first input data option 302 of N1 xN2 x N3 in the case of two channels (one channel for a first polarization and the other channel for a second polarization).

[0033] For space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the second NN model type (referred to herein as SF option 2 complex or SF-2c) corresponds to 2 . The second input data option 304 is N1 x N2 x N3.

[0034] For space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the third NN model type (referred to herein as SF option 3 complex or SF-3c) corresponds to N1 x 2 . A third input data option 306 of N2 x N3.

[0035] For space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the fourth NN model type (referred to herein as SF option 4 complex or SF-4c) corresponds to N1 x N2 x 2 . Fourth input data option 308 of N3. Since multipath in the time domain causes frequency selectivity in the frequency domain, the disclosed input formula is also applicable to space-delay domain compression, and in this setting, N3 means the number of delay taps of the wireless channel.

[0036] Space-frequency domain compression (real-valued NN)

[0037] The real-valued NN uses separate channels for I and Q (I / Q) values. For space-frequency domain compression using real-valued NN, the data input to the first NN model type (referred to herein as SF option 1 real or SF-1r) corresponds to a first input data option 302 of N1 x N2 x N3 with four channels (two channels (I / Q) for the first polarization and two channels (I / Q) for the second polarization).

[0038] For space-frequency domain compression using real-valued NN, in the case of two channels (I / Q) (each including two polarizations), the data input to the second NN model type (referred to herein as SF option 2 real or SF-2r) corresponds to 2 . The second input data option 304 is N1 xN2 xN3.

[0039] For space-frequency domain compression using real-valued NN, in the case of two channels (I / Q) (each including two polarizations), the data input to the third NN model type (referred to herein as SF option 3 real or SF-3r) corresponds to N1 x2 . A third input data option 306 of N2 x N3.

[0040] For space-frequency domain compression using real-valued NN, in the case of two channels (I / Q) (each including two polarizations), the data input to the fourth NN model type (referred to herein as SF option 4 real or SF-4r) corresponds to N1 x N2 x 2 . Fourth input data option 308 of N3. Since multipath in the time domain causes frequency selectivity in the frequency domain, the disclosed input formula is also applicable to space-delay domain compression, and in this setting, N3 means the number of delay taps of the wireless channel.

[0041] Example 3D CNN Autoencoder

[0042] Figure 4is a block diagram illustrating, at a high level, an example 3D CNN autoencoder that may be used according to certain embodiments. See, for example, “Unsupervised Spatial–Spectral Feature Learning by 3D Convolutional Autoencoder for Hyperspectral Classification,” Shaohui Mei, et al., IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 57, NO. 9, September 2019. As shown, 3D input data 402 is provided to an encoder (shown above the horizontal line), which includes a first 3D convolutional layer 404, which provides its output 406 to one or more subsequent layers until the nth 3D convolutional layer 408 provides its output 410 to a 3D max pooling layer 412, which is used for feature learning and is configured to reduce the number of training parameters in the CNN.

[0043] The encoder output 422 (features) is provided to a decoder (shown below the horizontal line), which includes a first 3D deconvolution layer 418, which provides its output 416 to one or more subsequent layers until the nth 3D deconvolution layer 414 provides reconstructed data 420. The reconstructed data 420 can be compared with the input data 402 (e.g., for training a CNN).

[0044] Time-space-frequency domain compression / prediction (complex-valued NN)

[0045] Time-space-frequency domain compression and prediction and Figure 3 The example shown uses the number of opportunities N4 together. Thus, for time-space-frequency domain compression using a complex-valued NN, in the case of two channels (one channel for a first polarization and the other channel for a second polarization), the data input to the first NN model type (referred to herein as time-space-frequency (TSF) option 1 complex or TSF-1c) corresponds to a first input data option 302 of N1 x N2 x N3 x N4 having dimensions N4.

[0046] For time-space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the second NN model type (referred to herein as TSF option 2 complex or TSF-2c) corresponds to 2. A second input data option 304 of N1 xN2 x N3 x N4 having dimension N4.

[0047] For time-space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the third NN model type (referred to herein as TSF option 3 complex or TSF-3c) corresponds to N1 x2 . A third input data option 306 of N2 x N3 x N4 having dimension N4.

[0048] For time-space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the fourth NN model type (referred to herein as TSF option 4 complex or TSF-4c) corresponds to N1 x N2 x 2 . A fourth input data option 308 of N3 x N4 having dimension N4.

[0049] For time-space-frequency domain compression using complex-valued NN, in the case of one channel (including two polarizations), the data input to the fifth NN model type (referred to herein as TSF option 5 complex or TSF-5c) corresponds to N1 x N2 x N3 x 2 . The first input data option 302 of N4 has data concatenated in N4 dimensions. Since multipath in the time domain causes frequency selectivity in the frequency domain, the disclosed input formula is also applicable to time-space-delay domain compression, and in this setting, N3 means the number of delay taps of the wireless channel. Through the duality between Doppler frequency and time domain selectivity, the disclosed input formula is also applicable to Doppler-space-frequency domain compression and Doppler-space-delay domain compression.

