Joint AI / ML-based denoising and compression of CSI feedback

Abstract

Joint denoising and compression of channel state information (CSI) feedback may be performed. An example apparatus may include a processor configured to perform one or more actions. The device may receive configuration information indicative of a hidden mode of operation and an encoder model. The device may receive a CSI reference signal from a network node. The device may generate an estimated channel matrix based on the CSI reference signal. The device may generate a hidden representation of the estimated channel matrix based on the hidden mode of operation and the encoder model. The device may send a hidden representation of the estimated channel matrix to the network node.

Description

[0001] Cross - reference to related applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 413,886, filed on October 6, 2022, the content of which is incorporated herein by reference. Background Art

[0003] Mobile communication using wireless communication continues to evolve. The fifth - generation mobile communication radio access technology (RAT) can be referred to as 5G New Radio (NR). The previous - generation (legacy) mobile communication RAT can be, for example, the fourth - generation (4G) Long - Term Evolution (LTE). Summary of the Invention

[0004] The systems, methods, and means described herein relate to joint denoising and compression of channel state information (CSI) feedback based on artificial intelligence / machine learning (AI / ML).

[0005] An example device (e.g., a wireless transmit - receive unit (WTRU)) can include a processor configured to perform actions. The device can receive configuration information indicating a latent operation mode and an encoder model. The device can receive a CSI reference signal from a network node. The device can generate an estimated channel matrix based on the CSI reference signal. The device can generate a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model. The device can send the latent representation of the estimated channel matrix to the network node.

[0006] The device can generate a vector representing a latent distribution associated with the estimated channel matrix.

[0007] The latent operation mode can be a multiple - latent mode. Generating a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model can involve: generating a vector representing a latent distribution associated with the estimated channel matrix; and sampling a Gaussian distribution based on the vector to generate a latent sample associated with the estimated channel matrix, wherein the latent representation of the estimated channel matrix includes the latent sample.

[0008] The latent operation mode can be a distribution mode. Generating a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model can involve generating a vector representing a latent distribution associated with the estimated channel matrix. The latent representation of the estimated channel matrix can include the vector.

[0009] The device may estimate the value of a training loss parameter based on attributes of an estimated channel matrix. The device may transmit the training loss parameter to a network node. The device may receive a gradient vector associated with a latent representation and the training loss parameter from the network node. The device may update an encoder model based on the gradient vector. The attributes of the estimated channel matrix include one or more of the following: Doppler spread, delay spread, signal-to-noise ratio (SNR), or channel rank.

[0010] The device may determine to perform CSI denoising based on the estimated channel matrix. The device may transmit an indication of the determination to the network node. A latent representation of the estimated channel matrix may be further generated based on the determination.

[0011] The device may receive a channel state information (CSI) reference signal including a noisy channel matrix. The noisy channel matrix may be encoded. Based on the encoded noisy channel matrix, a plurality of latent representation vectors may be output. A Gaussian distribution may be sampled based on the latent representation vectors. The device may output a latent representation based on the sampling. The plurality of latent representation vectors may be transmitted to a network entity for sampling the Gaussian distribution. An indication of a latent operation mode may be received. The device may determine whether to output a latent representation of the encoded noisy channel matrix based on the indication. The plurality of latent representation vectors may be generated by a neural network. The neural network may have been unsupervised trained based on an unbiased estimate of a decoder error. The unbiased estimate of the decoder error may include a Stein's unbiased risk estimate (SURE) of the decoder error. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In addition, like reference numerals in the drawings indicate like elements.

[0013] Figure 1A is a system diagram illustrating an example communication system in which one or more of the disclosed embodiments may be implemented.

[0014] Figure 1B is a system diagram of an example wireless transmit / receive unit (WTRU) that may be used within the communication system shown in Figure 1A in accordance with one embodiment.

[0015] Figure 1C is a system diagram of an example radio access network (RAN) and an example core network (CN) that may be used within the communication system shown in Figure 1A in accordance with one embodiment.

[0016] Figure 1D is a system diagram of another example RAN and another example CN that may be used within the communication system shown in Figure 1A in accordance with one embodiment.

[0017] Figure 2 Illustrates an example of channel state information (CSI) report settings, resource settings, and link configurations.

[0018] Figure 3 Illustrates an example of codebook-based precoding with feedback information.

[0019] Figure 4 Is a block diagram illustrating an example technique for joint CSI compression and denoising.

[0020] Figure 5 Is a flowchart illustrating an example technique for joint CSI compression and denoising.

[0021] Figure 6 Is a flowchart illustrating an example technique for online training of an encoder model for joint CSI compression and denoising.

[0022] Figure 7 Is a graph illustrating the number of delay taps versus the percentage of total power for indoor and outdoor datasets.

[0023] Figure 8A and Figure 8B Is a graph illustrating the mean squared error (MSE)-compression tradeoff in a supervised setting.

[0024] Figure 9A and Figure 9B Is a graph illustrating the reconstruction quality versus the effective signal-to-noise ratio (SNR).

[0025] Figure 10 Is a table illustrating the normalized reconstruction quality versus the compression ratio in high and low signal-to-noise ratio (SNR) states. Detailed Description

[0026] Figure 1A Is a diagram illustrating an example communication system 100 in which one or more of the disclosed embodiments may be implemented. The communication system 100 may be a multi-access system that provides content such as voice, data, video, messaging, broadcasts, etc. to multiple wireless users. The communication system 100 may enable multiple wireless users to access such content by sharing system resources including wireless bandwidth. For example, the communication system 100 may employ one or more channel access methods such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), single carrier FDMA (SC-FDMA), zero tail unique word DFT spread OFDM (ZT UW DTS-s OFDM), unique word OFDM (UW-OFDM), resource block filtered OFDM, filter bank multicarrier (FBMC), and the like.

[0027] As Figure 1AAs shown in [Fig. 0], the communication system 100 may include wireless transmit / receive units (WTRUs) 102a, 102b, 102c, 102d, a radio access network (RAN) 104 / 113, a core network (CN) 106 / 115, a public switched telephone network (PSTN) 108, the Internet 110, and other networks 112, but it will be appreciated that the disclosed embodiments contemplate any number of WTRUs, base stations, networks, and / or network elements. Each of the WTRUs 102a, 102b, 102c, 102d may be any type of device configured to operate and / or communicate in a wireless environment. By way of example, the WTRUs 102a, 102b, 102c, 102d - any of which may be referred to as a "station" and / or "STA" - may be configured to transmit and / or receive wireless signals and may include a user equipment (UE), a mobile station, a fixed or mobile subscriber unit, a subscription-based unit, a pager, a cellular phone, a personal digital assistant (PDA), a smartphone, a laptop, a netbook, a personal computer, a wireless sensor, a hotspot or Mi-Fi device, an Internet of Things (IoT) device, a watch or other wearable device, a head-mounted display (HMD), a vehicle, a drone, a medical device and application (e.g., remote surgery), an industrial device and application (e.g., a robot and / or other wireless devices operating in an industrial and / or automated processing chain environment), a consumer electronic device, a device operating on a commercial and / or industrial wireless network, and the like. Any of the WTRUs 102a, 102b, 102c, and 102d may be interchangeably referred to as a UE.

[0028] The communication system 100 may also include base station 114a and / or base station 114b. Each of the base stations 114a, 114b may be any type of device configured to wirelessly interface with at least one of the WTRUs 102a, 102b, 102c, 102d to facilitate access to one or more communication networks, such as the CN 106 / 115, the Internet 110, and / or other networks 112. By way of example, the base stations 114a, 114b may be a base transceiver station (BTS), a Node-B, an eNode B, a home Node B, a home eNode B, a gNB, an NR NodeB, a site controller, an access point (AP), a wireless router, and the like. Although the base stations 114a, 114b are each depicted as a single element, it will be appreciated that the base stations 114a, 114b may include any number of interconnected base stations and / or network elements.

[0029] Base station 114a may be part of RAN 104 / 113, which may also include other base stations and / or network elements (not shown), such as base station controllers (BSCs), radio network controllers (RNCs), relay nodes, etc. Base station 114a and / or base station 114b may be configured to transmit and / or receive wireless signals on one or more carrier frequencies, which may be referred to as cells (not shown). These frequencies may be in licensed spectrum, unlicensed spectrum, or a combination of licensed and unlicensed spectrum. A cell may provide coverage for wireless services to a specific geographical area, which may be relatively fixed or may vary over time. The cell may also be divided into cell sectors. For example, the cell associated with base station 114a may be divided into three sectors. Thus, in one embodiment, base station 114a may include three transceivers, i.e., one transceiver for each sector of the cell. In one embodiment, base station 114a may employ multiple-input multiple-output (MIMO) technology and may utilize multiple transceivers for each sector of the cell. For example, beamforming may be used to transmit and / or receive signals in a desired spatial direction.

[0030] Base stations 114a, 114b may communicate with one or more of WTRUs 102a, 102b, 102c, 102d via air interface 116, which may be any suitable wireless communication link (e.g., radio frequency (RF), microwave, centimeter wave, millimeter wave, infrared (IR), ultraviolet (UV), visible light, etc.). Any suitable radio access technology (RAT) may be used to establish air interface 116.

[0031] More specifically, as described above, communication system 100 may be a multi-access system and may employ one or more channel access schemes, such as CDMA, TDMA, FDMA, OFDMA, SC-FDMA, etc. For example, base station 114a in RAN 104 / 113 and WTRUs 102a, 102b, 102c may implement radio technologies such as Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (UTRA), which may use Wideband CDMA (WCDMA) to establish air interfaces 115 / 116 / 117. WCDMA may include communication protocols such as High-Speed Packet Access (HSPA) and / or Evolved HSPA (HSPA+). HSPA may include High-Speed Downlink (DL) Packet Access (HSDPA) and / or High-Speed UL Packet Access (HSUPA).

[0032] In one embodiment, base station 114a and WTRUs 102a, 102b, 102c may implement radio technologies such as evolved UMTS terrestrial radio access (E-UTRA) which may establish an air interface 116 using long term evolution (LTE) and / or advanced LTE (LTE-A) and / or LTE-A Pro.

[0033] In one embodiment, base station 114a and WTRUs 102a, 102b, 102c may implement radio technologies such as NR radio access which may establish an air interface 116 using new radio (NR).

[0034] In one embodiment, base station 114a and WTRUs 102a, 102b, 102c may implement multiple radio access technologies. For example, base station 114a and WTRUs 102a, 102b, 102c may implement LTE radio access and NR radio access together, for example using the dual connectivity (DC) principle. Thus, the air interface utilized by WTRUs 102a, 102b, 102c may be characterized by multiple types of radio access technologies and / or transmissions sent to / from multiple types of base stations (e.g., eNBs and gNBs).

[0035] In other embodiments, base station 114a and WTRUs 102a, 102b, 102c may implement radio technologies such as IEEE 802.11 (i.e., Wi-Fi), IEEE 802.16 (i.e., WiMAX), CDMA2000, CDMA2000 1X, CDMA2000 EV-DO, Interim Standard 2000 (IS-2000), Interim Standard 95 (IS-95), Interim Standard 856 (IS-856), Global System for Mobile Communications (GSM), Enhanced Data Rates for GSM Evolution (EDGE), GSM EDGE (GERAN), and the like.

