Joint learning of autoencoder pairs for wireless communication

Abstract

Various aspects of this disclosure generally relate to wireless communication. In some aspects, a client may use a first client autoencoder to determine a feature vector associated with one or more features, which are related to the client's environment. The client may use a second client autoencoder and determine a latent vector based at least in part on the feature vector. The client may transmit the feature vector and the latent vector. Numerous other aspects are provided.

Description

[0001] Cross-references to related applications

[0002] This patent application claims priority to Greek patent application No. 20200100498, filed on August 18, 2020, entitled "FEDERATED LEARNING OF AUTOENCODER PAIRS FOR WIRELESS COMMUNICATION," which has been assigned to the assignee of this application. The disclosure of the prior application is considered part of this patent application and is incorporated herein by reference. Technical Field

[0003] Various aspects of this disclosure generally relate to wireless communication and techniques and apparatus for reporting channel state information. Background Technology

[0004] Wireless communication systems are widely deployed to provide a variety of telecommunications services such as telephone, video, data, messaging, and broadcasting. Typical wireless communication systems employ multiple access technologies that can support communication with multiple users by sharing available system resources (e.g., bandwidth, transmit power). Examples of such multiple access technologies include Code Division Multiple Access (CDMA) systems, Time Division Multiple Access (TDMA) systems, Frequency Division Multiple Access (FDMA) systems, Orthogonal Frequency Division Multiple Access (OFDMA) systems, Single Carrier Frequency Division Multiple Access (SC-FDMA) systems, Time Division Synchronous Code Division Multiple Access (TD-SCDMA) systems, and Long Term Evolution (LTE). LTE / LTE-Advanced is an enhancement set of the Universal Mobile Telecommunications System (UMTS) mobile standard issued by the 3rd Generation Partnership Project (3GPP).

[0005] A wireless network may include several base stations (BSs) capable of supporting communication between several user equipments (UEs). UEs may communicate with the BS via downlink and uplink. A "downlink" (or forward link) refers to the communication link from the BS to the UE, and an "uplink" (or reverse link) refers to the communication link from the UE to the BS. As will be described in more detail herein, the BS may be referred to as a B-node, gNB, access point (AP), radio headend, transmit / receive point (TRP), new radio (NR) BS, 5G B-node, etc.

[0006] The multiple access technologies mentioned above have been adopted in various telecommunications standards to provide a common protocol enabling different user equipment to communicate at the city, country, region, and even global levels. NR (also known as 5G) is an enhancement set of the LTE mobile standard issued by 3GPP. NR is designed to better support mobile broadband Internet access by improving spectrum efficiency, reducing costs, improving service, utilizing new spectrum, and better integrating with other open standards that support beamforming, multiple-input multiple-output (MIMO) antenna technologies and carrier aggregation, using Orthogonal Frequency Division Multiplexing (OFDM) with a Cyclic Prefix (CP) on the downlink (DL) and CP-OFDM and / or SC-FDM (e.g., also known as Discrete Fourier Transform Extended OFDM (DFT-s-OFDM)) on the uplink (UL). However, with the continued growth in demand for mobile broadband access, there is a need for further improvements to LTE and NR technologies. Preferably, these improvements should be applicable to other multiple access technologies and telecommunications standards that employ these technologies. Summary of the Invention

[0007] In some aspects, a wireless communication method performed by a client includes: determining a feature vector associated with one or more features, the one or more features being associated with the client's environment, using a first client autoencoder. The method may include: determining a latent vector using a second client autoencoder and at least in part based on the feature vector. The method may further include: transmitting the feature vector and the latent vector.

[0008] In some aspects, a wireless communication method performed by a server includes: receiving from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment. The method may include: receiving from the client a latent vector at least partially based on the feature vector. The method may include: determining an observed wireless communication vector based at least partially on the feature vector and the latent vector. The method may further include: performing a wireless communication action based at least partially on the determined observed wireless communication vector.

[0009] In some aspects, a client for wireless communication includes: a memory and one or more processors coupled to the memory. The memory and the one or more processors are configured to: determine a feature vector associated with one or more features, the one or more features being associated with the client's environment, using a first client autoencoder. The memory and the one or more processors are further configured to: determine a latent vector using a second client autoencoder and at least partially based on the feature vector. The memory and the one or more processors are further configured to: transmit the feature vector and the latent vector.

[0010] In some aspects, a server for wireless communication includes: a memory and one or more processors coupled to the memory. The memory and the one or more processors are configured to: receive from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment. The memory and the one or more processors are further configured to: receive from the client a latent vector at least partially based on the feature vector. The memory and the one or more processors are further configured to: determine an observed wireless communication vector at least partially based on the feature vector and the latent vector. The memory and the one or more processors are further configured to: perform a wireless communication action at least partially based on the determined observed wireless communication vector.

[0011] In some aspects, a non-transient computer-readable medium storing a set of instructions for wireless communication, the set of instructions comprising one or more instructions, which, when executed by one or more processors of a client, cause the client to: determine a feature vector associated with one or more features, the one or more features being related to the client's environment, using a first client autoencoder. The one or more instructions may further cause the client to: determine a latent vector using a second client autoencoder and at least in part based on the feature vector. The one or more instructions may further cause the client to: transmit the feature vector and the latent vector.

[0012] In some aspects, a non-transient computer-readable medium storing a set of instructions for wireless communication, the set of instructions comprising one or more instructions, which, when executed by one or more processors of a server, cause the server to: receive from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment. The one or more instructions may further cause the server to: receive from the client a latent vector at least partially based on the feature vector. The one or more instructions may further cause the server to: determine an observed wireless communication vector at least partially based on the feature vector and the latent vector. The one or more instructions may further cause the server to: perform a wireless communication action at least partially based on the determined observed wireless communication vector.

[0013] In some aspects, an apparatus for wireless communication includes: means for determining a feature vector associated with one or more features using a first autoencoder, the one or more features being associated with the environment of the apparatus. The apparatus further includes: means for determining a latent vector using a second autoencoder and at least partially based on the feature vector. The apparatus further includes: means for transmitting the feature vector and the latent vector.

[0014] In some aspects, an apparatus for wireless communication includes: means for receiving from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment. The apparatus further includes: means for receiving from the client a latent vector at least partially based on the feature vector. The apparatus further includes: means for determining an observed wireless communication vector based at least partially on the feature vector and the latent vector. The apparatus further includes: means for performing a wireless communication action based at least partially on the determined observed wireless communication vector.

[0015] The aspects generally include, as described substantially with reference to the accompanying drawings and description and explained as such, methods, apparatus, systems, computer program products, non-transient computer-readable media, clients, servers, user equipment, base stations, wireless communication equipment, and / or processing systems.

[0016] The foregoing has broadly outlined the features and technical advantages of the examples according to this disclosure in an effort to facilitate a better understanding of the following detailed description. Additional features and advantages will be described thereafter. The disclosed concepts and specific examples can be readily used as the basis for modifying or designing other structures for implementing the same purposes as this disclosure. Such equivalent constructions do not depart from the scope of the appended claims. The characteristics of the concepts disclosed herein, in both their organization and manner of operation, and their associated advantages, will be better understood by considering the following description in conjunction with the accompanying drawings. Each drawing is provided for illustrative and descriptive purposes and does not define any limitation on the claims. Attached Figure Description

[0017] To gain a more detailed understanding of the features described above in this disclosure, reference can be made to various aspects of the above brief overview, some of which are illustrated in the accompanying drawings. However, it should be noted that the drawings illustrate only certain typical aspects of this disclosure and should not be considered as limiting its scope, as other equivalent aspects are permissible in this description. Identical reference numerals in different drawings may identify the same or similar elements.

[0018] Figure 1 This is a diagram illustrating an example of a wireless network according to this disclosure.

[0019] Figure 2 This is a diagram illustrating an example of communication between a base station and a user equipment (UE) in a wireless network according to this disclosure.

[0020] Figure 3 This is a diagram illustrating an example of wireless communication using an autoencoder according to the present disclosure.

[0021] Figure 4This is an illustration of an example of an oriented graphical model corresponding to an autoencoder according to the present disclosure.

[0022] Figure 5 This is a diagram illustrating an example of an autoencoder pair according to the present disclosure.

[0023] Figure 6 This is a diagram illustrating an example of wireless communication using an autoencoder according to the present disclosure.

[0024] Figure 7 This is a diagram illustrating an example of fully joint learning of autoencoder pairs according to this disclosure.

[0025] Figure 8 This is a diagram illustrating an example of partial joint learning of autoencoder pairs according to this disclosure.

[0026] Figure 9-17 This is a diagram illustrating an example of joint learning of autoencoder pairs according to this disclosure.

[0027] Figure 18 and 19 This is a diagram illustrating an example process associated with joint learning of autoencoder pairs according to this disclosure.

[0028] Figure 20 and 21 This is an example of an apparatus for wireless communication according to this disclosure.

[0029] Figure 22 and 23 These are illustrations illustrating an example of the hardware implementation of an apparatus employing a processing system according to this disclosure.

[0030] Figure 24 and 25 These are illustrations illustrating examples of the implementation of the code and circuitry system for the equipment according to this disclosure. Detailed Implementation

[0031] Clients operating within a network can measure reference signals and other information to report to a server. For example, a client may measure reference signals during beam management to achieve channel state feedback (CSF), measure the received power of reference signals from serving cells and / or neighboring cells, measure the signal strength of networks between radio access technologies (e.g., WiFi), measure sensor signals used to detect the location of one or more objects in the environment, and so on. However, reporting this information to the server can consume communication and / or network resources.

[0032] In some aspects described herein, a client (e.g., a UE, base station, transmit-receive point (TRP), network equipment, low Earth orbit (LEO) satellite, medium Earth orbit (MEO) satellite, geostationary orbit (GEO) satellite, highly elliptical orbit (HEO) satellite, etc.) may use one or more neural networks that can be trained to learn the dependence of the measured quality on individual parameters, to isolate the measured quality through the various layers (also referred to as "operations") of the one or more neural networks, and to compress the measurement in a manner that limits compression loss. The client may transmit the compressed measurement to a server (e.g., a TRP, another UE, base station, etc.).

[0033] The server can decode compressed measurements using one or more decompression and reconstruction operations associated with one or more neural networks. These decompression and reconstruction operations may be at least partially based on a feature set of the compressed dataset to produce reconstructed measurements. The server may then perform wireless communication actions based at least partially on the reconstructed measurements.

[0034] In some respects, clients and servers can use autoencoder pairs to compress and reconstruct information. In other respects, autoencoder pairs can be trained using federated learning. Federated learning is a machine learning technique that enables multiple clients to collaboratively learn a neural network model without the server collecting that data from these clients. Typically, federated learning involves training a single global neural network model from data stored on multiple clients. For example, in the FedAvg algorithm, the server sends a neural network model to the clients. Each client uses its own data to train the received neural network model and sends the updated model back to the server. The server then averages the updated neural network models from each client to obtain a new neural network model.

[0035] However, in some scenarios, some clients may operate in environments different from other clients (e.g., indoors / outdoors, stationary in a coffee shop / mobile on a highway, etc.). In some cases, different clients may suffer from different implementation aspects (e.g., different form factors, different RF impairments, etc.). As a result, from the perspective of physical layer link performance, finding a single neural network model that is well-suited for all devices may be difficult.

[0036] Based on various aspects of the techniques and apparatus described herein, partially federated learning and / or fully federated learning techniques can be used to provide and train personalized autoencoder models adapted to the corresponding clients. In this way, various aspects can contribute to better physical layer link performance. In some cases, a pair of autoencoders can be used at the client. This pair of autoencoders may include a first autoencoder and a second autoencoder.

[0037] A first autoencoder can be configured to extract features about the client's environment, which can be used to tune a second autoencoder to function well in the perceived environment. This feature vector and the observed wireless communication vector can be provided as input to the second autoencoder, which can be configured to perform wireless communication tasks, for example, by providing a latent vector. The client can provide the feature vector and the latent vector to a server, which can use a decoder corresponding to the second autoencoder to recover the observed wireless communication vector. Aspects of the techniques described herein can be used for any number of cross-node machine learning challenges, including facilitating channel state feedback, facilitating client localization, learning modulation and / or waveforms for wireless communication, and so on.

[0038] The autoencoder can be a regular autoencoder and / or a variational autoencoder. During training, in some respects, the autoencoder pairs can learn collaboratively using a joint learning process. During inference, the feature vector from the encoder of the first autoencoder can be used by the encoder of the second autoencoder when computing the latent vector. This feature vector can be considered as a learned label indicating the perceived environment. The client can provide this latent vector and the feature vector to the server, which can use the feature vector when decoding the received latent vector using the decoder of the second autoencoder. This feature vector configures the decoder of the second autoencoder to act as an "expert" on the environment perceived by the UE. In some respects, the format of the features is learned collaboratively from data available to each of the clients using joint learning, without exchanging data.

[0039] In some respects, a pair of autoencoders may be used at the client end. In other respects, one or more autoencoders may be used at the server end. The autoencoder used at the client end may be referred to herein as an "autoencoder" (when it is clear from the context that the autoencoder is used at the client end (as opposed to the server)) or a "client-side autoencoder." The autoencoder used at the server end may be referred to as an "autoencoder" (when it is clear from the context that the autoencoder is used at the server end (as opposed to the client)) or a "server-side autoencoder." In some respects, the server-side autoencoder may be a second client-side autoencoder, may be based on a second client-side autoencoder, may be similar to a second client-side autoencoder, and so on.

[0040] The various aspects of this disclosure are described more fully below with reference to the accompanying drawings. However, this disclosure may be implemented in many different forms and should not be construed as being limited to any specific structure or function given throughout this disclosure. Rather, these aspects are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art. Based on the teachings herein, those skilled in the art will appreciate that the scope of this disclosure is intended to cover any aspect of this disclosure disclosed herein, whether implemented independently of or in combination with any other aspect of this disclosure. For example, any number of aspects set forth herein may be used to implement an apparatus or practice. Furthermore, the scope of this disclosure is intended to cover such apparatuses or methods practiced using additional structures, functionalities, or structures and functionalities that complement or supplement the various aspects of this disclosure set forth herein. It should be understood that any aspect of this disclosure disclosed herein may be implemented by one or more elements of the claims.

[0041] Several aspects of a telecommunications system will now be described with reference to various devices and techniques. These devices and techniques will be described in the following detailed description and explained in the accompanying drawings by various blocks, modules, components, circuits, steps, processes, algorithms, etc. (collectively, "elements"). These elements can be implemented using hardware, software, or a combination thereof. Whether such elements are implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system.

[0042] It should be noted that although the aspects herein may be described using terms commonly associated with 5G or NR radio access technology (RAT), the aspects of this disclosure may be applied to other RATs, such as 3G RAT, 4G RAT, and / or RATs after 5G (e.g., 6G).

[0043] Figure 1 This is a diagram illustrating an example of a wireless network 100 according to this disclosure. The wireless network 100 may be a 5G (NR) network and / or an LTE network, etc., or may include its elements. The wireless network 100 may include several base stations 110 (shown as BS110a, BS 110b, BS 110c, and BS 110d) and other network entities. A base station (BS) is an entity that communicates with a user equipment (UE) and may also be referred to as an NR BS, B-node, gNB, 5G B-node (NB), access point, transmit / receive point (TRP), etc. Each BS may provide communication coverage for a specific geographic area. In 3GPP, the term "cell" may refer to the coverage area of ​​a BS and / or the BS subsystem serving that coverage area, depending on the context in which the term is used.

[0044] A BS can provide communication coverage for macrocells, picocells, femtocells, and / or another type of cell. Macrocells can cover a relatively large geographic area (e.g., a radius of several kilometers) and allow unrestricted access by UEs with a service subscription. Picocells can cover a relatively small geographic area and allow unrestricted access by UEs with a service subscription. Femtocells can cover a relatively small geographic area (e.g., a residential area) and allow restricted access by UEs associated with that femtocell (e.g., UEs in a Closed Subscriber Group (CSG)). A BS used for macrocells may be referred to as a macro BS. A BS used for picocells may be referred to as a pico BS. A BS used for femtocells may be referred to as a femto BS or a home BS. Figure 1 In the example shown, BS 110a can be a macro BS for macro cell 102a, BS 110b can be a pico BS for pico cell 102b, and BS 110c can be a femto BS for femto cell 102c. A BS may support one or more (e.g., three) cells. The terms “eNB,” “base station,” “NR BS,” “gNB,” “TRP,” “AP,” “B node,” “5G NB,” and “cell” are used interchangeably herein.

[0045] In some examples, the cell may not be stationary, and the geographical area of ​​the cell may move depending on the location of the mobile BS. In some examples, BSs may interconnect with each other and / or interconnect to one or more other BSs or network nodes (not shown) in the wireless network 100 via various types of backhaul interfaces, such as direct physical connections or virtual networks, using any suitable transport network.

[0046] The wireless network 100 may also include a relay station. A relay station is an entity capable of receiving data transmissions from an upstream station (e.g., a BS or a UE) and transmitting those data transmissions to a downstream station (e.g., a UE or a BS). A relay station may also be a UE capable of relaying transmissions for other UEs. Figure 1 In the example shown, relay BS 110d can communicate with macro BS 110a and UE 120d to facilitate communication between BS 110a and UE 120d. A relay BS can also be referred to as a relay station, relay base station, relay, etc.

