Qualification of machine learning based csi prediction

Abstract

Certain aspects of the present disclosure provide techniques for qualifying machine learning model based channel state information (CSI) prediction. An example method generally includes receiving, from a network entity, a channel state information (CSI) prediction model for quantized CSI, computing CSI based on downlink reference signal measurements, generating a quantized CSI difference based on quantization of a difference between the computed CSI and CSI predicted based on the CSI prediction model, and reporting the computed CSI and the quantized CSI difference to the network entity.

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to U.S. Application No. 16 / 888,593, filed May 29, 2020, which has been assigned to the assignee of this application and is incorporated herein in its entirety. Technical Field

[0003] Various aspects of this disclosure relate to wireless communications, and more specifically, to techniques for using machine learning models to predict channel state information (CSI) and to qualify CSI predictions made using machine learning models. Background Technology

[0004] Wireless communication systems are widely deployed to provide a variety of telecommunications services, such as telephone, video, data, messaging, and broadcasting. These wireless communication systems can employ multiple access technologies that enable communication with multiple users by sharing available system resources (e.g., bandwidth, transmit power, etc.). Examples of such multiple access systems include 3GPP Long Term Evolution (LTE) systems, LTE-A systems, Code Division Multiple Access (CDMA) systems, Time Division Multiple Access (TDMA) systems, Frequency Division Multiple Access (FDMA) systems, Orthogonal Frequency Division Multiple Access (OFDMA) systems, Single Carrier Frequency Division Multiple Access (SC-FDMA) systems, and Time Division Synchronous Code Division Multiple Access (TD-SCDMA) systems, to name just a few.

[0005] These multiple access technologies have been adopted in various telecommunications standards to provide a common protocol enabling different wireless devices to communicate at the city, national, regional, and even global levels. New radios (e.g., 5G NR) are examples of emerging telecommunications standards. NR is a collection of enhancements to the LTE mobile standard released by 3GPP. NR is designed to better support mobile broadband internet access by improving spectrum efficiency, reducing costs, improving service, utilizing new spectrum, and using OFDMA with a cyclic prefix (CP) on the downlink (DL) and uplink (UL). To this end, NR supports beamforming, multiple-input multiple-output (MIMO) antenna technology, and carrier aggregation.

[0006] However, with the continued growth in demand for mobile broadband access, further improvements to NR and LTE technologies are needed. Preferably, these improvements should be applicable to other multiple access technologies and telecommunications standards that employ these technologies. Summary of the Invention

[0007] The systems, methods, and apparatuses of this disclosure each have several aspects, each of which is responsible for more than just its desired properties. Without limiting the scope of this disclosure as set forth in the following claims, some features will now be briefly discussed. After considering this discussion, and particularly after reading the section entitled “Detailed Description,” it will be understood how the features of this disclosure provide advantages, including, for example, efficient modifications to the Synchronization Signal Block (SSB) pattern to select a reduced number of optimal beams, which can allow for reduced monitoring time by the UE and / or free up resources for data transmission.

[0008] Certain aspects of the subject matter described in this disclosure can be implemented in a method of wireless communication performed by a user equipment. This method typically includes: receiving from a network entity a CSI prediction model for quantized channel state information (CSI); calculating the CSI based on downlink reference signal measurements; generating a quantized CSI difference based on quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model; and reporting the calculated CSI and the quantized CSI difference to the network entity.

[0009] Certain aspects of the subject matter described in this disclosure can be implemented in a method of wireless communication performed by a network entity. This method typically includes: transmitting to a user equipment (UE) a CSI prediction model for predicting channel state information (CSI) and one or more reference signals for the UE to use in calculating the CSI and generating one or more CSI differences based on the calculated CSI and the CSI predicted based on the CSI prediction model; receiving the calculated CSI and the quantized CSI differences from the UE; and communicating with the UE based on the calculated CSI and the quantized CSI differences.

[0010] The aspects generally include methods, apparatus, systems, computer-readable media, and processing systems as generally described herein with reference to and illustrated by the accompanying drawings.

[0011] To achieve the foregoing and related objectives, one or more aspects include the features fully described below and specifically pointed out in the claims. The following description and drawings set forth certain illustrative features of one or more aspects in detail. However, these features indicate only a few of the various ways in which the principles of each aspect can be employed. Attached Figure Description

[0012] To enable a more detailed understanding of the features of this disclosure described above, reference can be made to the more specific description briefly outlined above, some of which are illustrated in the figures. However, it should be noted that the figures illustrate only certain typical aspects of this disclosure and should therefore not be considered as limiting its scope, as the description may acknowledge other equivalent aspects.

[0013] Figure 1 This is a block diagram conceptually illustrating an example wireless communication network according to certain aspects of this disclosure.

[0014] Figure 2 This is a block diagram conceptually illustrating the design of an example base station (BS) and user equipment (UE) according to certain aspects of this disclosure.

[0015] Figure 3 This is an example frame format for communication in a wireless communication network, based on certain aspects of this disclosure.

[0016] Figure 4 An example of a channel state information (CSI) prediction module based on machine learning (ML) is shown according to certain aspects of this disclosure.

[0017] Figure 5 Example operations for wireless communication performed by a user equipment according to certain aspects of this disclosure are shown for predicting channel state information (CSI) and qualifying the CSI prediction.

[0018] Figure 6 Example operations for wireless communication by a network entity, according to certain aspects of this disclosure, are shown for configuring a user equipment (UE) to predict channel state information (CSI) and to qualify the CSI prediction.

[0019] Figure 7 An example timeline is shown according to certain aspects of this disclosure, in which machine learning models can be used to predict channel state information (CSI) and to qualify it based on a qualification scheme and the measured CSI.

[0020] Figure 8 This is a flowchart illustrating the configuration call of a user equipment (UE) based on a machine learning model to predict channel state information (CSI) and to qualify the predicted CSI, according to certain aspects of this disclosure.

[0021] Figure 9 This is a flowchart illustrating an example of configuring a user equipment to qualify for CSI predictions based on machine learning models, according to certain aspects of this disclosure.

[0022] Figure 10 A communication device according to various aspects of the present disclosure is shown, which may include various components configured to perform operations for the techniques disclosed herein.

[0023] Figure 11A communication device according to various aspects of the present disclosure is shown, which may include various components configured to perform operations for the techniques disclosed herein.

[0024] To facilitate understanding, the same reference numerals have been used where possible to designate the same elements shared across the figures. It is anticipated that elements disclosed in one aspect may be beneficially applied to other aspects without specific description. Detailed Implementation

[0025] This disclosure provides apparatus, methods, processing systems, and computer-readable media for using machine learning models to predict Channel State Information (CSI) and for qualifying the predicted CSI based on the difference between the predicted CSI and the measured CSI. In some cases, predicting CSI based on machine learning models and reporting whether the CSI prediction is qualified or unqualified can reduce the overhead of CSI reporting and improve the accuracy of CSI reporting (e.g., in situations where channel conditions or interference change rapidly). Furthermore, predicting CSI based on machine learning models and reporting whether the CSI prediction is qualified or unqualified can provide improved reliability and reduced latency in the connection between network entities and user equipment (UE) (e.g., adjusting network connection parameters between the UE and network entities using the predicted CSI instead of waiting for the UE to report the CSI measurement).

[0026] The following description provides examples of using machine learning models to predict Channel State Information (CSI) and to qualify the predicted CSI based on the difference between the predicted CSI and the measured CSI, and is not intended to limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the elements discussed without departing from the scope of this disclosure. Various procedures or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined in some other examples. For example, any number of aspects set forth herein may be used to implement an apparatus or practice method. Furthermore, the scope of this disclosure is intended to cover such apparatuses or methods that are practiced using structures, functionalities, or structures and functionalities other than or different from the aspects of this disclosure set forth herein. It should be understood that any aspect of this disclosure may be embodied by one or more elements of the claims. The word “exemplary” as used herein means “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or superior to other aspects.

[0027] Generally, any number of wireless networks can be deployed in a given geographical area. Each wireless network can support a specific Radio Access Technology (RAT) and can operate on one or more frequencies. A RAT can also be referred to as a radio technology, air interface, etc. A frequency can also be referred to as a carrier, subcarrier, frequency channel, tone, subband, etc. Each frequency can support a single RAT in a given geographical area to avoid interference between wireless networks using different RATs.

[0028] The techniques described herein can be used in a variety of wireless network and radio technologies. Although the aspects may be described herein using terms commonly associated with 3G, 4G and / or newer radio technologies (e.g., 5G NR), the aspects of this disclosure can be applied to other generation-based communication systems.

[0029] NR access can support a variety of wireless communication services, such as enhanced mobile broadband (eMBB) for wide bandwidth (e.g., 80 MHz or above), millimeter wave (mmW) for high carrier frequencies (e.g., 25 GHz or above), massive machine-type communication (mMTC) for non-backward-compatible MTC technologies, and / or mission-critical ultra-reliable low-latency communication (URLLC). These services can include latency and reliability requirements. These services can also have different transmission time intervals (TTIs) to meet corresponding quality of service (QoS) requirements. Furthermore, these services can coexist in the same subframe. NR supports beamforming, and beam direction can be dynamically configured. Precoded MIMO transmissions can also be supported. MIMO configurations in DL can support up to 8 transmit antennas, with up to 8 streams in multi-layer DL transmission and up to 2 streams per UE. Multi-layer transmission with up to 2 streams per UE can be supported. Up to 8 serving cells can be utilized to support aggregation of multiple cells.