[0036] Figure 1AThe base station 114b therein can be, for example, a wireless router, a home Node B, a home eNode B, or an access point, and can utilize any suitable RAT to facilitate wireless connections in a local area such as a business premises, a home, a vehicle, a campus, an industrial facility, an air corridor (e.g., for drones), a road, and the like. In one embodiment, the base station 114b and the WTRUs 102c, 102d can implement a radio technology such as IEEE 802.11 to establish a wireless local area network (WLAN). In one embodiment, the base station 114b and the WTRUs 102c, 102d can implement a radio technology such as IEEE 802.15 to establish a wireless personal area network (WPAN). In yet another embodiment, the base station 114b and the WTRUs 102c, 102d can utilize a cellular-based RAT (e.g., WCDMA, CDMA2000, GSM, LTE, LTE-A, LTE-A Pro, NR, etc.) to establish a pico cell or a femto cell. As Figure 1A shown, the base station 114b can be directly connected to the Internet 110. Thus, the base station 114b may not need to access the Internet 110 via the CN 106 / 115.

[0037] The RAN 104 / 113 can communicate with the CN 106 / 115, which can be any type of network configured to provide voice, data, applications, and / or voice over Internet protocol (VoIP) services to one or more of the WTRUs 102a, 102b, 102c, 102d. The data can have different quality of service (QoS) requirements such as different throughput requirements, latency requirements, fault tolerance requirements, reliability requirements, data throughput requirements, mobility requirements, and the like. The CN 106 / 115 can provide call control, billing services, location-based services, prepaid calling, Internet connectivity, video distribution, etc., and / or perform advanced security functions such as user authentication. Although not shown in Figure 1A the figure, it will be appreciated that the RAN 104 / 113 and / or the CN 106 / 115 can communicate directly or indirectly with other RANs that employ the same RAT or a different RAT as the RAN 104 / 113. For example, in addition to being connected to the RAN 104 / 113 that may utilize the NR radio technology, the CN 106 / 115 can also communicate with another RAN (not shown) that employs a GSM, UMTS, CDMA 2000, WiMAX, E-UTRA, or WiFi radio technology.

[0038] CN 106 / 115 can also be used as a gateway for the WTRU 102a, 102b, 102c, 102d to access the PSTN 108, the Internet 110, and / or other networks 112. The PSTN 108 can include a circuit-switched telephone network that provides plain old telephone service (POTS). The Internet 110 can include a global system of interconnected computer networks and devices (which use common communication protocols such as the Transmission Control Protocol (TCP), the User Datagram Protocol (UDP), and / or the Internet Protocol (IP) in the TCP / IP Internet protocol suite). The network 112 can include wired and / or wireless communication networks owned and / or operated by other service providers. For example, the network 112 can include another CN connected to one or more RANs, which can employ the same RAT or a different RAT as the RAN 104 / 113.

[0039] Some or all of the WTRU 102a, 102b, 102c, 102d in the communication system 100 can include multi-mode capabilities (e.g., the WTRU 102a, 102b, 102c, 102d can include multiple transceivers for communicating with different wireless networks over different wireless links). For example, Figure 1A the WTRU 102c shown in can be configured to communicate with a base station 114a that can employ a cellular-based radio technology and with a base station 114b that can employ an IEEE 802 radio technology.

[0040] Figure 1B is a system diagram of an illustrative example WTRU 102. As Figure 1B shown, the WTRU 102 can include a processor 118, a transceiver 120, transmit / receive elements 122, a speaker / microphone 124, a keyboard 126, a display / touchpad 128, a non-removable memory 130, a removable memory 132, a power supply 134, a Global Positioning System (GPS) chipset 136, other peripheral devices 138, an encoder 140, and / or an artificial intelligence / machine learning (AI / ML) module 142, among other things. It will be appreciated that the WTRU 102 can include any sub-combination of the foregoing elements while remaining consistent with the embodiments.

[0041] The processor 118 can be a general-purpose processor, a dedicated processor, a conventional processor, a digital signal processor (DSP), multiple microprocessors, one or more microprocessors associated with a DSP core, a controller, a microcontroller, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) circuit, any other type of integrated circuit (IC), a state machine, and the like. The processor 118 can perform signal encoding, data processing, power control, input / output processing, and / or any other functions that enable the WTRU 102 to operate in a wireless environment. The processor 118 can be coupled to the transceiver 120, and the transceiver 120 can be coupled to the transmit / receive element 122. Although Figure 1B the processor 118 and the transceiver 120 are depicted as separate components, it will be appreciated that the processor 118 and the transceiver 120 can be integrated together in an electronic package or chip.

[0042] The transmit / receive element 122 can be configured to transmit signals to or receive signals from a base station (e.g., base station 114a) via the air interface 116. For example, in one embodiment, the transmit / receive element 122 can be an antenna configured to transmit and / or receive RF signals. In one embodiment, the transmit / receive element 122 can be a transmitter / detector configured to transmit and / or receive IR, UV, or visible light signals, for example. In yet another embodiment, the transmit / receive element 122 can be configured to transmit and / or receive both RF and optical signals. It will be appreciated that the transmit / receive element 122 can be configured to transmit and / or receive any combination of wireless signals.

[0043] Although in Figure 1B the transmit / receive element 122 is described as a single element, the WTRU 102 can include any number of transmit / receive elements 122. More specifically, the WTRU 102 can employ MIMO technology. Thus, in one embodiment, the WTRU 102 can include two or more transmit / receive elements 122 (e.g., multiple antennas) for transmitting and receiving wireless signals via the air interface 116.

[0044] The transceiver 120 can be configured to modulate the signals to be transmitted by the transmit / receive element 122 and demodulate the signals received by the transmit / receive element 122. As described above, the WTRU 102 can have multi-mode capabilities. Thus, the transceiver 120 can include multiple transceivers for enabling the WTRU 102 to communicate via multiple RATs (such as NR and IEEE 802.11), for example.

[0045] The processor 118 of the WTRU 102 may be coupled to and may receive user input data from: the speaker / microphone 124, the keyboard 126, and / or the display / touchpad 128 (e.g., a liquid crystal display (LCD) display unit or an organic light emitting diode (OLED) display unit). The processor 118 may also output user data to the speaker / microphone 124, the keyboard 126, and / or the display / touchpad 128. In addition, the processor 118 may access information from and store data in: any type of suitable memory (such as non-removable memory 130 and / or removable memory 132). The non-removable memory 130 may include random access memory (RAM), read-only memory (ROM), a hard disk, or any other type of memory storage device. The removable memory 132 may include a subscriber identity module (SIM) card, a memory stick, a secure digital (SD) memory card, and the like. In other embodiments, the processor 118 may access information from and store data in: a memory that is not physically located on the WTRU 102 (such as a server or a home computer (not shown)).

[0046] The processor 118 may receive power from a power supply 134 and may be configured to distribute and / or control power to other components in the WTRU 102. The power supply 134 may be any suitable device for powering the WTRU 102. For example, the power supply 134 may include one or more dry cells (e.g., nickel cadmium (NiCd), nickel zinc (NiZn), nickel metal hydride (NiMH), lithium ion (Li-ion), etc.), a solar cell, a fuel cell, and the like.

[0047] The processor 118 may also be coupled to a GPS chipset 136, which may be configured to provide location information (e.g., longitude and latitude) regarding the current location of the WTRU 102. In addition to or instead of the information from the GPS chipset 136, the WTRU 102 may receive location information from a base station (e.g., base stations 114a, 114b) via an air interface 116, and / or may determine its location based on the timing of signals received from two or more nearby base stations. It will be appreciated that the WTRU 102 may obtain location information by any suitable location determination means while remaining consistent with the embodiments.

[0048] The processor 118 can be further coupled to other peripheral devices 138, which can include one or more software and / or hardware modules that provide additional features, functionality, and / or wired or wireless connectivity. For example, the peripheral device 138 can include an accelerometer, an electronic compass, a satellite transceiver, a digital camera (for photos and / or videos), a universal serial bus (USB) port, a vibration device, a television transceiver, a hands-free headset, modules, a frequency modulation (FM) radio unit, a digital music player, a media player, an electronic game player module, an Internet browser, a virtual reality and / or augmented reality (VR / AR) device, an activity tracker, and the like. The peripheral device 138 can include one or more sensors, which can be one or more of the following: a gyroscope, an accelerometer, a Hall effect sensor, a magnetometer, an orientation sensor, a proximity sensor, a temperature sensor, a time sensor, a geographic location sensor, an altimeter, a light sensor, a touch sensor, a magnetometer, a barometer, an attitude sensor, a biosensor, and / or a humidity sensor.

[0049] The WTRU 102 can include a full-duplex radio for which the transmission and reception of some or all signals (e.g., associated with a particular subframe for both UL (e.g., for transmission) and downlink (e.g., for reception)) can be concurrent and / or simultaneous. The full-duplex radio can include an interference management unit to reduce and / or substantially eliminate self-interference via hardware (e.g., a choke) or via signal processing of a processor (e.g., a separate processor (not shown) or via the processor 118). In one embodiment, the WRTU 102 can include a half-duplex radio for which the transmission and reception of some or all signals (e.g., associated with a particular subframe for UL (e.g., for transmission) or downlink (e.g., for reception)).

[0050] The encoder 140 and the AI / ML module 142 can be configured to perform joint denoising and compression of channel state information (CSI) signals, as further explained herein. Joint denoising and compression can be performed using supervised or unsupervised learning. Although shown as separate components, in some examples, the encoder 140 and / or the AI / ML module 142 can be implemented as part of the processor 118.

[0051] Figure 1C is a system diagram of the RAN 104 and the CN 106 according to one embodiment. As described above, the RAN 104 can communicate with the WTRU 102a, 102b, 102c via the air interface 116 using E-UTRA radio technology. The RAN 104 can also communicate with the CN 106.

[0052] RAN 104 may include eNode-Bs 160a, 160b, 160c, although it will be appreciated that RAN 104 may include any number of eNode-Bs while remaining consistent with the embodiments. Each of eNode-Bs 160a, 160b, 160c may include one or more transceivers for communicating with WTRUs 102a, 102b, 102c via air interface 116. In one embodiment, eNode-Bs 160a, 160b, 160c may implement MIMO technology. Thus, for example, eNode-B 160a may use multiple antennas to transmit wireless signals to and / or receive wireless signals from WTRU 102a.

[0053] Each of eNode-Bs 160a, 160b, 160c may be associated with a particular cell (not shown) and may be configured to handle radio resource management decisions, handover decisions, user scheduling in UL and / or DL, and the like. As Figure 1C shown, eNode-Bs 160a, 160b, 160c may communicate with each other via the X2 interface.

[0054] Figure 1C The CN 106 shown may include a Mobility Management Entity (MME) 162, a Serving Gateway (SGW) 164, and a Packet Data Network (PDN) Gateway (or PGW) 166. Although each of the foregoing elements is depicted as part of the CN 106, it will be appreciated that any of these elements may be owned and / or operated by an entity other than the CN operator.

[0055] The MME 162 may be connected to each of eNode-Bs 162a, 162b, 162c in the RAN 104 via the S1 interface and may act as a control node. For example, the MME 162 may be responsible for authenticating users of WTRUs 102a, 102b, 102c, bearer activation / deactivation, selecting a particular serving gateway during the initial attachment of WTRUs 102a, 102b, 102c, and the like. The MME 162 may provide control plane functionality for interworking between the RAN 104 and other RANs (not shown) employing other radio technologies such as GSM and / or WCDMA.