[0050] Time-space-frequency domain compression / prediction (real-valued NN)

[0051] For time-space-frequency domain compression using real-valued NN, with four channels (two channels (real and imaginary) for the first polarization and two channels (real and imaginary) for the second polarization), the data input to the first NN model type (referred to herein as TSF option 1 real or TSF-1r) corresponds to a first input data option 302 with dimension N4 of N1 x N2 x N3 x N4.

[0052] For time-space-frequency domain compression using real-valued NN, the data input to the second NN model type (referred to herein as TSF option 2 real or TSF-2r) corresponds to 2 in the case of two channels (real and imaginary) each including two polarizations.. A second input data option 304 of N1 x N2 x N3 x N4 having dimension N4.

[0053] For time-space-frequency domain compression using real-valued NN, in the case of two channels (real and imaginary) each including two polarizations, the data input to the third NN model type (referred to herein as TSF option 3 real or TSF-3r) corresponds to N1 x 2 . A third input data option 306 of N2 x N3 x N4 having dimension N4.

[0054] For time-space-frequency domain compression using real-valued NN, the data input to the fourth NN model type (referred to herein as TSF option 4 real or TSF-4r) corresponds to N1 x N2 x 2 in the case of two channels (real and imaginary) each comprising two polarizations. . A fourth input data option 308 of N3 x N4 having dimension N4.

[0055] For time-space-frequency domain compression using real-valued NN, in the case of two channels (real and imaginary) each comprising two polarizations, the data input to the fifth NN model type (referred to herein as TSF option 5 real or TSF-5r) corresponds to N1 x N2 x N3 x 2 . A fourth input data option 308 of N4 with data concatenated in N4 dimensions. Since multipath in the time domain causes frequency selectivity in the frequency domain, the disclosed input formula is also applicable to time-space-delay domain compression, and in this setting, N3 means the number of delay taps of the wireless channel. By the duality between Doppler frequency and time domain selectivity, the disclosed input formula is also applicable to Doppler-space-frequency domain compression and Doppler-space-delay domain compression.

[0056] Long Short-Term Memory NN

[0057] In certain embodiments, a Long Short-Term Memory (LSTM) NN can be used with 3D input data for time-frequency domain compression. The LSTM NN includes a feedback connection. For the LSTM, one dimension can be considered as time. Similarly, for tensorflow, the data input to the LSTM NN can be 2D (ConvLSTM2D layer) or 3D (ConvLSTM3D layer). Therefore, LSTM with 2D can also be used for frequency domain compression, thereby considering input with images of N1×N2 as a building block for further options. The time dimension of the LSTM can be along N3.

[0058] Thus, in certain embodiments, LSTM architectures with 3D can also be used for time-frequency domain compression, treating inputs with images of N1×N2×N3 as building blocks, and further options may be available for 4D autoencoders. The time dimension of the LSTM may be along N4. Additionally, or in other embodiments, inputs of N1×N2×N4 may be used as building blocks, and further options may be available for 4D autoencoders. The time dimension of the LSTM is along N3.

[0059] Figure 5 is a flow chart illustrating a method 500 for ML-based CSI feedback according to certain embodiments. At block 502, for one or more NN model types (e.g., SF-1c, ..., SF-4c, SF-1r, ..., SF-4r, or other NN model types discussed herein), the UE reports support for input data formats in UE capability signaling.

[0060] In one embodiment, at block 504, for the selected NN model type, based on UE capability signaling, the network (NW) configures a single NN model to the UE. At block 506, the UE then generates CSI feedback according to the configured NN model.

[0061] In another embodiment, at box 508, for one or more selected NN model types, based on UE capability signaling, the NW configures multiple NN models to the UE. At box 510, the UE then selects one of the multiple configured NN models. At box 512, the UE generates CSI feedback based on the NN model selected by the UE. At box 514, the UE generates CSI feedback based on the NN model selected by the UE.

[0062] Figure 6 is a flow chart of a method 600 for a UE to provide ML-based CSI to a wireless network according to one embodiment. In box 602, for one or more NN model types, the method 600 includes: transmitting UE capability signaling to the wireless network to report one or more 3D or 4D input data formats to the CSI feedback encoder. In box 604, for one or more NN model types, the method 600 includes: processing NN model configuration information from the wireless network. In box 606, the method 600 includes: generating CSI feedback by providing DL channel data or DL ​​pre-decoder formatted according to one or more 3D or 4D input formats to the NN model corresponding to the NN model configuration information. In box 608, the method 600 includes: transmitting the CSI feedback to the wireless network.

[0063] In certain embodiments of method 600, the NN model configuration information corresponds only to the NN model selected by the wireless network.

[0064] In certain embodiments of method 600, the NN model configuration information corresponds to multiple NN models, and method 600 also includes: selecting a NN model from among the multiple NN models by the UE; and indicating the NN model selected by the UE to the wireless network in CSI feedback.

[0065] In certain embodiments of method 600, one or more 3D or 4D input formats include at least: a first dimension, which includes the number of antenna columns N1 at the base station; a second dimension, which includes the number of antenna rows N2 at the base station; and a third dimension, which includes the number of subbands N3 used for CSI feedback.

[0066] In certain embodiments, the NN model comprises a complex-valued NN for space-frequency domain compression, and in the case of a first channel for a first polarization and a second channel for a second polarization, one or more 3D or 4D input formats comprise N1 x N2 x N3.