[0030] The techniques presented in this article can be applied in various frequency bands used in NR. For example, for higher frequency bands (referred to as FR4) (e.g., 52.6 GHz–114.25 GHz), OFDM waveforms with very large subcarrier spacing (960 kHz–3.84 MHz) are required to combat severe phase noise. Due to the large subcarrier spacing, the slot length tends to be short. In lower frequency bands (referred to as FR2) (24.25 GHz to 52.6 GHz) with a 120 kHz SCS, the slot length is 125 microseconds (μSec), while in FR4 with 960 kHz, the slot length is 15.6 μSec.

[0031] Figure 1An example wireless communication network 100 in which aspects of this disclosure may be implemented is shown. For example, a base station (BS) 110a of the wireless communication network 100 may be configured with a CSI prediction module 112, which is configured to perform (or assist BS 110 in performing) aspects of this disclosure. Figure 6 Operation 600 configures UE 120 to predict Channel State Information (CSI) and qualify the predicted CSI. Similarly, UE 120a of wireless communication network 100 may be configured with a CSI prediction module 122, which is configured to perform (or assist UE 120 in performing) according to various aspects of this disclosure. Figure 5 Operation 500 is used to predict Channel State Information (CSI) and to qualify the predicted CSI.

[0032] The wireless communication network 100 can be an NR system (e.g., a 5G NR network). For example... Figure 1 As shown, the wireless communication network 100 can communicate with the core network 132. The core network 132 can communicate with one or more base stations (BS) 110 and / or user equipment (UE) 120 in the wireless communication network 100 via one or more interfaces.

[0033] like Figure 1 As shown, the wireless communication network 100 may include several BSs 110a-110z (each BS is also individually referred to herein as BS 110 or collectively as BS 110) and other network entities. BS 110 may provide communication coverage for a specific geographic area, sometimes referred to as a "cell," which may be stationary or mobile depending on the location of the mobile BS 110. In some examples, BS 110 may use any suitable transport network to interconnect with each other and / or with one or more other BSs or network nodes (not shown) within the wireless communication network 100 via various types of backhaul interfaces (e.g., direct physical connection, wireless connection, virtual network, etc.). Figure 1 In the example shown, BS 110a, 110b, and 110c can be macro BSs for macro cells 102a, 102b, and 102c, respectively. BS 110x can be a pico BS for pico cell 102x. BS 110y and 110z can be femto BSs for femto cells 102y and 102z, respectively. A BS can support one or more cells. Network controller 130 can be coupled to a set of BS 110s and provide coordination and control (e.g., via backhaul) for these BS 110s.

[0034] BS 110 communicates with UEs 120a-120y (each UE is also individually referred to herein as UE120 or collectively as UE 120) in the wireless communication network 100. UEs 120 (e.g., 120x, 120y, etc.) may be distributed throughout the wireless communication network 100, and each UE 120 may be stationary or mobile. The wireless communication network 100 may also include relay stations (e.g., relay station 110r, also referred to as a relay, etc.) that receive data and / or other information transmissions from upstream stations (e.g., BS 110a or UE 120r) and send data and / or other information transmissions to downstream stations (e.g., UE 120 or BS 110), or relay transmissions between UEs 120 to facilitate communication between devices.

[0035] Figure 2 BS 110a and UE 120a are shown (e.g., in...) Figure 1 Example components in a wireless communication network 100, which can be used to implement various aspects of this disclosure.

[0036] At BS 110a, the transmitting processor 220 can receive data from the data source 212 and control information from the controller / processor 240. This control information can be used for the Physical Broadcast Channel (PBCH), Physical Control Format Indicator Channel (PCFICH), Physical Hybrid ARQ Indicator Channel (PHICH), Physical Downlink Control Channel (PDCCH), Group Common PDCCH (GC PDCCH), etc. This data can be used for the Physical Downlink Shared Channel (PDSCH), etc. The Media Access Control (MAC)-Control Element (MAC-CE) is a MAC layer communication structure that can be used for exchanging control commands between wireless nodes. The MAC-CE can be carried in shared channels such as the Physical Downlink Shared Channel (PDSCH), Physical Uplink Shared Channel (PUSCH), or Physical Sidelink Shared Channel (PSSCH).

[0037] Processor 220 can process (e.g., encode and symbol map) the data and control information to obtain data symbols and control symbols, respectively. Transmit processor 220 can also generate reference symbols for signals such as the primary synchronization signal (PSS), secondary synchronization signal (SSS), and channel state information reference signal (CSI-RS). Where applicable, transmit (TX) multiple-input multiple-output (MIMO) processor 230 can perform spatial processing (e.g., precoding) on ​​the data symbols, control symbols, and / or reference symbols, and can provide output symbol streams to modulators (MODs) 232a-232t. Each modulator 232 can process its corresponding output symbol stream (e.g., for OFDM, etc.) to obtain an output sample stream. Each modulator can further process (e.g., convert to analog, amplify, filter, and up-convert) the output sample stream to obtain a downlink signal. The downlink signal from modulators 232a-232t can be transmitted via antennas 234a-234t, respectively.

[0038] At UE 120a, antennas 252a-252r can receive downlink signals from BS 110a and can provide the received signals to demodulators (DEMODs) in transceivers 254a-254r respectively. Each demodulator 254 can adjust (e.g., filter, amplify, downconvert, and digitize) the corresponding received signal to obtain an input sample. Each demodulator can further process the input sample (e.g., for OFDM, etc.) to obtain the received symbols. MIMO detector 256 can obtain the received symbols from all demodulators 254a-254r, perform MIMO detection on the received symbols where applicable, and provide the detected symbols. Receiver processor 258 can process (e.g., demodulate, deinterleave, and decode) the detected symbols, provide decoded data for UE 120a to data sink 260, and provide decoded control information to controller / processor 280.

[0039] On the uplink, at UE 120a, the transmitting processor 264 can receive and process data from data source 262 (e.g., for the Physical Uplink Shared Channel (PUSCH)) and control information from controller / processor 280 (e.g., for the Physical Uplink Control Channel (PUCCH)). The transmitting processor 264 can also generate reference symbols for reference signals (e.g., for the Sounding Reference Signal (SRS)). Where applicable, symbols from the transmitting processor 264 can be pre-encoded by the TXMIMO processor 266, further processed by demodulators in transceivers 254a-254r (e.g., for SC-FDM, etc.), and transmitted to BS 110a. At BS 110a, uplink signals from UE 120a can be received by antenna 234, processed by demodulator 232, detected by MIMO detector 236 where applicable, and further processed by receiving processor 238 to obtain decoded data and control information transmitted by UE 120a. The receiver processor 238 can provide decoded data to the data sink 239 and decoded control information to the controller / processor 240.

[0040] Memory 242 and 282 can store data and program code for BS 110a and UE 120a, respectively. Scheduler 244 can schedule UE for data transmission on downlink and / or uplink.

[0041] The antenna 252, processors 266, 258, 264, and / or controller / processor 280 of UE 120a, and / or the antenna 234, processors 220, 230, 238, and / or controller / processor 240 of BS110a, can be used to perform the various techniques and methods described herein. For example, as Figure 2 As shown, the controller / processor 240 of BS 110a has a CSI prediction module 241, and the controller / processor 280 of UE 120a has a CSI prediction module 281. The CSI prediction module 241 can be configured to perform... Figure 6 Operation 600, and / or CSI prediction module 281 can be configured to perform Figure 5 Operation 500. Although shown at the controller / processor, other components of UE 120a and BS 110a can be used to perform the operations described herein.

[0042] NR can utilize Orthogonal Frequency Division Multiplexing (OFDM) with a cyclic prefix (CP) on both the uplink and downlink. NR can support half-duplex operation using Time Division Duplex (TDD). OFDM and Single-Carrier Frequency Division Multiplexing (SC-FDM) divide the system bandwidth into multiple orthogonal subcarriers, which are often referred to as tones, bins, etc. Each subcarrier can be modulated with data. Modulation symbols can be transmitted in the frequency domain using OFDM and in the time domain using SC-FDM. The spacing between adjacent subcarriers can be fixed, and the total number of subcarriers can depend on the system bandwidth. The minimum resource allocation (called a resource block (RB)) can be 12 consecutive subcarriers. The system bandwidth can also be divided into subbands. For example, a subband can cover multiple RBs. NR can support a basic subcarrier spacing (SCS) of 15 kHz, and other SCSs can be defined relative to the basic SCS (e.g., 30 kHz, 60 kHz, 120 kHz, 240 kHz, etc.). The symbol and time slot length scale with the subcarrier spacing. The CP length also depends on the subcarrier spacing.

