Wireless device, network device, method for wireless device, and method for network device

Abstract

The present disclosure relates to DMRS overhead adaptation using AI-based channel estimation. A wireless device can be configured to receive, from a network device, downlink data transmitted using a DMRS pattern; perform AI-based downlink channel estimation based on the downlink data, including: inputting one or more received downlink DMRS symbols included in the received downlink data to a neural network model for downlink channel estimation stored in a memory of the wireless device to obtain, as an output of the neural network model, an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel; and report the optimal downlink DMRS pattern to the network device.

Description

TECHNICAL FIELD

[0001] The present application relates generally to wireless communication systems, including wireless devices and network devices for demodulation reference signal (DMRS) overhead adaptation using artificial intelligence (AI)-based channel estimation. BACKGROUND

[0002] Wireless mobile communication technology uses various standards and protocols to transmit data between base stations and wireless communication devices. Wireless communication system standards and protocols can include, for example, the Third Generation Partnership Project (3GPP) Long-Term Evolution (LTE) (such as 4G), 3GPP New Radio (NR) (such as 5G), and IEEE 802.11 standards (commonly referred to as Wi-Fi® within the industry organization) for wireless local area networks (WLANs).

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

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

[0005] A base station used by a RAN can correspond to that RAN. One example of an E-UTRAN base station is an Evolved Universal Terrestrial Radio Access Network (E-UTRAN) Node B (also denoted as an Evolved Node B, Enhanced Node B, eNodeB, or eNB). One example of an NG-RAN base station is a Next Generation Node B (sometimes also referred to as a gNodeB or gNB).

[0006] ​The RAN, through its interaction with the core network (CN), is responsible for the SUMMARY

[0007] A demodulation reference signal (DMRS) can be used for channel estimation and demodulation. Specifically, a DMRS symbol with a certain pattern can be carried by a resource block and transmitted from a transmitter to a receiver. The receiver can estimate a channel using a received DMRS symbol with the certain pattern known in advance by the receiver.

[0008] Recently, there is increasing interest in using artificial intelligence (AI) for air interface. One possible improvement can be DMRS overhead adaptation using AI-based channel estimation. DMRS overhead (e.g., ratio of transmitted DMRS symbols and total data transmitted) determined by DMRS pattern can have an impact on channel estimation performance. By using AI-based channel estimation, better performance can be achieved for DMRS overhead adaptation and / or channel estimation. However, procedures / mechanisms to support DMRS overhead adaptation using AI-based estimation are still under discussion.

[0009] Embodiments in this disclosure relate to devices and methods for DMRS overhead adaptation using AI-based estimation.

[0010] A wireless device according to some embodiments of the disclosure can be configured to receive, from a network device, downlink data transmitted using a downlink demodulation reference signal (DMRS) pattern; perform artificial intelligence (AI)-based downlink channel estimation based on the downlink data, including: inputting one or more received downlink DMRS symbols included in the received downlink data to a neural network model for downlink channel estimation stored in a memory of the wireless device to obtain, as an output of the neural network model, an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel; and report the optimal downlink DMRS pattern to the network device.

[0011] A network device according to some embodiments of the disclosure can be configured to configure a wireless device to enable artificial intelligence (AI)-based downlink channel estimation; and receive, from the wireless device, an optimal downlink DMRS pattern output from a neural network model for the AI-based downlink channel estimation, wherein the neural network model has an input of downlink DMRS symbols included in downlink data received by the wireless device and an output of an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel.

[0012] A network device according to some embodiments of the disclosure can be configured to receive, from a wireless device, uplink data transmitted using an uplink demodulation reference signal (DMRS) pattern; perform artificial intelligence (AI)-based uplink channel estimation based on the uplink data, including: inputting one or more received uplink DMRS symbols included in the received uplink data to a neural network model for uplink channel estimation stored in a memory of the network device to obtain an estimated uplink channel corresponding to the uplink data and a best downlink DMRS pattern for the estimated uplink channel as an output of the neural network model; and instructing the wireless device to use the best uplink DMRS pattern for uplink transmission.

[0013] A method for a wireless device according to some embodiments of the disclosure can include receiving, from a network device, downlink data transmitted using a downlink demodulation reference signal (DMRS) pattern; performing artificial intelligence (AI)-based downlink channel estimation based on the downlink data, including: inputting one or more received downlink DMRS symbols included in the received downlink data to a neural network model for downlink channel estimation to obtain an estimated downlink channel corresponding to the downlink data and a best downlink DMRS pattern for the estimated downlink channel as an output of the neural network model; and reporting the best downlink DMRS pattern to the network device.

[0014] A method for a network device according to some embodiments of the disclosure can include configuring a wireless device to enable artificial intelligence (AI)-based downlink channel estimation; and receiving, from the wireless device, a best downlink DMRS pattern output from a neural network model for the AI-based downlink channel estimation, wherein the neural network model has an input of downlink DMRS symbols included in downlink data received by the wireless device and an output of an estimated downlink channel corresponding to the downlink data and a best downlink DMRS pattern for the estimated downlink channel.

[0015] A method for a network device according to some embodiments of the present disclosure can include receiving, from a wireless device, uplink data transmitted using an uplink demodulation reference signal (DMRS) pattern; performing artificial intelligence (AI)-based uplink channel estimation based on the uplink data, including: inputting one or more received uplink DMRS symbols included in the received uplink data to a neural network model for uplink channel estimation to obtain an estimated uplink channel corresponding to the uplink data and a best downlink DMRS pattern for the estimated uplink channel as outputs of the neural network model; and instructing the wireless device to use the best uplink DMRS pattern for uplink transmission.

[0016] The techniques described herein can be implemented in and / or used with a number of different types of devices, including but not limited to cellular phones, tablet computers, wearable computing devices, portable media players, and any of a variety of other computing devices.

[0017] This summary is intended to provide a brief overview of some of the subject matter described in this document. Accordingly, it will be appreciated that the above-described features are merely examples and should not be construed to narrow the scope or spirit of the subject matter described herein. Other features, aspects, and advantages of the subject matter described herein will become apparent from the following DETAILED DESCRIPTION, Figures, and Claims. BRIEF DESCRIPTION OF DRAWINGS

[0018] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.

[0019] Figure 1 An exemplary architecture of a wireless communication system according to embodiments disclosed herein is shown.

[0020] Figure 2 A system for performing signaling between a wireless device and a network device according to embodiments disclosed herein is shown.