[0067] In some embodiments, the NN model includes a complex-valued NN for space-frequency domain compression, and for a channel including two polarizations, one or more 3D or 4D input formats include 2 . N1x N2 x N3.

[0068] In some embodiments, the NN model includes a complex-valued NN for space-frequency domain compression, and for a channel including two polarizations, one or more 3D or 4D input formats include N1 x2 . N2 x N3.

[0069] In some embodiments, the NN model comprises a complex-valued NN for space-frequency domain compression, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 xN2 x 2 . N3.

[0070] In certain embodiments, the NN model includes a real-valued NN for space-frequency domain compression, and in the case of a first channel for a first polarization comprising in-phase (I) values, a second channel for a second polarization comprising I values, a third channel for the first polarization comprising quadrature (Q) values, and a fourth channel for the second polarization comprising Q values, one or more 3D or 4D input formats include N1x N2 x N3.

[0071] In some embodiments, the NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprises 2 . N1 xN2 x N3.

[0072] In some embodiments, the NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprise N1 x2 . N2 x N3.

[0073] In some embodiments, the NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprise N1 xN2 x 2 . N3.

[0074] In certain embodiments of the method 600, the one or more 3D or 4D input formats further include a number N4 of opportunities to generate precoders for the number N3 of subbands.

[0075] In some embodiments, the NN model includes a complex-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for a first polarization and a second channel for a second polarization, one or more 3D or 4D input formats include N1 x N2 x N3 x N4.

[0076] In some embodiments, the NN model includes a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel including two polarizations, one or more 3D or 4D input formats include 2 . N1 x N2 x N3 x N4.

[0077] In some embodiments, the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

[0078] In some embodiments, the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x 2. N3 x N4.

[0079] In some embodiments, the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x N3 x 2 . N4.

[0080] In some embodiments, the NN model includes a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for a first polarization comprising a real part, a second channel for a second polarization comprising a real part, a third channel for the first polarization comprising an imaginary part, and a fourth channel for the second polarization comprising an imaginary part, one or more 3D or 4D input formats include N1 xN2 x N3 x N4.

[0081] In some embodiments, the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for two polarizations comprising a real part and a second channel for two polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3 x N4.

[0082] In some embodiments, the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for both polarizations comprising a real part and a second channel for both polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise N1 x 2 . N2x N3 x N4.

[0083] In some embodiments, the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for two polarizations comprising a real part and a second channel for two polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise N1 x N2 x2 . N3 x N4.

[0084] In some embodiments, the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for two polarizations comprising a real part and a second channel for two polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise N1 x N2 x N3 x 2 . N4.

[0085] In certain embodiments of method 600, the NN model comprises a long short-term memory (LSTM) NN.

[0086] Embodiments contemplated herein include an apparatus comprising means for performing one or more elements of the method 600. The apparatus may be, for example, an apparatus that is a UE (such as the wireless device 902 as a UE, as described herein).

[0087] Embodiments contemplated herein include one or more non-transitory computer-readable media including instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform one or more elements of method 600. The non-transitory computer-readable medium may be, for example, a memory of a UE (such as memory 906 of a wireless device 902 that is a UE, as described herein).

[0088] Embodiments contemplated herein include an apparatus comprising logical components, modules, or circuits for performing one or more elements of the method 600. The apparatus may be, for example, an apparatus of a UE (such as wireless device 902 as a UE, as described herein).

[0089] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of the method 600. The apparatus may be, for example, an apparatus of a UE (such as wireless device 902 as a UE, as described herein).

[0090] Embodiments contemplated herein include a signal as described in or associated with one or more elements of method 600 .

[0091] Embodiments contemplated herein include a computer program or computer program product including instructions, wherein execution of the program by a processor causes the processor to perform one or more elements of the method 600. The processor may be a processor of a UE (such as the processor 904 of the wireless device 902 as a UE, as described herein). The instructions may be located, for example, in a processor of the UE and / or on a memory (such as the memory 906 of the wireless device 902 as a UE, as described herein).

[0092] Figure 7It is a flowchart of a method 700 for a base station to configure one or more NN models for CSI feedback to a UE according to one embodiment. In box 702, for one or more NN model types, the method 700 includes: receiving a UE capability message indicating one or more 3D or 4D input data formats for a CSI feedback encoder at the UE. In box 704, for one or more NN model types, the method 700 includes: configuring one or more NN models to the UE based on the UE capability message to encode DL channel data or DL ​​predecoder formatted according to one or more 3D or 4D input formats. In box 706, the method 700 includes: receiving CSI feedback from the UE. In box 708, the method 700 includes: using a decoder at the base station to generate reconstructed data from the CSI feedback, the decoder being configured with a selected NN model from one or more NN models used by the encoder at the UE.

[0093] In certain embodiments of method 700, configuring one or more NN models to the UE includes: configuring a selected NN model to the UE, and the selected NN model is selected by a base station.

[0094] In certain embodiments of method 700, configuring one or more NN models to the UE includes: configuring multiple NN models to the UE, and method 700 also includes: receiving an indication of the selected NN model from the UE in the CSI feedback.