[0056] The SGW 164 can be connected via the S1 interface to each of the eNode-Bs 160a, 160b, 160c in the RAN 104. The SGW 164 can generally route and forward user data packets to / from the WTRUs 102a, 102b, 102c. The SGW 164 can perform other functions such as anchoring the user plane during handovers between eNode Bs, triggering paging when DL data is available for the WTRUs 102a, 102b, 102c, managing and storing the contexts of the WTRUs 102a, 102b, 102c, and the like.

[0057] The SGW 164 can be connected to the PGW 166, which can provide the WTRUs 102a, 102b, 102c with access to a packet switched network (such as the Internet 110) to facilitate communication between the WTRUs 102a, 102b, 102c and IP-enabled devices.

[0058] The CN 106 can facilitate communication with other networks. For example, the CN 106 can provide the WTRUs 102a, 102b, 102c with access to a circuit switched network such as the PSTN 108 to facilitate communication between the WTRUs 102a, 102b, 102c and traditional landline communication devices. For example, the CN 106 can include or communicate with the following: an IP gateway (e.g., an IP Multimedia Subsystem (IMS) server) that serves as an interface between the CN 106 and the PSTN 108. In addition, the CN 106 can provide the WTRUs 102a, 102b, 102c with access to other networks 112, which can include other wired and / or wireless networks owned and / or operated by other service providers.

[0059] Although the WTRU is described as a wireless terminal in Figures 1A - 1D it is contemplated that in some representative embodiments, such a terminal can (e.g., temporarily or permanently) use a wired communication interface with the communication network.

[0060] In a representative embodiment, another network 112 can be a WLAN.

[0061] In an infrastructure basic service set (BSS) mode, a WLAN can have an access point (AP) for the BSS and one or more stations (STAs) associated with the AP. The AP can have access to or be connected to a distributed system (DS) or another type of wired / wireless network that carries traffic into and / or out of the BSS. Traffic destined for an STA from outside the BSS can reach the STA through the AP and can be delivered to the STA. Traffic originating from an STA to a destination outside the BSS can be sent to the AP to be delivered to the corresponding destination. For example, traffic between STAs within the BSS can be sent through the AP, where the source STA can send the traffic to the AP and the AP can deliver the traffic to the destination STA. Traffic between STAs within the BSS can be considered and / or referred to as peer traffic. Peer traffic can be sent between a source STA and a destination STA using direct link setup (DLS) (e.g., directly between the source STA and the destination STA). In some representative embodiments, DLS can use 802.11e DLS or 802.11z tunnel DLS (TDLS). A WLAN using an independent BSS (IBSS) mode may not have an AP, and STAs within the IBSS or using the IBSS (e.g., all STAs) can communicate directly with each other. The IBSS communication mode is sometimes referred to as the "ad hoc" communication mode in this document.

[0062] When using an 802.11ac infrastructure operation mode or a similar operation mode, the AP can transmit beacons on a fixed channel, such as the primary channel. The primary channel can be of a fixed width (e.g., a 20 MHz wide bandwidth) or a width dynamically set via signaling. The primary channel can be the operating channel of the BSS and can be used by STAs to establish a connection with the AP. In some representative embodiments, for example, in an 802.11 system, carrier sense multiple access with collision avoidance (CSMA / CA) can be implemented. For CSMA / CA, STAs (e.g., each STA), including the AP, can sense the primary channel. If the primary channel is sensed / detected by a particular STA and / or determined to be busy, that particular STA can back off. Only one STA (e.g., only one station) can transmit at any given time in a given BSS.

[0063] High throughput (HT) STAs can communicate using a 40 MHz wide channel, e.g., by combining the primary 20 MHz channel with an adjacent or non - adjacent 20 MHz channel to form a 40 MHz wide channel.

[0064] A very high throughput (VHT) STA can support channels that are 20 MHz, 40 MHz, 80 MHz, and / or 160 MHz wide. 40 MHz and / or 80 MHz channels can be formed by combining adjacent 20 MHz channels. A 160 MHz channel can be formed by combining eight adjacent 20 MHz channels, or by combining two non - adjacent 80 MHz channels - this can be referred to as an 80+80 configuration. For the 80+80 configuration, after channel coding, the data can pass through a segment parser, which can split the data into two streams. The inverse fast Fourier transform (IFFT) processing and time - domain processing can be performed separately on each stream. These streams can be mapped to two 80 MHz channels, and the data can be transmitted by the transmitting STA. At the receiver of the receiving STA, the above operations for the 80+80 configuration can be reversed, and the combined data can be sent to the media access control (MAC).

[0065] 802.11af and 802.11ah support operating modes below 1 GHz. Relative to the operating modes used in 802.11n and 802.11ac, the channel operating bandwidth and carrier in 802.11af and 802.11ah are reduced. 802.11af supports 5 MHz, 10 MHz, and 20 MHz bandwidths in the television white space (TVWS) spectrum, and 802.11ah supports 1 MHz, 2 MHz, 4 MHz, 8 MHz, and 16 MHz bandwidths using non - TVWS spectrum. According to a representative embodiment, 802.11ah can support metering - type control / machine - type communication, such as MTC devices in a macro - coverage area. MTC devices can have certain capabilities (e.g., limited capabilities), which include supporting (e.g., only supporting) certain and / or limited bandwidths. MTC devices can include a battery with a battery life higher than a threshold (e.g., to maintain a very long battery life).

[0066] A WLAN system that can support multiple channels and channel bandwidths (such as 802.11n, 802.11ac, 802.11af, and 802.11ah) includes a channel that can be designated as the primary channel. The bandwidth of the primary channel can be equal to the maximum common operating bandwidth supported by all STAs in the BSS. The bandwidth of the primary channel can be set and / or restricted by the STA that supports the minimum bandwidth operating mode among all STAs operating in the BSS. In an example of 802.11ah, for an STA that supports (e.g., only supports) the 1MHz mode (e.g., an MTC type of device), the primary channel can be 1MHz wide, even if the AP and other STAs in the BSS support 2MHz, 4MHz, 8MHz, 16MHz, and / or other channel bandwidth operating modes. Carrier sensing and / or network allocation vector (NAV) settings can depend on the state of the primary channel. If the primary channel is busy, for example, due to an STA (which only supports the 1MHz operating mode) transmitting to the AP, the entire available frequency band can be considered busy, even if most of the frequency band remains idle and can be available.

[0067] In the United States, the available frequency band that can be used by 802.11ah is from 902MHz to 928MHz. In Korea, the available frequency band is from 917.5MHz to 923.5MHz. In Japan, the available frequency band is from 916.5MHz to 927.5MHz. The total bandwidth available for 802.11ah is 6MHz to 26MHz, depending on the country code.

[0068] Figure 1D is a system diagram illustrating RAN 113 and CN 115 according to one embodiment. As described above, RAN 113 can communicate with WTRUs 102a, 102b, 102c via air interface 116 using NR radio technology. RAN 113 can also communicate with CN 115.

[0069] The RAN 113 may include gNBs 180a, 180b, 180c, although it should be appreciated that the RAN 113 may include any number of gNBs while remaining consistent with the embodiments. The gNBs 180a, 180b, 180c may each include one or more transceivers for communicating with the WTRUs 102a, 102b, 102c via the air interface 116. In one embodiment, the gNBs 180a, 180b, 180c may implement MIMO technology. For example, the gNBs 180a, 108b may utilize beamforming to transmit signals to and / or receive signals from the gNBs 180a, 180b, 180c. Thus, for example, the gNB 180a may use multiple antennas to transmit wireless signals to the WTRU 102a and / or receive wireless signals from the WTRU 102a. In one embodiment, the gNBs 180a, 180b, 180c may implement carrier aggregation technology. For example, the gNB 180a may transmit multiple component carriers (not shown) to the WTRU 102a. A subset of these component carriers may be on unlicensed spectrum while the remaining component carriers may be on licensed spectrum. In one embodiment, the gNBs 180a, 180b, 180c may implement coordinated multi-point (CoMP) technology. For example, the WTRU 102a may receive coordinated transmissions from the gNB 180a and the gNB 180b (and / or gNB 180c).

[0070] The WTRUs 102a, 102b, 102c may communicate with the gNBs 180a, 180b, 180c using transmissions associated with scalable numerology. For example, the OFDM symbol interval and / or the OFDM subcarrier interval may be different for different transmissions, different cells, and / or different portions of the radio transmission spectrum. The WTRUs 102a, 102b, 102c may communicate with the gNBs 180a, 180b, 180c using subframes or transmission time intervals (TTIs) of various or scalable lengths (e.g., which contain different numbers of OFDM symbols and / or last for different lengths of absolute time).

[0071] gNBs 180a, 180b, 180c can be configured to communicate with WTRUs 102a, 102b, 102c in a stand-alone configuration and / or a non-stand-alone configuration. In a stand-alone configuration, WTRUs 102a, 102b, 102c can communicate with gNBs 180a, 180b, 180c without also being connected to another RAN (e.g., such as eNode-Bs 160a, 160b, 160c). In a stand-alone configuration, WTRUs 102a, 102b, 102c can use one or more of gNBs 180a, 180b, 180c as a mobility anchor. In a stand-alone configuration, WTRUs 102a, 102b, 102c can communicate with gNBs 180a, 180b, 180c using signals in an unlicensed band. In a non-stand-alone configuration, WTRUs 102a, 102b, 102c can communicate / connect with gNBs 180a, 180b, 180c while also communicating / connecting with another RAN such as eNode-Bs 160a, 160b, 160c. For example, WTRUs 102a, 102b, 102c can implement the DC principle to communicate with one or more gNBs 180a, 180b, 180c and one or more eNode-Bs 160a, 160b, 160c substantially simultaneously. In a non-stand-alone configuration, eNode-Bs 160a, 160b, 160c can act as a mobility anchor for WTRUs 102a, 102b, 102c, and gNBs 180a, 180b, 180c can provide additional coverage and / or throughput for serving WTRUs 102a, 102b, 102c.

[0072] Each of gNBs 180a, 180b, 180c can be associated with a specific cell (not shown) and can be configured to handle radio resource management decisions, handover decisions, user scheduling in UL and / or DL, network slicing support, dual connectivity, interworking between NR and E-UTRA, routing of user plane data to user plane functions (UPFs) 184a, 184b, routing of control plane information to access and mobility management functions (AMFs) 182a, 182b, and the like. As Figure 1D shown, gNBs 180a, 180b, 180c can communicate with each other via the Xn interface.

[0073] Figure 1DCN 115 shown in the figure may include at least one AMF 182a, 182b, at least one UPF 184a, 184b, at least one session management function (SMF) 183a, 183b, and possibly data networks (DN) 185a, 185b. Although each of the foregoing elements is depicted as part of CN 115, it will be appreciated that any of these elements may be owned and / or operated by an entity other than the CN operator.

[0074] AMF 182a, 182b may be connected to one or more of gNB 180a, 180b, 180c in RAN 113 via the N2 interface and may be used as a control node. For example, AMF 182a, 182b may be responsible for authenticating users of WTRU 102a, 102b, 102c, supporting network slicing (e.g., handling different PDU sessions with different requirements), selecting a particular SMF 183a, 183b, managing the registration area, terminating NAS signaling, mobility management, and the like. AMF 182a, 182b may use network slicing in order to customize the CN support for WTRU 102a, 102b, 102c based on the type of service that the WTRU 102a, 102b, 102c is utilizing. For example, different network slices may be established for different use cases, such as services that rely on ultra-reliable low-latency (URLLC) access, services that rely on enhanced massive mobile broadband (eMBB) access, services for machine type communication (MTC) access, and the like. AMF 162 may provide control plane functions for interworking between RAN 113 and other RANs (not shown) that employ other radio technologies, such as LTE, LTE-A, LTE-A Pro, and / or non-3GPP access technologies such as WiFi.