[0021] Figure 3 A flowchart of an exemplary method performed by a wireless device for downlink DMRS overhead adaptation using AI-based channel estimation according to some embodiments disclosed herein is shown.

[0022] Figures 4-5 A signaling diagram of an exemplary method for downlink DMRS overhead adaptation using AI-based channel estimation according to some embodiments disclosed herein is shown.

[0023] Figure 6An exemplary DMRS pattern for training a neural network model is shown in accordance with some embodiments disclosed herein.

[0024] Figure 7 A flow diagram of an exemplary method for downlink DMRS overhead adaptation using AI-based channel estimation performed by a network device is shown in accordance with some embodiments disclosed herein.

[0025] Figure 8 A flow diagram of an exemplary method for uplink DMRS overhead adaptation using AI-based channel estimation performed by a network device is shown in accordance with some embodiments disclosed herein. DETAILED DESCRIPTION

[0026] Embodiments are described in terms of a UE. However, the reference to a UE is provided for illustrative purposes only. Exemplary embodiments can be used with any electronic component that can establish a connection with a network and is configured with hardware, software, and / or firmware for exchanging information and data with the network. Thus, a UE as described herein is used to represent any appropriate electronic component.

[0027] Figure 1 An exemplary architecture of a wireless communication system 100 is shown in accordance with embodiments disclosed herein. The description provided below is directed to an exemplary wireless communication system 100 that operates in conjunction with LTE system standards and / or 5G or NR system standards provided by 3GPP Technical Specifications.

[0028] As shown in Figure 1 The wireless communication system 100 includes UE 102 and UE 104 (although any number of UEs can be used). In this example, the UE 102 and the UE 104 are shown as smartphones (e.g., handheld touchscreen mobile computing devices connectable to one or more cellular networks) but can also include any mobile or non-mobile computing devices configured for wireless communication.

[0029] The UE 102 and the UE 104 can be configured to communicatively couple with a RAN 106. In embodiments, the RAN 106 can be an NG-RAN, an E-UTRAN, etc. The UE 102 and the UE 104 utilize connections (or channels) with the RAN 106, shown as connections 108 and 110, respectively, wherein each connection (or channel) comprises a physical communications interface. The RAN 106 can comprise one or more base stations, such as base stations 112 and 114, implementing the connections 108 and 110.

[0030] In this example, the connections 108 and 110 are over-the-air interfaces that implement such communicative coupling, and can conform with the RAT(s) used by the RAN 106, such as, for example, LTE and / or NR.

[0031] In some implementations, the UEs 102 and 104 can also directly exchange communication data via a sidelink interface 116. The UE 104 is shown to be configured to access an access point (AP 118) via a connection 120. The connection 120 can comprise a local wireless connection, such as a connection consistent with any IEEE 802.11 protocol, Bluetooth® brand protocol, Code Division Multiple Access (CDMA) 1X, CDMA2000, Global System for Mobile Communications (GSM), Ix Evolution-Data Optimized (EV-DO), IEEE 802.16, Ix.21, Ix.30, Ix.60, or any other local wireless protocol. The AP 118 can comprise a router, hub, or switch, for example, and can provide connectivity to a network (not shown) such as the Internet or to other devices. The AP 118 can or can not be connected to the CN 124 through the network 122, in this example.

[0032] In implementations, the UEs 102 and 104 can be configured to communicate using Orthogonal Frequency-Division Multiplexing (OFDM) communication signals with each other or with the base stations 112 and / or 114 over a multicarrier communication channel in accordance with various communication techniques, such as, but not limited to, an Orthogonal Frequency-Division Multiple Access (OFDMA) communication technique (e.g., for downlink communications) or a Single Carrier Frequency Division Multiple Access (SC-FDMA) communication technique (e.g., for uplink and ProSe or sidelink communications), although the scope of the implementations is not limited in this respect. The OFDM signals can comprise a plurality of orthogonal subcarriers.

[0033] In some implementations, all or a portion of base station 112 or base station 114 can be implemented as one or more software entities running on a server computer as part of a virtual network. Additionally, or in other implementations, the base stations 112 or 114 can be configured to communicate with each other via interface 122. In implementations where the wireless communication system 100 is an LTE system (e.g., where CN 124 is an EPC), the interface 122 can be an X2 interface. The X2 interface can be defined between two or more base stations (e.g., two or more eNBs, etc.) connected to the EPC or between two eNBs connected to the EPC. In implementations where the wireless communication system 100 is an NR system (e.g., where CN 124 is a 5GC), the interface 122 can be an Xn interface. The Xn interface is defined between two or more base stations (e.g., two or more gNBs, etc.) connected to the 5GC, between a base station 112 (e.g., gNB) and an eNB connected to the 5GC, and / or between two eNBs connected to the 5GC (e.g., CN 124).

[0034] The RAN 106 is shown to be communicatively coupled to a CN 124. The CN 124 can include one or more network elements 126, which are configured to provide various data and telecommunications services to customers / subscribers (e.g., users of UEs 102 and 104) who are connected to the CN 124 via the RAN 106. The components of the CN 124 can be implemented in one physical device or each individually as physically separate devices.

[0035] In embodiments, the CN 124 can be an EPC, and the RAN 106 can be connected with the CN 124 via an S1 interface 128. In embodiments, the S1 interface 128 can be split into two parts, an S1 user plane (S1-U) interface, which carries traffic data between the base stations 112, 114 and serving gateways (S-GWs), and an S1-MME interface, which is a signaling interface between the base stations 112, 114 and mobility management entities (MMEs).

[0036] In embodiments, the CN 124 can be a 5GC, and the RAN 106 can be connected with the CN 124 via an NG interface 128. In embodiments, the NG interface 128 can be split into two parts: an NG user plane (NG-U) interface, which carries traffic data between the base stations 112, 114 and user plane functions (UPFs); and an SI control plane (NG-C) interface, which is a signaling interface between the base stations 112, 114 and access and mobility management functions (AMFs).

[0037] Generally, the application server 130 can be an element of a network that provides content, or enables various services, that are accessed by the UEs 102 and 104 using IP bearer resources of the CN 124. The application server 130 can also be configured to support one or more communication services (e.g., VoIP sessions, group communication sessions, etc.) for the UEs 102 and 104 via the CN 124. The application server 130 can communicate with the CN 124 through an IP communications interface 132.