[0095] In certain embodiments of method 700, one or more 3D or 4D input formats include at least: a first dimension, which includes the number of antenna columns N1 at the base station; a second dimension, which includes the number of antenna rows N2 at the base station; and a third dimension, which includes the number of subbands N3 used for CSI feedback.

[0096] In some embodiments, the selected NN model comprises a complex-valued NN for space-frequency domain compression, and in the case of a first channel for a first polarization and a second channel for a second polarization, one or more 3D or 4D input formats comprise N1 x N2 x N3.

[0097] In some embodiments, the selected NN model comprises a complex-valued NN for space-frequency domain compression, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprises 2 . N1 x N2 x N3.

[0098] In some embodiments, the selected NN model comprises a complex-valued NN for space-frequency domain compression, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprises N1 x 2 .N2 x N3.

[0099] In some embodiments, the selected NN model comprises a complex-valued NN for space-frequency domain compression, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3.

[0100] In certain embodiments, the selected NN model includes a real-valued NN for space-frequency domain compression, and in the case of a first channel for a first polarization comprising in-phase (I) values, a second channel for a second polarization comprising I values, a third channel for the first polarization comprising quadrature (Q) values, and a fourth channel for the second polarization comprising Q values, one or more 3D or 4D input formats include N1 x N2 x N3.

[0101] In some embodiments, the selected NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprises 2 . N1 x N2 x N3.

[0102] In some embodiments, the selected NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3.

[0103] In some embodiments, the selected NN model comprises a real-valued NN for space-frequency domain compression, and in the case of a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3.

[0104] In certain embodiments of the method 700, the one or more 3D or 4D input formats further include a number N4 of opportunities to generate precoders for the number N3 of subbands.

[0105] In some embodiments, the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for a first polarization and a second channel for a second polarization, one or more 3D or 4D input formats comprise N1 x N2 x N3 x N4.

[0106] In some embodiments, the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprises 2 . N1 x N2 x N3 x N4.

[0107] In some embodiments, the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

[0108] In some embodiments, the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3x N4.

[0109] In some embodiments, the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and for a channel comprising two polarizations, one or more 3D or 4D input formats comprise N1 x N2 x N3 x2 . N4.

[0110] In some embodiments, the selected NN model includes a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for a first polarization including a real part, a second channel for a second polarization including a real part, a third channel for the first polarization including an imaginary part, and a fourth channel for the second polarization including an imaginary part, one or more 3D or 4D input formats include N1 x N2 x N3 x N4.

[0111] In some embodiments, the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for both polarizations comprising a real part and a second channel for both polarizations comprising an imaginary part, one or more 3D or 4D input formats comprises 2 . N1 x N2 x N3 x N4.

[0112] In some embodiments, the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for both polarizations comprising a real part and a second channel for both polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

[0113] In some embodiments, the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for both polarizations comprising a real part and a second channel for both polarizations comprising an imaginary part, one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3 x N4.

[0114] In some embodiments, the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and in the case of a first channel for both polarizations comprising a real part and a second channel for both polarizations comprising an imaginary part, the one or more 3D or 4D input formats comprise N1 x N2 x N3 x 2 . N4.

[0115] In certain embodiments of method 700, the selected NN model comprises a long short-term memory (LSTM) NN.

[0116] Embodiments contemplated herein include an apparatus comprising means for performing one or more elements of method 700. The apparatus may be, for example, an apparatus that is a base station (such as network device 918 that is a base station, as described herein).

[0117] Embodiments contemplated herein include one or more non-transitory computer-readable media that 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 method 700. The non-transitory computer-readable medium may be, for example, a memory of a base station (such as memory 922 of network device 918 acting as a base station, as described herein).

[0118] Embodiments contemplated herein include an apparatus comprising logical components, modules, or circuits operable to perform one or more elements of method 700. The apparatus may be, for example, an apparatus that is a base station (such as network device 918 as a base station, as described herein).

[0119] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of method 700. The apparatus may be, for example, an apparatus that is a base station (such as network device 918 that is a base station, as described herein).

[0120] Embodiments contemplated herein include a signal as described in or associated with one or more elements of method 700 .

[0121] Embodiments contemplated herein include a computer program or computer program product including instructions, wherein execution of the program by a processing element causes the processing element to perform one or more elements of method 700. The processor may be a memory of a base station (such as processor 920 of network device 918 as a base station, as described herein). These instructions may be located, for example, in a processor and / or on a memory of a base station (such as memory 922 of network device 918 as a base station, as described herein).

[0122] Figure 8 An example architecture of a wireless communication system 800 according to embodiments disclosed herein is illustrated. The following description is provided for an example wireless communication system 800 operating in conjunction with the LTE system standard and / or the 5G or NR system standard provided in the 3GPP technical specifications.

[0123] like Figure 8 As shown, the wireless communication system 800 includes UE 802 and UE 804 (although any number of UEs may be used). In this example, UE 802 and UE 804 are illustrated as smartphones (e.g., handheld touch screen mobile computing devices capable of connecting to one or more cellular networks), but may also include any mobile or non-mobile computing device configured for wireless communication.