[0043] Figure 3 This is a diagram illustrating an example of frame format 300 for NR. The transmission timeline for each of the downlink and uplink can be divided in units of radio frames. Each radio frame can have a predetermined duration (e.g., 10 ms) and can be divided into 10 subframes, each 1 ms long, indexed from 0 to 9. Each subframe can include a variable number of time slots (e.g., 1, 2, 4, 8, 16… time slots), depending on the SCS. Each time slot can include a variable number of symbol periods (e.g., 7 or 14 symbols), depending on the SCS. The symbol periods within each time slot can be assigned an index. Micro-slots can be referred to as sub-slot structures, which refer to transmission time intervals with a duration less than a time slot (e.g., 2, 3, or 4 symbols).

[0044] Each symbol in a time slot can indicate the link direction used for data transmission (e.g., DL, UL, or flexible), and the link direction for each subframe can be dynamically switched. The link direction can be based on the time slot format. Each time slot can include DL / UL data as well as DL / UL control information.

[0045] In NR, a Synchronization Signal (SSB) block is transmitted. The SS block includes the PSS, SSS, and a two-symbol PBCH. The SS block can be transmitted at fixed time slot locations. The PSS and SSS are available to the UE for cell search and acquisition. The PSS provides half-frame timing, and the SS provides CP length and frame timing. The PSS and SSS provide cell identification. The PBCH carries basic system information such as downlink system bandwidth, intra-radio frame timing information, SS burst set periodicity, and system frame number.

[0046] Additional system information (such as Residual Minimum System Information (RMSI), System Information Block (SIB), and Other System Information (OSI)) can be transmitted in certain subframes on the Physical Downlink Shared Channel (PDSCH).

[0047] Example of Channel State Information (CSI) Prediction Based on ML

[0048] Channel State Information (CSI) can be used for various purposes, such as link adaptation via Adaptive Modulation and Coding (AMC), multi-user diversity (MUD) via Proportional Fairness (PF) scheduling, and multi-user (MU) Multiple-Input Multiple-Output (MIMO) (MU-MIMO) communication. To facilitate CSI feedback over wireless network links, the transmitter typically arranges a series of pilot timings for one or more receivers to measure the CSI, and the transmitter and receiver arrange a series of feedback timings to collect CSI feedback (i.e., for the receiver to send CSI feedback and for the transmitter to receive CSI feedback from the receiver). The transmitter can use the CSI feedback received at time t to send it to the receiver at time t+Δ2.

[0049] Typically, to minimize overhead, the periodicity of the feedback timing can be the same as or greater than the periodicity of the pilot timing. Furthermore, during the feedback timing, the UE can report compressed or quantized CSI. Typically, to report quantized CSI, the receiver can report the CSI measured before t-Δ1 at time t.

[0050] Because channel conditions can change between receiving the reference signal or pilot signal (e.g., during pilot timing) and transmitting the CSI report at time t, the time gap between the CSI measurement and the transmission of subsequent signals based on the CSI measurement may result in inaccurate CSI being used to perform transmissions between the transmitter and receiver. That is, a CSI measurement at time t-Δ1 and a transmission at time t+Δ2 may result in a time delay of Δ1+Δ2 between the measurement and transmission, and therefore, due to CSI aging (e.g., changes in channel conditions during the Δ1+Δ2 time delay), the transmitter may potentially use outdated or inaccurate CSI to perform a transmission at time t+Δ2. To account for CSI aging during the Δ1+Δ2 time delay between the CSI measurement and the transmission based on the measured CSI, the transmitter can use a back-off procedure to perform transmissions to the receiver. The back-off procedure typically allows the transmitter to use a more conservative transmission pattern than that indicated by CSI feedback (e.g., based on the CSI measurement at time t-Δ1). For example, the transmitter can use a lower modulation and decoding scheme (MCS), a reduced rank, etc., for transmission at time t+Δ2 than the CSI feedback generated based on CSI measurements at time t-Δ1 and reported at time t.

[0051] In some cases, CSI feedback can take into account CSI aging and the possible difference in channel state between a measurement at time t-Δ1 and a CSI feedback-based transmission at time t+Δ2. Various techniques can be used to account for CSI aging, and these techniques can assume, for example, the amount of change in CSI between a measurement at time t-Δ1 and a CSI feedback-based transmission at time t+Δ2. In some embodiments, machine learning models can be used to predict CSI at time t+Δ2 and account for CSI aging. These machine learning models can be, for example, Long Short-Term Memory (LSTM) models, Gated Recurrent Units (GRUs), neural networks, or other machine learning models that allow for prediction of CSI at future time points.

[0052] like Figure 4As shown, a machine learning (ML)-based (CSI prediction) model can be trained to learn the relationship between the CSI measured at a first time, the CSI measured at a second time, and the time delay between the first and second times. CSI measurements can be performed at the UE and reported to the gNB. These measurements can be added to a training dataset that can be used to train (or retrain) the ML-based CSI prediction model. Since the dataset (which correlates the CSI measured at the first and second times with the time delay between the first and second times) can be very large, sharing the entire dataset with the UE is not feasible. In some cases, instead of sharing the dataset, a more practical approach is to use the dataset to train a model such as a neural network (NN), and then share the neural network model and the parameters (e.g., weights, etc.) used for the trained model with the UE. The UE can then use the trained NN to predict CSI based on the time delay between the CSI measurement and the transmission of CSI feedback reported to the gNB.

[0053] Various techniques can be used to train such machine learning models to learn how to predict CSI based on the time delay between the measurement of CSI and the transmission of CSI feedback reported to the gNB. Given the input of a CSI measurement and the time delay between the measurement and the transmission of CSI feedback reported to the gNB, the trained machine learning model can predict the CSI at the time when the gNB is expected to perform the transmission.

[0054] In some embodiments, supervised learning techniques can be used to train a machine learning model. In this technique, an input dataset of n tuples {CSI measurements at time t-Δ1, CSI measurements at time t+Δ2, and time delay Δ1+Δ2} is used to train the machine learning model to identify the relationship between the time delay and the predicted CSI at time t+Δ2. The input dataset can be, for example, the CSI measurement performed at time t-Δ1 and the time delay Δ1+Δ2, and the machine learning model can be trained to output a prediction of the channel condition at time t+Δ2 (i.e., the expected time for the gNB to perform transmission to the UE based on CSI feedback provided by the UE and generated based on the CSI measurement at time t-Δ1). In some embodiments, the machine learning model can include a softmax layer that generates a probability score for each of the plurality of CSI values. The values(s) that can be identified as the predicted CSI at time t+Δ2 can therefore be n values ​​with the highest probability scores generated by the machine learning model.

[0055] The machine learning model can be trained offline and deployed to the user equipment (UE) to predict CSI and report CSI feedback based on the predicted CSI. The gNB can use this machine learning model to determine the parameters for sending data to the UE. During deployment, the UE can independently measure CSI at times t-Δ1 and t+Δ2 and determine the accuracy of the CSI prediction at t+Δ2. If the prediction is consistently inaccurate due to exceeding a threshold, the UE can determine that the model is inaccurate and can use the recorded CSI to update the training dataset and guide the gNB to reconfigure the UE with the updated, retrained machine learning model.

[0056] The machine learning described in this article can use any appropriate machine learning algorithm. In some non-limiting examples, the machine learning algorithm is a supervised learning algorithm, an unsupervised learning algorithm, a reinforcement learning algorithm, a deep learning algorithm, an artificial neural network algorithm, or other types of machine learning algorithm.

[0057] In some examples, machine learning is performed using deep convolutional networks (DCNs). A DCN is a network of convolutional networks configured with additional pooling and normalization layers. DCNs have achieved state-of-the-art performance on many tasks. DCNs can be trained using supervised learning, where both the input and output targets are known for many paradigms and are used to modify the network's weights using gradient descent. DCNs can be feedforward networks. Furthermore, as mentioned above, the connections from neurons in the first layer of a DCN to groups of neurons in the next higher layer are shared across neurons in the first layer. The feedforward and shared connections of a DCN can be leveraged for fast processing. For example, the computational cost of a DCN may be significantly less than that of a similarly sized neural network that includes recursive or feedback connections.

[0058] In some examples, machine learning is performed using neural networks. Neural networks can be designed with a wide variety of connection patterns. In feedforward networks, information is passed from lower to higher layers, where each neuron in a given layer communicates with neurons in higher layers. Hierarchical representations can be built into successive layers of a feedforward network. Neural networks can also have recursive or feedback (also known as top-down) connections. In recursive connections, the output from a neuron in a given layer can communicate with another neuron in the same layer. Recursive architectures can help identify patterns across more than one chunk of input data that is passed sequentially to the neural network. Connections from neurons in a given layer to neurons in lower layers are called feedback (or top-down) connections. Networks with many feedback connections can be helpful when recognizing high-level concepts can help discern specific low-level features of the input.