[0075] SMF 183a, 183b may be connected to AMF 182a, 182b in CN 115 via the N11 interface. SMF 183a, 183b may also be connected to UPF 184a, 184b in CN 115 via the N4 interface. SMF 183a, 183b may select and control UPF 184a, 184b and configure the traffic routing through UPF 184a, 184b. SMF 183a, 183b may perform other functions, such as managing and allocating UE IP addresses, managing PDU sessions, controlling policy enforcement and QoS, providing downlink data notifications, and the like. The PDU session type may be IP-based, non-IP-based, Ethernet-based, and the like.

[0076] UPF 184a and 184b can be connected to one or more of gNBs 180a, 180b, and 180c in RAN 113 via the N3 interface, which can provide access to a packet-switched network (such as the Internet 110) for WTRUs 102a, 102b, and 102c to facilitate communication between WTRUs 102a, 102b, and 102c and IP-enabled devices. UPFs 184a and 184b can perform other functions such as routing and forwarding packets, enforcing user plane policies, supporting multi-homed PDU sessions, handling user plane QoS, buffering downlink packets, providing mobility anchoring, and the like.

[0077] CN 115 can facilitate communication with other networks. For example, CN 115 can include or communicate with the following: an IP gateway (e.g., an IP Multimedia Subsystem (IMS) server) that acts as an interface between CN 115 and the PSTN 108. Additionally, CN 115 can provide access to other networks 112 for WTRUs 102a, 102b, and 102c, and the other networks 112 can include other wired and / or wireless networks owned and / or operated by other service providers. In one embodiment, WTRUs 102a, 102b, and 102c can be connected to local DNs 185a and 185b via the N3 interface to UPFs 184a and 184b and the N6 interface between UPFs 184a and 184b and data networks (DNs) 185a and 185b.

[0078] In view of Figures 1A - 1D and Figures 1A - 1D the corresponding descriptions, one or more or all of the functions described herein for one or more of WTRUs 102a-d, base stations 114a-b, eNode-Bs 160a-c, MME 162, SGW 164, PGW 166, gNBs 180a-c, AMFs 182a-b, UPFs 184a-b, SMFs 183a-b, DNs 185a-b, and / or any other device(s) described herein can be performed by one or more emulation devices (not shown). The emulation device(s) can be one or more devices configured to emulate one or more or all of the functions described herein. For example, emulation devices can be used to test other devices and / or simulate network and / or WTRU functions.

[0079] Emulation devices can be designed to perform one or more tests on other devices in a laboratory environment and / or an operator network environment. For example, one or more emulation devices can perform one or more or all functions while being fully or partially implemented and / or deployed as part of a wired and / or wireless communication network in order to test other devices within the communication network. One or more emulation devices can perform one or more or all functions while being temporarily implemented / deployed as part of a wired and / or wireless communication network. An emulation device can be directly coupled to another device for testing purposes and / or can perform tests using over-the-air wireless communication.

[0080] One or more emulation devices can perform one or more functions (including all functions) without being implemented / deployed as part of a wired and / or wireless communication network. For example, an emulation device can be used in test scenarios in a test laboratory and / or a non-deployed (e.g., test) wired and / or wireless communication network in order to implement testing of one or more components. One or more emulation devices can be test devices. An emulation device can transmit and / or receive data using direct RF coupling and / or wireless communication via an RF circuit (e.g., which can include one or more antennas).

[0081] This document provides (one or more) features associated with channel state information (CSI) reporting. CSI can include at least one of the following: a channel quality indicator (CQI), a rank indicator (RI), a precoding matrix indicator (PMI), L1 channel measurements (e.g., a reference signal received power (RSRP) such as L1-RSRP or a signal-to-interference-and-noise ratio (SINR)), a CSI reference signal (CSI-RS) resource indicator (CRI), a synchronization signal / physical broadcast channel (SS / PBCH) block resource indicator (SSBRI), a layer indicator (LI), and / or any other measurement quantity measured by a WTRU based on a configured reference signal (e.g., CSI-RS, SS / PBCH block, or any other reference signal).

[0082] This document provides example CSI report frameworks. A WTRU can be configured to report CSI over an uplink control channel (e.g., on a Physical Uplink Control Channel (PUCCH)). In some examples, the WTRU can be configured to report CSI on a UL PUSCH grant (e.g., upon request of the gNB). The CSI-RS can cover the entire bandwidth of a Bandwidth Part (BWP). The CSI-RS can cover a subset of the BWP. Whether the CSI-RS covers the entire bandwidth or a subset of the BWP can depend on the CSI-RS configuration. The CSI-RS can be configured in Physical Resource Blocks (PRBs) (e.g., in each PRB within the CSI-RS bandwidth). The CSI-RS can be configured in every other PRB within the CSI-RS bandwidth. The CSI-RS resources can be configured to be periodic, semi-persistent, or aperiodic (e.g., in the time domain). The semi-persistent CSI-RS can be similar to the periodic CSI-RS. In the semi-persistent CSI-RS, the resources can be activated (deactivated) by a Medium Access Control (MAC) Control Element (CE). In the semi-persistent CSI-RS, the WTRU can report the associated measurements if (e.g., only if) the resources are activated. For the aperiodic CSI-RS, a CSI report can be triggered. For example, the CSI report can be triggered by a request for the CSI report (e.g., in DCI). The periodic report can be carried over the PUCCH. The semi-persistent report can be carried over the PUCCH or the PUSCH. The scheduler can use the reported CSI. For example, the scheduler can use the reported CSI to allocate resource blocks (e.g., the best resource blocks). The scheduler can allocate resource blocks, determine a precoding matrix, beam, transmission mode, and / or select an appropriate Modulation and Coding Scheme (MCS) based on the time-frequency selectivity of the channel. The reliability, accuracy, and / or timeliness of the WTRU CSI report may be involved in meeting the Ultra-Reliable and Low-Latency Communication (URLLC) service criteria (e.g., requirements).

[0083] The WTRU can be configured with CSI measurement settings. For example, the WTRU can receive configuration information from the network (e.g., from the gNB). The configuration information can include one or more CSI measurement settings (e.g., CSI measurement setting information). Based on the configuration information, the WTRU can perform one or more actions (e.g., receive a signal, measure an aspect of the signal, estimate a channel based on the measurement, report a measurement and / or estimate of the channel to the network, and / or the like). The one or more actions can be indicated by the CSI measurement settings. The CSI measurement settings can include one or more CSI report settings, resource settings, and / or a link between the one or more CSI report settings and the one or more resource settings. Figure 2Illustrates an example of CSI report settings, resource settings, and the configuration of the link between one or more CSI report settings and one or more resource settings.

[0084] CSI measurement settings can include one or more configuration parameters. Example configuration parameters can include N CSI report settings (e.g., where N is greater than or equal to 1), M resource settings (e.g., where M is greater than or equal to 1), and / or CSI measurement settings that link the N CSI report settings to the M resource settings. Example CSI report settings can include one or more of the following: time domain behavior (e.g., aperiodic, periodic, and / or semi-persistent), frequency granularity (e.g., at least for PMI and CQI), CSI report type (e.g., PMI, CQI, RI, CRI, etc.), PMI type (e.g., type I or II, in the case where PMI is reported), and / or codebook configuration. Example resource settings can include one or more of the following: time domain behavior (e.g., aperiodic, periodic, and / or semi-persistent), RS type (e.g., for channel measurement and / or interference measurement), and / or S resource sets (e.g., where S is greater than or equal to 1). In some examples, a resource set (e.g., each of the S resource sets) can include K resources (e.g., where K is greater than or equal to 1).

[0085] Example CSI measurement settings can include one or more of the following: CSI report settings, resource settings, and / or reference transmission scheme settings (e.g., for CQI). For CSI reporting of a component carrier, one or more frequency granularities can be supported. Some example frequency granularities include wideband CSI, partial band CSI, and / or sub-band CSI.

[0086] This document provides one or more features associated with codebook-based precoding. Figure 3 Illustrates an example of codebook-based precoding with feedback information. The feedback information can include a precoding matrix index (PMI). The PMI can be referred to as the codeword index in the codebook.

[0087] As Figure 3 shown, the codebook can include a set of precoding vectors / matrices for one or more ranks (e.g., each rank) and the number of antenna ports. One or more precoding vectors / matrices (e.g., each of the precoding vectors / matrices) can have its own index (e.g., such that the receiver can notify the transmitter of the preferred precoding vector / matrix index). Codebook-based precoding may have performance degradation (e.g., as compared to non-codebook-based precoding, due to its limited number of precoding vectors / matrices). Codebook-based precoding can be associated with lower control signaling / feedback overhead. Table 1 shows an example codebook for 2Tx.

[0088] Table 1: 2Tx Downlink Codebook

[0089]

[0090]

[0091] This document provides example CSI processing criteria. A CSI processing unit (CPU) may be referred to as a minimum CSI processing unit, and a WTRU may support one or more CPUs (e.g., X CPUs). A WTRU with X CPUs may estimate X CSI feedback calculations in parallel. X may be a WTRU capability configuration. If the WTRU is requested to estimate more than X CSI feedbacks simultaneously, the WTRU may perform X high-priority CSI feedbacks (e.g., only X high-priority CSI feedbacks, and the rest may not be estimated).

[0092] The start and end of the CPU may be determined based on the CSI report type (e.g., aperiodic, periodic, or semi-persistent). For an aperiodic CSI report, the CPU may be occupied starting from the first orthogonal frequency division multiplexing (OFDM) symbol after PDCCH triggering until the last OFDM symbol of the PUSCH carrying the CSI report. For periodic and semi-persistent CSI reports, the CPU may be occupied starting from the first OFDM symbol of one or more associated measurement resources (e.g., not earlier than the CSI reference resource) until the last OFDM symbol of the CSI report.

[0093] Based on the following CSI measurement types (e.g., beam-based or non-beam-based), the number of occupied CPUs may vary: non-beam-related reports (e.g., when K s CSI-RS resources in a CSI-RS resource set are used for channel measurement, K s CPUs); beam-related reports (e.g., cri-RSRP, ssb-Index-RSRP, or none), e.g., since the CSI calculation complexity is low, 1 CPU may be used regardless of the number of CSI-RS resources in the CSI-RS resource set used for channel measurement, or no CPU may be available for P3 (e.g., downlink beam refinement procedure) operations or aperiodic tracking reference signal (TRS) transmissions; for an aperiodic CSI report with a single CSI-RS resource, 1 CPU may be occupied; or for a CSI report with K s CSI-RS resources, since the WTRU needs to perform CSI measurements on each CSI-RS resource, K s CPUs may be occupied.

[0094] If the number of unoccupied CPUs (N u ) is less than the number of CPUs to be used for CSI reporting (N r )(e.g., the number of CPUs required for CSI reporting), then the WTRU may discard CSI reports based on priority (e.g., in the case of UCI on PUSCH without data / HARQ), and / or the WTRU may report false information in the N r -N u CSI reports (e.g., to avoid rate matching handling of PUSCH based on priority in other cases).