[0124] UE 802 and UE 804 may be configured to be communicatively coupled to RAN 806. In an embodiment, RAN 806 may be NG-RAN, E-UTRAN, etc. UE 802 and UE 804 utilize connections (or channels) (shown as connection 808 and connection 810, respectively) with RAN 806, where each connection (or channel) includes a physical communication interface. RAN 806 may include one or more base stations (such as base station 812 and base station 814) that implement connection 808 and connection 810.

[0125] In this example, connection 808 and connection 810 are air interfaces that enable this communicative coupling and may conform to the RAT used by the RAN 806 , such as, for example, LTE and / or NR.

[0126] In some embodiments, UE 802 and UE 804 may also directly exchange communication data via side link interface 816. UE 804 is shown as being configured to access an access point (shown as AP 818) via connection 820. For example, connection 820 may include a local wireless connection, such as a connection conforming to any IEEE 802.11 protocol, wherein AP 818 may include In this example, AP 818 may not be connected to another network (eg, the Internet) through CN 824.

[0127] In an embodiment, UE 802 and UE 804 may be configured to communicate with each other or with base station 812 and / or base station 814 over a multi-carrier communication channel using orthogonal frequency division multiplexing (OFDM) communication signals according to various communication techniques, such as, but not limited to, orthogonal frequency division multiple access (OFDMA) communication techniques (e.g., for downlink communication) or single carrier frequency division multiple access (SC-FDMA) communication techniques (e.g., for uplink and ProSe or sidelink communication), although the scope of the embodiments is not limited in this respect. OFDM signals may include multiple orthogonal subcarriers.

[0128] In some embodiments, all or part of the base station 812 or the base station 814 may be implemented as one or more software entities running on a server computer as part of a virtual network. In addition, or in other embodiments, the base station 812 or the base station 814 may be configured to communicate with each other via the interface 822. In an embodiment where the wireless communication system 800 is an LTE system (e.g., when the CN 824 is an EPC), the interface 822 may be an X2 interface. The X2 interface may be defined between two or more base stations (e.g., two or more eNBs, etc.) connected to the EPC and / or between two eNBs connected to the EPC. In an embodiment where the wireless communication system 800 is an NR system (e.g., when the CN 824 is a 5GC), the interface 822 may be an Xn interface. The Xn interface is defined between two or more base stations (e.g., two or more gNBs, etc.) connected to the 5GC, between the base station 812 (e.g., gNB) and the eNB connected to the 5GC, and / or between two eNBs connected to the 5GC (e.g., CN 824).

[0129] RAN 806 is shown as being communicatively coupled to CN 824. CN 824 may include one or more network elements 826 configured to provide various data and telecommunication services to customers / subscribers (e.g., UE 802 and users of UE 804) connected to CN 824 via RAN 806. The components of CN 824 may be implemented in one physical device or separate physical devices including components for reading and executing instructions from a machine-readable or computer-readable medium (e.g., a non-transitory machine-readable storage medium).

[0130] In an embodiment, CN 824 may be an EPC, and RAN 806 may be connected to CN 824 via an S1 interface 828. In an embodiment, S1 interface 828 may be divided into two parts: an S1 user plane (S1-U) interface, which carries traffic data between base station 812 or base station 814 and a serving gateway (S-GW); and an S1-MME interface, which is a signaling interface between base station 812 or base station 814 and a mobility management entity (MME).

[0131] In an embodiment, CN 824 may be a 5GC, and RAN 806 may be connected to CN 824 via an NG interface 828. In an embodiment, NG interface 828 may be divided into two parts: an NG user plane (NG-U) interface, which carries traffic data between base station 812 or base station 814 and a user plane function (UPF); and an S1 control plane (NG-C) interface, which is a signaling interface between base station 812 or base station 814 and an access and mobility management function (AMF).

[0132] In general, the application server 830 may be an element that provides applications (e.g., packet-switched data services) that use Internet Protocol (IP) bearer resources with the CN 824. The application server 830 may also be configured to support one or more communication services (e.g., VoIP sessions, group communication sessions, etc.) for the UE 802 and UE 804 via the CN 824. The application server 830 may communicate with the CN 824 via an IP communication interface 832.

[0133] Fig. 9 A system 900 for performing signaling 934 between a wireless device 902 and a network device 918 according to embodiments disclosed herein is illustrated. The system 900 may be part of a wireless communication system as described herein. The wireless device 902 may be, for example, a UE of a wireless communication system. The network device 918 may be, for example, a base station (e.g., an eNB or a gNB) of a wireless communication system.

[0134] The wireless device 902 may include one or more processors 904. The processor 904 may execute instructions to perform various operations of the wireless device 902, as described herein. The processor 904 may include one or more baseband processors implemented using, for example, a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a controller, a field programmable gate array (FPGA) device, another hardware device, a firmware device, or any combination thereof, configured to perform the operations described herein.

[0135] The wireless device 902 may include a memory 906. The memory 906 may be a non-transitory computer-readable storage medium storing instructions 908 (which may include, for example, instructions executed by the processor 904). The instructions 908 may also be referred to as program code or a computer program. The memory 906 may also store data used by the processor 904 and results calculated by the processor.