[0038] Figure 2 A system 200 for performing signaling 234 between a wireless device 202 and a network device 218 is shown, in accordance with embodiments disclosed herein. The system 200 can be part of a wireless communication system, as described herein. The wireless device 202 can be, for example, a UE of the wireless communication system. The network device 218 can be, for example, a base station (e.g., an eNB or gNB) of the wireless communication system.

[0039] The wireless device 202 can include one or more processors 204. The processors 204 can execute instructions to perform various operations of the wireless device 202, as described herein. The processors 204 can include one or more baseband processors implemented using, for example, a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a controller, a field programmable gate array (FPGA) device, another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein.

[0040] The wireless device 202 can include a memory 206. The memory 206 can be a non-transitory computer-readable storage medium that stores instructions 208 (which can include, for example, instructions for execution by the processors 204). The instructions 208 can also be referred to as program code or computer programs. The memory 206 can also store data used by the processors 204 and results of operations performed by the processors.

[0041] The wireless device 202 can include one or more transceivers 210, which can include radio-frequency (RF) transmitter and / or receiver circuits that use the antennas 212 of the wireless device 202 to facilitate signaling (e.g., signaling 234) to and / or from the wireless device 202 with other devices (e.g., network devices 218) in accordance with a corresponding RAT.

[0042] The wireless device 202 can include one or more antennas 212 (e.g., one, two, four, or more). For embodiments with multiple antennas 212, the wireless device 202 can utilize spatial diversity of such multiple antennas 212 to transmit and / or receive multiple different data streams on the same time-frequency resources. This approach can be referred to as, for example, a multiple-input multiple-output (MIMO) approach (referring to the multiple antennas used in this regard at the transmitting device and the receiving device, respectively). MIMO transmission by the wireless device 202 can be implemented in accordance with precoding (or digital beamforming) applied to the wireless device 202 that multiplexes data streams among the antennas 212 such that each data stream is received at an appropriate signal strength relative to other streams and at a desired location in space (e.g., the location of the receiver associated with the data stream) in accordance with known or assumed channel characteristics. Certain embodiments can use a single-user MIMO (SU-MIMO) approach (where data streams are all directed to a single receiver) and / or a multi-user MIMO (MU-MIMO) approach (where individual data streams can be directed to individual (different) receivers at different locations in space).

[0043] In certain embodiments with multiple antennas, wireless device 202 can implement an analog beamforming technique whereby the phases of the signal transmitted by antennas 212 are adjusted relative so that the (joint) transmission by antennas 212 can have directivity (this is sometimes referred to as beam steering).

[0044] Wireless device 202 can include one or more interfaces 214. Interfaces 214 can be used to provide input to or output from wireless device 202. For example, a wireless device 202 that is a UE can include interfaces 214 such as a microphone, speaker, touchscreen, buttons, and so on in order to allow a user of the UE to input to and / or output from the UE. Other interfaces of such a UE can be made up of transmitters, receivers, and other circuitry (e.g., in addition to transceiver 210 / antennas 212 already described) that allow the UE to communicate with other devices, and can operate according to known protocols (e.g., Bluetooth®, Wi-Fi®, ZigBee®, etc.).

[0045] Network device 218 can include one or more processors 220. Processors 220 can execute instructions to perform various operations of network device 218 as described herein. Processors 204 can include one or more baseband processors implemented using, for example, CPUs, DSPs, ASICs, controllers, FPGA devices, another hardware devices, firmware devices, or any combination thereof configured to perform the operations described herein.

[0046] Network device 218 can include memory 222. Memory 222 can be a non-transitory computer-readable storage medium that stores instructions 224 (which can include, for example, instructions for execution by processors 220). Instructions 224 can also be referred to as program code or computer programs. Memory 222 can also store data used by processors 220 and results

[0047] Network device 218 can include one or more transceivers 226 that can include radio-frequency (RF) transmitter and / or receiver circuits that use antennas 228 of network device 218 to facilitate signaling (e.g., signaling 234) to and / or from network device 218 with other devices (e.g., wireless device 202) according to a corresponding RAT.

[0048] Network device 218 can include one or more antennas 228 (e.g., one, two, four, or more). In embodiments with multiple antennas 228, network device 218 can perform MIMO, digital beamforming, analog beamforming, beam steering, and so on as previously described.

[0049] ​The network equipment 218 can include one or more interfaces 230. The interfaces 230 can be used to provide input to or output from the network equipment 218. For example, the network equipment 218 as a base station can include an interface 230 consisting of a transmitter, receiver, and other circuitry (e.g., in addition to the transceivers 226 / antennas 228 already described) that enables the base station to communicate with other equipment in the core network and / or with external networks, computers, databases, etc. for the purpose of performing operations, management, and maintenance of the base station or other equipment with which it is in operable connection.

[0050] Figure 3 A flowchart of an exemplary method 300 for downlink DMRS overhead adaptation using AI-based channel estimation performed by a wireless device, in accordance with some embodiments disclosed herein, is shown. The wireless device can correspond to Figure 1 the UE 102, 104 described in Figure 2 any of the wireless devices 202 described in

[0051] In various embodiments, some of the elements of the illustrated method can be performed simultaneously, in a different order than shown, can be omitted, replaced by other elements, or combined with other elements. Additional elements can also be performed as desired.

[0052] At step 302, the wireless device receives downlink data transmitted using an initial DMRS pattern from a network equipment.

[0053] In some embodiments, the downlink data can include a physical downlink shared channel (PDSCH) and / or a physical downlink control channel (PDCCH). Further, the initial DMRS pattern can be any pattern used by the network equipment and indicated to the wireless device. For example, Figure 6 (a) shows an example of the initial DMRS pattern, where the shaded resource elements labeled “1” in the resource blocks correspond to the DMRS symbols that make up the DMRS pattern. It should be noted that the wording “initial” described herein only means to distinguish the currently used DMRS pattern from the best DMRS pattern after AI-based downlink channel estimation, and there is no other restriction for the DMRS pattern.

[0054] At step 304, the wireless device performs AI-based downlink channel estimation based on the downlink data. A neural network model can be used for the AI-based downlink channel estimation. The input of the neural network model can be one or more DMRS symbols included in the downlink data (e.g., Figure 6six DMRS symbols in the resource block shown in (a)), and the output of the neural network model can be an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel.

[0055] The type of neural network can be arbitrary according to actual design requirements. For example, the neural network can be, but is not limited to, a recurrent neural network (RNN), a long / short-term memory (LSTM), a convolutional neural network (CNN), and a deep neural network (DNN).