[0059] Artificial neural networks, which can be composed of interconnected groups of artificial neurons (e.g., neuron models), are computing devices or representations of methods performed by computing devices. These neural networks can be used in a variety of applications and / or devices, such as Internet Protocol (IP) cameras, Internet of Things (IoT) devices, autonomous vehicles, and / or service robots. Individual nodes in an artificial neural network can mimic biological neurons by taking input data and performing simple operations on that data. The results of the simple operations performed on the input data are selectively passed to other neurons. Weights are associated with each vector and node in the network, and these values ​​constrain the correlation between the input and output data. For example, the input data for each node can be multiplied by a corresponding weight value, and the products can be summed. The sum of the products can be adjusted by optional biases, and activation functions can be applied to the results, producing the node's output signal or "output activation." The weights can initially be determined from training data through an iterative flow of the network (e.g., the weights are established during a training phase, in which the network learns how to identify a particular category based on the typical characteristics of its input data).

[0060] Different types of artificial neural networks can be used to implement machine learning, such as recurrent neural networks (RNNs), multilayer perceptron (MLP) neural networks, and convolutional neural networks (CNNs). RNNs work by storing the output of a layer and feeding that output back into the input to help predict the outcome of that layer. In an MLP neural network, data can be fed into the input layer, and one or more hidden layers provide a hierarchy of abstraction for the data. Predictions can then be made on the output layer based on this abstraction. MLPs are particularly well-suited for classification prediction problems where the input is assigned a category or label. A convolutional neural network (CNN) is a type of feedforward artificial neural network. A CNN can comprise an array of artificial neurons, each with a receptive field (e.g., a localized region of the input space) that collectively tile the input space. CNNs have numerous applications. In particular, CNNs have been widely used in pattern recognition and classification. In a hierarchical neural network architecture, the output of the first layer of artificial neurons becomes the input of the second layer, the output of the second layer becomes the input of the third layer, and so on. CNNs can be trained to recognize hierarchical features. The computation in a convolutional neural network architecture can be distributed across a large number of processing nodes, which can be configured in one or more computation chains. These multi-layer architectures can be trained layer by layer at a time and can be fine-tuned using backpropagation.

[0061] Example of qualification assessment for machine learning-based channel state information prediction

[0062] Various aspects of this disclosure provide techniques for using machine learning models to predict channel state information (CSI) and for qualifying CSI predictions made using machine learning models. For example, these techniques can be used to, for instance, allow CSI predictions to be used in transmissions between network entities (e.g., gNodeB) and UEs, and can lead to more efficient use of wireless communication resources, improved reliability of communication between network entities and UEs, etc.

[0063] As discussed, using machine learning models to predict CSI at the time the network entity will send it to the UE can lead to improved accuracy of the CSI used to determine transmission parameters, and thus can lead to, for example, increased throughput, improved reliability, and other improvements to wireless communication compared to transmission parameters identified using a priori-defined CSI backoff schemes. Furthermore, network entities can improve network efficiency by learning the accuracy of CSI reports generated by the UE, and can do so faster based on the ratio of predicted CSI to the accuracy of CSI reported by the UE a posteriori (e.g., based on ACK / NACK statistics, which could result in a significant reduction in network throughput due to several failed transmissions and retransmissions).

[0064] However, because generating predictions based on machine learning models can be a more resource-intensive process than using prior-defined backoff schemes, reporting inaccurate predictions can lead to limited or no improvement in the use of computational resources and power for communication between network entities and the UE (or even a degradation in network performance). For example, reporting inaccurate CSI (which may include inaccurate early Hybrid Automatic Repeat Request (HARQ) feedback) can negatively impact network performance. For instance, a correct early NACK can trigger an early retransmission of a NACKed transmission, which may help meet latency budgets; however, an incorrect early NACK (e.g., a false alarm) may result in wasted retransmissions. In another example, a correct early ACK can trigger an earlier release of redundant resources (and correspondingly increase available resources, which can be beneficial in a loaded network); however, an incorrect early NACK may result in packet loss at Media Access Control (MAC) or other higher layers.

[0065] Therefore, aspects of this disclosure can allow for the prediction and qualification of CSI predictions to minimize, for example, false alarms, false detections, and other conditions that may degrade network performance (e.g., in high-load networks, where inaccurate predictions can have a significant negative impact on network performance). By predicting CSIs and identifying whether the prediction is qualified or unqualified (e.g., whether the predicted CSI is within an accuracy threshold, or whether the predicted CSI falls into one or more categories considered sufficiently accurate), network performance can be improved by increasing throughput, reducing the negative impact of inaccurate reporting, and so on.

[0066] Figure 5 An example operation 500 for wireless communication performed by a user equipment (UE) is shown. For example, operation 500 can be performed by... Figure 1 or Figure 2 The UE 120 is used to perform the function to predict the CSI in accordance with certain aspects of this disclosure and to quantify the accuracy of the predicted CSI.

[0067] As shown, operation 500 begins at box 502, where the UE receives a Channel State Information (CSI) prediction model from the network entity for the quantized CSI. The CSI prediction model can be a machine learning model, which the UE can execute to predict the CSI based on the time delay between the current CSI measurement and the time when the network entity will perform transmission based on the CSI feedback reported to the network entity.

[0068] At box 504, the UE calculates CSI based on downlink reference signal measurements.

[0069] At box 506, the UE generates a quantized CSI difference based on the quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model. Typically, the quantized CSI difference can be a value indicating the relative accuracy level of the CSI predicted based on the CSI prediction model. For example, the rule used to quantize the CSI difference can quantize the difference into one of two values ​​representing a coarse classification of the difference (e.g., within a threshold amount or not within a threshold amount). In another example, the rule used to quantize the CSI difference can quantize the difference into one of multiple values ​​representing a finer-grained classification of the difference (e.g., highly accurate, accurate, acceptable, inaccurate, highly inaccurate).

[0070] At box 508, the UE reports the difference between the calculated CSI and the quantized CSI to the network entity.

[0071] Figure 6 An example operation 600 for wireless communication performed by a network entity is shown. Operation 600 can be considered as an extension of what has been discussed above. Figure 5 This is a supplement to operation 500. For example, operation 600 can be... Figure 1 or Figure 2 The BS 110a (e.g., gNB) is used to configure the UE to predict the CSI and quantify the accuracy of the predicted CSI according to certain aspects of this disclosure, and to perform transmission to the UE using the predicted CSI and the quantized accuracy of the predicted CSI.

[0072] As shown, operation 600 begins at block 602, where the network entity sends a CSI prediction model for predicting Channel State Information (CSI) and one or more reference signals to the User Equipment (UE) for use in calculating the CSI and generating one or more CSI differences (e.g., generating one or more CSI differences based on the calculated CSI and the CSI predicted by the CSI prediction model). Typically, the network entity can configure the UE to predict CSI and quantify one or more CSI differences by sending a configuration message to the UE, which includes the CSI prediction model and a qualification scheme that the UE can use to quantify the accuracy of the CSI predicted using the CSI prediction model. In some embodiments, when the network entity is configured with the CSI prediction model and qualification rules for quantifying CSI differences, the network entity can implicitly configure the UE to report the calculated CSI and the quantified CSI differences. In some embodiments, the UE can explicitly configure itself to report the calculated CSI and the quantified CSI differences, for example, based on a qualification session, as discussed in further detail below.

[0073] At box 604, the network entity receives the calculated CSI and the quantized CSI difference from the UE. The quantized CSI difference can be a quantized value generated based on a CSI difference that represents the difference between the current CSI measurement and a predicted CSI generated using a previous CSI measurement and the time delay between the previous CSI measurement and the time when the current CSI measurement was generated. As discussed, the quantized CSI difference can be a value generated based on the qualification scheme discussed above with respect to box 602.

[0074] At box 606, the network entity communicates with the UE based on the difference between the calculated CSI and the quantized CSI. For example, the network entity may adjust the modulation and decoding scheme, the rank indicator, or other parameters used to perform the transmission with the UE.

[0075] Figure 5 and Figure 6 For operations of 500 and 600, please refer to the following: Figure 9 and Figure 10 Use the call flow graph shown to understand this. In other words, Figure 9 and Figure 10 The UE and gNB can execute Figure 6 and Figure 7 Operations 500 and 600.

[0076] Figure 7 Example timelines for measuring CSI and predicting CSI based on the measured CSI are shown. Timeline 700 shows four examples 702a-702d of CSI measurement and prediction.

[0077] Typically, CSI measurement and prediction instance 702 can be based on the expected time when the network entity will transmit data and the time when the UE will measure CSI and generate a CSI feedback report, which the network entity can use to identify the parameters used to send signals to the UE. These instances can represent CSI measurements performed during training or qualification processes (discussed in further detail below) or during communication with the network entity.

[0078] Given a time t representing the time when the network entity receives CSI feedback from the UE, the UE can measure the CSI and generate a CSI feedback report at time t-Δ1 and send it to the network entity. Time t+Δ2 represents the time when the network entity will perform the transmission to the UE. Because the channel conditions (and therefore the measured CSI) may change during the timing gap Δ1+Δ2 between the time the UE reports the CSI to the network entity and the time the network entity will perform the transmission to the UE, the UE can use a predictive model to predict the CSI at time t+Δ2, which takes the measured CSI at time t-Δ1 and the duration of the timing gap Δ1+Δ2 as input.