[0047] In some aspects, wireless network 100 may include one or more non-terrestrial network (NTN) deployments, wherein non-terrestrial wireless communication devices may include UEs (which are interchangeably referred to herein as “non-terrestrial UEs”), BSs (which are interchangeably referred to herein as “non-terrestrial BSs” and “non-terrestrial base stations”), relay stations (which are interchangeably referred to herein as “non-terrestrial relay stations”), and so on. As used herein, “NTN” may refer to a network to which access is facilitated by non-terrestrial UEs, non-terrestrial BSs, non-terrestrial relay stations, and so on.

[0048] Wireless Network 100 may include any number of non-terrestrial wireless communication devices. Non-terrestrial wireless communication devices may include satellites, unmanned aerial vehicle (UAS) platforms, etc. Satellites may include low Earth orbit (LEO) satellites, medium Earth orbit (MEO) satellites, geostationary orbit (GEO) satellites, highly elliptical orbit (HEO) satellites, etc. UAS platforms may include high-altitude platform stations (HAPS) and may include balloons, airships, aircraft, etc. Non-terrestrial wireless communication devices may be part of an NTN separate from Wireless Network 100. Alternatively, the NTN may be part of Wireless Network 100. Satellites may use satellite communications to communicate directly and / or indirectly with other entities in Wireless Network 100. Other entities may include UEs (e.g., terrestrial UEs and / or non-terrestrial UEs), one or more other satellites in one or more NTN deployments, other types of BSs (e.g., stationary or terrestrial BSs), relay stations, one or more components and / or devices included in the core network of Wireless Network 100, etc.

[0049] Wireless network 100 can be a heterogeneous network comprising different types of Base Stations (BSs) such as macro BSs, pico BSs, femto BSs, relay BSs, etc. These different types of BSs may have different transmit power levels, different coverage areas, and different effects on interference in wireless network 100. For example, macro BSs may have high transmit power levels (e.g., 5 to 40 watts), while pico BSs, femto BSs, and relay BSs may have lower transmit power levels (e.g., 0.1 to 2 watts).

[0050] Network controller 130 can be coupled to a set of BSs and can provide coordination and control over these BSs. Network controller 130 can communicate with each BS via backhaul. These BSs can also communicate with each other directly or indirectly, for example, via wireless or wired backhaul.

[0051] UE 120 (e.g., 120a, 120b, 120c) may be distributed throughout the wireless network 100, and each UE may be stationary or mobile. UE may also be referred to as an access terminal, terminal, mobile station, subscriber unit, station, etc. UE may be a cellular phone (e.g., a smartphone), personal digital assistant (PDA), wireless modem, wireless communication device, handheld device, laptop computer, cordless phone, wireless local loop (WLL) station, tablet, camera, gaming device, netbook, smartbook, ultrabook, medical device or equipment, biometric sensor / device, wearable device (smartwatch, smart clothing, smart glasses, smart wristband, smart jewelry (e.g., smart ring, smart bracelet)), entertainment device (e.g., music or video device, or satellite radio), vehicle component or sensor, smart meter / sensor, industrial manufacturing equipment, GPS device, or any other suitable device configured to communicate via wireless or wired media.

[0052] Some UEs can be considered machine-type communication (MTC) devices, or evolved or enhanced machine-type communication (eMTC) UEs. MTC and eMTC UEs include, for example, robots, drones, remote devices, sensors, instruments, monitors, and / or location tags that can communicate with a base station, another device (e.g., a remote device), or some other entity. Wireless nodes can provide connectivity to or to a network (e.g., a wide area network, such as the Internet or a cellular network) via wired or wireless communication links, for example. Some UEs can be considered Internet of Things (IoT) devices, and / or can be implemented as NB-IoT (Narrowband Internet of Things) devices. Some UEs can be considered customer premises equipment (CPE). UE 120 can be included within a housing that houses components of UE 120, such as processor components and / or memory components. In some aspects, the processor components and memory components can be coupled together. For example, the processor components (e.g., one or more processors) and memory components (e.g., memory) can be operatively coupled, communicatively coupled, electronically coupled, and / or electrically coupled.

[0053] Generally, any number of wireless networks can be deployed in a given geographical area. Each wireless network can support a specific RAT and can operate on one or more frequencies. A RAT can also be referred to as a radio technology, air interface, etc. A frequency can also be referred to as a carrier, frequency channel, etc. Each frequency can support a single RAT in a given geographical area to avoid interference between wireless networks using different RATs. In some cases, NR or 5G RAT networks can be deployed.

[0054] In some respects, two or more UEs 120 (e.g., shown as UE 120a and UE 120e) may communicate directly using one or more sidelink channels (e.g., without using base station 110 as an intermediary). For example, UEs 120 may communicate using peer-to-peer (P2P) communication, device-to-device (D2D) communication, vehicle-to-everything (V2X) protocols (e.g., which may include vehicle-to-vehicle (V2V) protocols or vehicle-to-infrastructure (V2I) protocols), and / or mesh networks. In some respects, UEs 120 may perform scheduling operations, resource selection operations, and / or other operations described elsewhere herein as performed by base station 110.

[0055] Devices of the wireless network 100 can communicate using the electromagnetic spectrum, which can be subdivided into various categories, bands, channels, etc., based on frequency or wavelength. For example, devices of the wireless network 100 can communicate using an operating band with a first frequency range (FR1) and / or an operating band with a second frequency range (FR2), the first frequency range (FR1) spanning from 410 MHz to 7.125 GHz and the second frequency range (FR2) spanning from 24.25 GHz to 52.6 GHz. The frequencies between FR1 and FR2 are sometimes referred to as intermediate frequency bands. Although a portion of FR1 is greater than 6 GHz, FR1 is often referred to as the "sub-6 GHz band." Similarly, although different from the extremely high frequency (EHF) band (30 GHz – 300 GHz) designated as the "millimeter wave" band by the International Telecommunication Union (ITU), FR2 is often referred to as the "millimeter wave" band. Therefore, unless otherwise stated, it should be understood that, if used herein, the terms sub-6 GHz, etc., can broadly refer to frequencies less than 6 GHz, frequencies within FR1, and / or intermediate frequency band frequencies (e.g., greater than 7.125 GHz). Similarly, unless otherwise stated, it should be understood that, if used herein, the terms "millimeter wave," etc., can broadly refer to frequencies within the EHF band, frequencies within FR2, and / or intermediate frequency band frequencies (e.g., less than 24.25 GHz). It is conceivable that the frequencies included in FR1 and FR2 can be modified, and the techniques described herein are applicable to those modified frequency ranges.

[0056] like Figure 1As shown, UE 120 may include a first communication manager 140. As described in more detail elsewhere herein, the first communication manager 140 may use a first autoencoder to determine a feature vector associated with one or more features related to the environment of the UE; use a second autoencoder and at least in part based on the feature vector to determine a latent vector; and transmit the feature vector and the latent vector. Additionally or alternatively, the first communication manager 140 may perform one or more other operations described herein.

[0057] In some aspects, base station 110 may include a second communication manager 150. As described elsewhere in this document in more detail, the second communication manager 150 may receive from the UE a feature vector associated with one or more features related to the UE's environment; receive from the UE a latent vector at least partially based on the feature vector; determine an observed wireless communication vector at least partially based on the feature vector and the latent vector; and perform a wireless communication action at least partially based on the determined observed wireless communication vector. Additionally or alternatively, the second communication manager 150 may perform one or more other operations described herein.

[0058] As indicated above, Figure 1 This is provided merely as an example. Other examples may differ from those provided. Figure 1 The example described.

[0059] Figure 2 This is a diagram illustrating an example 200 of communication between a base station 110 and a UE 120 in a wireless network 100 according to this disclosure. The base station 110 may be equipped with... T One antenna 234a to 234t, while the UE 120 can be equipped with R Antennas 252a to 252r, of which generally T ≥ 1 and R ≥ 1.

[0060] At base station 110, transmit processor 220 can receive data destined for one or more UEs from data source 212, select one or more modulation and coding schemes (MCS) for each UE based at least in part on channel quality indicators (CQI) received from each UE, process (e.g., encode and modulate) the data for each UE based at least in part on the MCS selected for each UE, and provide data symbols for all UEs. Transmit processor 220 can also process system information (e.g., semi-static resource allocation information (SRPI)) and control information (e.g., CQI requests, grants, and / or upper-layer signaling), and provide overhead symbols and control symbols. Transmit processor 220 can also generate reference symbols for reference signals (e.g., cell-specific reference signals (CRS) or demodulation reference signals (DMRS)) and synchronization signals (e.g., primary synchronization signal (PSS) or secondary synchronization signal (SSS)). The transmit (TX) multiple-input multiple-output (MIMO) processor 230 can perform spatial processing (e.g., precoding) on ​​data symbols, control symbols, overhead symbols, and / or reference symbols, where applicable, and can provide T output symbol streams to T modulators (MODs) 232a to 232t. Each modulator 232 can process its respective output symbol stream (e.g., for OFDM) to obtain an output sample stream. Each modulator 232 can further process (e.g., convert to analog, amplify, filter, and up-convert) the output sample stream to obtain a downlink signal. The signals from modulators 232a to 232t... T Each downlink signal can be transmitted via T Antennas 234a to 234t were used for transmission.

[0061] At UE 120, antennas 252a to 252r can receive downlink signals from base station 110 and / or other base stations and can provide the received signals to demodulators (DEMODs) 254a to 254r respectively. Each demodulator 254 can condition (e.g., filter, amplify, downconvert, and digitize) the received signal to obtain an input sample. Each demodulator 254 can further process the input sample (e.g., for OFDM) to obtain the received symbol. MIMO detector 256 can obtain signals from all... RThe receiver processor 258 receives the received symbols from demodulators 254a to 254r, performs MIMO detection on these received symbols where applicable, and provides detected symbols. The receiver processor 258 can process (e.g., demodulate and decode) these detected symbols, providing decoded data for UE 120 to data sink 260, and providing decoded control and system information to controller / processor 280. The term "controller / processor" can refer to one or more controllers, one or more processors, or a combination thereof. The channel processor can determine the Reference Signal Received Power (RSRP) parameter, the Received Signal Strength Indicator (RSSI) parameter, the Reference Signal Received Quality (RSRQ) parameter, and / or the CQI parameter. In some aspects, one or more components of UE 120 may be included in a housing.

[0062] Network controller 130 may include communication unit 294, controller / processor 290, and memory 292. Network controller 130 may include one or more devices, such as those in the core network. Network controller 130 may communicate with base station 110 via communication unit 294.

[0063] Antennas (e.g., antennas 234a to 234t and / or antennas 252a to 252r) may include or be included within one or more antenna panels, antenna groups, antenna element assemblies, and / or antenna arrays. Antenna panels, antenna groups, antenna element assemblies, and / or antenna arrays may include one or more antenna elements. Antenna panels, antenna groups, antenna element assemblies, and / or antenna arrays may include coplanar antenna element assemblies and / or non-coplanar antenna element assemblies. Antenna panels, antenna groups, antenna element assemblies, and / or antenna arrays may include antenna elements within a single housing and / or multiple antenna elements within housings. Antenna panels, antenna groups, antenna element assemblies, and / or antenna arrays may include elements coupled to one or more transmission and / or reception components (such as...). Figure 2 One or more antenna elements (one or more components).

[0064] On the uplink, at UE 120, transmit processor 264 can receive and process data from data source 262 and control information from controller / processor 280 (e.g., reports including RSRP, RSSI, RSRQ, and / or CQI). Transmit processor 264 can also generate reference symbols for one or more reference signals. Symbols from transmit processor 264 may be pre-encoded by TX MIMO processor 266 where applicable, further processed by modulators 254a to 254r (e.g., for DFT-s-OFDM or CP-OFDM), and transmitted to base station 110. In some aspects, modulators and demodulators (e.g., MOD / DEMOD 254) of UE 120 may be included in the modem of UE 120. In some aspects, UE 120 includes a transceiver. The transceiver may include any combination of antennas 252, modulators and / or demodulators 254, MIMO detectors 256, receiver processors 258, transmitter processors 264, and / or TX MIMO processors 266. The transceiver may be used by a processor (e.g., controller / processor 280) and memory 282 to perform aspects of any of the methods described herein.

[0065] At base station 110, uplink signals from UE 120 and other UEs can be received by antenna 234, processed by demodulator 232, detected by MIMO detector 236 where applicable, and further processed by receiver processor 238 to obtain decoded data and control information transmitted by UE 120. Receiver processor 238 can provide the decoded data to data sink 239 and the decoded control information to controller / processor 240. Base station 110 may include communication unit 244 and communicate with network controller 130 via communication unit 244. Base station 110 may include scheduler 246 to schedule downlink and / or uplink communications of UE 120. In some aspects, the modulator and demodulator (e.g., MOD / DEMOD 232) of base station 110 may be included in the modem of base station 110. In some aspects, base station 110 includes transceiver. The transceiver may include any combination of antennas 234, modulators and / or demodulators 232, MIMO detectors 236, receiver processors 238, transmitter processors 220, and / or TX MIMO processors 230. The transceiver may be used by a processor (e.g., controller / processor 240) and memory 242 to perform aspects of any of the methods described herein.

[0066] The controller / processor 240 of base station 110, the controller / processor 280 of UE 120, and / or Figure 2Any other component may perform one or more techniques associated with joint learning of the autoencoder pairs for wireless communication, as described in more detail elsewhere herein. For example, the controller / processor 240 of base station 110, the controller / processor 280 of UE 120, and / or Figure 2 Any other component of (such as) can execute or direct, for example Figure 9 The process 900 Figure 10 Process 1000 Figure 11 Process 1100 Figure 12 Process 1200 Figure 13 Process 1300 Figure 14 Process 1400 Figure 15 Process 1500 Figure 16 Process 1600 Figure 17 Process 1700 Figure 18 Process 1800, Figure 19 The process 1900, and / or the operation of other processes as described herein. Memory 242 and 282 may store data and program code for base station 110 and UE 120, respectively. In some aspects, memory 242 and / or memory 282 may include: a non-transitory computer-readable medium storing one or more instructions (e.g., code and / or program code) for wireless communication. For example, when these one or more instructions are executed by one or more processors of base station 110 and / or UE 120 (e.g., directly executed, or executed after compilation, transformation, and / or interpretation), they may cause the one or more processors, UE 120, and / or base station 110 to perform or direct, for example... Figure 9 The process 900 Figure 10 Process 1000 Figure 11 Process 1100 Figure 12 Process 1200 Figure 13 Process 1300 Figure 14 Process 1400 Figure 15 Process 1500 Figure 16 Process 1600 Figure 17 Process 1700 Figure 18 Process 1800, Figure 19 The process 1900, and / or other processes as described herein. In some respects, the execution instructions may include run instructions, translate instructions, compile instructions, and / or interpret instructions, etc.

[0067] In some aspects, the client (e.g., UE 120) may include means for determining a feature vector associated with one or more features related to the environment of the UE using a first autoencoder, means for determining a latent vector using a second autoencoder and at least partially based on the feature vector, means for transmitting the feature vector and the latent vector, etc. Additionally or alternatively, UE 120 may include means for performing one or more other operations described herein. In some aspects, such means may include a communication manager 140. Additionally or alternatively, such means may include a combination of... Figure 2 One or more other components of the described UE 120, such as controller / processor 280, transmit processor 264, TX MIMO processor 266, MOD 254, antenna 252, DEMOD 254, MIMO detector 256, receive processor 258, etc.

[0068] In some aspects, the server (e.g., base station 110) may include means for receiving from the UE a feature vector associated with one or more features related to the UE's environment; means for receiving from the UE a latent vector at least partially based on the feature vector; means for determining an observed wireless communication vector at least partially based on the feature vector and the latent vector; means for performing a wireless communication action at least partially based on the determined observed wireless communication vector; and so on. Additionally or alternatively, base station 110 may include means for performing one or more other operations described herein. In some aspects, such means may include a communication manager 150. In some aspects, such means may include a combination of... Figure 2 One or more other components of the described base station 110, such as antenna 234, DEMOD 232, MIMO detector 236, receiver processor 238, controller / processor 240, transmitter processor 220, TX MIMO processor 230, MOD 232, antenna 234, etc.

[0069] although Figure 2 The boxes in the diagram are interpreted as different components, but the functions described above with respect to these boxes can be implemented by a single hardware component, software component, or combination of components. For example, the functions described with respect to transmit processor 264, receive processor 258, and / or TX MIMO processor 266 can be performed by controller / processor 280 or under the control of controller / processor 280.

[0070] As indicated above, Figure 2 This is provided merely as an example. Other examples may differ from those provided. Figure 2 The example described.

[0071] Clients operating within a network can measure reference signals and other information to report to a server. For example, a client may measure reference signals during beam management to achieve channel state feedback (CSF), measure the received power of reference signals from serving cells and / or neighboring cells, measure the signal strength of networks between radio access technologies (e.g., WiFi), measure sensor signals used to detect the location of one or more objects in the environment, and so on. However, reporting this information to the server can consume communication and / or network resources.

[0072] In some aspects described herein, a client (e.g., a UE, base station, transmit-receive point (TRP), network equipment, low Earth orbit (LEO) satellite, medium Earth orbit (MEO) satellite, geostationary orbit (GEO) satellite, highly elliptical orbit (HEO) satellite, etc.) may use one or more neural networks that can be trained to learn the dependence of the measured quality on individual parameters, to isolate the measured quality through the various layers (also referred to as "operations") of the one or more neural networks, and to compress the measurement in a manner that limits compression loss. The client may transmit the compressed measurement to a server (e.g., a TRP, another UE, base station, etc.).