[0056] At step 306, the wireless device reports the optimal downlink DMRS pattern to the network device.

[0057] According to the method 300, downlink DMRS overhead adaptation using AI-based channel estimation is performed by the wireless device to obtain an optimal downlink DMRS pattern for future transmissions. The optimal downlink DMRS pattern can be an optimal pattern for obtaining a preferred channel estimation performance, and / or achieving a preferred DMRS overhead reduction, and / or achieving an optimal spectral efficiency for data transmission. Details of downlink DMRS overhead adaptation using AI-based channel estimation will be further explained with reference to the signaling diagram between the wireless device and the network device shown in Figures 4-5

[0058] Figures 4-5 A signaling diagram of an exemplary method for downlink DMRS overhead adaptation using AI-based channel estimation is shown according to some embodiments disclosed herein.

[0059] Figures 4-5 The wireless device shown in can correspond to the wireless device performing the method 300 described in Figure 3 Figures 4-5 The network device shown in can correspond to the network device communicating with the wireless device in Figure 3

[0060] Reference is now made to Figure 4 In some embodiments, prior to performing AI-based downlink channel estimation, such as performing the method 300 described in Figure 3 In some embodiments, prior to performing AI-based downlink channel estimation, such as performing the method 300 described in Figure 4 This process is shown in steps 402 and 404 of

[0061] ​​​At step 402, the wireless device reports the capability of performing AI-based downlink channel estimation to the network device. The capability can show that the wireless device supports using AI-based estimation for DMRS overhead adaptation. In some embodiments, the wireless device can report the capability to the network device by using higher layer signaling, such as radio resource control (RRC) signaling.

[0062] In some embodiments, the capability can include the capability of using a neural network model to perform AI-based downlink channel estimation. The capability can include the capability of using the neural network model to evaluate and determine the best DMRS pattern. The capability can include the capability of reporting the best DMRS pattern to the network device. The capability can include the capability of supporting different DMRS patterns signaled by the network device. The wireless device can include one or more of the above capabilities. In addition, the wireless device can include other capabilities for supporting DMRS overhead adaptation using AI-based estimation.

[0063] At step 404, the network device configures the wireless device to enable AI-based channel estimation in response to receiving the report of the capability of the wireless device in step 402. Step 404 serves as the feedback of the network device to the capability.

[0064] In some embodiments, the network device can broadcast (e.g., in a system information block (SIB)) that the network device supports AI-based downlink channel estimation to a plurality of wireless devices (e.g., UEs in a cell controlled by the same gNB) in order to facilitate one or more of the plurality of wireless devices that have the capability of performing AI-based downlink channel estimation to enable AI-based downlink channel estimation.

[0065] In some embodiments, if the wireless device has the capability, it can enable the AI-based downlink channel estimation mode by itself. Alternatively, if the wireless device has the capability, it can decide whether to enable the mode. For example, the wireless device can decide whether to enable the mode based on the evaluation of the channel estimation performance. For example, the wireless device can decide to enable the mode when the channel characteristics belong to a set of characteristics in the training data used to train the neural network model. Otherwise, the wireless device can decide not to enable the mode, and the wireless device will not change the initial DMRS pattern.

[0066] In some embodiments, in step 404, the network device can configure the wireless device to enable AI-based downlink channel estimation using wireless device-specific signaling, such as UE-specific signaling (e.g., higher layer signaling, such as RRC signaling).

[0067] In some cases, AI-based channel estimation can be completely up to wireless device implementation, and network devices have no knowledge about the detailed structure or algorithm of the block implemented on the wireless device for performing AI-based channel estimation.

[0068] In these cases, a neural network model can be trained by the wireless device. The training data for training the neural network model can include one or more predefined downlink DMRS patterns and one or more known downlink channels. Based on the training data, known DMRS symbols can be generated, which are inputs to the neural network model. In addition, the known downlink channels can be used as ground truth for training the neural network model.

[0069] In some embodiments, the known downlink channels can be obtained by using certain channel models. Alternatively, the known downlink channels can be obtained from field measurements.

[0070] In some embodiments, the training data can be generated by the wireless device. In some other embodiments, the training data can be provided to the wireless device from the network device, e.g., over the air interface. In addition, the training data can further include decision feedback data, which is decoded by the wireless device from the received downlink data. For example, after decoding a PDSCH, the decoded data of the PDSCH becomes known symbols to the wireless device. The decoded data can be used as decision feedback data to further train the neural network model. By doing so, the neural network model can be trained and adjusted in real time, which can save the data transmission overhead of the training data from the network device to the wireless device.

[0071] In some embodiments, the neural network model can be trained so as to improve channel estimation performance and / or reduce DMRS overhead.

[0072] Generally, the more DMRS symbols transmitted, the better the channel estimation performance. In contrast, the more DMRS symbols transmitted, the greater the DMRS overhead (e.g., the ratio of DMRS symbols in a PDSCH and the total data in the PDSCH). Thus, there is a tradeoff between channel estimation performance and DMRS overhead. By taking into account the tradeoff, the neural network model can be trained to take into account both channel estimation performance and DMRS overhead so as to achieve optimized DMRS overhead adaptation using AI-based channel estimation. In some embodiments, a certain metric (e.g., related to information bit quality) can be developed, and the neural network model can be trained to minimize or maximize the metric to achieve the tradeoff.

[0073] In some embodiments, the two conditions above, i.e., channel estimation performance and optimal DMRS pattern, can be considered separately. For example, two separate neural network models can be used for channel estimation and optimal DMRS pattern. The neural network model for channel estimation can be trained using training data such that the estimated channel can be as close as possible to the known channel used as ground truth. In addition, the neural network model for optimal DMRS pattern can be trained using training data including pre-known optimal DMRS patterns such that the neural network model outputs optimal DMRS patterns as close as possible to the pre-known optimal DMRS patterns.

[0074] In some embodiments, the optimal DMRS pattern as the output of the neural network model can be generated among a plurality of pre-defined DMRS patterns. For example, the neural network model can be trained such that the optimal DMRS pattern can be selected from the plurality of pre-defined DMRS patterns. In some embodiments, the DMRS pattern can span one or more time slots. In addition, the DMRS overhead can vary from time slot to time slot. In some embodiments, the DMRS overhead in the spatial domain for different antenna ports can be different.