[0079] In each instance 702, the UE can measure the CSI at time t+Δ2 and calculate the difference between the measured CSI and the predicted CSI at time t+Δ2. The difference between the measured CSI and the predicted CSI at time t+Δ2 can be quantified according to a qualification scheme that categorizes the difference into one of several categories. In some aspects, the quantified difference between the measured CSI and the predicted CSI at time t+Δ2 can be reported individually for each instance 702 in timeline 700.

[0080] In some aspects, the conformity assessment scheme can specify that the quantified difference will be generated based on a moving average over a time window W. For example, as Figure 7As shown, window W can cover four instances 702 of CSI measurement and prediction. In this case, for each instance 702a-702d in window W, the difference between the measured CSI and the predicted CSI at time t+Δ2 can be averaged, and a qualification scheme can be used to quantify the average difference between the measured CSI and the predicted CSI at time t+Δ2. By averaging the difference between the measured CSI and the predicted CSI at time t+Δ2 over the mobile window W, the UE can consider the CSI prediction made by the UE and may not report the CSI prediction to the network entity (e.g., when the CSI of the first carrier is based on a measurement on the second carrier, etc., because the UE is configured with a low-overhead configuration for CSI feedback).

[0081] In some aspects, the qualification scheme can specify that the quantified difference will be generated as the average of the most recent n CSI predictions. In this case, the number of CSI predictions used to generate the quantified difference can remain constant, unlike the use of a sliding time window W, in which any number of predictions can be considered.

[0082] In some aspects, the qualification scheme can define the difference between the calculated (or measured) CSI and the predicted CSI as the difference between the calculated CSI and the CSI predicted based on the CSI prediction model, filtered by an infinite impulse response (IIR). The IIR-filtered difference can be used to account for the previously calculated difference between the measured CSI and the predicted CSI at time t+Δ2 by using these previously calculated differences entering the IIR filter.

[0083] The quantized CSI difference can be quantized using a qualification scheme that quantizes the CSI difference into at least one of two categories. A one-bit indicator used for the quantized CSI difference can represent a coarse classification, where one value indicates that the CSI difference is less than a priori defined threshold (e.g., indicating that the CSI difference corresponds to an accurate prediction), which can be a value defined by the network configuration. Another value of the one-bit indicator can indicate that the CSI difference exceeds a priori defined threshold (e.g., indicating that the CSI difference corresponds to an inaccurate prediction). A larger number of bits used to represent the quantized CSI difference can allow for increased granularity in representing the accuracy level of the CSI difference. Typically, one or more values ​​of the quantized CSI difference can correspond to predicted CSIs that are "qualified" predictions (e.g., predictions that meet the accuracy threshold level), and other values ​​of the quantized CSI difference can correspond to predicted CSIs that are "unqualified" predictions (e.g., predictions that do not meet the accuracy threshold level).

[0084] In some aspects, the UE can be configured to report predicted CSIs based on a gating function. If the predicted CSI corresponds to a qualified prediction, the gating function can, for example, allow reporting the predicted CSI. If the predicted CSI corresponds to an unqualified prediction, it can choose not to report the predicted CSI. In some embodiments, the gating function can indicate that several consecutive qualified CSI predictions may be required before the UE can report the predicted CSI to a network entity.

[0085] If the UE does not report the predicted CSI or reports an unqualified prediction, the network entity can determine that the CSI prediction model did not accurately predict the CSI at time t+Δ2. Therefore, the network entity can use other techniques to account for CSI aging. For example, as discussed above, the network entity can use a priori defined backoff schemes to send to the UE a modulation and decoding scheme or rank indicator that is different from the one associated with the reported CSI measurement.

[0086] In some respects, the predicted CSI can be used for early HARQ feedback. In this case, the qualification scheme can be defined in terms of the false detection rate or false alarm rate. As discussed above, the false detection rate can represent the ratio of erroneous early ACKs sent to network entities for packets (e.g., ACKs are sent for packets that are not correctly received, which may result in packet loss due to early release of resources). The false alarm rate can represent the ratio of erroneous early NACKs sent to network entities for packets (e.g., NACKs are sent for packets that are correctly received). If the false detection rate or false alarm rate is below a threshold (e.g., the false detection rate or false alarm rate is a qualified rate), the qualification scheme can configure the UE to report the predicted CSI, and if the false detection rate or false alarm rate is above a threshold (e.g., the false detection rate or false alarm rate is an unqualified rate), then the predicted CSI is not reported. The false detection rate or false alarm rate can be calculated as a moving average over a time window W, the average false detection rate or false alarm rate of N predictions, or a value filtered by IIR. In some cases, network entities can configure the UE to report early HARQ feedback when the feedback regarding the false detection rate or both the false detection rate and the false alarm rate is qualified.

[0087] Figure 8 The diagram shows a call flow of messages that can be exchanged between UE 802 and gNB 804 (or other network entities) to configure UE 802 to predict CSI based on a CSI prediction model and to quantify the accuracy of CSI predictions based on a qualification scheme.

[0088] As shown, in order to configure UE 802 to predict CSI based on a CSI prediction model, gNB 804 sends the CSI prediction model and a qualification scheme 810 to the UE. As discussed above, the CSI prediction model can be a machine learning model configured to be executed on the UE and predict future CSI given a current CSI measurement and a time delay as input. As discussed above, the qualification scheme typically identifies qualification rules used to determine whether a predicted CSI is a qualified prediction or an unqualified prediction.

[0089] After UE 802 is configured with the CSI prediction model and qualification scheme 810, UE 802 can be ready to perform CSI measurements and predict CSI based on the measured CSI. For this purpose, gNB 804 can send a reference signal 812 to UE 802. As discussed above, gNB 804 can configure UE 802 with information regarding the timing of the measurement when the reference signal is sent to UE 802 and the timing of the report when a CSI feedback report will be sent to gNB 804. gNB 804 can modify one or more transmission parameters based on the received CSI feedback report and use these transmission parameters for transmission to UE 802.

[0090] At 814, the UE measures the CSI based on a reference signal. The UE also uses a predictive model to predict the CSI at future times. As discussed, the predictive model uses the measured CSI and the time gap between the measurement based on reference signal 812 and the expected time when another transmission will be received from gNB 804 (e.g., in this example, when gNB 804 sends reference signal 816 to UE 802).

[0091] Subsequently, gNB 804 sends reference signal 816 to UE 802. At 818, the UE measures the CSI based on the reference signal. The UE also calculates the difference between the predicted CSI generated at 814 and the measured CSI generated by measuring the reference signal 816 and quantizes the difference based on the qualification scheme received from gNB 804 in message 810. As discussed, UE 802 can generate one of several values ​​from the quantized difference based on rules included in the qualification scheme used to classify the difference. Some of the quantized values ​​may correspond to qualified predictions, and others may correspond to unqualified predictions.

[0092] After calculating and quantizing the difference between the predicted CSI generated at 814 and the measured CSI generated by the measurement reference signal 816, UE 802 generates and transmits message 820, which includes the measured CSI and the quantized difference 820. gNB 804 can use the measured CSI and the quantized difference 820 to determine whether the CSI prediction model accurately generates the prediction of channel state information. If the CSI prediction model accurately generates the prediction of channel state information, gNB 804 can then use the received CSI measurement from UE 802 to predict the CSI based on the received measurement and the time delay between the measurement and the transmission of the signal to UE 802, and adjust the transmission parameters used for transmission to UE (e.g., adjust the modulation and decoding scheme, rank indicator, etc.).

[0093] In some aspects, the gNB can arrange a session to qualify CSI predictions for the UE. During the session, the UE can send a series of predicted and measured CSI values ​​to the gNB (which can be referred to as "prediction and ground-truth pairs"). The gNB can configure the UE with a CSI prediction model before initiating the session to qualify the CSI predictions, but it can configure the UE without a qualification scheme. When the session is active, the gNB can qualify the predictions based on the qualification scheme and can add information about the predicted CSIs, measured CSIs, and whether the predictions are qualified to the training dataset for retraining and refining the CSI prediction model. After the session ends, the gNB can configure the UE to report qualified predictions (e.g., by sending the UE information identifying predictions made during the session that were determined to be qualified).

[0094] In some respects, varying parameters can be used to generate a series of prediction and benchmark-truth pairs. For example, different time delays (e.g., different values ​​of Δ1+Δ2) between measurements at the UE and the transmission of signals from the gNB can be used to generate prediction and benchmark-truth pairs. In another example, a series of prediction and benchmark-truth pairs can be based on one or more of the following: early HARQACK / NACK statistics, precoding matrix indicator (PMI), channel quality indicator (CQI), rank indicator (RI), signal-to-noise ratio (SNR), or other measurements or statistics that can provide information.

[0095] Figure 9 The diagram shows a call flow diagram of messages that can be exchanged between UE 902 and gNB 904 (or other network entities) to arrange a session for qualification of CSI prediction for the UE, as discussed above.

[0096] As shown, gNB 904 can send CSI prediction model 910 to UE 902. In some aspects, sending CSI prediction model 910 may be accompanied by an explicit indication that gNB 904 is initiating a session to qualify the CSI prediction for UE 902. In some aspects, sending CSI prediction model in the absence of a qualification scheme may implicitly indicate that gNB 904 is initiating a session to qualify the CSI prediction for UE 902.