[0095] Artificial intelligence (AI) may refer to behaviors exhibited by machines. Such behaviors can mimic cognitive functions to perceive, infer, adapt, and / or act. Machine learning (ML) can refer to a type of algorithm for solving problems based on learning through experience (e.g., data) without being explicitly programmed to do so (e.g., through a configured set of rules). ML can be considered a subset of AI.

[0096] Based on the nature of the data or feedback available to the learning algorithm, different machine learning paradigms can be envisioned. For example, supervised learning methods can involve learning a function that maps inputs to outputs based on labeled training examples (e.g., where each training example can include an input and a corresponding output). For example, unsupervised learning methods can involve detecting patterns in data without pre-existing labels. For example, reinforcement learning methods can involve performing a sequence of actions in an environment to increase (e.g., maximize) cumulative reward. ML algorithms can be applied using combinations or interpolations of the above learning methods. For example, semi-supervised learning methods can use a combination of a small amount of labeled data and a large amount of unlabeled data during training. In this regard, semi-supervised learning lies between unsupervised learning (e.g., no labeled training data) and supervised learning (e.g., only labeled training data).

[0097] Deep learning can refer to a class of ML algorithms that employ artificial neural networks loosely inspired by biological systems (e.g., deep neural networks (DNN)). For example, DNNs can include a class of ML models inspired by the human brain. In a DNN, inputs can be linearly transformed. In a DNN, inputs can pass through (one or more) non-linear activation functions multiple times. A DNN can include multiple layers. For example, a layer (e.g., each layer) can include a linear transformation and / or (one or more) non-linear activation functions.

[0098] A DNN can be trained using training data (e.g., via backpropagation algorithms). The DNN can be used in various fields (e.g., speech, vision, natural language, etc.) and various machine learning settings (e.g., supervised, unsupervised, semi-supervised, and / or the like). The term AI / ML-based method / process can include achieving behavior and / or compliance through data-based learning (e.g., without explicit configuration of a series of action steps). Such methods can enable a machine to learn complex behaviors (e.g., which may be difficult to specify and / or implement when using other methods).

[0099] This document provides examples of deep learning using the evidence lower bound (ELBO). The probability distribution of data with practical significance can be problematic. Based on variational inference (e.g., choosing a prior probability mass or density function with an appropriate choice), the lower bound of the divergence between two probability distributions can allow for effective estimation. The ELBO can be used in deep learning. For example, the ELBO can be used to create influential generative models (e.g., such as variational autoencoders and many variants). In supervised learning, the ELBO can be used to estimate mutual information (e.g., when the prior distribution is chosen to be a multivariate Gaussian distribution).

[0100] Deep learning can be used for downlink CSI compression and / or reconstruction in large-scale MIMO CSI feedback. This use of deep learning can outperform other (e.g., existing) compressive sensing-based methods (e.g., which rely on signal sparsity in the angular-delay domain, which may not always hold in complex real-world wireless environments). For example, compared to compressive sensing-based methods, some methods can demonstrate improved reconstruction performance of CSI from the angular-delay domain (e.g., in terms of the normalized mean square error (NMSE)).

[0101] This document provides an example loss function for deep learning-based methods for jointly denoising and compressing CSI feedback in an unsupervised learning manner. The loss function and the associated estimator can be adopted on existing deep learning-based models. For example, the loss function and the associated estimator can be used without additional parameters. The loss function and the associated estimator can be adopted without using (e.g., requiring) a centralized system design.

[0102] For compression, a variational-inference-based mutual information estimator for supervised classification can be used to control (e.g., explicitly control) the correlation-compression trade-off. A low-dimensional latent representation of the noisy CSI can be formed. The low-dimensional latent representation of the noisy CSI can retain relevant information for reconstruction while discarding the noise in the signal.

[0103] Example settings and data sets can be provided. A WTRU can receive CSI-RS (e.g., CSI-RS symbols) from a network node (e.g., a gNB). The WTRU can generate an estimated channel matrix based on the CSI-RS. For example, the WTRU can perform channel estimation on the CSI-RS. The channel estimation can be a noisy version of the true channel H. The channel estimation can be written as where W refers to the added noise. The noisy channel estimation can be used as training data. The noisy channel estimation can be a matrix. The size of the matrix can depend on the number of subcarriers, the number of transmit / receive antennas, and / or the number of orthogonal frequency division multiplexing (OFDM) symbols. An example of the effective matrix dimension can be N C ×N R ×N T (e.g., which represents the number of subcarriers (N C ), the number of transmit antennas (N T ), and the number of receive antennas (N R ).

[0104] Reference symbols (e.g., CSI-RS) received at the WTRU (e.g., during downlink communication) or at the gNB (e.g., during uplink communication) may be corrupted by different levels of noise. The reference symbols can be used for CSI estimation. As a result, the estimated CSI may be affected by the noise. Consequently, the estimated CSI may not be suitable for evaluating precoders, CSI compression, and / or combiners.

[0105] As described herein, when the input data is noisy, a CSI compression model can be trained (e.g., either online or offline). For example, the estimated CSI can be compressed and denoised simultaneously. For example, a processor configured with an AI / ML algorithm can compress and denoise the estimated CSI. For example, an AI / ML algorithm that compresses the estimated CSI can also denoise the estimated CSI.

[0106] ML-based solutions for CSI compression and denoising can be trained using a data set (e.g., for a variety of channel conditions). It may be difficult to generate such a large data set and ensure that a single model can operate effectively under channel conditions (e.g., all channel conditions). An online training scheme can be used to fine-tune or retrain the model for specific channel conditions. General deep learning models can use noisy inputs and noise-free references so that the channel can be effectively compressed and denoised. However, a noise-free channel may not be available over the air.

[0107] Therefore, techniques are provided herein for simultaneously denoising and compressing CSI data. Techniques are provided for online training of AI / ML models without noise-free reference data.

[0108] In estimation theory, in a standard noisy signal model (e.g., y = x + n, where the noise n is additive Gaussian), there exists an unbiased estimate of the mean squared error between an estimator (e.g., and the true signal (e.g., X). The unbiased estimate can be referred to as Stein's Unbiased Risk Estimate (SURE). SURE can be used for unsupervised image denoising (e.g., for unsupervised image denoising, where the estimator is parameterized with some deep learning model). In this case, unsupervised learning can be possible because an unbiased estimate of the mean squared error (MSE) can be obtained without knowing the ground truth labels. MSE can be used to evaluate the quality of ML.

[0109] SURE may be applicable to specific noise distributions (e.g., the exponential family). The first moment of SURE can be bounded. In some cases, SURE may not be used (e.g., the use of SURE may be restricted in image processing). In some other cases, the use of SURE may not be restricted (e.g., in wireless communication with additive Gaussian noise).

[0110] This document provides (one or more) features associated with preprocessing. Noisy channel estimation (e.g., noisy CSI) can be preprocessed before being used in a training process. An example method for preprocessing is the fast Fourier transform (FFT) of the channel matrix. For example, the FFT can be applied along any dimension of the channel matrix. For example, the FFT can be applied along the transmit (Tx) and receive (Rx) antenna axes. The FFT can be applied along all available axes.

[0111] This document provides (one or more) features associated with a deep learning-based encoder. CSI can be a complex-valued matrix (e.g., having real and imaginary parts representing I-Q samples). Due to the orthogonality of the I-Q channel, an example representation of the CSI matrix can be a real-valued tensor with the last dimension equal to 2 (e.g., an image with two channels). CSI data can be encoded with a neural network (e.g., a deep convolutional neural network (CNN) or a low-dimensional latent representation).

[0112] This document provides (one or more) features associated with joint denoising and compression. For unsupervised denoising, SURE can be used as an unbiased estimate of the MSE between the output of the decoder and the unavailable true CSI. For example, the unbiased estimate of the MSE can be expressed as Equation 1 below:

[0113]

[0114] where Div(·) represents the divergence, f θ (·) represents the encoder, σ 2denotes the per-element noise power, and Φ(·) denotes a composite function (e.g., including an encoder and a decoder).

[0115] For compression, a first surrogate upper bound on the mutual information (e.g., assuming a Gaussian prior) can be expressed as Equation 2 below:

[0116]

[0117] where the encoder output vector serves as the mean and variance (independently) for Gaussian sampling y = μ θ + v θ n, n ∼ N(0, I).

[0118] Denoted as a second surrogate upper bound can be expressed as Equation 3 below (e.g., where the variational prior is parameterized as a Gaussian mixture):

[0119]

[0120] where denotes the pairwise distance function shown in Equation 4.

[0121]

[0122] A first joint objective function (e.g., referred to as variational information bottleneck (VIB) + SURE or VIB mode) can be expressed as Equation 5.

[0123]

[0124] A second joint objective function can be referred to as non-linear information bottleneck (NIB) + SURE or NIB mode.

[0125] This document provides one or more features associated with online training. An example forward pass at a WTRU (e.g., an encoder) can be provided. The WTRU can be configured by a network node (e.g., a gNB) to perform CSI compression and denoising based on feedback using an AI / ML encoder. The WTRU can receive reference signals (e.g., CSI-RS, DM-RS) from a network node (e.g., a gNB). The WTRU can generate an estimated channel matrix based on the reference signals. For example, the WTRU can perform channel estimation using the reference signals. The WTRU can generate a dataset (e.g., which can be used for model training).

[0126] The WTRU may receive a trigger (e.g., from the gNB) to initiate an online training process. In one example, the WTRU may receive a flag (e.g., an information bottleneck (IB)-flag) (e.g., from the gNB) indicating the structure of the information bottleneck to be utilized. For example, the structure of the information bottleneck to be utilized may be (pre)-configured or signaled dynamically. The flag may indicate whether the model is trained and / or operated using a first joint objective function or a second joint objective function.

[0127] The WTRU may determine which structure of the information bottleneck to use. In this case, the WTRU may transmit a flag (e.g., to the gNB) to inform the gNB of the structure of the information bottleneck determined to be used. Physical downlink control channel (PDCCH), PUCCH, DCI, and / or other control signals may be used to transmit the trigger and / or the flag. The flag may indicate whether the encoder of the WTRU (e.g., encoder 140) should be in a multiple latent mode or a distribution mode.

[0128] In the distribution mode, the WTRU may use the encoder model provided herein to encode the noisy channel matrix. If the latent operation mode is the distribution mode, the WTRU may generate a latent representation of the estimated channel matrix by generating a vector representing the latent distribution associated with the estimated channel matrix. For example, in the distribution mode, the WTRU may obtain a vector as the output of the encoder model. The vector may be a latent representation of the estimated channel matrix. In the distribution mode, the WTRU may transmit the vector to the gNB.

[0129] In the multiple latent mode, the WTRU may use the encoder model provided herein to encode the estimated noisy channel matrix. If the latent operation mode is the multiple latent mode, the WTRU may generate a latent representation of the estimated channel matrix by generating a vector representing the latent distribution associated with the estimated channel matrix. For example, in the multiple latent mode, the WTRU may obtain a vector (e.g., as the output of the encoder model). In the multiple latent mode, the WTRU may sample a Gaussian distribution (e.g., )). For example, the WTRU may sample the Gaussian distribution based on the vector and / or based on a determined number of latent representations. The WTRU may sample the Gaussian distribution based on the vector to generate a latent sample associated with the estimated channel matrix. The latent representation of the estimated channel matrix may include the latent sample. The WTRU may transmit the latent sample (e.g., multiple latent representations) to the gNB.