[0073] The server can decode compressed measurements using one or more decompression and reconstruction operations associated with one or more neural networks. These decompression and reconstruction operations may be at least partially based on a feature set of the compressed dataset to produce reconstructed measurements. The server may then perform wireless communication actions based at least partially on the reconstructed measurements.

[0074] In some respects, clients and servers can use autoencoder pairs to compress and reconstruct information. In other respects, autoencoder pairs can be trained using federated learning. Federated learning is a machine learning technique that enables multiple clients to collaboratively learn a neural network model without the server collecting that data from user devices. Typically, federated learning involves training a single global neural network model from data stored on multiple clients. For example, in the FedAvg algorithm, the server sends a neural network model to the clients. Each client uses its own data to train the received neural network model and sends the updated model back to the server. The server then averages the updated neural network models from each client to obtain a new neural network model.

[0075] However, in some scenarios, some clients may operate in environments different from other clients (e.g., indoors / outdoors, stationary in a coffee shop / mobile on a highway, etc.). In some cases, different clients may suffer from different implementation aspects (e.g., different form factors, different RF impairments, etc.). As a result, from the perspective of physical layer link performance, finding a single neural network model that is well-suited for all devices may be difficult.

[0076] Based on aspects of the techniques and apparatus described herein, partially joint learning and / or fully joint learning techniques can be used to provide and train personalized autoencoder models adapted to the corresponding client. In this way, aspects can contribute to better physical layer link performance. In some aspects, a pair of autoencoders can be used at the client. The first autoencoder can be configured to extract features about the client's environment, which can be used to tune the second autoencoder to work well in the perceived environment. This feature vector and the observed wireless communication vector can be provided as input to the second autoencoder, which can be configured to perform wireless communication tasks, for example, by providing a latent vector. The client can provide the feature vector and the latent vector to a server, which can use a decoder corresponding to the second autoencoder to recover the observed wireless communication vector. Aspects of the techniques described herein can be used for any number of cross-node machine learning challenges, including facilitating channel state feedback, facilitating client localization, learning modulation and / or waveforms for wireless communication, etc.

[0077] The autoencoder can be a regular autoencoder and / or a variational autoencoder. During training, in some respects, the autoencoder pairs can learn collaboratively using a joint learning process. During inference, the feature vector from the encoder of the first autoencoder can be used by the encoder of the second autoencoder when computing the latent vector. This feature vector can be considered as a learned label indicating the perceived environment. The client can provide this latent vector and the feature vector to the server, which can use the feature vector when decoding the received latent vector using the decoder of the second autoencoder. This feature vector configures the decoder of the second autoencoder to act as an "expert" on the environment perceived by the UE. In some respects, the format of the features is learned collaboratively from data available to each of the clients using joint learning, without exchanging data.

[0078] Figure 3 This is a diagram illustrating an example 300 of wireless communication using a self-encoder according to the present disclosure. As shown, client 302 and server UE 304 can communicate with each other. In some aspects, client 302 and server 304 can communicate via a wireless network (e.g., Figure 1The wireless networks 100 shown communicate with each other. In some respects, more than one client 302 and / or more than one server 304 can communicate with each other.

[0079] In some aspects, client 302 may be, similar to, include, or be included in a wireless communication device (e.g., UE 120, base station 110, IAB node, etc.). In some aspects, server 304 may be, similar to, include, or be included in a wireless communication device (e.g., UE 120, base station 110, IAB node, etc.). For example, in some aspects, client 302 may be UE 120 and server 304 may be a base station, and client 302 and server 304 may communicate via an access link. In some aspects, client 302 and server 304 may be UE 120 communicating via a side link.

[0080] As shown, client 302 may include communication manager 306 (e.g., Figure 1 The communication manager 140 shown can be configured to perform one or more wireless communication tasks using a first client autoencoder 308 and a second client autoencoder 310. In some aspects, the first client autoencoder 308 may be, similar to, include, or be included in... Figure 5 The first autoencoder 510 shown and described below. In some aspects, the second client autoencoder 310 may be, similar to, include, or be included in. Figure 5 As shown and in the second autoencoder 520 described below.

[0081] As shown, the first client autoencoder 308 may include a first encoder 312, which is configured to receive observed environment vectors. f As input and providing feature vectors y As output, the first client-side autoencoder 308 also includes a first decoder 314, which is configured to receive feature vectors. y It serves as input and provides the observed environment vector as output.

[0082] As shown, the second client autoencoder 310 may include a second encoder 316 configured to receive observed wireless communication vectors. x and eigenvectors y As input, a latent vector is provided. h As output, the second client autoencoder 310 may also include a second decoder 318 configured to receive the latent vector. h and eigenvectors y As input, provide the observed wireless communication vector. xAs output, in operation, as indicated by the features represented by dashed lines, client 302 can determine the feature vector using the first and second encoders 312 and 316 of the first and second client autoencoders 308 and 310, respectively. y and latent vectors h .

[0083] like Figure 3 As shown, server 304 may include communication manager 320 (e.g., communication manager 150), which may be configured to perform one or more wireless communication tasks using server autoencoder 322. For example, in some aspects, server autoencoder 322 may correspond to a second client autoencoder 310. In some aspects, server autoencoder 322 may be, similar to, include, or be included in... Figure 5 As shown and in the second autoencoder 520 described below.

[0084] In some aspects, the server autoencoder 322 may include an encoder 324 configured to receive observed wireless communication vectors. x and eigenvectors y As input, a latent vector is provided. h As output, the server autoencoder 322 may also include a decoder 326 configured to receive the latent vector. h and eigenvectors y As input, provide the observed wireless communication vector. x As output, in operation, as indicated by the features represented by dashed lines, server 304 can utilize the decoder 326 of server autoencoder 322 to determine the observed wireless communication vector. x .

[0085] like Figure 3 As shown, client 302 may include a transceiver (shown as "Tx / Rx") 328 that facilitates wireless communication with transceiver 330 of server 304. As indicated by reference numeral 332, server 304 may, for example, use transceiver 330 to transmit wireless communication to client 302. In some aspects, the wireless communication may include reference signals (such as Channel State Information Reference Signal (CSI-RS)). Transceiver 328 of client 302 may receive the wireless communication. Communication manager 306 may determine the observed wireless communication vector at least in part based on the wireless communication. x For example, in the aspect where wireless communication is CSI-RS, the observed wireless communication vectors x It may include channel state information (CSI).

[0086] As shown, the communication manager 306 can obtain the observed environment vector. f and the observed environment vectorf A first encoder 312 is provided to a first client autoencoder 308. A communication manager 306 can obtain observed environment vectors from memory, from one or more sensors, etc. As shown, the first encoder 312 can at least partially base its operation on the observed environment vectors. f Determine the feature vector y As shown, the communication manager 306 can transmit feature vectors y and observed wireless communication vectors x The second encoder 316 is provided as input to the second client autoencoder 310. The second encoder 316 of the second client autoencoder 310 may be at least partially based on the feature vector. y and observed wireless communication vectors x To determine the hidden vector h .

[0087] As indicated by reference numeral 334 in the attached figure, the communication manager 306 can transmit feature vectors. y and latent vectors h The feature vector is provided to transceiver 328 for transmission. As indicated by reference numeral 336, transceiver 328 can transmit and transceiver 330 of server 304 can receive the feature vector. y and latent vectors h As shown, the communication manager 320 of server 304 can transmit feature vectors. y and latent vectors h The decoder 326 is provided as input to the server autoencoder 322. The decoder 326 may be based at least partially on the feature vector. y and latent vectors h To determine (e.g., reconstruct) the observed wireless communication vectors x In some respects, server 304 may be based at least in part on observed wireless communication vectors. x To perform wireless communication actions. For example, in the observed wireless communication vector x Including CSI aspects, the communication manager 320 of server 304 can use CSI for communication grouping, beamforming, etc.

[0088] As indicated above, Figure 3 This is provided merely as an example. Other examples may differ from those provided. Figure 3 The example described.

[0089] Figure 4 This is an illustration of example 400 of an oriented graphical model corresponding to an autoencoder according to the present disclosure. In some aspects, for example, the oriented graphical model may correspond to... Figure 3 The client-side self-encoders 308 and 310 shown are... Figure 3The server autoencoder 322, etc., shown in the figure.

[0090] In some aspects, the client s The data distribution at this location can be derived from Figure 4 The directional graphical model shown is used to represent this. This model provides a client-side solution. s ; Observed wireless communication vector x Hidden vectors h ; Observed environment vector f ; and eigenvectors y Examples of representations of relationships. In some aspects, the observed environment vectors. f This may include observable clients that facilitate learning. s One or more variables of the environment. In some respects, feature vectors y This may include summarizing, aggregating, and / or otherwise characterizing observed environment vectors. f The information in the text is learned through "tags" or features.

[0091] In some respects, latent vectors h It can be associated with wireless communication tasks. In some aspects, wireless communication tasks may include: determining channel state feedback (CSF), determining location information associated with a client, determining modulation associated with wireless communication, determining waveforms associated with wireless communication, and so on.

[0092] For example, in some aspects, paired autoencoders can be used to compress CSF to feed CSI back to the server. In some aspects, observed wireless communication vectors x This may include the propagation channel estimated at least in part by the client (e.g., UE 120) based on the received CSI-RS. Latent vector h This may include compressed CSF that is to be fed back to the server (e.g., base station 110).

[0093] In some respects, the observed environment vectors f This may include clients (e.g., Figure 3 The client 302 shown can obtain any number of different types of information about its environment. This information may include information about the client itself (e.g., device information, configuration information, capability information, etc.), information about the state associated with the client (e.g., operating state, power state, activation state, etc.), information about the client's location (e.g., location information, orientation information, geographic information, motion information, etc.), information about the client's surrounding environment (e.g., weather information, information about obstacles affecting wireless signals around the client, information about materials near the client, etc.), and so on. Observed environment vector fIt can be formed by cascading one or more information indicators (such as those listed above).

[0094] In some aspects, such as the observed environment vectors f This may include client identifiers (IDs), client antenna configurations, large-scale channel characteristics, CSI-RS configurations, images obtained through imaging equipment, a portion of the estimated propagation channel, and so on. In some aspects, for example, large-scale channel characteristics may indicate channel-associated delay spread, channel-associated power delay profiles, channel-associated Doppler measurements, channel-associated Doppler spectrum, channel-associated signal-to-noise ratio (SNR), channel-associated signal-to-noise plus-interference ratio (SINR), reference signal received power (RSRP), received signal strength indicator (RSSI), and so on.

[0095] In some respects, based on the aspects discussed in this article, Figure 4 The graphical model shown can be used to derive probabilistic expressions related to the use of the corresponding autoencoder. For example, in some respects, the graphical model can indicate the following conditional probabilities:

[0096] ,

[0097] in p It's probability. x It is the observed wireless communication vector. f It is the observed environment vector. h It is a hidden vector. y It is at least partially based on f eigenvectors, and s It is the client. In some respects, the conditional probabilities indicated by the directional graphical model can be used to configure the training of the autoencoder.

[0098] As indicated above, Figure 4 This is provided merely as an example. Other examples may differ from those provided. Figure 4 The example described.

[0099] Figure 5 This is a diagram illustrating an example 500 of an autoencoder pair according to this disclosure. Aspects of example 500 can be provided by a client (e.g., Figure 3 The client 302 shown), server (e.g., Figure 3 This is achieved through methods such as the server 304 shown in the diagram. Figure 5 The autoencoder pair shown can be used to implement and Figure 4 The diagram illustrates the various aspects associated with the directional graphical model. An autoencoder associated with the client can be referred to as a client-side autoencoder. An autoencoder associated with the server can be referred to as a server-side autoencoder.

[0100] As shown, the autoencoder pair includes a first autoencoder 510 and a second autoencoder 520. The first autoencoder 510 and / or the second autoencoder 520 may be a regular autoencoder or a variational autoencoder. The first autoencoder 510 may include a first encoder 530, which is configured to receive an observed environment vector. f As input and providing feature vectors y As output, the first autoencoder 510 may also include a first decoder 540 configured to receive the feature vector. y As input, and provide (e.g., recover) the observed environment vector. f As output.

[0101] As shown, the second autoencoder 520 may include a second encoder 550 configured to receive the observed wireless communication vector. x and eigenvectors y As input, a latent vector is provided. h As output, the second autoencoder 520 may also include a second decoder 560 configured to receive the latent vector. h and eigenvectors y As input, it provides (e.g., recovers) the observed wireless communication vector. x As output.

[0102] Depending on various aspects, the autoencoder pair (such as autoencoder 510 and autoencoder 520) can be trained before being used for inference. In some aspects, for example, the autoencoder pair can be trained based on the use of reparameterization. For example, during training, the first encoder 530 can be used to sample or compute feature vectors using reparameterization techniques. y This can be used to train both the first autoencoder 510 and the second autoencoder 520. In some aspects, training the first autoencoder 510 and / or the second autoencoder 520 may include: determining a set of neural network parameters that maximizes the variational lower bound function (which may be interchangeably referred to as the evidence lower bound (ELBO) function) corresponding to the first autoencoder 510 and the second autoencoder 520. θ and .

[0103] In some respects, in order to establish a proof function Find ELBO above, and you can introduce variational distributions. And ELBO can be written as:

[0104] ,

[0105] in 1This represents the neural network parameters associated with the first encoder 530. 2 This represents the neural network parameters associated with the second encoder 550. θ 1 This represents the neural network parameters associated with the first decoder 540. θ 2 This represents the neural network parameters associated with the second decoder 560. = [ 1, [2] represents the neural network parameters associated with the first encoder 530 and the second encoder 550, and θ = [θ1, θ2] represents the neural network parameters associated with the first decoder 540 and the second decoder 560. ELBO It can be used to train and optimize neural network parameters θ and In some respects, ( 1,θ1) can be referred to as the set of autoencoder parameters associated with the first autoencoder 510, and ( 2, θ2) can be referred to as the set of autoencoder parameters associated with the second autoencoder 520. In some aspects, training the first and second autoencoders 510 and 520 may include: finding the parameter set that makes ELBO, Maximize the neural network parameters θ and In some respects, the negative variational lower bound function may correspond to the sum of a first loss function associated with the first autoencoder 510 and a second loss function associated with the second autoencoder 520. For example, the first loss function may include a first reconstruction loss and a first regularization term for the first autoencoder 510, and the second loss function may include a second reconstruction loss and a second regularization term for the second autoencoder 520.

[0106] For example, ELBO can be decomposed into two parts:

[0107] ,in:

[0108] )),as well as

[0109] ,

[0110] in Parameterized by the first encoder 530, Parameterized by the first decoder 540, Parameterized by the second encoder 550, Parameterized by the second decoder 560, The second autoencoder 520 was trained by parameterizing the previous network. It can be assumed to be a Gaussian distribution (e.g., ) ),

[0111] It is used for the reconstruction loss of the first autoencoder 510.

[0112] It is the reconstruction loss used for the second autoencoder 520.

[0113] It is a regularization term used for the first autoencoder 510, and It is a regularization term used for the second autoencoder 520.

[0114] Here, by The given negative ELBO (also known as variational free energy) can be used as the loss function during training, and the stochastic gradient descent (SGD) algorithm can be used to optimize the neural network parameters θ and Correspondingly, this loss can be decomposed into ,in It is the loss of the first autoencoder 510, and This is the loss of the second encoder 520. Negative ELBO This can be referred to as the "total loss". In some respects, the regularization term KL... and It may be optional.

[0115] If the regularization term is not included in the loss function, the autoencoders 510 and 520 (which may have originally been variational autoencoders) are simplified to regular autoencoders.

[0116] In some respects, the feature vectors from the first encoder 530 y The second autoencoder 520 is adjustable to perform well in the observed environment. This can be considered as expert selection. Feature vector y The format can be learned by training both the first autoencoder 510 and the second autoencoder 520 together. In this way, the feature vectors y The learning process may not require manual input. In some respects, the first autoencoder 510 and / or the second autoencoder 520 can be trained using unsupervised learning procedures. The first autoencoder 510 and / or the second autoencoder 520 can use fully joint learning procedures (e.g., such as...). Figure 7 and 9 As shown in -14), some joint learning procedures (e.g., such as Figure 8 and 15 (as shown in -17), etc., to train.

[0117] As indicated above, Figure 5This is provided merely as an example. Other examples may differ from those provided. Figure 5 The example described.

[0118] Figure 6 This is a diagram illustrating an example 600 of wireless communication using a self-encoder according to the present disclosure. As shown, client 605 and server UE 610 can communicate with each other. In some aspects, client 605 may be, similar to, include, or be included in... Figure 3 In the client 302 shown. In some respects, server 610 may be, similar to, include, or be included in. Figure 3 The server 304 shown is shown.

[0119] As indicated by reference numeral 615 in the accompanying drawings, server 610 can transmit and client 605 can receive wireless communications. In some aspects, the wireless communications may include reference signals (e.g., CSI-RS, etc.), data communications, control communications, etc. In some aspects, the wireless communications may be carried using a physical downlink control channel (PDCCH), a physical downlink shared channel (PDSCH), a physical sidelink control channel (PSCCH), a physical sidelink shared channel (PSSCH), etc.