[0097] Subsequently, gNB 904 can send reference signal 912 to UE 902. At 914, the UE can measure CSI based on reference signal 912 and can use a prediction model to predict CSI at a future time. As discussed, the prediction model uses the measured CSI and the time gap between the measurement based on reference signal 912 and the expected time from which another transmission will be received from gNB 904. In some aspects, the time gap can be a hypothetical amount of time specified by gNB 904 for predicting CSI.

[0098] Subsequently, gNB 904 sends a reference signal 916 to UE 902. At 918, the UE measures the CSI based on the reference signal. The UE also calculates the difference between the predicted CSI generated at 914 and the measured CSI generated by measuring the reference signal 916. Then, UE 902 can send a message 920 including the predicted CSI and the measured CSI to gNB 904 for gNB to use in configuring an authentication scheme for the UE.

[0099] Typically, reference signals 912 and 916, and their corresponding operations 914 and 918 at UE 902, can be repeated until the session for certifying the CSI prediction for UE 902 is completed. After the session completes, at 922, the gNB determines whether the difference reported by the UE indicates whether the corresponding CSI prediction is certified or uncertified. After determining whether the prediction is certified or uncertified, gNB 904 can generate a certified scheme configuration and send the certified scheme configuration 924 to UE 902 for subsequent use.

[0100] Figure 10 A communication device 1000 (e.g., a user equipment) is shown, which may include operations configured to perform the techniques disclosed herein (such as...). Figure 5The communication device 1000 includes various components (e.g., corresponding to component plus functional components) shown in the diagram. The communication device 1000 includes a processing system 1002 coupled to a transceiver 1008 (e.g., a transmitter and / or receiver). The transceiver 1008 is configured to transmit and receive signals for the communication device 1000 via an antenna 1010, such as the various signals described herein. The processing system 1002 may be configured to perform processing functions for the communication device 1000, including processing signals received by and / or to be transmitted by the communication device 1000.

[0101] Processing system 1002 includes processor 1004 coupled to computer-readable medium / memory 1012 via bus 1006. In some aspects, computer-readable medium / memory 1012 is configured to store instructions (e.g., computer-executable code) that, when executed by processor 1004, cause processor 1004 to perform. Figure 5 The operations shown herein or other operations used to perform the various techniques discussed herein. In some aspects, the computer-readable medium / memory 1012 stores code 1014 for receiving a CSI prediction model for quantized channel state information (CSI), code 1016 for calculating the CSI based on downlink reference signal measurements, code 1018 for generating a quantized CSI difference based on the quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model, and code 1020 for reporting the calculated CSI and the quantized CSI difference. In some aspects, the processor 1004 has circuitry configured to implement the code stored in the computer-readable medium / memory 1012. Processor 1004 includes circuitry 1022 for receiving a CSI prediction model for quantized channel state information (CSI), circuitry 1024 for calculating CSI based on downlink reference signal measurements, circuitry 1026 for generating a quantized CSI difference based on quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model, and circuitry 1028 for reporting the calculated CSI and the quantized CSI difference.

[0102] Figure 11 A communication device 1100 (e.g., a network entity, such as a gNodeB) is shown, which may include operations configured to perform the techniques disclosed herein (such as...). Figure 6The communication device 1100 includes various components (e.g., corresponding to component plus functional components) shown in the diagram. The communication device 1100 includes a processing system 1102 coupled to a transceiver 1108 (e.g., a transmitter and / or receiver). The transceiver 1108 is configured to transmit and receive signals for the communication device 1100 via an antenna 1110, such as the various signals described herein. The processing system 1102 may be configured to perform processing functions for the communication device 1100, including processing signals received by and / or to be transmitted by the communication device 1100.

[0103] Processing system 1102 includes processor 1104 coupled to computer-readable medium / memory 1112 via bus 1106. In some aspects, computer-readable medium / memory 1112 is configured to store instructions (e.g., computer-executable code) that, when executed by processor 1104, cause processor 1104 to perform. Figure 6 The operations shown herein or other operations used to perform the various techniques discussed herein. In some aspects, the computer-readable medium / memory 1112 stores code 1114 for transmitting to a user equipment (UE) a CSI prediction model for predicting channel state information (CSI) and one or more reference signals; code 1116 for receiving a calculated CSI and a quantized CSI difference; and code 1118 for communicating with the UE based on the calculated CSI and the quantized CSI difference. In some aspects, the processor 1104 has circuitry configured to implement the code stored in the computer-readable medium / memory 1112. The processor 1104 includes circuitry 1120 for transmitting to the user equipment (UE) a CSI prediction model for predicting channel state information (CSI) and one or more reference signals; circuitry 1122 for receiving a calculated CSI and a quantized CSI difference; and circuitry 1124 for communicating with the UE based on the calculated CSI and the quantized CSI difference.

[0104] The techniques described herein can be used in various wireless communication technologies, such as NR (e.g., 5G NR), 3GPP Long Term Evolution (LTE), LTE Advanced (LTE-A), Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single Carrier Frequency Division Multiple Access (SC-FDMA), Time Division Synchronous Code Division Multiple Access (TD-SCDMA), and other networks. The terms "network" and "system" are often used interchangeably. CDMA networks can implement radio technologies such as Universal Terrestrial Radio Access (UTRA), cdma2000, etc. UTRA includes Wideband CDMA (WCDMA) and other variations of CDMA. cdma2000 covers the IS-2000, IS-95, and IS-856 standards. TDMA networks can implement radio technologies such as the Global System for Mobile Communications (GSM). OFDMA networks can implement radio technologies such as NR (e.g., 5G RA), evolved UTRA (E-UTRA), Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, Flash-OFDMA, etc. UTRA and E-UTRA are part of the Universal Mobile Telecommunications System (UMTS). LTE and LTE-A are UMTS versions using E-UTRA. UTRA, E-UTRA, UMTS, LTE, LTE-A, and GSM are described in documents from an organization called the 3rd Generation Partnership Project (3GPP). cdma2000 and UMB are described in documents from an organization called the 3rd Generation Partnership Project 2 (3GPP2). NR is an emerging wireless communication technology under development.

[0105] In 3GPP, the term "cell" can refer to the coverage area of ​​a Node B (NB) and / or the NB subsystem serving that coverage area, depending on the context in which the term is used. In NR systems, the term "cell" is used interchangeably with BS, Next Generation Node B (gNB or gNodeB), Access Point (AP), Distributed Unit (DU), Carrier, or Transmitter Receiver Point (TRP). A BS can provide communication coverage for macrocells, picocells, femtocells, and / or other types of cells. A macrocell can cover a relatively large geographic area (e.g., a radius of several kilometers) and can allow unrestricted access by UEs with a service subscription. A picocell can cover a relatively small geographic area and can allow unrestricted access by UEs with a service subscription. A femtocell can cover a relatively small geographic area (e.g., a home) and can allow restricted access by UEs associated with that femtocell (e.g., UEs in a Closed Subscriber Group (CSG), UEs for users in a home, etc.). A BS used for a macrocell can be called a macro BS. A BS used for a picocell can be called a pico BS. A BS used for a femtocell can be called a femtocell BS or a home BS.

[0106] A UE can also be referred to as a mobile station, terminal, access terminal, subscriber unit, station, customer premises equipment (CPE), cellular phone, smartphone, personal digital assistant (PDA), wireless modem, wireless communication device, handheld device, laptop computer, cordless phone, wireless local loop (WLL) station, tablet computer, camera, gaming device, netbook, smartbook, ultrabook, home appliance, medical device or medical equipment, biosensor / device, wearable device (such as smartwatch, smart clothing, smart glasses, smart wristband, smart jewelry (e.g., smart media, smart bracelet, etc.)), entertainment device (e.g., music device, video device, satellite radio, etc.), vehicle component or sensor, smart meter / sensor, industrial manufacturing equipment, GPS device, or any other suitable device configured to communicate via wireless or wired media. Some UEs can be considered machine-type communication (MTC) devices or evolved MTC (eMTC) devices. MTC and eMTC UEs include, for example, robots, drones, remote devices, sensors, instruments, monitors, location tags, etc., which can communicate with the BS, another device (e.g., a remote device), or some other entity. Wireless nodes can provide connectivity, for example, to or from a network (e.g., a wide area network such as the Internet or cellular networks), via wired or wireless communication links. Some UEs can be considered Internet of Things (IoT) devices, which can be narrowband IoT (NB-IoT) devices.

[0107] In some examples, access to the air interface can be scheduled. A scheduling entity (e.g., a BS) allocates resources for communication between some or all devices and equipment within its service area or cell. The scheduling entity can be responsible for scheduling, allocating, reconfiguring, and releasing resources for one or more subordinate entities. That is, for scheduled communication, the subordinate entities utilize the resources allocated by the scheduling entity. A base station is not the only entity that can act as a scheduling entity. In some examples, a UE can act as a scheduling entity and can schedule resources for one or more subordinate entities (e.g., one or more other UEs), and other UEs can utilize the resources scheduled by the UE for wireless communication. In some examples, a UE can act as a scheduling entity in peer-to-peer (P2P) networks and / or mesh networks. In mesh network examples, UEs can communicate directly with each other in addition to communicating with a scheduling entity.