[0075] In some embodiments, in addition to the initial training data used to train the neural network model, the wireless device can need additional training data to refine the neural network model. In some embodiments, the wireless device can request the network device to provide the additional training data. Figure 4 This process is shown in steps 406 and 408.

[0076] At step 406, the wireless device requests the network device to provide additional training data for training the neural network model.

[0077] In some embodiments, the wireless device can determine whether to request additional training data based on the evaluation of the channel estimation performance. The evaluation can be based on Doppler estimation, speed of the wireless device, sudden change in channel characteristics, etc.

[0078] At step 408, the network device provides the additional training data to the wireless device for training the neural network model.

[0079] In some embodiments, step 408 can be performed in response to the request by the wireless device in step 406. In these embodiments, prior to step 406, the network device can indicate the capability to provide additional training data via broadcast or UE-specific signaling.

[0080] In some other embodiments, in step 408, the network device can provide additional training data to the wireless device based on its own criteria. For example, the network device can provide additional training data periodically or based on channel characteristics (e.g., the network device detects significant change in channel characteristics). In some embodiments, step 406 can be omitted.

[0081] In some embodiments, the network device can indicate to the wireless device additional training data in downlink control information (DCI). For example, the network device can indicate in the DCI whether a transmission (e.g., PDSCH) includes training data. Alternatively, for example, when there will be training data sent to the wireless device, the network device can semi-statically configure (e.g., through RRC signaling) the transmission timing of the additional training data.

[0082] In some embodiments, the additional training data provided by the network device can include at least one of: additional downlink data that includes symbols known to the wireless device in addition to the DMRS symbols included in the downlink data; or additional downlink data that uses a DMRS pattern that includes more DMRS symbols than the DMRS symbols included in the received downlink data. Reference is made to Figure 6 Some examples of additional training data can be explained.

[0083] Figure 6 An exemplary DMRS pattern for training a neural network model is shown in accordance with some embodiments disclosed herein. Figure 6 Each of (a) through (d) represents a resource block, and the shaded resource elements labeled “1” in the resource block correspond to DMRS symbols that make up the DMRS pattern.

[0084] Figure 6 (a) shows an initial DMRS pattern used for transmitting downlink data. Figure 6 (b) through (d) show resource blocks in which additional known symbols are included. In Figure 6 In (b), the shaded resource elements labeled “2” in the resource block correspond to symbols known to the wireless device in addition to the DMRS symbols labeled “1.” Similarly, in Figure 6 In (c), more known symbols labeled “2” are placed in the resource elements. In addition, in Figure 6 In (d), all other symbols in the resource block in addition to the DMRS symbols labeled “1” are set to known symbols. By providing more known symbols to the wireless device as additional training data, the wireless device can obtain more information for training to refine the neural network model.

[0085] In some embodiments, the known symbols labeled “2” can also be DMRS symbols. For example, in Figure 6 (b) through (d), the known symbols labeled “2” are DMRS symbols in addition to the DMRS symbols labeled “1” in Figure 6(a) Compared to, a denser DMRS pattern is used as the training data pattern. Alternatively, other symbols known to the wireless device can be used as known symbols in the additional training data. In addition, it should be noted that the training data pattern in the present disclosure is not limited to Figure 6 (b) to (d) shown in the patterns, and can be designed according to actual needs.

[0086] In some embodiments, the network device can dynamically indicate (e.g., in DCI) which type of additional training data (e.g., Figure 6 (b) to (d) shown in the training data patterns) will be sent to the wireless device.

[0087] Referring back to Figure 4 , at step 410, the wireless device reports the best downlink DMRS pattern obtained from the trained neural network model to the network device. In Figure 3 The process for obtaining the best downlink DMRS pattern from the neural network model has been described in the method 300.

[0088] At step 412, the network device indicates the best DMRS pattern to the wireless device as the new DMRS pattern. In some embodiments, the network device sends signaling to the wireless device to change the DMRS pattern from the initial DMRS pattern to the new DMRS pattern. The signaling can be dynamic (e.g., through DCI) or semi-static (e.g., through RRC signaling or MAC CE (Media Access Control Control Element)). In some other embodiments, the wireless device can request to change the DMRS pattern, and the network device can provide an acknowledgment to the wireless device to follow the request and start using the new DMRS pattern after a predefined application delay.

[0089] At step 414, the network device transmits downlink data using the best DMRS pattern.

[0090] The above Figure 4 A signaling diagram of an exemplary method for downlink DMRS overhead adaptation using AI-based channel estimation is shown, in which the neural network model is trained at the wireless device. In some cases, considering that the network device can generally have stronger computing capability, the neural network model can be trained at the network device and provided to the wireless device. These embodiments can be described with reference to Figure 5 .

[0091] Figure 5 A signaling diagram of an exemplary method for downlink DMRS overhead adaptation using AI-based channel estimation is shown, in accordance with some embodiments disclosed herein. Figure 4 With Figure 5The difference between the two is in step 506, and the other steps 502, 504, 508-512 correspond to steps 402, 404, 410-414, respectively, and the description thereof will be omitted.

[0092] In Figure 5 The method of FIG. 6 can be performed by a network device, such as the network device 200, as described herein. The network device can be, for example, a base station, a gNB, an eNB, a gNB-CU, a gNB-DU, a gNB- CU-UP, a gNB-DU-UP, or the like. Figure 4 In the method of FIG. 6, the neural network model is trained by the network device. The training process can be similar to the training process performed in the wireless device described with reference to FIG. 5. After the training, at step 506, the network device sends the trained neural network model for the wireless device to download.

[0093] In some embodiments, the network device can send the following information of the neural network model to the wireless device, which includes one or more of the following: the type of the neural network, the number of layers, the input and output, the parameters (e.g., weights and biases) used in the neural network. The wireless device can download the information of the neural network to reconstruct the neural network model on its own side. In some embodiments, some of the above information can be predefined or uniform on both the wireless device and the network device. In this case, the wireless device can download less information from the network device.

[0094] In some embodiments, the wireless device can download the neural network model from the network device via broadcast (e.g., SIB) or UE-specific signaling (e.g., RRC signaling), and start to use the neural network model to perform the above-described AI-based channel estimation.

[0095] In some embodiments, although the neural network model is trained and provided by the network device, the neural network model can be further trained by the wireless device using the training data to update the neural network model. In addition, the wireless device can request additional training data from the network device to further refine the neural network model. The updated neural network model can be a feedback to the network device. In addition, the network device can further train the updated neural network model. By doing so, the neural network model can be trained on both the wireless device and the network device to achieve better performance.