[0120] As indicated by reference numeral 620, client 605 can determine a feature vector associated with one or more features, which are related to the environment of client 605. For example, in some aspects, client 605 may use a first autoencoder (e.g., Figure 5 The first self-encoder 510 shown Figure 3 The first client autoencoder 308 shown in the figure, etc., determines the feature vector. As indicated by reference numeral 625, the client 605 can determine the latent vector. For example, in some aspects, the client can use a second autoencoder (e.g., Figure 5 The second self-encoder 520 shown Figure 3 The second client autoencoder 310, etc. shown in the figure, is used to determine the latent vector.

[0121] As indicated by reference numeral 630, client 605 can transmit and server 610 can receive the feature vector and the latent vector. In some aspects, the feature vector and the latent vector may be carried in the Physical Uplink Control Channel (PUCCH), Physical Uplink Shared Channel (PUSCH), etc. As indicated by reference numeral 635, server 610 can determine the observed wireless communication vector. In some aspects, server 610 can determine the observed wireless communication vector based at least in part on the feature vector and the latent vector. For example, in some aspects, server 610 may use an autoencoder (e.g., Figure 5The second self-encoder 520 shown Figure 3 The decoder (such as the server autoencoder 322 shown) determines the observed wireless communication vector. As indicated by reference numeral 640, the server can perform wireless communication actions based at least in part on the determined observed wireless communication vector.

[0122] As indicated above, Figure 6 This is provided merely as an example. Other examples may differ from those provided. Figure 6 The example described.

[0123] Figure 7 This is a diagram illustrating an example 700 of fully joint learning of an autoencoder pair according to the present disclosure. As shown, server 705, first client 710, and second client 715 can communicate with each other. In some aspects, server 705 can be, similar to, include, or be included in. Figure 6 The server 610 shown Figure 3 The server 304, etc., shown in the example. In some respects, client 710 and / or client 715 may be, similar to, include, or be included in. Figure 6 The client 605 shown Figure 3 As shown in client 302, etc. In some respects, any number of additional clients can communicate with server 705.

[0124] In some respects, fully joint learning may include jointly training the first autoencoder (e.g., Figure 5 The first autoencoder 510 shown) and the second autoencoder (e.g., Figure 5 The second autoencoder 520 is shown in the diagram. In fully joint learning, both autoencoders are trained by the client. In some aspects, fully joint learning may include alternating between training the first autoencoder and training the second autoencoder. In some aspects, the feature vector... y The mapping may change slowly because of the observed vector. f The selected client features may be relatively static or change slowly. As a result, in some respects, alternating between training the first autoencoder and training the second autoencoder may include: performing a first number of training iterations associated with the first autoencoder according to a first training frequency; and performing a second number of training iterations associated with the second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0125] As indicated by reference numeral 720, server 705 may transmit neural network parameter sets (e.g., parameters θ and ...) to a set of clients (e.g., clients 710 and 715). As mentioned above Figure 5(As described). These neural network parameters may correspond to a first client autoencoder and a second client autoencoder. As indicated by reference numeral 725, the first client 710 may determine the updated neural network parameter set. As indicated by reference numeral 730, the first client 710 may transmit and the server 705 may receive the updated neural network parameter set.

[0126] As indicated by reference numeral 735, the second client 715 can also determine the updated neural network parameter set. As indicated by reference numeral 740, the second client 715 can transmit, and the server 705 can receive, the updated neural network parameter set. As indicated by reference numeral 745, the server 705 can determine the "final" updated neural network parameter set. The server 705 can determine the final updated neural network parameter set by averaging the updated neural network parameters received from clients 710 and 715. The server 705 can use the final updated neural network parameter set to update the server autoencoder, the first client autoencoder, and the second client autoencoder.

[0127] As indicated above, Figure 7 This is provided merely as an example. Other examples may differ from those provided. Figure 7 The example described.

[0128] Figure 8 This is a diagram illustrating an example 800 of partial joint learning of an autoencoder pair according to this disclosure. As shown, server 805, client 810, and client 815 can communicate with each other. In some aspects, server 805 can be, similar to, include, or be included in... Figure 6 The server 610 shown Figure 3 The server 304, etc., shown in the example. In some respects, client 810 and / or client 815 may be, similar to, include, or be included in. Figure 6 The client 605 shown Figure 3 As shown in client 302, etc. In some respects, any number of additional clients can communicate with server 805.

[0129] In some aspects, a partially joint learning procedure may include centralized learning for training the first autoencoder in an autoencoder pair, while the second autoencoder is trained locally. For example, in some aspects, the server 805 may train the first autoencoder (e.g., Figure 5 The first autoencoder 510 shown is provided to clients 810 and 815. Clients 810 and 815 can use joint learning to train a second autoencoder (e.g., ...). Figure 5 The second self-encoder 520 shown in the figure.

[0130] In some respects, server 805 may update the first autoencoder at a lower frequency than the second autoencoder. For example, clients 810 and 815 may provide server 805 with observed environment vectors for infrequent training of the first autoencoder. For example, in some respects, performing a partially joint learning procedure may include: performing a first set of training iterations associated with the first autoencoder according to a first training frequency; and performing a second set of training iterations associated with the second autoencoder according to a second training frequency higher than the first training frequency.

[0131] As indicated by reference numeral 820 in the accompanying drawings, server 805 can transmit and clients 810 and 815 can receive neural network parameters (e.g., parameters θ and φ). As mentioned above Figure 5 As described. As indicated by reference numeral 825, client 810 can determine the updated parameters associated with the second autoencoder (shown as "AE2 parameters"). As indicated by reference numeral 830, client 810 can transmit and server 805 can receive the updated parameters associated with the second autoencoder.

[0132] As shown by reference numeral 835, client 815 can determine the updated parameters (shown as "AE2 parameters") associated with the second autoencoder. As shown by reference numeral 840, client 815 can transmit, and server 805 can receive, the updated parameters associated with the second autoencoder. As shown by reference numeral 845, server 805 can determine the last updated parameters associated with the second autoencoder. Server 805 can use the last updated parameters associated with the second autoencoder to update both the server autoencoder and the second client autoencoder.

[0133] As indicated by reference numeral 850, clients 810 and 815 may transmit, and server 805 may receive, observed environment vectors. In some aspects, clients 810 and 815 may provide observed environment vectors to server 805 for server 805 to train a first autoencoder. As indicated by reference numeral 855, server 805 may train the first autoencoder. Server 805 may train the first autoencoder at least in part based on the environment vectors. As indicated by reference numeral 860, server 805 may transmit, and clients 810 and 815 may receive, the first autoencoder.

[0134] As indicated above, Figure 8 This is provided merely as an example. Other examples may differ from those provided. Figure 8 The example described.

[0135] Figure 9This is a diagram illustrating an example process 900 of fully joint learning of an autoencoder pair according to this disclosure. Process 900 can be, for example, performed by a client (e.g., Figure 7 The client 710 shown Figure 7 The client 715 shown Figure 6 The client 605 shown Figure 3 The aspects of the fully federated learning process performed by the client (302, etc.) shown in the figure.

[0136] Process 900 may include receiving neural network parameters from a server (box 905). In some aspects, these neural network parameters may include a set of neural network parameters associated with a first autoencoder and a second autoencoder. The process may further include obtaining the observed environment vector and the observed wireless communication vector (box 910). For example, the client may obtain the observed environment training vector. f and observed wireless communication training vectors x .

[0137] As shown, process 900 may include determining the loss of the first autoencoder (box 915). For example, in some aspects, the client may train the loss by taking the observed environment training vectors. f Input the first encoder of the first autoencoder to determine the training feature vector y This determines the first loss corresponding to the first autoencoder. The client can further process the trained feature vectors. y A first decoder is input to a first autoencoder to determine a first training output of the first autoencoder. The client may determine a first loss based at least in part on the first training output. In some aspects, the first loss may be associated with the parameter set of the neural network. In some aspects, the client may determine a first regularization term corresponding to the first autoencoder. The client may determine the first loss based at least in part on the first regularization term.

[0138] Process 900 may further include determining the loss of the second autoencoder (box 920). In some aspects, for example, the client can do this by using the trained feature vectors... y and observed wireless communication training vectors x The second encoder, which is input to the second autoencoder, determines the training latent vector. h This determines the second loss corresponding to the second autoencoder. The client can then use the trained feature vectors... y and training hidden vectors hA second decoder is input to the second autoencoder to determine a second training output of the second autoencoder. The client may determine a second loss associated with the second autoencoder based at least in part on the second training output. In some aspects, the second loss may be associated with the set of parameters of the neural network. In some aspects, the client may determine a second regularization term corresponding to the second autoencoder. The client may determine the second loss based at least in part on the second regularization term.

[0139] Process 900 may further include updating the neural network parameters (box 925). In some aspects, the client may update the neural network parameters by determining the total loss by summing the first loss and the second loss. The client may determine multiple gradients of the total loss relative to the neural network parameter set, and update the neural network parameter set at least in part based on these multiple gradients.

[0140] In some aspects, process 900 may further include repeating operations 910-925 a specified number of times (box 930). For example, the client may update the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set. Process 900 may include providing the updated neural network parameters to the server (box 935). For example, the client may send the final updated autoencoder parameter set to the server.

[0141] As indicated above, Figure 9 This is provided merely as an example. Other examples may differ from those provided. Figure 9 The example described.

[0142] Figure 10 This is a diagram illustrating an example process 1000 for joint learning of autoencoder pairs according to this disclosure. Example process 1000 can be, for example, provided by a server (e.g., Figure 7 The server 705 shown Figure 6 The server 610 shown Figure 3 The aspects of the fully federated learning process performed by the server (e.g., 304, etc.) shown.

[0143] Process 1000 may include selecting a set of clients from which to receive updates (box 1005). For example, the server may select a set of clients from which to receive updates associated with a first client autoencoder and a second client autoencoder. Process 1000 may include transmitting a neural network parameter set to the set of clients (box 1010). For example, the server may transmit a neural network parameter set to the set of clients (…). ,θ).

[0144] Process 1000 may include receiving multiple updated neural network parameter sets from the client set (box 1015). For example, the server may receive updated neural network parameter sets associated with a first client autoencoder and a second client autoencoder (box 1015). Process 1000 may further include determining the final updated neural network parameter set (box 1020). For example, the server may determine the final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder based at least in part on multiple updated neural network parameters. As shown, process 1000 may include returning to box 1005 (box 1025) and repeating process 1000 using one or more additional client sets.

[0145] As indicated above, Figure 10 This is provided merely as an example. Other examples may differ from those provided. Figure 10 The example described.

[0146] Figure 11 This is a diagram illustrating example 1100 of joint learning of autoencoder pairs according to this disclosure. Example 1100 is, for example, capable of being learned by a client (e.g., Figure 7 The client 710 shown Figure 7 The client 715 shown Figure 6 The client 605 shown Figure 3 The aspects of the fully federated learning procedure executed by the client (302, etc.) shown in the figure.

[0147] Process 1100 may include receiving neural network parameters from a server (box 1105). In some aspects, these neural network parameters may include a set of neural network parameters associated with a first autoencoder and a second autoencoder. The process may further include obtaining the observed environment training vector and the observed wireless communication training vector (box 1110). For example, the client may obtain the observed environment training vector. f and observed wireless communication training vectors x .

[0148] As shown, process 1100 may include determining the loss of the first client autoencoder (box 1115). For example, in some aspects, the client may train the loss by taking the observed environment training vectors. f Input the first encoder of the first client autoencoder to determine the training feature vector y This is used to determine the first loss corresponding to the first client autoencoder. The client can further train the feature vectors... yA first decoder is input to the first client autoencoder to determine a first training output of the first client autoencoder. The client may determine a first loss based at least in part on the first training output. In some aspects, the client may determine a first regularization term corresponding to the first client autoencoder. The client may determine the first loss based at least in part on the first regularization term.

[0149] Process 1100 may further include determining the loss of the second client autoencoder (box 1120). In some aspects, for example, the client can determine the loss by training feature vectors. y and observed wireless communication training vectors x Input the second encoder of the second client autoencoder to determine the training latent vectors. h To determine the second loss corresponding to the second client autoencoder. This client can train the feature vector. y and training hidden vectors h A second decoder is input to the second client autoencoder to determine a second training output of the second client autoencoder. The client may determine a second loss associated with the second client autoencoder based at least in part on the second training output. In some aspects, the second loss may be associated with the neural network parameter set. In some aspects, the client may determine a second regularization term corresponding to the second client autoencoder. The client may determine the second loss based at least in part on the second regularization term.

[0150] Process 1100 may further include updating the autoencoder parameter set associated with the first client autoencoder (box 1125). In some aspects, the client may update the neural network parameters by determining the total loss by summing the first loss and the second loss. The client may determine the total loss relative to the autoencoder parameter set (box 1125). Multiple gradients associated with θ1).

[0151] In some aspects, process 1100 may further include repeating operations 1110-1125 a specified number of times (box 1130). For example, the client may update the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set. Process 1100 may include providing the updated autoencoder parameters to the server (box 1135).

[0152] As indicated above, Figure 11 This is provided merely as an example. Other examples may differ from those provided. Figure 11 The example described.

[0153] Figure 12 This is a diagram illustrating an example process 1200 for joint learning of autoencoder pairs according to this disclosure. Example process 1200 is, for example, capable of being performed by a client (e.g., Figure 7 The client 710 shown Figure 7 The client 715 shown Figure 6 The client 605 shown Figure 3 The aspects of the fully federated learning process performed by the client (302, etc.) shown in the figure.

[0154] Process 1200 may include receiving neural network parameters from a server (box 1205). In some aspects, these neural network parameters may include a set of neural network parameters associated with a first client-side autoencoder and a second client-side autoencoder. The process 1200 may further include obtaining the observed environment training vectors and determining the feature training vectors (box 1210). For example, the client may obtain the observed environment training vectors. f And it can be trained, at least in part, based on observed environmental vectors. f Determine feature training vectors y .

[0155] Process 1200 may include obtaining the observed wireless communication training vectors (box 1215) and determining the loss of the second client autoencoder (box 1220). For example, the client may obtain the observed wireless communication training vectors. x and the feature training vector y and observed wireless communication training vectors x Input the second encoder of the second client autoencoder to determine the training latent vectors. h This client can train feature vectors. y and training hidden vectors h The second decoder of the second autoencoder is input to determine the second training output of the second autoencoder. The client can determine the loss associated with the second autoencoder, at least in part, based on the second training output. In some respects, this loss can be related to the autoencoder parameter set (…). The autoencoder parameter set is associated with a second autoencoder (θ2, θ2). Process 1200 may include updating the autoencoder parameter set associated with the second client autoencoder (box 1225). In some aspects, the client may determine multiple gradients of the loss relative to the autoencoder parameter set associated with the second client autoencoder. The client may update the autoencoder parameter set at least in part based on these multiple gradients.

[0156] In some aspects, process 1200 may further include repeating operations 1210-1225 a specified number of times (box 1230). For example, the client may update the autoencoder parameter set associated with the second client autoencoder a specified number of times to determine the last updated autoencoder parameter set associated with the second client autoencoder. Process 1200 may include providing the updated autoencoder parameters to the server (box 1235).

[0157] As indicated above, Figure 12 This is provided merely as an example. Other examples may differ from those provided. Figure 12 The example described.

[0158] Figure 13 This is a diagram illustrating an example process 1300 for joint learning of autoencoder pairs according to this disclosure. Example process 1300 can be, for example, provided by a server (e.g., Figure 7 The server 705 shown Figure 6 The server 610 shown Figure 3 The aspects of the fully federated learning process performed by the server (e.g., 304, etc.) shown.

[0159] Process 1300 may include selecting a set of clients from which to receive updates (box 1305). Process 1300 may include transmitting a neural network parameter set to the client set (box 1310). For example, the server may transmit a neural network parameter set to the client set ( , θ).

[0160] Process 1300 may include receiving multiple updated autoencoder parameter sets associated with the first client autoencoder. 1, θ1) (Box 1315). Process 1300 may further include determining the last updated set of autoencoder parameters associated with the first client autoencoder (Box 1320) and returning to Box 1305 (Box 1325) to repeat process 1300.

[0161] As indicated above, Figure 13 This is provided merely as an example. Other examples may differ from those provided. Figure 13 The example described.

[0162] Figure 14 This is a diagram illustrating example 1400 of joint learning of autoencoder pairs according to this disclosure. Example 1400 is, for example, capable of being generated by a server (e.g., Figure 7 The server 705 shown Figure 6 The server 610 shown Figure 3 The aspects of the fully federated learning procedure executed by the server (e.g., 304, etc.) shown.

[0163] Process 1400 may include selecting a set of clients from which to receive updates (box 1405). Process 1400 may include transmitting a neural network parameter set to the client set (box 1410). For example, the server may transmit a neural network parameter set to the client set ( , θ).

[0164] Process 1400 may include receiving multiple updated autoencoder parameter sets associated with the second client autoencoder. 2, θ2) (Box 1415). Process 1400 may further include determining the last updated set of autoencoder parameters associated with the second client autoencoder (Box 1420) and returning to Box 1405 (Box 1425) to repeat process 1400.

[0165] As indicated above, Figure 14 This is provided merely as an example. Other examples may differ from those provided. Figure 14 The example described.

[0166] Figure 15 This is a diagram illustrating an example process 1500 for joint learning of autoencoder pairs according to this disclosure. Example process 1500 is, for example, capable of being performed by a client (e.g., Figure 8 The client 810 shown Figure 8 The client 815 shown Figure 6 The client 605 shown Figure 3 The client 302, etc. shown in the figure, is an aspect of the joint learning procedure executed by the client.