[0108] The methods disclosed herein include one or more steps or actions for implementing these methods. The method steps and / or actions may be interchanged without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims.

[0109] As used herein, the phrase “at least one” in the list of referenced items refers to any combination of these items, including a single member. As an example, “at least one of a, b, or c” is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiples of the same element (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, cc, and cccc, or any other ordering of a, b, and c).

[0110] As used herein, the term "determine" encompasses a wide variety of actions. For example, "determine" can include calculation, operation, processing, derivation, research, searching (e.g., searching in a table, database, or other data structure), ascertaining, etc. Furthermore, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Additionally, "determine" can include parsing, selecting, choosing, establishing, etc.

[0111] The foregoing description is provided to enable any person skilled in the art to practice the various aspects described herein. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects. Therefore, the claims are not intended to be limited to the aspects shown herein, but rather to be consistent with the full scope of the language of the claims, wherein, unless specifically stated otherwise, reference to elements in the singular form is not intended to mean “one and only one”, but rather “one or more”. Unless otherwise specifically stated otherwise, the term “some” refers to one or more. All structural and functional equivalents of elements of the various aspects described throughout this disclosure that are known or will be known by those skilled in the art are expressly incorporated herein by reference and are intended to be covered by the claims. Furthermore, nothing disclosed herein is intended to be offered to the public, whether or not it is explicitly recited in the claims. No element of any claim is to be construed under 35 U.S.SC §112(f) unless it is expressly recited using the phrase “for a component of”, or, in the case of a method claim, using the phrase “for a step of”.

[0112] The various operations of the methods described above can be performed by any suitable component capable of performing the corresponding function. Components may include various hardware and / or software components and / or modules, including but not limited to circuits, application-specific integrated circuits (ASICs), or processors. Typically, when operations are illustrated in a figure, these operations may have corresponding components plus functional components with similar numbering.

[0113] The various illustrative logic blocks, modules, and circuits described in connection with this disclosure can be implemented or performed using a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but alternatively, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a combination of multiple microprocessors, a combination of one or more microprocessors with a combined DSP core, or any other combination of such configurations.

[0114] If implemented in hardware, an example hardware configuration could include a processing system in a wireless node. The processing system could be implemented with a bus architecture. The bus could include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the processing system. The bus could link together various circuitry, including a processor, machine-readable media, and a bus interface. The bus interface could be used to connect network adapters and other components to the processing system via the bus. The network adapter could be used to implement signal processing functions at the PHY layer. In the user terminal (see...), Figure 1 In this case, a user interface (e.g., keyboard, display, mouse, joystick, etc.) can also be connected to the bus. The bus can also link various other circuits, such as timing sources, peripherals, voltage regulators, power management circuits, etc., which are known in the art and therefore will not be described further. The processor can be implemented with one or more general-purpose and / or special-purpose processors. Examples include microprocessors, microcontrollers, DSP processors, and other circuits capable of executing software. Those skilled in the art will recognize how the described functionality for processing the system can be optimally implemented based on the specific application and the overall design constraints imposed on the system as a whole.

[0115] If implemented in software, functionality can be stored or sent as one or more instructions or code to a computer-readable medium. Software should be interpreted broadly to mean instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Computer-readable media includes both computer storage media and communication media, with communication media encompassing any medium that facilitates the transfer of computer programs from one place to another. The processor may be responsible for managing the bus and general processing, including the execution of software modules stored on machine-readable storage media. The computer-readable storage media may be coupled to the processor, allowing the processor to read information from and write information to it. Alternatively, the storage media may be integrated into the processor. For example, machine-readable media may include transmission lines, carrier waves modulated by data, and / or computer-readable storage media on which instructions are stored, separate from the wireless node, all accessible to the processor via a bus interface. Alternatively or additionally, the machine-readable media, or any portion thereof, may be integrated into the processor, such as in cases where it may have a cache and / or a general-purpose register file. Examples of machine-readable storage media may include RAM (random access memory), flash memory, ROM (read-only memory), PROM (programmable read-only memory), EPROM (erasable programmable read-only memory), EEPROM (electrically erasable programmable read-only memory), registers, disks, optical disks, hard drives, or any other suitable storage media or any combination thereof. Machine-readable media may be embodied in computer program products.

[0116] Software modules can comprise a single instruction or a number of instructions, and can be distributed across several different code segments, different programs, and multiple storage media. Computer-readable media can include several software modules. Software modules include instructions that, when executed by a device such as a processor, cause the processing system to perform various functions. Software modules can include send modules and receive modules. Each software module can reside in a single storage device or be distributed across multiple storage devices. For example, when a trigger event occurs, a software module can be loaded from the hard drive into RAM. During the execution of a software module, the processor can load some instructions into the cache to increase access speed. Then, one or more cache lines are loaded into a general-purpose register file for execution by the processor. When referring to the functionality of a software module below, it will be understood that such functionality is implemented by the processor when instructions from that software module are executed.

[0117] Furthermore, any connection is properly referred to as a computer-readable medium. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies (such as infrared (IR), radio, and microwave), then coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies (such as infrared, radio, and microwave) are all included in the definition of medium. As used herein, optical discs and disks include compact discs (CDs), laser discs, optical discs, digital versatile discs (DVDs), floppy disks, and... Disks, where magnetic disks typically reproduce data magnetically, and optical disks reproduce data optically using lasers. Therefore, in some aspects, computer-readable media can include non-transitory computer-readable media (e.g., tangible media). Furthermore, in other aspects, computer-readable media can include transient computer-readable media (e.g., signals). Combinations of the above should also be included within the scope of computer-readable media.

[0118] Therefore, certain aspects may include computer program products for performing the operations presented herein. For example, such computer program products may include computer-readable media on which instructions are stored (and / or encoded) that can be executed by one or more processors to perform the operations described herein, such as, for performing the operations described herein and... Figure 5 and Figure 6 The instructions for the operation are shown in the image.

[0119] Furthermore, it should be understood that modules and / or other suitable components for performing the methods and techniques described herein can be downloaded and / or otherwise obtained by the user terminal and / or base station (if applicable). For example, the device can be coupled to a server to facilitate the transfer of components for performing the methods described herein. Alternatively, the various methods described herein can be provided via storage components (e.g., RAM, ROM, physical storage media such as CDs or floppy disks, etc.), so that the user terminal and / or base station can obtain the various methods when coupled to the device or when providing storage components to the device. Furthermore, any other suitable techniques for providing the methods and techniques described herein to the device can be utilized.

[0120] It will be understood that the claims are not limited to the precise configuration and components shown above. Various modifications, alterations, and variations may be made to the arrangement, operation, and details of the methods and apparatus described above without departing from the scope of the claims.

Claims

1. A method for wireless communication performed by a user equipment (UE), comprising: The network entity receives a CSI prediction model for Channel State Information (CSI) and a qualification scheme for classifying CSIs predicted based on the CSI prediction model, wherein the prediction model is a machine learning model. Receive a first reference signal from the network entity at the first moment; The CSI is predicted at a second time based on the CSI prediction model, wherein the prediction is based on the first reference signal and the time interval between the second time and the first time. At the second moment, a second reference signal is received from the network entity; CSI is calculated based on the second reference signal; A quantized CSI difference is generated based on the quantization of the difference between the calculated CSI and the predicted CSI, wherein the difference is quantized based on the qualification scheme received from the network entity. as well as The network entity reports the difference between the calculated CSI and the quantified CSI.

2. The method according to claim 1, wherein, The quantified CSI difference is generated as one of a plurality of values ​​based on the conformity assessment scheme, the plurality of values ​​corresponding to an indication of whether the CSI predicted based on the CSI prediction model is a conformity assessment prediction or an unconformity assessment prediction based on the conformity assessment scheme.

3. The method according to claim 2, wherein, The qualified predictions include predictions for differences that are less than a threshold amount.

4. The method according to claim 2, further comprising: The predicted CSI is reported based on the determination that the predicted CSI is a qualified prediction.

5. The method according to claim 2, wherein, The predicted CSI was used for early hybrid automatic repeat request (HARQ) feedback.

6. The method according to claim 5, wherein, The qualification assessment scheme includes the determination of the false detection rate or false alarm rate.

7. The method according to claim 6, wherein, When the false detection rate or the false alarm rate is less than a threshold, the prediction is considered to be qualified.

8. The method according to claim 6, wherein, When the false detection rate or the false alarm rate exceeds a threshold, the prediction is considered to be unqualified.

9. The method according to claim 2, further comprising: Report the calculated CSI and the predicted CSI to the network entity; as well as The network entity receives configuration information that guides the UE to report a qualified CSI prediction based on the qualification scheme.