[0130] The WTRU can be configured to utilize one or more estimated channel attributes (e.g., Doppler, delay spread, SNR, channel rank, etc.). The WTRU can estimate the value of a training loss parameter based on the attributes of the estimated channel. For example, the WTRU can be configured to use a rule-based or ML-based model to estimate the value of the training loss parameter (e.g., SURE-related parameters, loss weighting, and / or noise variance / power). The WTRU can estimate (e.g., continuously calculate) the training loss parameter value. The WTRU can transmit the training loss parameter to a network node. For example, the WTRU can send an updated value in each frame. The WTRU can send the updated value asynchronously (e.g., when the new value differs by a large margin from the previously signaled value).

[0131] From the network side (e.g., the gNB side), in the distribution mode, the gNB can obtain a vector from the WTRU. The gNB can use the vector to form one or more latent representations. The gNB can then use a decoder to estimate the decompressed channel. The gNB can use the (one or more) training loss parameters signaled by the WTRU to calculate a loss function and / or (one or more) gradients. The gNB can transmit (one or more) gradient vectors to the WTRU (e.g., two gradient vectors corresponding to each of the vectors). The WTRU can receive the (one or more) gradient vectors associated with the latent representation and the training loss parameter from the network node. The (one or more) gradient vectors can be used (e.g., by the WTRU) to update the encoder model.

[0132] In the multiple latent mode, the gNB can receive multiple latent representations. In this case, for each received latent representation, the gNB can use a decoder to estimate the decompressed channel. The gNB can use the parameters signaled by the WTRU to calculate a loss function and gradients. For each latent representation, the gNB can transmit the gradient vector back to the WTRU. The WTRU can use the gradient vector to update the encoder model.

[0133] The encoder neural network can use a noisy channel estimate (e.g., ) as an input. The encoder neural network can output a mean vector (e.g., ) and a second output (e.g., ). The dimension of may depend on the IB-Flag. In the VIB mode, a diagonal covariance matrix can be used. In the NIB mode,

[0134] The total forward pass flow at the WTRU can be expressed as Equation 6 below.

[0135]

[0136] One or more example data transmissions from a WTRU to a gNB are provided. The WTRU may receive configuration information indicating a latent operation mode and an encoder model. For example, the WTRU may (e.g., at the start of training) receive a flag to indicate whether the latent operation mode (e.g., encoder output) should be in a multiple latent mode or a distribution mode. The WTRU may generate a latent representation (e.g., a vector or latent sample) of an estimated channel matrix based on the latent operation mode and the encoder model. The dimension of the data transmitted from the WTRU to the gNB may vary depending on the latent operation mode. For example, in the distribution mode, the WTRU may use the encoder model (proposed herein) to encode a noisy channel matrix to obtain a mean (e.g., μ θ ) vector and a variance (e.g., ) vector as the output of the encoder. The WTRU may transmit the output vector to the gNB. In the multiple latent mode, the WTRU may use the encoder model (proposed herein) to encode the estimated noisy channel matrix to obtain a mean (e.g., μ θ ) vector and a variance (e.g., ) vector. The WTRU may sample a Gaussian distribution (e.g., ) based on a number of latent representations configured for the WTRU. The number of latent representations may be expressed by Equation 7 below:

[0137] z d = μ θ + ν θ · ε d Equation 7

[0138] where ε d ~ N(0, I d ) is a random sample from a zero-mean Gaussian distribution with covariance equal to the d-dimensional identity matrix. One or more latent representations (e.g., in a desired format) of the estimated channel matrix may be sent (e.g., transmitted) to a network node.

[0139] The WTRU may transmit other data and / or parameters (e.g., to enable loss and gradient calculations). For example, the WTRU may transmit the received noise variance / power, the input channel matrix and / or other parameters associated with the SURE loss.

[0140] Examples of forward passes at the gNB (e.g., decoder) are provided herein. In some examples, (e.g., depending on the latent operation mode) the gNB may utilize the received sampled vectors and generate multiple desired samples. In some examples (e.g., in the multiple latent mode), the gNB decoder can use a random Gaussian sampling module to generate samples (e.g., z d = μ d + v d · ε d ), where ε d ~ N(0, I d ) is a random sample from a zero-mean Gaussian distribution with covariance equal to the d-dimensional identity matrix).

[0141] The sampling mechanism can be executed multiple times. For example, if the sampling is executed S times, the latent representation (e.g., generated by the gNB decoder) can be expressed as Equation 8 below.

[0142]

[0143] The decoder can reconstruct the noisy CSI with a deep architecture where z d is used as the input, is used as the output, and parameterized by φ. The reconstructed noisy CSI can be expressed as Equation 9 below.

[0144]

[0145] This document provides one or more features associated with training loss and gradient backpropagation. The WTRU can estimate the value of the training loss parameter based on the attributes of the estimated channel matrix. For example, given z d , and the training loss parameter (e.g., the loss function corresponding to VIB+SURE or NIB+SURE) can be calculated (e.g., by the WTRU). The training loss parameter can be transmitted to the network node. The loss parameter can be passed to gradient descent-based learning (e.g., standard gradient descent-based learning) for backpropagation.

[0146] This document provides one or more features associated with gradient flow. The decoder (e.g., at the network node) can update the parameter φ. The network node can determine one or more gradient vectors associated with the latent representation and the training loss parameter. The decoder can transmit the one or more gradient vectors back to the encoder (e.g., the WTRU). The data transmitted by the gNB may depend on the latent operation mode. For example, in the distribution mode, the gNB can transmit two gradient vectors corresponding to each of the μ θ and ν θ vectors. For example, in the multiple latent mode, the gNB can send two gradient vectors corresponding to each z dGradient. The WTRU may update the encoder model based on the gradient vector(s) (e.g., the gradient(s) may be used by the WTRU to update the encoder model).

[0147] One or more example inferences are provided. The WTRU may be configured (e.g., by the gNB) to perform CSI feedback compression and denoising (e.g., using an AI / ML encoder). The WTRU may be configured to operate in a specified latent operation mode (e.g., multiple latent mode or distribution mode).

[0148] The WTRU may determine to perform CSI denoising (e.g., in conjunction with CSI compression). For example, the WTRU may determine to perform CSI denoising based on an estimated channel matrix. The WTRU may transmit an indication of the determination to a network node. A latent representation of the estimated channel matrix may be generated based on the determination. The WTRU may receive a trigger (e.g., from the gNB). The trigger may indicate that the WTRU perform (e.g., initiate) CSI feedback with denoising.

[0149] The WTRU may receive a reference signal (e.g., CSI-RS, DM-RS) from the gNB. The WTRU may use the reference signal to perform channel estimation. The encoded CSI feedback may be transmitted to the gNB (e.g., based on the latent operation mode). The mean (e.g., μ θ ) vector and variance (e.g., ν θ ) vector may be (re)transmitted (e.g., if operating in the distribution mode). One or more (e.g., multiple) sampled vectors may be transmitted (e.g., if operating in the multiple latent mode). A decoder (e.g., at the gNB) may utilize the received CSI feedback to reconstruct the channel (e.g., represented by ).

[0150] Figure 4 is a block diagram illustrating an example technique for joint CSI compression and denoising. As shown, the estimated channel matrix (e.g., the noisy channel matrix) may be an input to a denoising and encoding network (e.g., a joint denoising and compression network). The output of the denoising and encoding network may be a mean vector and a variance vector (e.g., μ θ and ν θ ), as shown. The vectors may be used for sampling (e.g., to generate latent samples).

[0151] Figure 5It is a flowchart illustrating an example technique for joint CSI compression and denoising. As shown, a network node may send configuration information to a WTRU. The configuration information may indicate an encoder model and a latent operation mode. The network node may trigger the WTRU to perform joint CSI compression and denoising. The WTRU may determine (e.g., based on the trigger) to perform joint CSI compression and denoising. The WTRU may notify the network node of the joint CSI compression and denoising decision.

[0152] The network node may send a reference signal (e.g., CSI-RS) to the WTRU. The WTRU may generate an estimated channel matrix based on the reference signal. The WTRU may generate a latent representation of the estimated channel based on the latent operation mode and the encoder model (e.g., by performing joint CSI compression and denoising). The WTRU may send the latent representation of the estimated channel matrix to the network node.

[0153] In the distribution mode, the latent representation may be a vector representing a latent distribution associated with the estimated channel matrix (e.g., a mean vector and a variance vector μ θ and ν θ ). In this case, the network may generate latent samples based on the latent representation.

[0154] In the multiple latent mode, the WTRU may sample a Gaussian distribution based on a vector to generate latent samples associated with the estimated channel matrix. In this case, the latent representation of the estimated channel matrix may be the latent samples.

[0155] Figure 6 It is a flowchart illustrating an example technique for online training of an encoder model for joint CSI compression and denoising. As shown, a network node may send configuration information to a WTRU. The configuration information may indicate an encoder model and a latent operation mode. Although Figure 6 not shown in, the network node may trigger the WTRU to perform joint CSI compression and denoising; the WTRU may determine (e.g., based on the trigger) to perform joint CSI compression and denoising; and the WTRU may notify the network node of the joint CSI compression and denoising decision, as Figure 5 shown in.

[0156] Online training may involve the WTRU repeating one or more actions (e.g., in a loop). For example, the WTRU may receive a reference signal (e.g., CSI-RS) from the network node. The WTRU may generate an estimated channel matrix based on the reference signal. The WTRU may generate a latent representation of the estimated channel based on the latent operation mode and the encoder model (e.g., by performing joint CSI compression and denoising). The WTRU may estimate the value of a loss parameter. The WTRU may send the latent representation of the estimated channel matrix and the estimated loss parameter (e.g., the value of the estimated loss parameter) to the network node.

[0157] As described above, in the distribution mode, a network node may generate latent samples based on latent representations (e.g., a mean vector and a variance vector). The network node may generate a gradient vector based on the latent representation and a loss parameter. The network node may send the gradient vector to a WTRU. The WTRU may update an encoder model based on the gradient vector.

[0158] This document provides example results. Figure 8A and Figure 8B Illustrates results of using VIB+SURE and / or NIB+SURE on a baseline model (e.g., CsiNet). Figure 8A Illustrates the inference performance of the VIB and NIB modes (e.g., inference performance based on an indoor dataset). The CsiNet model may rely on the true CSI (H). Aspects of the present disclosure contemplate the case where only noisy CSI (e.g., only noisy CSI) is available. If only noisy CSI is available, the models described herein may treat the noisy CSI as the true CSI (e.g., ground truth). The noisy CSI may be represented as “CsiNet (noisy)”. The reconstruction quality of the CSI may be measured by the normalized mean squared error (NMSE). The NMSE may be calculated between the reconstructed and noiseless CSI (H). Joint denoising and compression methods (e.g., VIB+SURE and NIB+SURE) may outperform the baseline model (e.g., within the range of the evaluated effective SNR).

[0159] In one example, a network node (e.g., a base station (BS) or a gNB) may be equipped with N T antennas for OFDM transmission over N C subcarriers. In this case, a user (e.g., a single-antenna user) may (e.g., at the receiver side) observe a signal that may be represented by the following equation:

[0160]

[0161] where at the nth subcarrier, y n represents the received noisy symbol, represents the channel vector with the superscript H as the Hermitian operator, represents the precoding vector, represents the transmitted symbol, and represents the additive noise. For example, the observed signal may include a CSI reference signal. For example, the CSI reference signal may cover the entire bandwidth of a bandwidth part (BWP) or a small portion of the BWP. The CSI-RS resources may be configured (e.g., in the time domain) to be periodic, semi-persistent, or aperiodic.