[0167] Process 1500 may include receiving neural network parameters from a server (box 1505). In some aspects, these neural network parameters may include a set of neural network parameters associated with a first client-side autoencoder and a second client-side autoencoder. Process 1500 may further include obtaining the observed environment training vectors. f And determine the feature training vector y (box 1510). For example, the client can obtain the observed environment training vector. f and at least in part based on observed environment training vectors f Determine feature training vectors y .

[0168] As shown, process 1500 may include obtaining observed wireless communication training vectors (box 1515) and determining the loss of the second client autoencoder (box 1520). In some aspects, for example, the client may train the feature vectors... y and observed wireless communication training vectors xInput the second encoder of the second client autoencoder to determine the hidden training vectors. h To determine the second loss corresponding to the second client autoencoder. This client can train feature vectors. y and latent vectors h The second decoder of the second client autoencoder is input to determine the second training output of the second client autoencoder. The client can determine the loss associated with the second client autoencoder based at least in part on the second training output. In some respects, this loss can be related to the autoencoder parameter set (…). The autoencoder parameter set is associated with a second client autoencoder (θ2, θ2). In some aspects, the client can determine a second regularization term corresponding to the second client autoencoder. The client can determine a second loss based at least in part on the second regularization term.

[0169] Process 1500 may include updating the autoencoder parameter set associated with the second client autoencoder (box 1525). In some aspects, the client may determine multiple gradients of the loss relative to the autoencoder parameter set associated with the second client autoencoder. The client may update the autoencoder parameter set based at least in part on these multiple gradients.

[0170] In some aspects, process 1500 may further include repeating operations 1510-1525 a specified number of times (box 1530). For example, the client may update the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set. Process 1500 may include providing the server with the updated autoencoder parameter set associated with the second client autoencoder (box 1535).

[0171] As indicated above, Figure 15 This is provided merely as an example. Other examples may differ from those provided. Figure 15 The example described.

[0172] Figure 16 This is a diagram illustrating an example process 1600 for joint learning of autoencoder pairs according to this disclosure. Example process 1600 is, for example, capable of being performed by a server (e.g., Figure 8 The server 805 shown Figure 6 The server 610 shown Figure 3 The server 304, etc. shown in the figure represents aspects of the joint learning process performed by the server.

[0173] Process 1600 may include selecting the observed environment vectors to be received from it. f The client set (box 1605). Process 1600 may include receiving observed environment training vectors from this client set. f(Box 1610). As shown, process 1600 may include determining the loss of the first client autoencoder (Box 1615). For example, in some aspects, the server may do so by training vectors from the observed environment. f Input the first encoder of the first client autoencoder to determine the feature training vectors y This determines the first loss corresponding to the first client autoencoder. The server can further process the feature training vectors... y A first decoder is input to the first client-side autoencoder to determine a first training output of the first client-side autoencoder. The server may determine a first loss based at least in part on the first training output. In some aspects, the server may determine a first regularization term corresponding to the first client-side autoencoder. The server may determine the first loss based at least in part on the first regularization term.

[0174] Process 1600 may further include determining the last updated set of autoencoder parameters associated with the first client autoencoder (box 1620). In some aspects, the server can determine this by determining the loss of the first autoencoder relative to the set of autoencoder parameters associated with the first client autoencoder (box 1620). Multiple gradients (θ1, θ1) are used to update the neural network parameters. In some aspects, process 1600 may further include repeating operations 1610-1620 a specified number of times (box 1625) and returning to box 1605 to repeat process 1600 (box 1630).

[0175] As indicated above, Figure 16 This is provided merely as an example. Other examples may differ from those provided. Figure 16 The example described.

[0176] Figure 17 This is a diagram illustrating example 1700 of the joint learning of autoencoder pairs according to this disclosure. Example process 1700 is, for example, capable of being performed by a server (e.g., Figure 8 The server 805 shown Figure 6 The server 610 shown Figure 3 The server 304, etc. shown in the figure represents a portion of the joint learning procedure executed.

[0177] Process 1700 may include selecting a set of clients from which to receive updates (box 1705). For example, the server may select a set of clients from which to receive updates associated with a second client autoencoder. Process 1700 may include transmitting a set of neural network parameters to the set of clients (box 1710). For example, the server may transmit a set of neural network parameters to the set of clients. , θ).

[0178] Process 1700 may include receiving from the client set a plurality of updated autoencoder parameter sets associated with the second client autoencoder (box 1715). For example, the server may receive an updated autoencoder parameter set associated with the second client autoencoder ( 2, θ2). Process 1700 may further include determining the last updated set of autoencoder parameters associated with the second client autoencoder (box 1720). For example, the server may determine the last updated set of autoencoder parameters associated with the second client autoencoder by averaging the corresponding updated parameters received from the client. As shown, process 1700 may include returning to box 1705 (box 1725) and repeating process 1700 using one or more additional client sets.

[0179] As indicated above, Figure 17 This is provided merely as an example. Other examples may differ from those provided. Figure 17 The example described.

[0180] Figure 18 This is a diagram illustrating an example procedure 1800 performed by a client, for example, according to this disclosure. Example procedure 1800 is where the client (e.g., Figure 3 The example shown is client 302 performing operations associated with joint learning of autoencoder pairs for wireless communication.

[0181] like Figure 18 As shown, in some aspects, process 1800 may include using a first client autoencoder to determine a feature vector associated with one or more features, which are related to the client's environment (box 1810). For example, the client (e.g., using...) Figure 20 The Communication Manager (2004) can use a first autoencoder to determine a feature vector associated with one or more features, which are related to the client’s environment, as described above.

[0182] like Figure 18 As further shown, in some aspects, process 1800 may include determining a latent vector using a second client autoencoder and at least in part based on the feature vector (box 1820). For example, the client (e.g., using...) Figure 20 The Communication Manager 2004 can use a second client autoencoder and at least in part based on the feature vector to determine the latent vector, as described above.

[0183] like Figure 18 As further shown, in some aspects, process 1800 may include transmitting the feature vector and the latent vector (box 1830). For example, the client (e.g., using...) Figure 20The transmission component 2006 can transmit the feature vector and the latent vector, as described above.

[0184] Process 1800 may include additional aspects, such as any single aspect or any combination of aspects described below and / or in conjunction with one or more other processes described elsewhere herein.

[0185] In a first aspect, the first client-side autoencoder includes: a first encoder configured to receive an observed environment vector as input and provide the feature vector as output, and a first decoder configured to receive the feature vector as input and provide the observed environment vector as output.

[0186] In a second aspect, either alone or in combination with the first aspect, the second autoencoder includes: a second encoder configured to receive an observed wireless communication vector and the feature vector as input and provide the latent vector as output; and a second decoder configured to receive the latent vector and the feature vector as input and provide the observed wireless communication vector as output.

[0187] In the third aspect, determining the feature vector, either alone or in combination with one or more of the first and second aspects, includes providing the observed environment vector as input to the first autoencoder.

[0188] In the fourth aspect, either alone or in combination with one or more of the first to third aspects, the observed environment vector includes one or more feature components, wherein the one or more feature components indicate: client vendor identifier, client antenna configuration, large-scale channel characteristics, channel state information reference signal configuration, image obtained by an imaging device, a portion of the estimated propagation channel, or a combination thereof.

[0189] In the fifth aspect, alone or in combination with one or more of the first to fourth aspects, the large-scale channel characteristics indicate: channel-associated delay spread, channel-associated power delay profile, channel-associated Doppler measurement, channel-associated Doppler spectrum, channel-associated signal-to-noise ratio, channel-associated signal-to-noise plus-interference ratio, reference signal received power, received signal strength indicator, or a combination thereof.

[0190] In the sixth aspect, the latent vector is associated with the wireless communication task, either alone or in combination with one or more of the first to fifth aspects.

[0191] In the seventh aspect, alone or in combination with one or more of the first to sixth aspects, the wireless communication task includes: determining channel state feedback (CSF), determining location information associated with the client, determining modulation associated with wireless communication, determining waveform associated with wireless communication, or a combination thereof.

[0192] In the eighth aspect, alone or in combination with one or more of the first to seventh aspects, the wireless communication task includes: determining the CSF, and process 1800 includes: receiving a channel state information (CSI) reference signal (CSI-RS), determining the CSI based at least in part on the CSI-RS; and providing the CSI as input to a second autoencoder.

[0193] In the ninth aspect, either alone or in combination with one or more of the first to eighth aspects, the implicit vector includes compressed channel state feedback.

[0194] In the tenth aspect, transmitting the feature vector and the latent vector, either alone or in combination with one or more of the first to ninth aspects, includes transmitting the feature vector and the latent vector using a physical uplink control channel, a physical uplink shared channel, or a combination thereof.

[0195] In the eleventh aspect, alone or in combination with one or more of the first to tenth aspects, at least one of the first autoencoder or the second autoencoder includes a variational autoencoder.

[0196] In the twelfth aspect, alone or in combination with one or more of the first to eleventh aspects, process 1800 includes: training at least one of the first autoencoder or the second autoencoder.

[0197] In the thirteenth aspect, training at least one of the first autoencoder or the second autoencoder, either alone or in combination with one or more of the first to twelfth aspects, includes using reparameterization.

[0198] In the fourteenth aspect, training at least one of the first autoencoder or the second autoencoder, either alone or in combination with one or more of the first to thirteenth aspects, comprises: determining a set of neural network parameters that maximizes the variational lower bound function corresponding to the first autoencoder and the second autoencoder.

[0199] In the fifteenth aspect, either alone or in combination with one or more of the first to fourteenth aspects, the negative variational lower bound function corresponds to the sum of the first loss function associated with the first autoencoder and the second loss function associated with the second autoencoder.

[0200] In the sixteenth aspect, either alone or in combination with one or more of the first to fifteenth aspects, the first loss function includes a first reconstruction loss and a first regularization term for the first autoencoder, and the second loss function includes a second reconstruction loss and a second regularization term for the second autoencoder.

[0201] In the seventeenth aspect, either alone or in combination with one or more of the first to sixteenth aspects, the first autoencoder and the second autoencoder are regular autoencoders, and the variational lower bound function does not include a regularization term.

[0202] In the eighteenth aspect, training at least one of the first autoencoder or the second autoencoder, alone or in combination with one or more of the first to seventeenth aspects, includes: training at least one of the first autoencoder or the second autoencoder using a joint learning procedure.

[0203] In the nineteenth aspect, training at least one of the first autoencoder or the second autoencoder, alone or in combination with one or more of the first to eighteenth aspects, comprises: training the first autoencoder and the second autoencoder to determine the format of the feature vector.

[0204] In the twentieth aspect, training at least one of the first autoencoder or the second autoencoder, alone or in combination with one or more of the first to nineteenth aspects, includes using an unsupervised learning procedure.

[0205] In aspect 21, training at least one of the first autoencoder or the second autoencoder, alone or in combination with one or more of aspects 1 to 20, includes performing a fully joint learning procedure.

[0206] In aspect 22, the fully joint learning procedure is performed, either alone or in combination with one or more of aspects 1 to 21, including jointly training the first autoencoder and the second autoencoder.

[0207] In aspect 23, the fully joint learning procedure is performed, either alone or in combination with one or more of aspects 1 to 22, including alternating between training the first autoencoder and training the second autoencoder.

[0208] In the twentieth aspect, alternating between training the first autoencoder and training the second autoencoder, either alone or in combination with one or more of the first to twenty-third aspects, includes: performing a first plurality of training iterations associated with the first autoencoder according to a first training frequency, and performing a second plurality of training iterations associated with the second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0209] In aspect 25, training at least one of the first autoencoder or the second autoencoder, either alone or in combination with one or more of aspects 1 to 24, comprises: performing a partially joint learning procedure.

[0210] In the twenty-sixth aspect, performing a partial joint learning procedure, either alone or in combination with one or more of the first to twenty-fifth aspects, includes: providing an observed environment vector to a server, and receiving a first autoencoder from the server, wherein the first autoencoder is at least partially based on the observed environment vector.

[0211] In the twenty-seventh aspect, either alone or in combination with one or more of the first to twenty-sixth aspects, the first autoencoder is based at least in part on at least one additional environment vector associated with at least one additional client.

[0212] In aspect twenty-eight, performing a partial joint learning procedure, either alone or in combination with one or more of aspects one through twenty-seven, includes: updating the second autoencoder to determine the updated neural network parameter set, and transmitting the updated neural network parameter set to the server.

[0213] In the twenty-ninth aspect, either alone or in combination with one or more of the first to twenty-eighth aspects, performing a partial joint learning procedure includes: performing a first plurality of training iterations associated with a first autoencoder according to a first training frequency, wherein performing the training iterations in the first plurality of training iterations includes: providing an observed environment vector to a server, and receiving an updated first autoencoder from the server, wherein the updated first autoencoder is at least partially based on the observed environment vector; and performing a second plurality of training iterations associated with a second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0214] In the thirtieth aspect, either alone or in combination with one or more of the first to twenty-ninth aspects, performing a partially joint learning procedure includes: receiving a set of neural network parameters associated with a first autoencoder and a second autoencoder from a server; obtaining an observed environment training vector; inputting the observed environment training vector into a first encoder of the first autoencoder to determine a training feature vector; obtaining an observed wireless communication training vector; inputting the training feature vector and the observed wireless communication training vector into a second encoder of the second autoencoder to determine a training latent vector; inputting the training feature vector and the training latent vector into a second decoder of the second autoencoder to determine a second training output of the second autoencoder; and determining a loss associated with the second autoencoder based at least in part on the second training output, wherein the loss is associated with the set of neural network parameters.

[0215] In the thirty-first aspect, alone or in combination with one or more of the first to thirtieth aspects, process 1800 includes: determining a plurality of gradients of the loss relative to a set of autoencoder parameters, wherein the set of autoencoder parameters corresponds to a second autoencoder, and updating the set of autoencoder parameters at least in part based on the plurality of gradients.

[0216] In the thirty-second aspect, alone or in combination with one or more of the first to thirty-first aspects, process 1800 includes: updating the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set.

[0217] In aspect thirty-three, alone or in combination with one or more of aspects one through thirty-two, process 1800 includes: transmitting the final updated set of autoencoder parameters to the server.

[0218] In the thirty-fourth aspect, alone or in combination with one or more of the first to thirty-third aspects, process 1800 includes: determining a first loss corresponding to a first autoencoder, wherein determining the first loss includes: receiving a set of neural network parameters associated with a first autoencoder and a second autoencoder from a server; obtaining an observed environment training vector; inputting the observed environment training vector into a first encoder of the first autoencoder to determine a training feature vector; inputting the training feature vector into a first decoder of the first autoencoder to determine a first training output of the first autoencoder; and determining the first loss based at least in part on the first training output, wherein the first loss is associated with the set of neural network parameters.

[0219] In the thirty-fifth aspect, alone or in combination with one or more of the first to thirty-fourth aspects, process 1800 includes: determining a second loss corresponding to the second autoencoder, wherein determining the second loss includes: obtaining an observed wireless communication training vector; inputting the training feature vector and the observed wireless communication training vector into a second encoder of the second autoencoder to determine a training latent vector; inputting the training feature vector and the training latent vector into a second decoder of the second autoencoder to determine a second training output of the second autoencoder; and determining a second loss associated with the second autoencoder based at least in part on the second training output, wherein the second loss is associated with the set of parameters of the neural network.

[0220] In the thirty-sixth aspect, alone or in combination with one or more of the first to thirty-fifth aspects, process 1800 includes: determining a first regularization term corresponding to a first autoencoder; and determining a second regularization term corresponding to a second autoencoder, wherein determining the first loss includes: determining the first loss based at least in part on the first regularization term, and wherein determining the second loss includes: determining the second loss based at least in part on the second regularization term.

[0221] In the thirty-seventh aspect, alone or in combination with one or more of the first to thirty-sixth aspects, process 1800 includes determining the total loss by summing the first loss and the second loss.

[0222] In the thirty-eighth aspect, alone or in combination with one or more of the first to thirty-seventh aspects, process 1800 includes: determining a plurality of gradients of the total loss relative to the set of parameters of the neural network, and updating the set of parameters of the neural network at least in part based on the plurality of gradients.

[0223] In the thirty-ninth aspect, alone or in combination with one or more of the first to thirty-eighth aspects, process 1800 includes: updating the neural network parameter set a specified number of times to determine the final updated neural network parameter set.

[0224] In the fortieth aspect, alone or in combination with one or more of the first to thirty-ninth aspects, process 1800 includes: transmitting the final updated set of neural network parameters to the server.

[0225] In the forty-first aspect, alone or in combination with one or more of the first to fortieth aspects, process 1800 includes: determining a first plurality of gradients of the total loss relative to a first set of autoencoder parameters associated with the first autoencoder, and updating the first set of autoencoder parameters at least in part based on the first plurality of gradients.

[0226] In aspect 42, alone or in combination with one or more of aspects 1 to 41, process 1800 includes: updating the first autoencoder parameter set a specified number of times to determine the final updated first autoencoder parameter set.

[0227] In aspect 43, alone or in combination with one or more of aspects 1 to 42, process 1800 includes: transmitting the last updated set of first autoencoder parameters to the server.