[0136] The wireless device 902 may include one or more transceivers 910, which may include radio frequency (RF) transmitter and / or receiver circuitry that uses an antenna 912 of the wireless device 902 to facilitate signaling (e.g., signaling 934) to and / or from the wireless device 902 with other devices (e.g., network device 918) in accordance with a corresponding RAT.

[0137] The wireless device 902 may include one or more antennas 912 (e.g., one, two, four, or more). For implementations with multiple antennas 912, the wireless device 902 may utilize the spatial diversity of such multiple antennas 912 to transmit and / or receive multiple different data streams on the same time-frequency resources. This behavior may be referred to as, for example, MIMO behavior (referring to the multiple antennas used at each of the transmitting device and the receiving device to implement this aspect). MIMO transmission by the wireless device 902 may be implemented based on precoding (or digital beamforming) applied at the wireless device 902, which multiplexes data streams between antennas 912 based on known or assumed channel characteristics, so that each data stream is received at an appropriate signal strength relative to the other streams and at a desired location in the spatial domain (e.g., the location of a receiver associated with the data stream). Certain implementations may use a single-user MIMO (SU-MIMO) approach (where all data streams are directed to a single receiver) and / or a multi-user MIMO (MU-MIMO) approach (where separate data streams may be directed to separate (different) receivers in different locations in the spatial domain).

[0138] In certain embodiments with multiple antennas, the wireless device 902 may implement analog beamforming techniques whereby the phases of signals transmitted by the antennas 912 are relatively adjusted such that the (joint) transmissions of the antennas 912 are directional (this is sometimes referred to as beam steering).

[0139] The wireless device 902 may include one or more interfaces 914. The interface 914 may be used to provide input to the wireless device 902 or provide output from the wireless device. For example, a wireless device 902 as a UE may include an interface 914, such as a microphone, a speaker, a touch screen, buttons, etc., to allow a user of the UE to input and / or output to the UE. Other interfaces of such a UE may be composed of transmitters, receivers, and other circuits (e.g., in addition to the transceiver 910 / antenna 912 described above), which allow communication between the UE and other devices and may be performed according to known protocols (e.g., etc.) to perform the operation.

[0140] The wireless device 902 may include a CSI feedback module 916. The CSI feedback module 916 may be implemented via hardware, software, or a combination thereof. For example, the CSI feedback module 916 may be implemented as a processor, circuit, and / or instructions 908 stored in the memory 906 and executed by the processor 904. In some examples, the CSI feedback module 916 may be integrated within the processor 904 and / or the transceiver 910. For example, the CSI feedback module 916 may be implemented by a combination of software components (e.g., executed by a DSP or a general purpose processor) and hardware components (e.g., logic gates and circuits) within the processor 904 or the transceiver 910.

[0141] The CSI feedback module 916 may be used in various aspects of the present disclosure, for example, Figure 5 and Figure 6 In addition, the CSI feedback module 916 may include an encoder such as Figure 1 The encoder 102 or Figure 4 The encoder shown.

[0142] The network device 918 may include one or more processors 920. The processor 920 may execute instructions to perform various operations of the network device 918, as described herein. The processor 920 may include one or more baseband processors configured to perform the operations described herein, the one or more baseband processors being implemented using, for example, a CPU, a DSP, an ASIC, a controller, an FPGA device, another hardware device, a firmware device, or any combination thereof.

[0143] The network device 918 may include a memory 922. The memory 922 may be a non-transitory computer-readable storage medium storing instructions 924 (which may include, for example, instructions executed by the processor 920). The instructions 924 may also be referred to as program code or a computer program. The memory 922 may also store data used by the processor 920 and results calculated by the processor.

[0144] The network device 918 may include one or more transceivers 926, which may include RF transmitter and / or receiver circuits that use an antenna 928 of the network device 918 to facilitate signaling (e.g., signaling 934) sent or received by the network device 918 with other devices (e.g., wireless device 902) according to the corresponding RAT.

[0145] The network device 918 may include one or more antennas 928 (e.g., one, two, four, or more). In embodiments with multiple antennas 928, the network device 918 may perform MIMO, digital beamforming, analog beamforming, beam steering, etc. as described.

[0146] The network device 918 may include one or more interfaces 930. The interface 930 may be used to provide input to or output from the network device 918. For example, the network device 918 as a base station may include an interface 930 composed of a transmitter, a receiver and other circuits (e.g., in addition to the transceiver 926 / antenna 928 described above), which enables the base station to communicate with other equipment in the core network and / or enables the base station to communicate with external networks, computers, databases, etc., in order to achieve the purpose of performing operations, management and maintenance of the base station or other equipment operably connected thereto.

[0147] The network device 918 may include a CSI feedback module 932. The CSI feedback module 932 may be implemented via hardware, software, or a combination thereof. For example, the CSI feedback module 932 may be implemented as a processor, circuit, and / or instructions 924 stored in the memory 922 and executed by the processor 920. In some examples, the CSI feedback module 932 may be integrated within the processor 920 and / or the transceiver 926. For example, the CSI feedback module 932 may be implemented by a combination of software components (e.g., executed by a DSP or a general purpose processor) and hardware components (e.g., logic gates and circuits) within the processor 920 or the transceiver 926.