[0162] By arranging the channel vectors across all subcarriers, the channel matrix can be expressed as Therefore The channel matrix H can be obtained by transmitting a reference signal (e.g., v n x n , also referred to as a pilot signal) from the transmitter and estimating the reference signal at the receiver. In a frequency-division duplexing (FDD)-massive multiple-input multiple-output (MIMO) system, N T >> 1. In this case, the channel matrix H can be a high-dimensional matrix (e.g., which can be burdensome for the system due to a large amount of CSI feedback). H can have a sparse representation in the angular-delay domain. This can reduce (e.g., significantly reduce) the CSI feedback burden. By applying a 2D discrete Fourier transform (DFT) to H, the angular-delay form ( ) can be expressed as:

[0163]

[0164] where and are two DFT matrices.

[0165] In the sense that (e.g., most of) the signal power can be concentrated in the first N a << N T angles of arrival (e.g., other angles of arrival can be neglected), (e.g., compared to H) can be sparse. Therefore, can be truncated to its first N a rows without losing too much information from H.

[0166] The noisy truncated angular-delay domain CSI can be expressed as:

[0167]

[0168] where W is the additive noise (e.g., for each entry and where represents the truncation process).

[0169] can be reconstructed using the noisy compressed CSI feedback . Assuming sparsity, H can be expressed as the truncated CSI in the angular-delay domain. can represent the noisy estimate of the truncated CSI in the angular-delay domain.

[0170] This document provides one or more features associated with the joint compression and denoising of CSI. The joint compression and denoising of CSI feedback can be formulated as the following Markov chain: Provides a technique for finding an encoder and a decoder f θ , g φ parameterized by a class of learning models θ ∈ Θ, φ ∈ Φ. The reconstruction can be reduced (e.g., minimized) with respect to the mean squared error (MSE) between the reconstructed and the true . The mutual information between the noisy and the low-dimensional latent representation can be compressed (e.g., where L ≤ N and at a predetermined level C0 > 0).

[0171] The original complex CSI can (e.g., by convention) be split into real and imaginary image channels. The problem to be solved can be expressed in the constrained optimization form shown in Equation 13.

[0172]

[0173] The encoder can access (e.g., only access) the noisy CSI to form the latent representation. The MSE can be calculated with respect to (e.g., the unknown / hidden) true CSI. Using the Lagrange multiplier, Equation 13 can be rewritten as the following loss function:

[0174]

[0175] where γ is a trade-off parameter greater than 0. The trade-off parameter can be used to control the weights between the two objectives.

[0176] In the case of not knowing the true CSI (e.g., H), Equation 14 may be difficult to solve. Using SURE, the MSE term in Equation 14 can be estimated with (e.g., only) the noisy CSI (e.g., ).

[0177] The noisy CSI (e.g., only the noisy CSI) may be (e.g., in practice) obtainable. In this case, joint denoising and compression of CSI feedback may be difficult. SURE can be used for unsupervised denoising (e.g., unsupervised image denoising can be accomplished using SURE).

[0178] A k-dimensional linear model can be considered (e.g., y = x + n, where the noise n is additive Gaussian). The noise n can be independent and identically distributed Gaussian . In this case, given an estimator of the moments of x that is bounded given access (e.g., only access) to the observations y, (e.g., where ), an unbiased estimator of the MSE (e.g., V(X, ψ(Y)) := E[||ψ(Y) - X|| 2 ) can be expressed as:

[0179]

[0180] The last term in Equation 15 represents the divergence ψ(y) of the estimator.

[0181] SURE can be extended to noise statistics belonging to the exponential family applicable to colored noise. It can be assumed that It can be assumed that and H and are vectorized. Under these assumptions, the MSE can be expressed as shown in Equation 16.

[0182]

[0183] Equation 16 is independent of the true CSI (H). One can use (e.g., only use) the noisy to estimate Equation 16. The divergence term in Equation 16 may be difficult to optimize. Using Monte-Carlo estimation techniques, one can generate a set of standard normal Gaussian samples (e.g., ). Then the divergence term can be estimated (e.g., for small values δ > 0), as shown in Equation 17.

[0184]

[0185] This paper provides one or more features associated with estimating the mutual information. The mutual information (e.g., ) may be intractable in some cases (e.g., most cases). Using variational inference methods (e.g., methods that have been successful in supervised classification tasks and unsupervised clustering tasks), one can apply an alternative loss upper bound. The alternative loss upper bound can allow the mutual information to be estimated.

[0186] This paper provides one or more features associated with determining the alternative loss upper bound (e.g., features similar to those used in deriving the evidence lower bound (ELBO)). Given the observations X as input, one can construct an encoder for the predicted mean and variance pair (e.g., μ(x), v 2 (x)). A set of independent and identically distributed standard normal samples ε can be shifted and scaled to produce an output (e.g., z = μ + vε). For example, given some input, the output of the encoder can be distributed as Using this reparameterization method, the upper bound of the mutual information can be expressed by the following closed-form equation:

[0187]

[0188] where d is the dimension of the latent representation and r(Z) is the d-dimensional standard normal density function (e.g., used as a reference functional). The reparameterized mean and variance is a function of (e.g., although for clarity of presentation, the notation may be simplified).

[0189] Equation 18 can be estimated by Monte-Carlo sampling of batch (e.g., mini-batch) training data. Such sampling can be expressed as:

[0190]

[0191] where B is the size of the batch (e.g., mini-batch).

[0192] The mutual information can also be estimated by assuming that the reference density function r(Z) is a Gaussian mixture. The mutual information estimate can be expressed as:

[0193]

[0194] where H r (Z) represents the Gaussian mixture entropy.

[0195] If the Gaussian mixture has the same variance across components and between elements (e.g., v 2 )(e.g., ), then the compositional upper bound of the Gaussian mixture entropy can be expressed in closed form. If the upper bound is considered together with the reparameterization result, the following upper bound of the mutual information can be derived:

[0196]

[0197] where d(·||·) is the pairwise distance function defined as follows:

[0198]

[0199] This paper provides example estimators. One or more (e.g., two) estimators can be used in supervised or unsupervised settings (e.g., regarding the system's knowledge of the true CSI (H)). In the unsupervised scenario, Equation 17 and Equation 19 can be substituted into Equation 14 to obtain the following estimator expressed in Equation 23:

[0200]

[0201] where b denotes the batch (e.g., mini-batch) index, i denotes the index of the CSI and its reconstruction, j denotes the index of the latent dimension, and w b,i represents the sampled standard normal noise in the b-th batch (e.g., mini-batch) of the i-th element of the CSI sample. The estimator can be a loss function, which can be used to solve the problem expressed in Equation 13.

[0202] The solution of Equation 23 can be estimated without knowing the true CSI(H) (e.g., because of noise). is the only input to Equation 23). The solution of Equation 23 can be estimated in an unsupervised manner. If the true CSI(H) is known, the estimator can be adjusted for the supervised setting. For example, the estimator can be adjusted for the supervised setting by replacing the SURE estimator with the standard MSE:

[0203]

[0204] In Equations 23 and 24, the estimator of the mutual information I(Z;X) can be replaced with Equation 21. Then, the estimator can be used to obtain a replacement loss function that can be used to solve the problem in Equation 13.

[0205] Minimizing (e.g., explicitly minimizing) the mutual information I(Z;X) can be used as a regularization of the learning model and trade off the reconstruction quality (e.g., correlation) and / or the complexity of the latent representation. Experimental results can be provided. The experimental results can show that there is an optimal trade off when the hidden target (e.g., the true CSI) is hidden while the noisy estimate of the CSI is available.

[0206] In the first example, the trade off parameter γ of the estimator (e.g., two variational inference based mutual information estimators) can be selected in the supervised learning setting. The trade off parameter γ can vary in the low SNR state and the high SNR state. In the second example, the SURE estimator can be incorporated for unsupervised learning. The γ with the best performance in the first example can be selected. Then, the step size of the Monte-Carlo estimate of the divergence in the SURE estimator can be varied. Given two hyperparameters, the number of dimensions of the latent representation layer can vary in the low SNR state and the high SNR state (e.g., thus varying the compression ratio).

[0207] This paper provides the feature(s) associated with the implementation and the dataset. The loss function provided in this paper can be applied without changing the decoder architecture. The loss function provided in this paper can be applied with some (e.g., minimal) modification to the architecture of the encoder.

[0208] In some examples, CsiNet can be adopted as a baseline. The fully connected bottleneck layer of CsiNet can be replaced with a variational encoder (e.g., in Equations 21 and 22). Thus, the compression ratio can be expressed as:

[0209]

[0210] where d is the output dimension of the bottleneck layer, and K is the dimension of the CSI data sample (e.g., separating the real part and the complex part).

[0211] The indoor dataset used in CsiNet can be used as a real CSI matrix (e.g., for comparison purposes). Zero-mean circularly complex Gaussian noise can be added to each entry with a controlled noise variance, as shown in Equation 26:

[0212]

[0213] As a low SNR scenario, the value of σ can be 10 -1 (e.g., effective SNR ≈ -7.43 dB). For a high SNR scenario, the value of σ can be 10 -2 (e.g., effective SNR ≈ 12.6 dB). There may be a mapping between σ and the SNR of the indoor CsiNet data. As Figure 7 shown, the effective SNR can be the percentage of the signal power that first reaches 99% with respect to the increasing delay taps. Figure 7 The figure illustrates the sparsity in the angular delay domain of the indoor / outdoor dataset. The average squared norm of each CSI sample of the test indoor data may be approximately E in ≈ 0.93. The average squared norm of each CSI sample of the test outdoor data may be approximately E out ≈ 1.64.

[0214] This paper provides the feature(s) associated with compression, such as regularization with noisy CSI. The examples provided in this paper may have one or more (e.g., two) hyperparameters to select. For example, the hyperparameters may include the MSE compression trade-off multiplier γ and the step size ε for Monte-Carlo numerical divergence estimation.

[0215] Image denoising techniques can be used to select the value of ε. For example, a high ε may result in significant estimation errors. For example, a small ε may result in numerical instability. The techniques provided in this paper may experience similar trade-offs. An empirical study of γ (e.g., only γ) can be provided (e.g., instead of jointly evaluating two hyperparameters). The proposed loss function is applicable to the supervised learning setting.

[0216] When the observations are noiseless (e.g., as in the case of training a standard autoencoder using fixed neurons in the bottleneck layer), the signal without explicit compression can retain information (e.g., most of the information) for reconstruction. When the observations are noisy, in addition to dimensional compression (e.g., dimensionality reduction), compression of hidden features can also be provided.

[0217] Regularization can be used for the loss function during the training phase (e.g., to avoid overfitting, generalization accuracy). For example, the compression term in the IB method may have a regularization effect. Therefore, the loss function provided in this paper can balance between the reconstruction quality and the generalization error.

[0218] The average SNR of injecting noise into the indoor dataset of CsiNet can be fixed. The examples provided in this paper can be trained with a series of trade-off parameters (e.g., γ ∈ [0, 10 -2 ). If γ = 0, the loss function can depend on MSE (e.g., only MSE) (e.g., SURE for the unsupervised case).