[0228] In the forty-fourth aspect, alone or in combination with one or more of the first to forty-third aspects, process 1800 includes: determining a second plurality of gradients of the second loss relative to a second set of autoencoder parameters associated with the second autoencoder, and updating the second set of autoencoder parameters at least in part based on the second plurality of gradients.

[0229] In aspect 45, alone or in combination with one or more of aspects 1 to 44, process 1800 includes: updating the second autoencoder parameter set a specified number of times to determine the final updated second autoencoder parameter set.

[0230] In aspect 46, either alone or in combination with one or more of aspects 1 to 45, process 1800 includes: transmitting the final updated set of second autoencoder parameters to the server.

[0231] although Figure 18 An example box of process 1800 is shown, but in some respects, process 1800 may include... Figure 18 The boxes depicted in the process are compared to additional boxes, fewer boxes, different boxes, or boxes arranged differently. Additionally or alternatively, two or more boxes in process 1800 can be executed in parallel.

[0232] Figure 19 This is a diagram illustrating an example process 1900 performed by a server, for example, according to this disclosure. Example process 1900 is where the server (e.g., Figure 3 The example shown is server 304 performing operations associated with joint learning of autoencoder pairs for wireless communication.

[0233] like Figure 19 As shown, in some aspects, process 1900 may include receiving a feature vector from a client associated with one or more features, which are related to the client's environment (box 1910). For example, the server (e.g., using...) Figure 22 The receiving component 2202 can receive feature vectors associated with one or more features from the client, which are associated with the client's environment, as described above.

[0234] like Figure 19 As further shown, in some aspects, process 1900 may include receiving a latent vector from the client that is at least partially based on the feature vector (box 1920). For example, the server (e.g., using...) Figure 22 The receiving component 2202 can receive, from the client, a latent vector at least in part based on the feature vector, as described above.

[0235] like Figure 19 As further shown, in some aspects, process 1900 may include determining the observed wireless communication vector based at least in part on the feature vector and the latent vector (box 1930). For example, the server (e.g., using...) Figure 22 The communication manager 2204 can determine the observed wireless communication vector based at least in part on the feature vector and the latent vector, as described above.

[0236] like Figure 19 As further shown, in some aspects, process 1900 may include performing a wireless communication action at least in part based on determining an observed wireless communication vector (box 1940). For example, the server (e.g., using...) Figure 22 The communication manager 2204 can perform wireless communication actions, at least in part, based on the determined observed wireless communication vectors, as described above.

[0237] Process 1900 may include additional aspects, such as any single aspect or any combination of aspects described below and / or in conjunction with one or more other processes described elsewhere herein.

[0238] In the first aspect, the feature vector is at least partially based on the observed environment vector.

[0239] In the second aspect, either alone or in combination with the first aspect, the observed environment vector includes one or more feature components, wherein the one or more feature components indicate: client vendor identifier, client antenna configuration, large-scale channel characteristics, channel state information reference signal configuration, image obtained by an imaging device, a portion of the estimated propagation channel, or a combination thereof.

[0240] In a third aspect, alone or in combination with one or more of the first and second aspects, the large-scale channel characteristics indicate: channel-associated delay spread, channel-associated power delay profile, channel-associated Doppler measurement, channel-associated Doppler spectrum, channel-associated signal-to-noise ratio, channel-associated signal-to-noise plus-interference ratio, reference signal received power, received signal strength indicator, or a combination thereof.

[0241] In the fourth aspect, either alone or in combination with one or more of the first to third aspects, the observed wireless communication vectors are associated with the wireless communication task.

[0242] In the fifth aspect, alone or in combination with one or more of the first to fourth aspects, the wireless communication task includes: receiving channel state feedback (CSF), receiving location information associated with the UE, receiving modulation associated with wireless communication, receiving waveforms associated with wireless communication, or combinations thereof.

[0243] In the sixth aspect, alone or in combination with one or more of the first to fifth aspects, the wireless communication task includes receiving a CSF, and the process 1900 includes transmitting a channel state information reference signal (CSI-RS), wherein the CSF corresponds to the CSI-RS.

[0244] In the seventh aspect, either alone or in combination with one or more of the first to sixth aspects, the implicit vector includes compressed channel state feedback.

[0245] In the eighth aspect, the feature vector and the latent vector are carried, either alone or in combination with one or more of the first to seventh aspects, in the physical uplink control channel, the physical uplink shared channel, or a combination thereof.

[0246] In the ninth aspect, determining the observed wireless communication vector, either alone or in combination with one or more of the first to eighth aspects, includes inputting the feature vector and the latent vector into the decoder of the autoencoder.

[0247] In the tenth aspect, alone or in combination with one or more of the first to ninth aspects, the autoencoder includes a regular autoencoder or a variational autoencoder.

[0248] In the eleventh aspect, alone or in combination with one or more of the first to tenth aspects, the autoencoder corresponds to a second client autoencoder in a pair of client autoencoders, the pair of client autoencoders including the first client autoencoder and the second client autoencoder.

[0249] In the twelfth aspect, alone or in combination with one or more of the first to eleventh aspects, the first client autoencoder includes: a first encoder configured to receive an observed environment vector as input and provide the feature vector as output, and a first decoder configured to receive the feature vector as input and provide the observed environment vector as output.

[0250] In the thirteenth aspect, alone or in combination with one or more of the first to twelfth aspects, the second client autoencoder includes: a second encoder configured to receive an observed wireless communication vector and the feature vector as input and provide the latent vector as output; and a second decoder configured to receive the latent vector and the feature vector as input and provide the observed wireless communication vector as output.

[0251] In the fourteenth aspect, alone or in combination with one or more of the first to thirteenth aspects, process 1900 includes: performing a partial joint learning procedure, which includes training a first client autoencoder and transmitting the first client autoencoder to the client.

[0252] In the fifteenth aspect, either alone or in combination with one or more of the first to fourteenth aspects, performing a partial joint learning procedure includes: performing a first plurality of training iterations associated with a first client autoencoder according to a first training frequency, wherein performing the training iterations in the first plurality of training iterations includes: receiving observed environment training vectors from the client, and transmitting an updated first client autoencoder to the client, wherein the updated first client autoencoder is at least partially based on the observed environment training vectors, and performing a second plurality of training iterations associated with a second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0253] In the sixteenth aspect, alone or in combination with one or more of the first to fifteenth aspects, process 1900 includes: transmitting a first client self-encoder to at least one additional client.

[0254] In the seventeenth aspect, training the first client autoencoder, either alone or in combination with one or more of the first to sixteenth aspects, includes using an unsupervised learning procedure.

[0255] In the eighteenth aspect, alone or in combination with one or more of the first to seventeenth aspects, process 1900 includes: receiving observed environment training vectors from the client, wherein training the first client autoencoder includes: training the first client autoencoder at least in part based on the observed environment training vectors.

[0256] In the nineteenth aspect, alone or in combination with one or more of the first to eighteenth aspects, process 1900 includes: receiving at least one additional observed environment training vector from at least one additional client, wherein training the first client autoencoder includes: training the first client autoencoder at least in part based on the at least one additional observed environment training vector.

[0257] In the twentieth aspect, alone or in combination with one or more of the first to nineteenth aspects, process 1900 includes: selecting a set of clients from which to receive a set of observed environment training vectors, wherein the set of clients includes the client and the at least one additional client.

[0258] In the twenty-first aspect, training the first client autoencoder, either alone or in combination with one or more of the first to twentieth aspects, comprises: determining a loss corresponding to the first client autoencoder, wherein determining the loss comprises: obtaining a set of neural network parameters associated with the first client autoencoder and the second client autoencoder; receiving an observed environment training vector from the client; inputting the observed environment training vector into the encoder of the first client autoencoder to determine a training feature vector; inputting the training feature vector into the decoder of the first client autoencoder to determine a training output of the first client autoencoder; and determining the loss at least in part based on the training output, wherein the loss is associated with the set of neural network parameters.

[0259] In the twenty-second aspect, alone or in combination with one or more of the first to twenty-first aspects, process 1900 includes: determining a regularization term corresponding to the first client autoencoder, wherein determining the loss includes: determining the loss based at least in part on the regularization term.

[0260] In the twenty-third aspect, alone or in combination with one or more of the first to twenty-second aspects, process 1900 includes: determining a second plurality of gradients of the loss relative to the autoencoder parameter set; and updating the autoencoder parameter set at least in part based on the plurality of gradients, wherein the autoencoder parameter set corresponds to a first client autoencoder.

[0261] In the twenty-fourth aspect, alone or in combination with one or more of the first to twenty-third aspects, process 1900 includes: updating the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set.

[0262] In the twentieth aspect, either alone or in combination with one or more of the first to twenty-fourth aspects, performing a partial joint learning procedure includes: selecting a set of clients from which to receive updates associated with the second client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting neural network parameter sets associated with the first client autoencoder and the second client autoencoder to the set of clients; receiving multiple updated autoencoder parameter sets associated with the second client autoencoder from the set of clients; and determining a final updated autoencoder parameter set associated with the second client autoencoder based at least in part on the multiple updated autoencoder parameter sets.

[0263] In the twenty-sixth aspect, alone or in combination with one or more of the first to twenty-fifth aspects, process 1900 includes: transmitting to the client set the last updated set of autoencoder parameters associated with the second client autoencoder.

[0264] In the twenty-seventh aspect, determining the final updated autoencoder parameter set associated with the second client autoencoder, either alone or in combination with one or more of the first to twenty-sixth aspects, includes averaging multiple updated autoencoder parameter sets.

[0265] In the twentieth aspect, alone or in combination with one or more of the first to twenty-seventh aspects, process 1900 includes: selecting a set of clients from which to receive updates associated with the first client autoencoder and the second client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting a set of neural network parameters associated with the first client autoencoder and the second client autoencoder to the set of clients; receiving from the set of clients a plurality of updated neural network parameter sets associated with the first client autoencoder and the second client autoencoder; and determining, at least in part, a final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder based on the plurality of updated neural network parameters.

[0266] In the twenty-ninth aspect, alone or in combination with one or more of the first to twenty-eighth aspects, process 1900 includes: transmitting to the client set the final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder.

[0267] In the thirtieth aspect, determining the final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder, either alone or in combination with one or more of the first to twenty-ninth aspects, includes averaging multiple updated neural network parameter sets.

[0268] In the thirty-first aspect, alone or in combination with one or more of the first to thirtieth aspects, process 1900 includes: selecting a set of clients from which to receive updates associated with the first client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting autoencoder parameter sets associated with the first client autoencoder and the second client autoencoder to the set of clients; receiving a plurality of updated autoencoder parameter sets associated with the first client autoencoder from the set of clients; and determining, at least in part, a final updated autoencoder parameter set associated with the first client autoencoder based on the plurality of updated autoencoder parameter sets.

[0269] In the thirty-second aspect, alone or in combination with one or more of the first to thirty-first aspects, process 1900 includes: transmitting to the client set the last updated set of autoencoder parameters associated with the first client autoencoder.

[0270] In aspect thirty-three, determining the final updated autoencoder parameter set associated with the first client autoencoder, either alone or in combination with one or more of aspects one through thirty-two, includes averaging multiple updated autoencoder parameter sets.

[0271] although Figure 19 An example box of process 1900 is shown, but in some respects, process 1900 may include... Figure 19 The boxes depicted in the process are compared to additional boxes, fewer boxes, different boxes, or boxes arranged differently. Additionally or alternatively, two or more boxes in process 1900 can be executed in parallel.

[0272] Figure 20 This is a block diagram of an example device 2000 for wireless communication according to the present disclosure. Device 2000 may be a client, or a client may include device 2000. In some aspects, device 2000 includes a receiving component 2002, a communication manager 2004, and a transmitting component 2006, which can communicate with each other (e.g., via one or more buses). As shown, device 2000 can use the receiving component 2002 and the transmitting component 2006 to communicate with another device 2008 (such as a server, client, UE, base station, or another wireless communication device).

[0273] In some respects, Equipment 2000 can be configured to perform the actions described in this article. Figure 3-17 The described one or more operations. Additionally or alternatively, Equipment 2000 may be configured to perform one or more processes described herein (such as...). Figure 18 The process 1800). In some aspects, equipment 2000 may include the above combination. Figure 2 One or more components of the first UE as described.

[0274] Receiver component 2002 may provide means for receiving communications (such as reference signals, control information, data communications, or combinations thereof) from equipment 2008. Receiver component 2002 may provide the received communications to one or more other components of equipment 2000 (such as communication manager 2004). In some aspects, receiver component 2002 may provide means for performing signal processing on the received communications (such as filtering, amplification, demodulation, analog-to-digital conversion, demultiplexing, deinterleaving, demapping, equalization, interference cancellation, or decoding, and other examples), and may provide the processed signals to one or more other components. In some aspects, receiver component 2002 may include combinations of the above. Figure 2 The first UE described includes one or more antennas, demodulators, MIMO detectors, receiver processors, controllers / processors, memory, or combinations thereof.

[0275] The transmission component 2006 may provide means for transmitting communications (such as reference signals, control information, data communications, or combinations thereof) to the equipment 2008. In some aspects, the communication manager 2004 may generate communications and transmit the generated communications to the transmission component 2006 for transmission to the equipment 2008. In some aspects, the transmission component 2006 may provide means for performing signal processing (such as filtering, amplification, modulation, digital-to-analog conversion, multiplexing, interleaving, mapping, or encoding, etc.) on the generated communications and transmit the processed signals to the equipment 2008. In some aspects, the transmission component 2006 may include combinations of the above. Figure 2 The first UE described includes one or more antennas, modulators, transmit MIMO processors, transmit processors, controllers / processors, memory, or combinations thereof. In some aspects, the transmit component 2006 may coexist with the receive component 2002 in a transceiver.

[0276] In some aspects, the communication manager 2004 may provide means for determining a feature vector associated with one or more features, which are related to the client's environment, using a first autoencoder; means for determining a latent vector using a second autoencoder and at least in part based on the feature vector; and means for transmitting the feature vector and the latent vector. In some aspects, the communication manager 2004 may include a combination of the above. Figure 2 The controller / processor, memory, or combination thereof of the first UE described.

[0277] In some respects, one or more components of the Communication Manager 2004 and / or the component set may include or may be implemented within hardware (e.g., in combination with...). Figure 24 (One or more of the components in the described circuit system). In some aspects, one or more components in the Communication Manager 2004 and / or the component set may include or may be combined with the above. Figure 2 The described UE 120 is implemented within the controller / processor, memory, or a combination thereof.

[0278] In some respects, one or more components of Communication Manager 2004 and / or the component set can be implemented in code (e.g., as software or firmware stored in memory), such as in combination with Figure 24 The described code. For example, the Communication Manager 2004 and / or its components (or a portion thereof) may be implemented as instructions or code stored in a non-transient computer-readable medium and executable by a controller or processor to perform the functions or operations of the Communication Manager 2004 and / or the component. If implemented in code, the functionality of the Communication Manager 2004 and / or its components may be achieved by combining the above. Figure 2 The controller / processor, memory, scheduler, communication unit, or a combination thereof of the described UE 120 are used to perform this action.

[0279] Figure 20 The number and arrangement of components shown are provided as an example. In practice, different arrangements may exist. Figure 20 The components shown are compared to additional components, fewer components, different components, or components arranged differently. Furthermore, Figure 20 The two or more components shown can be implemented within a single component, or Figure 20 The single component shown can be implemented as multiple distributed components. Additionally or alternatively, Figure 20 The component set shown (e.g., one or more components) is executable and described as being composed of... Figure 20 The other set of components shown performs one or more functions.

[0280] Figure 21 This is a diagram illustrating an example 2100 of the hardware implementation of device 2105 using processing system 2110. Device 2105 can be a client.

[0281] Processing system 2110 can be implemented with a bus architecture generally represented by bus 2115. Depending on the specific application and overall design constraints of processing system 2110, bus 2115 may include any number of interconnect buses and bridges. Bus 2115 links together various circuits including one or more processors and / or hardware components (represented by processor 2120, the described components, and computer-readable medium / memory 2125). Bus 2115 may also link various other circuits, such as timing sources, peripheral devices, voltage regulators, power management circuits, etc.

[0282] Processing system 2110 may be coupled to transceiver 2130. Transceiver 2130 is coupled to one or more antennas 2135. Transceiver 2130 provides means for communicating with various other equipment via a transmission medium. Transceiver 2130 receives signals from one or more antennas 2135, extracts information from the received signals, and provides the extracted information to processing system 2110 (specifically, receiving component 2002). Additionally, transceiver 2130 receives information from processing system 2110 (specifically, transmission component 2006) and generates signals to be applied to one or more antennas 2135, at least in part, based on the received information.

[0283] Processing system 2110 includes a processor 2120 coupled to a computer-readable medium / memory 2125. The processor 2120 is responsible for general processing, including the execution of software stored on the computer-readable medium / memory 2125. When executed by the processor 2120, the software causes the processing system 2110 to perform the various functions described herein for any particular apparatus. The computer-readable medium / memory 2125 may also be used to store data manipulated by the processor 2120 during software execution. The processing system further includes at least one of the described components. Each component may be a software module running in the processor 2120, a software module residing in / stored in the computer-readable medium / memory 2125, one or more hardware modules coupled to the processor 2120, or some combination thereof.