10. The method according to claim 9, wherein, The calculated CSI and the predicted CSI include one or more of the following: early hybrid automatic repeat request (HARQ) ACK / NACK, HARQ-ACK / NACK, precoding matrix indicator (PMI), channel quality indicator (CQI), rank indicator (RI), or signal-to-noise ratio (SNR).

11. The method according to claim 1, wherein, The quantified CSI difference includes one of several values ​​that indicate the magnitude of the difference between the calculated CSI and the predicted CSI.

12. The method according to claim 1, wherein, The quantified CSI difference is determined based on the difference between the calculated CSI and the predicted CSI over the time window.

13. The method according to claim 12, wherein, The quantified CSI difference includes the average difference between the calculated CSI and the predicted CSI over the time window.

14. The method according to claim 1, wherein, The difference between the calculated CSI and the predicted CSI is based on the difference between the previously calculated CSI and the corresponding predicted CSI, filtered by infinite impulse response (IIR).

15. A method for wireless communication performed by a user equipment (UE), comprising: Receive CSI prediction models from network entities for Channel State Information (CSI); Receive a first reference signal from the network entity at the first moment; The CSI is predicted at a second time based on the CSI prediction model, wherein the prediction is based on the first reference signal and the time interval between the second time and the first time. At the second moment, a second reference signal is received from the network entity; CSI is calculated based on the second reference signal; A quantized CSI difference is generated based on the quantization of the difference between the calculated CSI and the predicted CSI, wherein the quantized CSI difference is based on the average of the differences between the calculated CSI and the predicted CSI over a time window; as well as The network entity reports the difference between the calculated CSI and the quantized CSI.

16. A method for wireless communication performed by a network entity, comprising: Sending a CSI prediction model for Channel State Information (CSI) and a qualification scheme for classifying CSI predicted based on the CSI prediction model to the User Equipment (UE), wherein the prediction model is a machine learning model; Send a first reference signal to the UE at the first moment; At the second moment, a second reference signal is sent to the UE; The UE receives a calculated CSI and a quantized CSI difference, wherein the quantized CSI difference is based on the quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model, wherein the difference is quantized based on the qualification scheme, wherein the calculated CSI is based on the second reference signal, and wherein the predicted CSI is predicted based on the first reference signal and the time interval between the second time and the first time. The accuracy of the CSI prediction model in generating channel state information predictions is determined based on the difference between the calculated CSI and the quantized CSI. as well as If the CSI prediction model accurately generates predictions of channel state information, then: The CSI is predicted based on subsequently received CSI measurements and the time delay between the measurement and the signal transmission to the UE; and Adjust the transmission parameters used for sending.

17. The method according to claim 16, wherein, The quantified CSI difference is generated as one of a plurality of values ​​based on the conformity assessment scheme, the plurality of values ​​corresponding to an indication of whether the CSI predicted based on the CSI prediction model is a conformity assessment prediction or an unconformity assessment prediction based on the conformity assessment scheme.

18. The method according to claim 17, wherein, The qualified predictions include predictions for differences that are less than a threshold amount.

19. The method of claim 17, further comprising: Receive the CSI predicted based on the CSI prediction model; as well as Based on the received CSI predicted by the CSI prediction model, it is determined that the CSI predicted by the CSI prediction model is a qualified prediction.

20. The method of claim 17, wherein, The CSI predicted based on the CSI prediction model was used for early hybrid automatic repeat request (HARQ) feedback.

21. The method according to claim 20, wherein, The qualification assessment scheme includes a scheme for classifying CSIs predicted by the CSI prediction model based on the false detection rate or false alarm rate.

22. The method of claim 21, further comprising: The CSI predicted by the CSI prediction model is determined based on the received CSI, including the CSI prediction made when the false detection rate or the false alarm rate is less than a threshold.

23. The method of claim 16, further comprising: The CSI calculated from the UE and the CSI predicted based on the CSI prediction model are received. as well as Send configuration information to the UE that guides the UE to report a qualified CSI prediction based on the qualification scheme.

24. The method according to claim 23, wherein, The calculated CSI and the CSI predicted based on the CSI prediction model include one or more of the following: early hybrid automatic repeat request (HARQ) ACK / NACK, HARQ-ACK / NACK, precoding matrix indicator (PMI), channel quality indicator (CQI), rank indicator (RI), or signal-to-noise ratio (SNR).

25. The method according to claim 16, wherein, The quantified CSI difference includes one of several values ​​that indicate the magnitude of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model.

26. The method according to claim 16, wherein, The difference between the calculated CSI and the CSI predicted based on the CSI prediction model is based on the difference between the previously calculated CSI and the corresponding predicted CSI, filtered by an infinite impulse response (IIR).

27. A method for wireless communication performed by a network entity, comprising: Send a CSI prediction model for Channel State Information (CSI) to the User Equipment (UE); Send a first reference signal to the UE at the first moment; At the second moment, a second reference signal is sent to the UE; The UE receives a calculated CSI and a quantized CSI difference, wherein the quantized CSI difference is based on the average difference between the calculated CSI and the CSI predicted based on the prediction model over a time window, wherein the calculated CSI is based on the second reference signal, and wherein the predicted CSI is predicted based on the first reference signal and the time interval between the second time and the first time. as well as The communication with the UE is based on the calculated CSI and the quantized CSI difference.

28. An apparatus for wireless communication performed by a user equipment (UE), comprising: The processor is configured as follows: The network entity receives a CSI prediction model for Channel State Information (CSI) and a qualification scheme for classifying the CSI predicted based on the CSI prediction model, wherein the prediction model is a machine learning model. Receive a first reference signal from the network entity at the first moment; The CSI is predicted at a second time based on the CSI prediction model, wherein the prediction is based on the first reference signal and the time interval between the second time and the first time. At the second moment, a second reference signal is received from the network entity; CSI is calculated based on the second reference signal; A quantized CSI difference is generated based on the quantization of the difference between the calculated CSI and the predicted CSI, wherein the difference is quantized based on the eligibility assessment scheme received from the network entity. Report the calculated CSI and the quantized CSI difference to the network entity; as well as Memory.

29. An apparatus for wireless communication by a network entity, comprising: The processor is configured as follows: The system sends a CSI prediction model for Channel State Information (CSI) and a qualification scheme for classifying the CSI predicted based on the CSI prediction model to the User Equipment (UE), wherein the prediction model is a machine learning model. Send a first reference signal to the UE at the first moment; At the second moment, a second reference signal is sent to the UE; The UE receives a calculated CSI and a quantized CSI difference, wherein the quantized CSI difference is based on the quantization of the difference between the calculated CSI and the CSI predicted based on the CSI prediction model, wherein the difference is quantized based on the qualification scheme, wherein the calculated CSI is based on the second reference signal, and wherein the predicted CSI is predicted based on the first reference signal and the time interval between the second time and the first time. The accuracy of the CSI prediction model in generating channel state information predictions is determined based on the difference between the calculated CSI and the quantized CSI. as well as If the CSI prediction model accurately generates predictions of channel state information, then: The CSI is predicted based on subsequently received CSI measurements and the time delay between the measurement and the signal transmission to the UE; and Adjust the transmission parameters used for sending; and Memory.

30. An apparatus for wireless communication by a user equipment, comprising: The processor is configured as follows: Receive CSI prediction models from network entities for Channel State Information (CSI); Receive a first reference signal from the network entity at the first moment; The CSI is predicted at a second time based on the CSI prediction model, wherein the prediction is based on the first reference signal and the time interval between the second time and the first time. At the second moment, a second reference signal is received from the network entity; CSI is calculated based on the second reference signal; A quantized CSI difference is generated based on the quantization of the difference between the calculated CSI and the predicted CSI, wherein the quantized CSI difference is based on the average of the differences between the calculated CSI and the predicted CSI over a time window; as well as Report the calculated CSI and the quantized CSI difference to the network entity; and Memory.

31. An apparatus for wireless communication by a network entity, comprising: The processor is configured as follows: Send a CSI prediction model for Channel State Information (CSI) to the User Equipment (UE); Send a first reference signal to the UE at the first moment; At the second moment, a second reference signal is sent to the UE; The UE receives a calculated CSI and a quantized CSI difference, wherein the quantized CSI difference is based on the average difference between the calculated CSI and the CSI predicted based on the prediction model over a time window, wherein the calculated CSI is based on the second reference signal, and wherein the predicted CSI is predicted based on the first reference signal and the time interval between the second time and the first time. as well as The communication with the UE is based on the calculated CSI and the quantized CSI difference. and Memory.

32. An apparatus for wireless communication by a user equipment, comprising components for implementing the steps of the method according to any one of claims 1-15.

33. An apparatus for wireless communication by a network entity, comprising components for implementing the steps of the method according to any one of claims 16-27.

34. A non-transitory storage medium having instructions stored thereon, which, when executed by a processor of a user equipment, implement the method according to any one of claims 1-15.

35. A non-transitory storage medium having instructions stored thereon, which, when executed by a processor of a network entity, implement the method according to any one of claims 16-27.

36. A program product comprising instructions that, when executed by a processor of a user equipment, implement the method according to any one of claims 1-15.

37. A program product comprising instructions that, when executed by a processor of a network entity, implement the method according to any one of claims 16-27.

Images