[0096] Embodiments contemplated herein include an apparatus comprising means for performing one or more elements of the method 300. The apparatus can be, for example, an apparatus of a UE (such as the wireless device 202 as a UE, as described herein).

[0097] Embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform one or more elements of the method 300. The non-transitory computer-readable media can be, for example, a memory of a UE (such as the memory 206 of the wireless device 202 as a UE, as described herein).

[0098] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry to perform one or more elements of method 300. The apparatus can be, for example, an apparatus of a UE (such as wireless device 202 as a UE, as described herein).

[0099] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions to, when executed by the one or more processors, cause the one or more processors to perform one or more elements of method 300. The apparatus can be, for example, an apparatus of a UE (such as wireless device 202 as a UE, as described herein).

[0100] Embodiments contemplated herein include a signal as described in or related to one or more elements of method 300.

[0101] Embodiments contemplated herein include a computer program or computer program product comprising instructions, where the program is executed by a processor to cause the processor to carry out one or more elements of method 300. The processor can be a processor of a UE (such as processor 204 of wireless device 202 as a UE, as described herein). The instructions may, for example, be located within the processor of the UE and / or on a memory of the UE (e.g., memory 206 of wireless device 202 as a UE, as described herein).

[0102] Figure 7 A flow diagram of an exemplary method 700 for downlink DMRS overhead adaptation using AI-based channel estimation performed by a network device is shown, in accordance with some embodiments disclosed herein. The network device can correspond to any of the base stations 112, 114 described in Figure 1 “Network device” described in Figure 2 “Network device” 218 described in

[0103] At step 702, the network device configures a wireless device to enable AI-based downlink channel estimation.

[0104] At step 704, the network device receives, from the wireless device, an optimal downlink DMRS pattern output from a neural network model for AI-based downlink channel estimation.

[0105] The processing in steps 702 and 704 corresponds to steps 404 and 410 of Figure 4 and details will be omitted here.

[0106] The above Figures 3-5 and Figure 7Methods and signaling diagrams for downlink DMRS overhead adaptation using AI-based channel estimation are shown. Next, an example method 700 performed by a network device for downlink DMRS overhead adaptation using AI-based channel estimation will be described with reference to Figure 8 An uplink DMRS overhead adaptation using AI-based channel estimation is described.

[0107] Figure 8 A flowchart of an example method 800 performed by a network device for uplink DMRS overhead adaptation using AI-based channel estimation is shown, according to some embodiments disclosed herein. The network device can correspond to Figure 1 any of the base stations 112, 114 described in Figure 2 any of the network devices 218 described in

[0108] At step 802, the network device receives uplink data transmitted using an initial DMRS pattern from a wireless device.

[0109] In some embodiments, the uplink data can include a physical uplink shared channel (PUSCH) and / or a physical uplink control channel (PUCCH). Further, the initial DMRS pattern can be any pattern used by the network device and indicated to the wireless device. For example, Figure 6 (a) shows an example of the initial DMRS pattern, where the shaded resource elements labeled as “1” in the resource blocks correspond to the DMRS symbols that make up the DMRS pattern.

[0110] At step 804, the network device performs AI-based uplink channel estimation based on the uplink data. A neural network model can be used for the AI-based uplink channel estimation. The input to the neural network model can be one or more DMRS symbols included in the uplink data, and the output of the neural network model can be an estimated uplink channel corresponding to the uplink data and an optimal uplink DMRS pattern for the estimated uplink channel.

[0111] The type of neural network can be arbitrary according to actual design requirements. For example, the neural network can be, but is not limited to, a recurrent neural network (RNN), a long / short-term memory (LSTM), a convolutional neural network (CNN), a deep neural network (DNN).

[0112] At step 806, the network device instructs the wireless device to use the optimal uplink DMRS pattern for uplink transmission. In some embodiments, the network device sends signaling to the wireless device to change the DMRS pattern from the initial DMRS pattern to a new DMRS pattern. The signaling can be dynamic (e.g., through DCI) or semi-static (e.g., through RRC signaling or MAC CE). In some other embodiments, the wireless device can apply the new DMRS pattern after a pre-defined application delay.

[0113] In some embodiments, the network device can configure the wireless device to enable the DMRS pattern adaptation for uplink transmission in response to receiving a report from the wireless device of a capability to support the DMRS pattern adaptation for uplink transmission.

[0114] In some embodiments, the neural network model can be trained based on training data including one or more predefined uplink DMRS patterns and one or more known uplink channels. The known uplink channels can be obtained by using certain channel models.

[0115] In some embodiments, the training data can be generated by the network device. In addition, the training data can also include decision feedback data decoded by the network device from the received uplink data. For example, after decoding the PUSCH, the decoded data of the PUSCH becomes known to the network device. The decoded data can be used as the decision feedback data to further train the neural network model. By doing so, the neural network model can be trained and adjusted in real time.

[0116] In some embodiments, by using the training data, the neural network model can be trained to improve the channel estimation performance and / or reduce the DMRS overhead.

[0117] Similarly, as described with respect to the case of downlink channel estimation, there can be a trade-off between the channel estimation performance and the DMRS overhead. By taking into account the trade-off, the neural network model can be trained to take into account both the channel estimation performance and the DMRS overhead in order to achieve an optimized DMRS overhead adaptation using AI-based channel estimation.

[0118] Alternatively, in some embodiments, the two conditions described above, i.e., the channel estimation performance and the optimal DMRS pattern, can be considered separately. For example, two separate neural network models can be used for the channel estimation and the optimal DMRS pattern. The neural network model for the channel estimation can be trained using the training data such that the estimated channel can be as similar as possible to the known channel used as the ground truth. In addition, the neural network model for the optimal DMRS pattern can be trained using the training data including the pre-known optimal DMRS pattern such that the optimal DMRS pattern output by the neural network model is as close as possible to the pre-known optimal DMRS pattern.

[0119] In some embodiments, the neural network model trained by the network device for AI-based uplink channel estimation can be provided to the wireless device for performing AI-based downlink channel estimation. This can be particularly effective if the same waveform (e.g., CP-OFDM (Cyclic Prefix-OFDM)) and the same DMRS pattern are used in the downlink and the uplink.

[0120] For TDD (time division duplex), there is channel reciprocity between downlink and uplink. Thus, a neural network model trained at the network device for uplink can be easily used for downlink at the wireless device.