[0284] In some aspects, processing system 2110 may be a component of UE 120 and may include memory 282 and / or at least one of the following: TX MIMO processor 266, RX processor 258, and / or controller / processor 280. In some aspects, apparatus 2105 for wireless communication provides means for: determining a feature vector associated with one or more features, the one or more features being associated with the client's environment, using a first client autoencoder; determining a latent vector using a second client autoencoder and at least in part based on the feature vector; and transmitting the feature vector and the latent vector. The aforementioned means may be one or more components of apparatus 2000 and / or processing system 2110 of apparatus 2105 configured to perform the functions described by the aforementioned means. As described elsewhere herein, processing system 2110 may include TX MIMO processor 266, RX processor 258, and / or controller / processor 280. In one configuration, the aforementioned apparatus may be a TX MIMO processor 266, an RX processor 258, and / or a controller / processor 280 configured to perform the functions and / or operations described herein.

[0285] Figure 21 This is provided as an example. Other examples may differ from this combination. Figure 21 The example described.

[0286] Figure 22This is a block diagram of an example device 2200 for wireless communication according to the present disclosure. Device 2200 may be a server, or a server may include device 2200. In some aspects, device 2200 includes a receiving component 2202, a communication manager 2204, and a transmitting component 2206, which can communicate with each other (e.g., via one or more buses). As shown, device 2200 can use the receiving component 2202 and the transmitting component 2206 to communicate with another device 2208 (such as a client, UE, server, base station, or another wireless communication device).

[0287] In some respects, Equipment 2200 can be configured to perform the actions described in this article. Figure 3-17 One or more operations described herein. Additionally or alternatively, equipment 2200 may be configured to perform one or more processes described herein (such as...). Figure 19 (Process 1900). In some respects, equipment 2200 may include the above combination. Figure 2 One or more components of the base station described.

[0288] Receiver component 2202 may provide means for receiving communications (such as reference signals, control information, data communications, or combinations thereof) from equipment 2208. Receiver component 2202 may provide the received communications to one or more other components of equipment 2200 (such as communication manager 2204). In some aspects, receiver component 2202 may provide means for performing signal processing on the received communications (such as filtering, amplification, demodulation, analog-to-digital conversion, demultiplexing, deinterleaving, demapping, equalization, interference cancellation, or decoding, and other examples), and may provide the processed signals to one or more other components. In some aspects, receiver component 2202 may include a combination of the above. Figure 2 The described base station includes one or more antennas, demodulators, MIMO detectors, receiver processors, controllers / processors, memory, or combinations thereof.

[0289] Transmission component 2206 may provide means for transmitting communications (such as reference signals, control information, data communications, or combinations thereof) to equipment 2208. In some aspects, communication manager 2204 may generate communications and transmit the generated communications to transmission component 2206 for transmission to equipment 2208. In some aspects, transmission component 2206 may provide means for performing signal processing (such as filtering, amplification, modulation, digital-to-analog conversion, multiplexing, interleaving, mapping, or encoding, etc.) on the generated communications and transmit the processed signals to equipment 2208. In some aspects, transmission component 2206 may include combinations of the above. Figure 2The described base station includes one or more antennas, modulators, transmit MIMO processors, transmit processors, controllers / processors, memory, or combinations thereof. In some aspects, the transmit component 2206 may coexist with the receive component 2002 in a transceiver.

[0290] The communication manager 2204 may provide means for: receiving from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment; receiving from the client a latent vector at least partially based on the feature vector; determining an observed wireless communication vector at least partially based on the feature vector and the latent vector; and performing a wireless communication action at least partially based on the determined observed wireless communication vector. In some aspects, the communication manager 2204 may include a combination of the above. Figure 2 The described base station's controller / processor, memory, scheduler, communication unit, or a combination thereof.

[0291] In some respects, the communication manager 2204 and / or one or more components in the component set may include or may be implemented within the hardware (e.g., in combination with...). Figure 25 (One or more of the components in the described circuit system). In some aspects, the communication manager 2204 and / or one or more components in the component set may include or may combine the above. Figure 2 The BS 110 is implemented within its controller / processor, memory, or a combination thereof.

[0292] In some respects, the communication manager 2204 and / or one or more components of the component set can be implemented in code (e.g., as software or firmware stored in memory), such as in combination with Figure 25 The described code. For example, the communication manager 2204 and / or a component (or a portion thereof) may be implemented as instructions or code stored in a non-transient computer-readable medium and executable by a controller or processor to perform the functions or operations of the communication manager 2204 and / or that component. If implemented in code, the functionality of the communication manager 2204 and / or the component can be achieved by combining the above. Figure 2 The controller / processor, memory, scheduler, communication unit, or a combination thereof of the BS 110 described herein shall be used to perform the execution.

[0293] Figure 22 The number and arrangement of components shown are provided as an example. In practice, different arrangements may exist. Figure 22 The components shown are compared to additional components, fewer components, different components, or components arranged differently. Furthermore, Figure 22 The two or more components shown can be implemented within a single component, or Figure 22The single component shown can be implemented as multiple distributed components. Additionally or alternatively, Figure 22 The component set shown (e.g., one or more components) can perform actions described as being performed by Figure 22 The other set of components shown performs one or more functions.

[0294] Figure 23 This is a diagram illustrating an example 2300 of the hardware implementation of device 2305 using processing system 2310. Device 2305 may be a server.

[0295] Processing system 2310 can be implemented with a bus architecture generally represented by bus 2315. Depending on the specific application and overall design constraints of processing system 2310, bus 2315 may include any number of interconnect buses and bridges. Bus 2315 links together various circuits including one or more processors and / or hardware components (represented by processor 2320, the described components, and computer-readable medium / memory 2325). Bus 2315 may also link various other circuits, such as timing sources, peripheral devices, voltage regulators, power management circuits, etc.

[0296] Processing system 2310 may be coupled to transceiver 2330. Transceiver 2330 is coupled to one or more antennas 2335. Transceiver 2330 provides means for communicating with various other equipment via a transmission medium. Transceiver 2330 receives signals from one or more antennas 2335, extracts information from the received signals, and provides the extracted information to processing system 2310 (specifically, receiving component 2202). Additionally, transceiver 2330 receives information from processing system 2310 (specifically, transmission component 2206) and generates signals to be applied to one or more antennas 2335, at least in part, based on the received information.

[0297] Processing system 2310 includes a processor 2320 coupled to a computer-readable medium / memory 2325. The processor 2320 is responsible for general processing, including the execution of software stored on the computer-readable medium / memory 2325. When executed by the processor 2320, the software causes the processing system 2310 to perform the various functions described herein for any particular apparatus. The computer-readable medium / memory 2325 may also be used to store data manipulated by the processor 2320 during software execution. The processing system further includes at least one of the described components. Each component may be a software module running in the processor 2320, a software module residing in / stored in the computer-readable medium / memory 2325, one or more hardware modules coupled to the processor 2320, or some combination thereof.

[0298] In some aspects, processing system 2310 may be a component of base station 110 and may include memory 242 and / or at least one of the following: TX MIMO processor 230, RX processor 238, and / or controller / processor 240. In some aspects, apparatus 2305 for wireless communication includes means for: receiving from a client a feature vector associated with one or more features, the one or more features being associated with the client's environment; receiving from the client a latent vector at least partially based on the feature vector; determining an observed wireless communication vector at least partially based on the feature vector and the latent vector; and performing a wireless communication action at least partially based on the determined observed wireless communication vector. The aforementioned means may be one or more components of apparatus 2200 and / or processing system 2310 of apparatus 2305 configured to perform the functions described by the aforementioned means. As described elsewhere herein, processing system 2310 may include TX MIMO processor 230, receiver processor 238, and / or controller / processor 240. In one configuration, the aforementioned apparatus may be a TX MIMO processor 230, a receiver processor 238, and / or a controller / processor 240 configured to perform the functions and / or operations described herein.

[0299] Figure 23 This is provided as an example. Other examples may differ from this combination. Figure 23 The example described.

[0300] Figure 24 This is a diagram illustrating the implementation of the code and circuitry system used in device 2405. Device 2405 can be a client.

[0301] like Figure 24 As further shown, the apparatus may include a circuit system (circuit system 2420) for determining feature vectors associated with one or more features, which are associated with the client's environment, using a first client autoencoder. For example, the apparatus may include a circuit system for enabling the apparatus to determine feature vectors associated with one or more features, which are associated with the client's environment, using a first autoencoder.

[0302] like Figure 24 As further shown, the apparatus may include a circuit system (circuit system 2425) for determining a latent vector using a second autoencoder and at least partially based on the feature vector. For example, the apparatus may include a circuit system for determining a latent vector using a second autoencoder and at least partially based on the feature vector.

[0303] like Figure 24As further shown, the apparatus may include a circuit system (circuit system 2430) for transmitting the feature vector and the latent vector. For example, the apparatus may include a circuit system for enabling the apparatus to transmit the feature vector and the latent vector.

[0304] like Figure 24 As further shown, the apparatus may include code (code 2435) stored in computer-readable medium 2125 for determining feature vectors associated with one or more features, which are associated with the environment of the client, using a first autoencoder. For example, the apparatus may include code that, when executed by processor 2120, enables processor 2120 to determine feature vectors associated with one or more features, which are associated with the environment of the client, using a first autoencoder.

[0305] like Figure 24 As further shown, the apparatus may include code (code 2440) stored in computer-readable medium 2125 for determining a latent vector using a second autoencoder and at least in part based on the feature vector. For example, the apparatus may include code that, when executed by processor 2120, enables processor 2120 to determine a latent vector using a second autoencoder and at least in part based on the feature vector.

[0306] like Figure 24 As further shown, the apparatus may include code (code 2445) stored in computer-readable medium 2125 for transmitting the feature vector and the latent vector. For example, the apparatus may include code that, when executed by processor 2120, enables transceiver 2130 to transmit the feature vector and the latent vector.

[0307] Figure 24 This is provided as an example. Other examples may differ from this combination. Figure 24 The example described.

[0308] Figure 25 This is a diagram illustrating the implementation of the code and circuitry system used in the device 2505. The device 2505 can be a server.

[0309] like Figure 25 As further shown, the apparatus may include a circuit system (circuit system 2520) for receiving feature vectors associated with one or more features from a client, the one or more features being associated with the client's environment. For example, the apparatus may include a circuit system for enabling the apparatus to receive feature vectors associated with one or more features from a client, the one or more features being associated with the client's environment.

[0310] like Figure 25As further shown, the apparatus may include a circuit system (circuit system 2525) for receiving a latent vector at least partially based on the feature vector from the client. For example, the apparatus may include a circuit system that enables the apparatus to receive a latent vector at least partially based on the feature vector from the client.

[0311] like Figure 25 As further shown, the apparatus may include a circuit system (circuit system 2530) for determining the observed wireless communication vector based at least in part on the feature vector and the latent vector. For example, the apparatus may include a circuit system for enabling the apparatus to determine the observed wireless communication vector based at least in part on the feature vector and the latent vector.

[0312] like Figure 25 As further shown, the apparatus may include a circuit system (circuit system 2535) for performing wireless communication actions at least in part based on a determined observed wireless communication vector. For example, the apparatus may include a circuit system for causing the apparatus to perform wireless communication actions at least in part based on a determined observed wireless communication vector.

[0313] like Figure 25 As further shown, the apparatus may include code (code 2540) stored in computer-readable medium 2325 for receiving feature vectors associated with one or more features related to the client's environment. For example, the apparatus may include code that, when executed by processor 2320, enables transceiver 2330 to receive feature vectors associated with one or more features related to the client's environment.

[0314] like Figure 25 As further shown, the apparatus may include code (code 2545) stored in computer-readable medium 2325 for receiving a latent vector at least partially based on the feature vector from the client. For example, the apparatus may include code that, when executed by processor 2320, enables transceiver 2330 to receive a latent vector at least partially based on the feature vector from the client.

[0315] like Figure 25 As further shown, the apparatus may include code (code 2550) stored in computer-readable medium 2325 for determining the observed wireless communication vector based at least in part on the feature vector and the latent vector. For example, the apparatus may include code that, when executed by processor 2320, enables processor 2320 to determine the observed wireless communication vector based at least in part on the feature vector and the latent vector.

[0316] like Figure 25As further shown, the apparatus may include code (code 2555) stored in computer-readable medium 2325 for performing wireless communication actions at least in part based on a determined observed wireless communication vector. For example, the apparatus may include code that, when executed by processor 2320, enables processor 2320 to perform wireless communication actions at least in part based on a determined observed wireless communication vector.

[0317] Figure 25 This is provided as an example. Other examples may differ from this combination. Figure 25 The example described.

[0318] The following provides an overview of some aspects of this disclosure:

[0319] Aspect 1: A wireless communication method performed by a client, comprising: determining a feature vector associated with one or more features, the one or more features being associated with the client’s environment, using a first autoencoder; determining a latent vector using a second autoencoder and at least in part based on the feature vector; and transmitting the feature vector and the latent vector.

[0320] Aspect 2: The method of aspect 1, wherein the first autoencoder includes: a first encoder configured to receive an observed environment vector as input and provide the feature vector as output; and a first decoder configured to receive the feature vector as input and provide the observed environment vector as output.

[0321] Aspect 3: The method of aspect 2, wherein the second autoencoder comprises: a second encoder configured to receive an observed wireless communication vector and the feature vector as input and provide the latent vector as output; and a second decoder configured to receive the latent vector and the feature vector as input and provide the observed wireless communication vector as output.

[0322] Aspect 4: The method of any of Aspects 1-3, wherein determining the feature vector includes: providing the observed environment vector as input to the first autoencoder.

[0323] Aspect 5: The method of aspect 4, wherein the observed environment vector includes one or more feature components, wherein the one or more feature components indicate: client vendor identifier, client antenna configuration, large-scale channel characteristics, channel state information reference signal configuration, image obtained by an imaging device, a portion of the estimated propagation channel, or a combination thereof.

[0324] Aspect 6: The method of aspect 5, wherein the large-scale channel characteristics indicate: channel-associated delay spread, channel-associated power delay profile, channel-associated Doppler measurement, channel-associated Doppler spectrum, channel-associated signal-to-noise ratio, channel-associated signal-to-noise plus-interference ratio, reference signal received power, received signal strength indicator, or a combination thereof.

[0325] Aspect 7: The method of any of Aspects 1-6, wherein the implicit vector is associated with the wireless communication task.

[0326] Aspect 8: The method of aspect 7, wherein the wireless communication task includes: determining channel state feedback (CSF), determining location information associated with the client, determining modulation associated with wireless communication, determining waveform associated with wireless communication, or a combination thereof.

[0327] Aspect 9: The method of aspect 8, wherein the wireless communication task includes determining the CSF, and wherein the method further includes: receiving a channel state information (CSI) reference signal (CSI-RS); determining the CSI based at least in part on the CSI-RS; and providing the CSI as input to a second autoencoder.

[0328] Aspect 10: The method of aspect 9, wherein the implicit vector includes compressed channel state feedback.

[0329] Aspect 11: A method of any of Aspects 1-10, wherein transmitting the feature vector and the latent vector comprises: using a physical uplink control channel, a physical uplink shared channel, or a combination thereof to transmit the feature vector and the latent vector.

[0330] Aspect 12: The method of any of Aspects 1-11, wherein at least one of the first autoencoder or the second autoencoder includes a variational autoencoder.

[0331] Aspect 13: The method of any of Aspects 1-12 further includes: training at least one of a first autoencoder or a second autoencoder.

[0332] Aspect 14: The method of aspect 13, wherein training at least one of the first autoencoder or the second autoencoder includes: using reparameterization.

[0333] Aspect 15: A method as in any of Aspects 13 or 14, wherein training at least one of a first autoencoder or a second autoencoder comprises: determining a set of neural network parameters that maximizes a variational lower bound function corresponding to the first autoencoder and the second autoencoder.

[0334] Aspect 16: The method of aspect 15, wherein the negative variational lower bound function corresponds to the sum of a first loss function associated with a first autoencoder and a second loss function associated with a second autoencoder.

[0335] Aspect 17: The method of aspect 16, wherein the first loss function includes a first reconstruction loss and a first regularization term for the first autoencoder, and wherein the second loss function includes a second reconstruction loss and a second regularization term for the second autoencoder.

[0336] Aspect 18: The method of either Aspect 16 or 17, wherein the first autoencoder and the second autoencoder are regular autoencoders, and wherein the variational lower bound function does not include a regularization term.

[0337] Aspect 19: The method of any of Aspects 13-18, wherein training at least one of the first autoencoder or the second autoencoder comprises: training at least one of the first autoencoder or the second autoencoder using a joint learning procedure.

[0338] Aspect 20: The method of any of Aspects 13-19, wherein training at least one of the first autoencoder or the second autoencoder comprises: training the first autoencoder and the second autoencoder to determine the format of the feature vector.

[0339] Aspect 21: The method of any of Aspects 13-20, wherein training at least one of the first autoencoder or the second autoencoder comprises: using an unsupervised learning procedure.

[0340] Aspect 22: The method of any of Aspects 13-21, wherein training at least one of the first autoencoder or the second autoencoder comprises: performing a fully joint learning procedure.

[0341] Aspect 23: The method of aspect 22, wherein performing the fully joint learning procedure includes: jointly training the first autoencoder and the second autoencoder.

[0342] Aspect 24: The method of aspect 22, wherein performing the fully joint learning procedure includes alternating between training the first autoencoder and training the second autoencoder.

[0343] Aspect 25: The method of aspect 24, wherein alternating between training the first autoencoder and training the second autoencoder includes: performing a first plurality of training iterations associated with the first autoencoder according to a first training frequency; and performing a second plurality of training iterations associated with the second autoencoder according to a second training frequency higher than the first training frequency.