[0148] The CSI feedback module 932 may be used in various aspects of the present disclosure, for example, Figure 5 and Figure 7 In addition, the CSI feedback module 932 may include a decoder such as Figure 1The decoder 104 shown or Figure 4 The decoder shown.

[0149] For one or more embodiments, at least one of the components set forth in one or more of the foregoing figures may be configured to perform one or more operations, techniques, processes, and / or methods as described herein. For example, a baseband processor as described herein in conjunction with one or more of the foregoing figures may be configured to operate according to one or more of the examples set forth herein. For another example, circuits associated with a UE, a base station, a network element, etc. as 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 set forth herein.

[0150] Unless otherwise expressly stated, any of the above embodiments may be combined with any other embodiment (or combination of embodiments). 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 form disclosed. In view of the above teachings, modifications and variations are possible, or modifications and variations can be obtained from the practice of various embodiments.

[0151] Embodiments and implementations of the systems and methods described herein may include various operations that may be embodied in machine executable instructions to be executed by a computer system. A computer system may include one or more general or special purpose computers (or other electronic devices). A computer system may include hardware components that include specific logic components for performing operations; or may include a combination of hardware, software, and / or firmware.

[0152] It should be appreciated that the systems described herein include descriptions of specific embodiments. These embodiments may be combined into a single system, partially combined into other systems, separated into multiple systems, or otherwise divided or combined. In addition, it is contemplated that parameters, attributes, aspects, etc. of another embodiment may be used in one embodiment. For clarity, these parameters, attributes, aspects, etc. are described only in one or more embodiments, and it should be appreciated that these parameters, attributes, aspects, etc. may be combined with or substituted for parameters, attributes, aspects, etc. of another embodiment unless expressly stated herein.

[0153] It is well known 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.

[0154] Although the foregoing has been described in considerable detail for the sake of clarity, it will be apparent that certain changes and modifications may be made without departing from the principles of the invention. It should be noted that there are many alternative ways to implement both the processes and the apparatus described herein. Therefore, the embodiments of the present invention should be regarded as illustrative rather than restrictive, and the specification is not limited to the details given herein, but may be modified within the scope and equivalents of the appended claims.

Claims

1. A method for a user equipment (UE) to provide machine learning (ML) based channel state information (CSI) to a wireless network, the method comprising: transmitting UE capability signaling to the wireless network to report one or more three-dimensional (3D) or four-dimensional (4D) input data formats to a CSI feedback encoder for one or more neural network (NN) model types; processing NN model configuration information from the wireless network for the one or more NN model types; generating CSI feedback by providing downlink (DL) channel data or a DL precoder formatted according to the one or more 3D or 4D input formats to a NN model corresponding to the NN model configuration information; as well as The CSI feedback is transmitted to the wireless network.

2. The method of claim 1, wherein the NN model configuration information corresponds only to the NN model selected by the wireless network.

3. The method according to claim 1, wherein the NN model configuration information corresponds to a plurality of NN models, and wherein the method further comprises: Selecting, by the UE, the NN model from among the multiple NN models; as well as The NN model selected by the UE is indicated to the wireless network in the CSI feedback.

4. The method of claim 1, wherein the one or more 3D or 4D input formats include at least: A first dimension, the first dimension comprising the number N1 of antenna columns at the base station; a second dimension, the second dimension comprising the number N2 of antenna rows at the base station; and a third dimension, the third dimension comprising the number N3 of subbands used for the CSI feedback.

5. The method of claim 4, wherein the NN model comprises a complex-valued NN for space-frequency domain compression, and wherein in the case of a first channel for a first polarization and a second channel for a second polarization, the one or more 3D or 4D input formats comprise N1 x N2 x N3.

6. The method of claim 4, wherein the NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3.

7. The method of claim 4, wherein the NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x 2 . N2xN3.

8. The method of claim 4, wherein the NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x N2 x2 . N3.

9. The method of claim 4, wherein the NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 in the case of a first channel comprising in-phase (I) values ​​for a first polarization, a second channel comprising the I values ​​for a second polarization, a third channel comprising quadrature (Q) values ​​for the first polarization, and a fourth channel comprising the Q values ​​for the second polarization.

10. The method of claim 4, wherein the NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3.

11. The method of claim 4, wherein the NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3.

12. The method of claim 4, wherein the NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x 2 with a first channel comprising in-phase (I) values ​​for two polarizations and a second channel comprising quadrature (Q) values ​​for the two polarizations. . N3.

13. The method of claim 4, wherein the one or more 3D or 4D input formats further comprises a number N4 of opportunities to generate precoders for the number N3 of subbands.

14. The method of claim 13, wherein the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein in the case of a first channel for a first polarization and a second channel for a second polarization, the one or more 3D or 4D input formats comprise N1 x N2 x N3 x N4.

15. The method of claim 13, wherein the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3 x N4.

16. The method of claim 13, wherein the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

17. The method of claim 13, wherein the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3 x N4.

18. The method of claim 13, wherein the NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x N2 x N3 x 2 . N4.

19. The method of claim 13, wherein the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 x N4 in the case of a first channel for a first polarization comprising a real part, a second channel for a second polarization comprising the real part, a third channel for the first polarization comprising an imaginary part, and a fourth channel for the second polarization comprising the imaginary part.