[0121] For FDD (frequency division duplex), even without channel reciprocity, many channel characteristics are similar between downlink and uplink. For example, if a neural network model trained for uplink is applicable to a wide range of SINR values, it can also be used for downlink.

[0122] In these embodiments, since a neural network model trained by a network device for uplink is directly provided to a wireless device for performing AI-based downlink channel estimation, the overhead of training a neural network model at the wireless device or the network device for downlink can be saved while the performance can be maintained.

[0123] Embodiments contemplated herein include an apparatus comprising means for performing one or more elements of method 700 / 800. The apparatus can be, for example, an apparatus of a base station (such as network device 218 as a base station, as described herein).

[0124] Embodiments contemplated herein include one or more non-transitory computer- readable media comprising instructions to, when executed by one or more processors of an electronic device, cause the electronic device to perform one or more elements of method 700 / 800. The non-transitory computer-readable media can be, for example, a memory of a base station (such as memory 222 of network device 218 as a base station, as described herein).

[0125] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry for performing one or more elements of method 700 / 800. The apparatus can be, for example, an apparatus of a base station (such as network device 218 as a base station, as described herein).

[0126] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions to, when executed by the one or more processors, cause the one or more processors to perform one or more elements of method 700 / 800. The apparatus can be, for example, an apparatus of a base station (such as network device 218 as a base station, as described herein).

[0127] Embodiments contemplated herein include a signal as described in or related to one or more elements of method 700 / 800.

[0128] Embodiments contemplated herein include a computer program or computer program product including instructions, where execution of the program by a processing element causes the processing element to perform one or more elements of the method 700 / 800. The processor can be a processor of a base station, such as the processor 220 of the network device 218 of the base station, as described herein. These instructions can be, for example, located in the processor of the UE and / or on a memory of the UE, such as the memory 222 of the network device 218 of the base station, as described herein.

[0129] For one or more embodiments, at least one of the components shown in one or more of the preceding figures can be configured to perform one or more operations, techniques, processes, and / or methods as described herein. For example, a baseband processor as described herein in connection with one or more of the preceding figures can be configured to operate in accordance with one or more of the examples described herein. For another example, circuitry associated with a UE, base station, network element, etc. as described above in connection with one or more of the preceding figures can be configured to operate in accordance with one or more of the examples shown herein.

[0130] Any of the above-described embodiments can be combined with any other embodiment (or combination of embodiments) unless explicitly stated otherwise. The foregoing description of one or more implementations provides functionality and / or technical advantages, but do not limit the implementations to the precise form described. Modifications and alterations, or additional functions, can be made to the implementations, as would be apparent to one of ordinary skill in the art based on the teachings of the present disclosure. Further, the description should not be interpreted as [including any particular step sequence], as such, the order of the steps can be changed while remaining within the scope of the implementations.

[0131] Embodiments and implementations of the systems and methods described herein can include various operations, which can be embodied in machine-executable instructions to be executed by a computer system. The computer system can include one or more general-purpose or special-purpose computers (or other electronic devices). The computer system can include hardware components, including specific logic for performing the operations, or can include a combination of hardware, software, and / or firmware.

[0132] It should be appreciated that the systems described herein include descriptions of specific embodiments. These embodiments can be combined into a single system, incorporated into other systems, divided into multiple systems, or otherwise divided or combined. Further, it is contemplated that parameters, attributes, aspects, etc. of one embodiment can be used in another embodiment. For clarity, these parameters, attributes, aspects, etc. are described in one or more embodiments only and it should be appreciated that these parameters, attributes, aspects, etc. can be combined with or substituted for those of another embodiment unless specifically stated otherwise.

[0133] It is well understood that the use of personally identifiable information should follow privacy policies and practices that are generally recognized as meeting or exceeding industry or governmental requirements for maintaining the privacy of users. In particular, personally identifiable information data should be managed and handled in a manner that allows registered users to exercise control over their personal information and to exercise choices regarding its use in ways that meet or exceed industry or governmental requirements for maintaining the privacy of users.

[0134] While the foregoing has been described in some detail for purposes of clarity and the best description possible, it will be apparent that certain changes and modifications might be made that will come within the principles described. It should be noted that there are many alternative ways of implementing both the processes and apparatuses described herein. Accordingly, the embodiments of the application are to be considered illustrative and not restrictive, and the description is not be limited to the details given herein, but might be modified within the scope and equivalents of the appended claims.

Claims

1. A wireless device, comprising: a memory having instructions stored therein; and at least one processor configured to execute the instructions stored in the memory to receive, from a network device, downlink data transmitted using a downlink demodulation reference signal (DMRS) pattern; perform artificial intelligence (AI)-based downlink channel estimation based on the downlink data, including: inputting one or more received downlink DMRS symbols included in the received downlink data to a neural network model for downlink channel estimation stored in the memory of the wireless device to obtain, as an output of the neural network model, an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel; and report the optimal downlink DMRS pattern to the network device.

2. The wireless device of claim 1, wherein the at least one processor is further configured to execute the instructions stored in the memory to report a capability of performing the AI-based downlink channel estimation to the network device; and enable the AI-based channel estimation based on feedback of the capability by the network device.

3. The wireless device of claim 1, wherein the neural network model is trained by the wireless device based on training data including one or more predefined downlink DMRS patterns and one or more known downlink channels, so as to improve channel estimation performance and / or reduce DMRS overhead.

4. The wireless device of claim 3, wherein the training data further includes decision feedback data decoded by the wireless device from the received downlink data.

5. The wireless device of claim 3, wherein the at least one processor is further configured to execute the instructions stored in the memory to request the network device to provide additional training data for training the neural network model.

6. The wireless device of claim 5, wherein the at least one processor is further configured to execute the instructions stored in the memory to determine whether to request the additional training data based on an evaluation of the channel estimation performance.

7. The wireless device of claim 5, wherein the additional training data includes at least one of: additional downlink data including symbols known to the wireless device in addition to the DMRS symbols included in the received downlink data; or additional downlink data using a DMRS pattern including more DMRS symbols than the DMRS symbols included in the received downlink data.

8. The wireless device of claim 1, wherein ​ The neural network model is trained by the network device based on training data comprising one or more predefined downlink DMRS patterns and one or more known downlink channels in order to improve the channel estimation performance and / or reduce the DMRS overhead, and The at least one processor is further configured to execute the instructions stored in the memory to download the neural network model from the network device.