[0219] Figure 8A and Figure 8B illustrate the MSE-compression trade-off in the supervised setting. Figure 8A and Figure 8B illustrate the effect of γ (e.g., in both high SNR and low SNR states). In Figure 8A and Figure 8B , the horizontal lines in each can respectively correspond to the cases of γ0 = 0 for low SNR and γ1 = 0 for high SNR. The latent dimensions of the two methods (e.g., VIB mode and NIB mode) can be 128. The compression ratio can be 1 / 8. γ = 0 can correspond to the case where MSE is the loss function (e.g., only the loss function) involved in the training stage of the model. In Figure 8A and Figure 8B , there are non-zero γ values such that the reconstruction quality (e.g., for the explored range) is optimized. If the achieved NMSE is lower than the NMSE of the line corresponding to γ i = 0, i ∈ {0, 1}, then there exists a trade-off parameter γ * value (e.g., the optimal choice) that achieves the reconstruction quality (e.g., the optimal reconstruction quality) for testing CSI samples. The trade-off parameter can be selected. For example, for different latent operation modes, the selected trade-off parameter may be different (e.g., for the VIB-based mode, γ v = 10 -5 ; and for the NIB-based mode, γ n = 10 -6 ).

[0220] This paper provides examples of the change from supervised CSI denoising to unsupervised CSI denoising. The value of the trade-off parameter γ (e.g., the optimal value) can be selected. The implicit operation modes can be compared at different SNRs (e.g., by controlled additive Gaussian noise). The SURE estimator can be the unbiased MSE with respect to the noisy CSI (e.g., assuming knowledge of the noise power is demonstrated through the noise level estimation phase). Thus, the SURE estimator can enable unsupervised learning. To compare CsiNet in an unsupervised scenario, the loss function of CsiNet can be replaced by SURE. The resulting unsupervised comparison scheme can be called CsiSURE. SURE may introduce an additional hyperparameter ε for the Monte-Carlo estimation of the divergence. An appropriate selection method can be used to select the value of ε.

[0221] The reconstruction quality can be compared among three modes (e.g., CsiNet, NIB mode, and VIB mode) within different SNR states. For the three modes (e.g., CsiNet, NIB mode, and VIB mode), the performance change between supervised learning and unsupervised learning can be compared.

[0222] Figure 9A and Figure 9B Illustrates the comparison of three modes (e.g., CsiNet, NIB mode, and VIB mode). Figure 9A Illustrates the comparison of the methods provided in this paper (e.g., VIB mode and NIB mode) with CsiNet in a supervised setting. As shown, the NIB mode and VIB mode can perform better at high SNR states. This may be due to explicit compression.

[0223] Figure 9B Illustrates the comparison of the modes provided in this paper (e.g., VIB mode and NIB mode) with CsiSURE in an unsupervised setting. As shown, the performance gain (e.g., due to explicit compression) can persist at high SNR states. The VIB mode can extend the improvement to the unsupervised setting (e.g., see the unsupervised low SNR case in Figure 9B ). Figure 9B Illustrates the comparison of the modes provided in this paper (e.g., VIB mode and NIB mode) with CsiNet trained with noisy CSI (e.g., only noisy CSI) in an unsupervised setting. As shown, using SURE can improve the reconstruction quality in all SNR states (e.g., all SNR states considered).

[0224] Examples of the compression ratio with noisy CSI can be provided. The comparison of the overall reconstruction quality at different compression ratios can be provided. Figure 10 is a table summarizing the above methods. As shown, Figure 10The tables in can be divided into a supervised group and an unsupervised group. For each group, each method can be evaluated in high SNR states and low SNR states (e.g., for a fixed compression ratio). The methods provided in this paper (e.g., VIB mode and NIB mode) are superior to CsiNet in high SNR cases (e.g., with non-negligible improvements). This may imply the advantages of explicit compression at different compression ratios.

[0225] In the unsupervised scenario, the modes provided in this paper (e.g., VIB mode and NIB mode) are superior to CsiSURE. The combination of SURE and explicit compression enables unsupervised training. This combination achieves higher reconstruction quality from noisy CSI over a wider range of SNR states and compression ratios.

[0226] Although the above features and elements have been described in a particular combination, each feature or element can be used alone without the other features and elements of the preferred embodiment, or in various combinations with or without other features and elements.

[0227] Although the implementations described herein may consider 3GPP-specific protocols, it should be understood that the implementations described herein are not limited to this scenario and can be applied to other wireless systems. For example, although the solutions described herein consider LTE, LTE-A, New Radio (NR), or 5G-specific protocols, it should be understood that the solutions described herein are not limited to this scenario and are also applicable to other wireless systems.

[0228] The above process can be implemented in a computer program, software, and / or firmware, which is incorporated in a computer-readable medium for execution by a computer and / or a processor. Examples of computer-readable media include, but are not limited to, electronic signals (transmitted via wired and / or wireless connections) and / or computer-readable storage media. Examples of computer-readable storage media include, but are not limited to, read-only memory (ROM), random access memory (RAM), registers, cache memories, semiconductor memory devices, magnetic media such as internal hard disks and removable disks, magneto-optical media, and / or optical media such as compact disc (CD)-ROM discs and / or digital versatile disc (DVD). A processor associated with the software can be used to implement a radio frequency transceiver for use in a WTRU, a terminal, a base station, an RNC, and / or any host computer.

[0229] It should be understood that an entity that executes the processes described herein can be a logical entity implemented in the form of software (e.g., computer-executable instructions) stored in the memory of a mobile device, network node, or computer system and executed on the processor of the mobile device, network node, or computer system. That is, these processes can be implemented in the form of software (e.g., computer-executable instructions) stored in the memory of a mobile device and / or network node (such as a node or computer system), and these computer-executable instructions, when executed by the processor of the node, execute the processes being discussed. It should also be understood that any transmission and reception processes shown in the figures can be executed by the communication circuitry of the node under the control of the processor of the node and the computer-executable instructions (e.g., software) it executes.

[0230] The various techniques described herein can be implemented in hardware or software, or in a combination of both where appropriate. Thus, an implementation of the subject matter described herein, or certain aspects or portions thereof, may take the form of program code (e.g., instructions) embodied in a tangible medium including any other machine-readable storage medium, where, when the program code is loaded into and executed by a machine such as a computer, the machine becomes an apparatus for practicing the subject matter described herein. In cases where the program code is stored on a medium, it may be the case that the program code being discussed is stored on one or more media that together execute the actions being discussed, that is, the one or more media together contain the code for performing the actions, but - in cases where there is more than one single medium - no particular portion of the code is required to be stored on any particular medium. In cases where the program code is executed on a programmable device, the computing device generally includes a processor, a storage medium readable by the processor (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device. One or more programs can be implemented or utilized, for example, by using an API, reusable controls, or the like to implement or utilize the processes described in connection with the subject matter described herein. Such programs are preferably implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the (one or more) programs can be implemented in assembly language or machine language. In any case, the language can be a compiled or interpreted language and combined with a hardware implementation.

[0231] Although example embodiments may relate to leveraging aspects of the subject matter described herein in the context of one or more standalone computing systems, the subject matter described herein is not so limited, but may be implemented in conjunction with any computing environment, such as a network or distributed computing environment. Further, aspects of the subject matter described herein may be implemented in or across multiple processing chips or devices, and storage may similarly be implemented across multiple devices. Such devices may include personal computers, network servers, handheld devices, supercomputers, or computers integrated into other systems, such as motor vehicles and airplanes.

[0232] In describing the preferred embodiments of the subject matter of this disclosure, specific terms are used for clarity, as shown in the accompanying drawings. However, the claimed subject matter is not intended to be limited to the specific terms so chosen, and it is to be understood that each specific element includes all technical equivalents that operate in a similar manner to achieve a similar purpose.

Claims

1. A wireless transmit / receive unit (WTRU) comprising: a processor configured to: receive configuration information, wherein the configuration information indicates a latent operation mode and an encoder model; receive a channel state information (CSI) reference signal from a network node; generate an estimated channel matrix based on the CSI reference signal; generate a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model; and send the latent representation of the estimated channel matrix to the network node.

2. The WTRU according to claim 1, wherein the processor is further configured to generate a vector representing a latent distribution associated with the estimated channel matrix.

3. The WTRU according to claim 1, wherein the latent operation mode includes a multiple latent mode, and the processor being configured to generate a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model includes the processor being configured to: generate a vector representing a latent distribution associated with the estimated channel matrix; and sample a Gaussian distribution based on the vector to generate a latent sample associated with the estimated channel matrix, wherein the latent representation of the estimated channel matrix includes the latent sample.

4. The WTRU according to claim 1, wherein the latent operation mode includes a distribution mode, and the processor being configured to generate a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model includes the processor being configured to generate a vector representing a latent distribution associated with the estimated channel matrix, wherein the latent representation of the estimated channel matrix includes the vector.

5. The WTRU according to claim 1, wherein the processor is further configured to: estimate a training loss parameter value based on an attribute of the estimated channel matrix; transmit the training loss parameter value to the network node; receive a gradient vector associated with the latent representation and the training loss parameter value from the network node; and update the encoder model based on the gradient vector.

6. The WTRU according to claim 5, wherein the attribute of the estimated channel matrix includes one or more of the following: Doppler spread, delay spread, signal-to-noise ratio (SNR), or channel rank.

7. The WTRU according to claim 1, wherein the processor is further configured to: determine to perform CSI denoising based on the estimated channel matrix; and transmit an indication of the determination to the network node, wherein the latent representation of the estimated channel matrix is further generated based on the determination.

8. A method comprising: receiving configuration information, wherein the configuration information indicates a latent operation mode and an encoder model; receiving a channel state information (CSI) reference signal from a network node; generating an estimated channel matrix based on the CSI reference signal; generating a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model; and sending the latent representation of the estimated channel matrix to the network node.

9. The method according to claim 8, wherein the method further includes generating a vector representing a latent distribution associated with the estimated channel matrix.

10. The method according to claim 8, wherein the latent operation mode includes a multiple latent mode, and generating a latent representation of the estimated channel matrix based on the latent operation mode and the encoder model includes: Generate a vector representing a latent distribution associated with an estimated channel matrix; and Sample a Gaussian distribution based on the vector to generate a latent sample associated with the estimated channel matrix, where the latent representation of the estimated channel matrix includes the latent sample.

11. The method according to claim 8, wherein the latent operation mode includes a distribution mode, and generating a latent representation of the estimated channel matrix based on the latent operation mode and an encoder model includes generating a vector representing a latent distribution associated with the estimated channel matrix, wherein the latent representation of the estimated channel matrix includes the vector.

12. The method according to claim 8, wherein the method further comprises: Estimate a training loss parameter value based on an attribute of the estimated channel matrix; Transmit the training loss parameter value to a network node; Receive, from the network node, a gradient vector associated with the latent representation and the training loss parameter value; and Update the encoder model based on the gradient vector.

13. The method according to claim 12, wherein the attribute of the estimated channel matrix includes one or more of the following: Doppler spread, delay spread, signal-to-noise ratio (SNR), or channel rank.

14. The method according to claim 8, wherein the method further comprises: Determine to perform CSI denoising based on the estimated channel matrix; and Transmit an indication of the determination to the network node, wherein the latent representation of the estimated channel matrix is further generated based on the determination.

Images