20. The method of claim 13, wherein the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3 x N4.

21. The method of claim 13, wherein the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

22. The method of claim 13, wherein the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x 2 with a first channel for two polarizations comprising a real part and a second channel for the two polarizations comprising an imaginary part. . N3 x N4.

23. The method of claim 13, wherein the NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 x 2 with a first channel for two polarizations comprising a real part and a second channel for the two polarizations comprising an imaginary part. . N4.

24. The method of claim 1, wherein the NN model comprises a long short-term memory (LSTM) NN.

25. A method for a base station to configure one or more neural network (NN) models for channel state information (CSI) feedback to a user equipment (UE), the method comprising: receiving a UE capability message indicating one or more three-dimensional (3D) or four-dimensional (4D) input data formats for a CSI feedback encoder at the UE for one or more NN model types; configuring, for the one or more NN model types, one or more NN models to the UE based on the UE capability message to encode downlink (DL) channel data or a DL precoder formatted according to the one or more 3D or 4D input formats; receiving the CSI feedback from the UE; as well as Reconstructed data or a reconstructed DL precoder is generated from the CSI feedback using a decoder at the base station, the decoder being configured with a selected NN model from the one or more NN models used by the encoder at the UE.

26. The method of claim 25, wherein configuring one or more NN models to the UE comprises: The selected NN model is configured to the UE, and the selected NN model is selected by the base station.

27. The method of claim 25, wherein configuring one or more NN models to the UE comprises: Configuring multiple NN models to the UE, and wherein the method further comprises: An indication of the selected NN model is received from the UE in the CSI feedback.

28. The method of claim 25, wherein the one or more 3D or 4D input formats include at least: A first dimension, the first dimension comprising the number N1 of antenna columns at the base station; a second dimension, the second dimension comprising the number N2 of antenna rows at the base station; and a third dimension, the third dimension comprising the number N3 of subbands used for the CSI feedback.

29. The method of claim 28, wherein the selected NN model comprises a complex-valued NN for space-frequency domain compression, and wherein in the case of a first channel for a first polarization and a second channel for a second polarization, the one or more 3D or 4D input formats comprise N1 x N2 x N3.

30. The method of claim 28, wherein the selected NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise 2 . N1 xN2 x N3.

31. The method of claim 28, wherein the selected NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x2 . N2 x N3.

32. The method of claim 28, wherein the selected NN model comprises a complex-valued NN for space-frequency domain compression, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 xN2 x 2 . N3.

33. The method of claim 28, wherein the selected NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 in the case of a first channel comprising in-phase (I) values ​​for a first polarization, a second channel comprising the I values ​​for a second polarization, a third channel comprising quadrature (Q) values ​​for the first polarization, and a fourth channel comprising the Q values ​​for the second polarization.

34. The method of claim 28, wherein the selected NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3.

35. The method of claim 28, wherein the selected NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3.

36. The method of claim 28, wherein the selected NN model comprises a real-valued NN for space-frequency domain compression, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x 2 with a first channel for two polarizations comprising in-phase (I) values ​​and a second channel for the two polarizations comprising quadrature (Q) values. . N3.

37. The method of claim 28, wherein the one or more 3D or 4D input formats further comprises a number N4 of opportunities to generate precoders for the number N3 of subbands.

38. The method of claim 37, wherein the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein in the case of a first channel for a first polarization and a second channel for a second polarization, the one or more 3D or 4D input formats comprise N1 x N2 x N3 x N4.

39. The method of claim 37, wherein the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3 xN4.

40. The method of claim 37, wherein the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

41. The method of claim 37, wherein the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x N2 x 2 . N3xN4.

42. The method of claim 37, wherein the selected NN model comprises a complex-valued NN for time-space-frequency domain compression or prediction, and wherein for a channel comprising two polarizations, the one or more 3D or 4D input formats comprise N1 x N2 x N3 x2 . N4.

43. The method of claim 37, wherein the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 xN4 in the case of a first channel for a first polarization comprising a real part, a second channel for a second polarization comprising the real part, a third channel for the first polarization comprising an imaginary part, and a fourth channel for the second polarization comprising the imaginary part.

44. The method of claim 37, wherein the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise 2 . N1 x N2 x N3 xN4.

45. The method of claim 37, wherein the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x 2 . N2 x N3 x N4.

46. ​​The method of claim 37, wherein the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x 2 with a first channel for two polarizations comprising a real part and a second channel for the two polarizations comprising an imaginary part. . N3xN4.

47. The method of claim 37, wherein the selected NN model comprises a real-valued NN for time-space-frequency domain compression or prediction, and wherein the one or more 3D or 4D input formats comprise N1 x N2 x N3 x2 with a first channel for two polarizations comprising a real part and a second channel for the two polarizations comprising an imaginary part. . N4.

48. The method of claim 25, wherein the selected NN model comprises a long short-term memory (LSTM) NN.

49. An apparatus comprising means for performing the method of any one of claims 1 to 48.

50. A computer-readable medium comprising instructions, which, when executed by one or more processors of an electronic device, cause the electronic device to perform the method according to any one of claims 1 to 48.

51. An apparatus comprising logic components, modules or circuits for performing the method of any one of claims 1 to 48.

Images