9. The wireless device of claim 8, wherein The at least one processor is further configured to execute the instructions stored in the memory to further train the neural network model using the training data to update the neural network model; and feed back the updated neural network model to the network device.

10. The wireless device of claim 1, wherein The received downlink data comprises a physical downlink shared channel (PDSCH) and / or a physical downlink control channel (PDCCH).

11. A network device, comprising: a memory, wherein instructions are stored; and at least one processor configured to execute the instructions stored in the memory to configure a wireless device to enable artificial intelligence (AI)-based downlink channel estimation; and receive, from the wireless device, an optimal downlink DMRS pattern output from a neural network model used for the AI-based downlink channel estimation, wherein the neural network model has as input a received downlink DMRS symbol included in downlink data received by the wireless device, and has as output an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel.

12. The network device of claim 11, wherein The neural network model is trained by the wireless device based on training data comprising one or more predefined downlink DMRS patterns and one or more known downlink channels in order to improve the channel estimation performance and / or reduce the DMRS overhead, and The at least one processor is further configured to execute the instructions stored in the memory to periodically or based on channel characteristics provide additional training data to the wireless device for training the neural network model.

13. The network device of claim 11, wherein The neural network model is trained by the wireless device based on training data comprising one or more predefined downlink DMRS patterns and one or more known downlink channels in order to improve the channel estimation performance and / or reduce the DMRS overhead, and The at least one processor is further configured to execute the instructions stored in the memory to indicate to the wireless device the capability to provide additional training data; and provide the additional training data to the wireless device in response to a request from the wireless device.

14. The network device of claim 12 or 13, wherein the at least one processor is further configured to execute the instructions stored in the memory to indicate the additional training data to the wireless device in downlink control information (DCI).

15. The network device of claim 12 or 13, wherein the at least one processor is further configured to execute the instructions stored in the memory to semi-statically configure a transmission timing of the additional training data.

16. The network device of claim 12 or 13, wherein the additional training data comprises at least one of: additional downlink data comprising symbols known to the wireless device in addition to the DMRS symbols comprised in the downlink data; or additional downlink data using a DMRS pattern comprising more DMRS symbols than the DMRS symbols comprised in the received downlink data.

17. The network device of claim 11, wherein the neural network model is trained by the network device based on training data comprising one or more predefined downlink DMRS patterns and one or more known downlink channels in order to improve the channel estimation performance and / or reduce the DMRS overhead, and the at least one processor is further configured to execute the instructions stored in the memory to provide the neural network model to the wireless device.

18. The network device of claim 17, wherein the at least one processor is further configured to execute the instructions stored in the memory to receive from the wireless device an updated neural network model further trained by the wireless device.

19. The network device of claim 11, wherein the at least one processor is further configured to execute the instructions stored in the memory to broadcast to a plurality of wireless devices that the network device supports the AI-based downlink channel estimation in order to facilitate one or more wireless devices in the plurality of wireless devices having a capability to perform the AI-based downlink channel estimation to enable the AI-based downlink channel estimation.

20. The network device of claim 11, wherein the at least one processor is further configured to execute the instructions stored in the memory to in response to receiving a report of a capability to perform the AI-based downlink channel estimation from the wireless device, configure the wireless device to enable the AI-based downlink channel estimation using signaling specific to the wireless device.

21. A network device, comprising: a memory, wherein instructions are stored; and at least one processor configured to execute the instructions stored in the memory to receive from a wireless device uplink data transmitted using an uplink demodulation reference signal (DMRS) pattern; performing artificial intelligence (AI)-based uplink channel estimation based on the uplink data, including: inputting one or more received uplink DMRS symbols included in the received uplink data to a neural network model for uplink channel estimation stored in the memory of the network device to obtain an estimated uplink channel corresponding to the uplink data and an optimal uplink DMRS pattern for the estimated uplink channel as an output of the neural network model; and indicating the wireless device to use the optimal uplink DMRS pattern for uplink transmission.

22. The network device of claim 21, wherein the at least one processor is further configured to execute the instructions stored in the memory to configure the wireless device to enable DMRS pattern adaptation for uplink transmission in response to receiving a report of a capability to support DMRS pattern adaptation for uplink transmission from the wireless device.

23. The network device of claim 21, wherein the neural network model is trained based on training data including one or more predefined uplink DMRS patterns and one or more known uplink channels in order to improve the channel estimation performance and / or reduce the DMRS overhead.

24. The network device of claim 23, wherein the training data further includes decision feedback data decoded by the network device from the received uplink data.

25. The network device of claim 21, wherein the received uplink data includes a physical uplink shared channel (PUSCH) and / or a physical uplink control channel (PUCCH).

26. The network device of claim 21, wherein the at least one processor is further configured to execute the instructions stored in the memory to provide the neural network model for the AI-based uplink channel estimation to the wireless device for execution of AI-based downlink channel estimation by the wireless device.

27. A method for a wireless device, comprising: receiving downlink data transmitted using a downlink demodulation reference signal (DMRS) pattern from a network device; performing artificial intelligence (AI)-based downlink channel estimation based on the downlink data, including: inputting one or more received downlink DMRS symbols included in the received downlink data to a neural network model for downlink channel estimation to obtain an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel as an output of the neural network model; and reporting the optimal downlink DMRS pattern to the network device.

28. A method for a network device, the method comprising: configuring a wireless device to enable artificial intelligence (AI)-based downlink channel estimation; and receiving, from the wireless device, an optimal downlink DMRS pattern output from a neural network model for the AI-based downlink channel estimation, wherein the neural network model has as input downlink DMRS symbols included in downlink data received by the wireless device and has as output an estimated downlink channel corresponding to the downlink data and an optimal downlink DMRS pattern for the estimated downlink channel.

29. A method for a network device, the method comprising: receiving, from a wireless device, uplink data transmitted using an uplink demodulation reference signal (DMRS) pattern; performing artificial intelligence (AI)-based uplink channel estimation based on the uplink data, including: inputting one or more received uplink DMRS symbols included in the received uplink data to a neural network model for uplink channel estimation to obtain, as output of the neural network model, an estimated uplink channel corresponding to the uplink data and an optimal uplink DMRS pattern for the estimated uplink channel; indicating the wireless device to use the optimal uplink DMRS pattern for uplink transmission.

30. A non-transitory computer-readable storage medium storing program instructions, wherein the program instructions, when executed by a computer system, cause the computer system to perform the method of any one of claims 27 to 29.

31. A computer program product comprising program instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 27 to 29.

Images