[0344] Aspect 26: The method of any of Aspects 13-21, wherein training at least one of the first autoencoder or the second autoencoder comprises: performing a partially joint learning procedure.

[0345] Aspect 27: The method of aspect 26, wherein performing a partial joint learning procedure includes: providing an observed environment vector to a server; and receiving a first autoencoder from the server, wherein the first autoencoder is at least partially based on the observed environment vector.

[0346] Aspect 28: The method of aspect 27, wherein the first autoencoder is based at least in part on at least one additional environment vector associated with at least one additional client.

[0347] Aspect 29: The method of any of Aspects 26-28, wherein performing a partial joint learning procedure includes: updating a second autoencoder to determine an updated neural network parameter set; and transmitting the updated neural network parameter set to a server.

[0348] Aspect 30: A method of any of Aspects 26-29, wherein performing a partial joint learning procedure includes: performing a first plurality of training iterations associated with a first autoencoder according to a first training frequency, wherein performing the training iterations in the first plurality of training iterations includes: providing an observed environment vector to a server; and receiving an updated first autoencoder from the server, wherein the updated first autoencoder is at least partially based on the observed environment vector; and performing a second plurality of training iterations associated with a second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0349] Aspect 31: A method of any of Aspects 26-30, wherein performing a partial joint learning procedure includes: receiving a set of neural network parameters associated with a first autoencoder and a second autoencoder from a server; obtaining an observed environment training vector; inputting the observed environment training vector into a first encoder of the first autoencoder to determine a training feature vector; obtaining an observed wireless communication training vector; inputting the training feature vector and the observed wireless communication training vector into a second encoder of the second autoencoder to determine a training latent vector; inputting the training feature vector and the training latent vector into a second decoder of the second autoencoder to determine a second training output of the second autoencoder; and determining a loss associated with the second autoencoder based at least in part on the second training output, wherein the loss is associated with the set of neural network parameters.

[0350] Aspect 32: The method of aspect 31 further includes: determining a plurality of gradients of the loss relative to an autoencoder parameter set, wherein the autoencoder parameter set corresponds to a second autoencoder; and updating the autoencoder parameter set at least in part based on the plurality of gradients.

[0351] Aspect 33: The method of aspect 32 further includes: updating the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set.

[0352] Aspect 34: The method of aspect 33 further includes: transmitting the last updated autoencoder parameter set to the server.

[0353] Aspect 35: The method of any of Aspects 1-34 further includes: determining a first loss corresponding to a first autoencoder, wherein determining the first loss includes: receiving from a server a set of neural network parameters associated with a first autoencoder and a second autoencoder; obtaining an observed environment training vector; inputting the observed environment training vector into a first encoder of the first autoencoder to determine a training feature vector; inputting the training feature vector into a first decoder of the first autoencoder to determine a first training output of the first autoencoder; and determining the first loss based at least in part on the first training output, wherein the first loss is associated with the set of neural network parameters.

[0354] Aspect 36: The method of aspect 35 further includes: determining a second loss corresponding to the second autoencoder, wherein determining the second loss includes: obtaining an observed wireless communication training vector; inputting the training feature vector and the observed wireless communication training vector into a second encoder of the second autoencoder to determine a training latent vector; inputting the training feature vector and the training latent vector into a second decoder of the second autoencoder to determine a second training output of the second autoencoder; and determining a second loss associated with the second autoencoder based at least in part on the second training output, wherein the second loss is associated with the set of parameters of the neural network.

[0355] Aspect 37: The method of aspect 36 further includes: determining a first regularization term corresponding to a first autoencoder; determining a second regularization term corresponding to a second autoencoder, wherein determining the first loss includes: determining the first loss at least in part based on the first regularization term, and wherein determining the second loss includes: determining the second loss at least in part based on the second regularization term.

[0356] Aspect 38: The method of any of Aspects 36 or 37 further includes: determining the total loss by summing the first loss and the second loss.

[0357] Aspect 39: The method of aspect 38 further includes: determining multiple gradients of the total loss relative to the set of parameters of the neural network; and updating the set of parameters of the neural network at least in part based on the multiple gradients.

[0358] Aspect 40: The method of aspect 39 further includes: updating the neural network parameter set a specified number of times to determine the final updated neural network parameter set.

[0359] Aspect 41: The method of aspect 40 further includes: transmitting the final updated set of neural network parameters to the server.

[0360] Aspect 42: The method of any of Aspects 38-41 further includes: determining a first plurality of gradients of the total loss relative to a first set of autoencoder parameters associated with the first autoencoder; and updating the first set of autoencoder parameters at least in part based on the first plurality of gradients.

[0361] Aspect 43: The method of aspect 42 further includes: updating the first autoencoder parameter set a specified number of times to determine the final updated first autoencoder parameter set.

[0362] Aspect 44: The method of aspect 43 further includes: transmitting the last updated first autoencoder parameter set to the server.

[0363] Aspect 45: The method of aspect 44 further includes: determining a second plurality of gradients of the second loss relative to the second autoencoder parameter set associated with the second autoencoder; and updating the second autoencoder parameter set at least in part based on the second plurality of gradients.

[0364] Aspect 46: The method of aspect 45 further includes: updating the second autoencoder parameter set a specified number of times to determine the final updated second autoencoder parameter set.

[0365] Aspect 47: The method of aspect 46 further includes: transmitting the last updated set of second autoencoder parameters to the server.

[0366] Aspect 48: A wireless communication method performed by a server, comprising: receiving from a client a feature vector associated with one or more features, the one or more features being associated with an environment of the client; receiving from the client a latent vector at least partially based on the feature vector; determining an observed wireless communication vector at least partially based on the feature vector and the latent vector; and performing a wireless communication action at least partially based on the determined observed wireless communication vector.

[0367] Aspect 49: The method of aspect 48, wherein the feature vector is at least partially based on the observed environment vector.

[0368] Aspect 50: The method of aspect 49, wherein the observed environment vector includes one or more feature components, wherein the one or more feature components indicate: client vendor identifier, client antenna configuration, large-scale channel characteristics, channel state information reference signal configuration, image obtained by an imaging device, a portion of the estimated propagation channel, or a combination thereof.

[0369] Aspect 51: The method of aspect 50, wherein the large-scale channel characteristics indicate: channel-associated delay spread, channel-associated power delay profile, channel-associated Doppler measurement, channel-associated Doppler spectrum, channel-associated signal-to-noise ratio, channel-associated signal-to-noise plus-interference ratio, reference signal received power, received signal strength indicator, or a combination thereof.

[0370] Aspect 52: The method of any of Aspects 48-51, wherein the observed wireless communication vector is associated with the wireless communication task.

[0371] Aspect 53: The method of aspect 52, wherein the wireless communication task includes: receiving channel state feedback (CSF), receiving location information associated with the UE, receiving modulation associated with wireless communication, receiving waveforms associated with wireless communication, or a combination thereof.

[0372] Aspect 54: A method of any of Aspects 52 or 53, wherein the wireless communication task includes: receiving the CSF, and wherein the method further includes: transmitting a Channel State Information Reference Signal (CSI-RS), wherein the CSF corresponds to the CSI-RS.

[0373] Aspect 55: The method of aspect 54, wherein the implicit vector includes compressed channel state feedback.

[0374] Aspect 56: The method of any of Aspects 48-55, wherein the feature vector and the latent vector are carried in: a physical uplink control channel, a physical uplink shared channel, or a combination thereof.

[0375] Aspect 57: The method of any of Aspects 48-56, wherein determining the observed wireless communication vector comprises: inputting the feature vector and the latent vector into the decoder of the autoencoder.

[0376] Aspect 58: The method of aspect 57, wherein the autoencoder includes a regular autoencoder or a variational autoencoder.

[0377] Aspect 59: The method of any of Aspects 57 or 58, wherein the autoencoder corresponds to a second client autoencoder in a pair of client autoencoders, the pair of client autoencoders including a first client autoencoder and a second client autoencoder.

[0378] Aspect 60: The method of aspect 59, wherein the first client autoencoder includes: a first encoder configured to receive an observed environment vector as input and provide the feature vector as output; and a first decoder configured to receive the feature vector as input and provide the observed environment vector as output.

[0379] Aspect 61: The method of aspect 60, wherein the second client autoencoder comprises: a second encoder configured to receive an observed wireless communication vector and the feature vector as input and provide the latent vector as output; and a second decoder configured to receive the latent vector and the feature vector as input and provide the observed wireless communication vector as output.

[0380] Aspect 62: The method of aspect 61 further includes performing a partial joint learning procedure, including: training a first client autoencoder; and transmitting the first client autoencoder to the client.

[0381] Aspect 63: The method of aspect 62, wherein performing a partial joint learning procedure includes: performing a first plurality of training iterations associated with a first client autoencoder according to a first training frequency, wherein performing the training iterations in the first plurality of training iterations includes: receiving observed environment training vectors from the client; and transmitting an updated first client autoencoder to the client, wherein the updated first client autoencoder is at least partially based on the observed environment training vectors; and performing a second plurality of training iterations associated with a second autoencoder according to a second training frequency, the second training frequency being higher than the first training frequency.

[0382] Aspect 64: The method of any of Aspects 62 or 63 further includes: transmitting a first client autoencoder to at least one additional client.

[0383] Aspect 65: The method of any of Aspects 62-64, wherein training the first client autoencoder includes: using an unsupervised learning procedure.

[0384] Aspect 66: The method of any of Aspects 62-65 further includes: receiving observed environment training vectors from the client, wherein training the first client autoencoder includes: training the first client autoencoder at least in part based on the observed environment training vectors.

[0385] Aspect 67: The method of aspect 66 further includes: receiving at least one additional observed environment training vector from at least one additional client, wherein training the first client autoencoder includes: training the first client autoencoder at least in part based on the at least one additional observed environment training vector.

[0386] Aspect 68: The method of aspect 67 further includes: selecting a set of clients from which to receive a set of observed environment training vectors, wherein the set of clients includes the client and the at least one additional client.

[0387] Aspect 69: A method of any of Aspects 62-68, wherein training a first client autoencoder comprises: determining a loss corresponding to the first client autoencoder, wherein determining the loss comprises: obtaining a set of neural network parameters associated with the first client autoencoder and a second client autoencoder; receiving an observed environment training vector from the client; inputting the observed environment training vector into the encoder of the first client autoencoder to determine a training feature vector; inputting the training feature vector into the decoder of the first client autoencoder to determine a training output of the first client autoencoder; and determining the loss at least in part based on the training output, wherein the loss is associated with the set of neural network parameters.

[0388] Aspect 70: The method of aspect 69 further includes: determining a regularization term corresponding to the first client autoencoder, wherein determining the loss includes: determining the loss based at least in part on the regularization term.

[0389] Aspect 71: The method of any of Aspects 69 or 70 further includes: determining a plurality of gradients of the loss relative to the autoencoder parameter set; and updating the autoencoder parameter set at least in part based on the plurality of gradients, wherein the autoencoder parameter set corresponds to a first client autoencoder.

[0390] Aspect 72: The method of aspect 71 further includes: updating the autoencoder parameter set a specified number of times to determine the final updated autoencoder parameter set.

[0391] Aspect 73: A method of any of Aspects 62-72, wherein performing a partial joint learning procedure includes: selecting a set of clients from which to receive updates associated with a second client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting neural network parameter sets associated with a first client autoencoder and a second client autoencoder to the set of clients; receiving a plurality of updated autoencoder parameter sets associated with the second client autoencoder from the set of clients; and determining a final updated autoencoder parameter set associated with the second client autoencoder based at least in part on the plurality of updated autoencoder parameter sets.

[0392] Aspect 74: The method of aspect 73 further includes: transmitting to the client set the last updated autoencoder parameter set associated with the second client autoencoder.

[0393] Aspect 75: The method of any of Aspects 73 or 74, wherein determining the last updated autoencoder parameter set associated with the second client autoencoder includes: averaging multiple updated autoencoder parameter sets.

[0394] Aspect 76: The method of any of Aspects 61-75 further includes: selecting a set of clients from which to receive updates associated with a first client autoencoder and a second client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting a set of neural network parameters associated with the first client autoencoder and the second client autoencoder to the set of clients; receiving from the set of clients a plurality of updated neural network parameter sets associated with the first client autoencoder and the second client autoencoder; and determining, at least in part, a final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder based on the plurality of updated neural network parameters.

[0395] Aspect 77: The method of aspect 76 further includes: transmitting to the client set the last updated neural network parameter set associated with the first client autoencoder and the second client autoencoder.

[0396] Aspect 78: The method of any of Aspects 76 or 77, wherein determining the final updated neural network parameter set associated with the first client autoencoder and the second client autoencoder comprises: averaging multiple updated neural network parameter sets.

[0397] Aspect 79: The method of any of Aspects 61-78 further includes: selecting a set of clients from which to receive updates associated with a first client autoencoder, wherein the set of clients includes the client and at least one additional client; transmitting autoencoder parameter sets associated with the first client autoencoder and a second client autoencoder to the set of clients; receiving a plurality of updated autoencoder parameter sets associated with the first client autoencoder from the set of clients; and determining a final updated autoencoder parameter set associated with the first client autoencoder based at least in part on the plurality of updated autoencoder parameter sets.

[0398] Aspect 80: The method of aspect 79 further includes: transmitting to the client set the last updated set of autoencoder parameters associated with the first client autoencoder.

[0399] Aspect 81: The method of any of Aspects 79 or 80, wherein determining the last updated autoencoder parameter set associated with the first client autoencoder includes: averaging multiple updated autoencoder parameter sets.

[0400] Aspect 82: An apparatus for wireless communication at a device, comprising: a processor; a memory coupled to the processor; and instructions stored in the memory and executable by the processor to cause the apparatus to perform one or more of the methods of aspects 1-47.

[0401] Aspect 83: An apparatus for wireless communication, comprising a memory and one or more processors coupled to the memory, the one or more processors being configured to perform one or more of the methods of aspects 1-47.

[0402] Aspect 84: An apparatus for wireless communication, comprising at least one means for performing one or more methods as described in aspects 1-47.

[0403] Aspect 85: A non-transient computer-readable medium storing code for wireless communication, the code including instructions executable by a processor to perform one or more of the methods of aspects 1-47.

[0404] Aspect 86: A non-transient computer-readable medium storing a set of instructions for wireless communication, the set of instructions including one or more instructions which, when executed by one or more processors of a device, cause the device to perform one or more methods as described in aspects 1-47.

[0405] Aspect 87: An apparatus for wireless communication at a device, comprising: a processor; a memory coupled to the processor; and instructions stored in the memory and executable by the processor to cause the apparatus to perform one or more of the methods of aspects 48-81.

[0406] Aspect 88: An apparatus for wireless communication, comprising a memory and one or more processors coupled to the memory, the one or more processors being configured to perform one or more of the methods of aspects 48-81.

[0407] Aspect 89: An apparatus for wireless communication, comprising at least one means for performing one or more methods as described in aspects 48-81.

[0408] Aspect 90: A non-transient computer-readable medium storing code for wireless communication, the code including instructions executable by a processor to perform one or more methods as described in aspects 48-81.

[0409] Aspect 91: A non-transient computer-readable medium storing a set of instructions for wireless communication, the set of instructions including one or more instructions which, when executed by one or more processors of a device, cause the device to perform one or more methods as described in aspects 48-81.

[0410] The foregoing disclosure provides explanations and descriptions, but is not intended to be exhaustive or to limit the aspects to the precise forms disclosed. Modifications and variations may be made in light of the foregoing disclosure or may be obtained through practice.

[0411] As used herein, the term "component" is intended to be interpreted broadly as hardware, firmware, and / or a combination of hardware and software. As used herein, a processor is implemented using hardware, firmware, and / or a combination of hardware and software. It will be apparent that the systems and / or methods described herein can be implemented in various forms as hardware, firmware, and / or combinations of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods is not limited in any aspect. Thus, the operation and behavior of these systems and / or methods are described herein without reference to any specific software code—it is understood that software and hardware can be designed to implement these systems and / or methods, at least in part, based on the descriptions herein.

[0412] As used in this article, depending on the context, a threshold can refer to a value greater than the threshold, greater than or equal to the threshold, less than the threshold, less than or equal to the threshold, equal to the threshold, not equal to the threshold, etc.

[0413] Although specific combinations of features are described in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of aspects. In fact, many of these features can be combined in ways not specifically described in the claims and / or not disclosed in the specification. Although each dependent claim listed below may be directly subordinated to only one claim, the disclosure of aspects includes each dependent claim being combined with each other claim in this set of claims. As used herein, the phrase “at least one of” refers to any combination of these items, including single members. As an example, “at least one of a, b, or c” is intended to cover: a, b, c, ab, ac, bc, and abc, as well as any combination having multiple identical elements (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other ordering of a, b, and c).

[0414] The elements, actions, or instructions used herein should not be construed as critical or necessary unless explicitly stated otherwise. Furthermore, as used herein, the articles “a” and “a certain” are intended to include one or more items and may be used interchangeably with “one or more.” Additionally, as used herein, the article “the” is intended to include one or more items referenced in conjunction with the article “the” and may be used interchangeably with “one or more.” Furthermore, as used herein, the terms “set” and “group” are intended to include one or more items (e.g., related items, unrelated items, or a combination of related and unrelated items) and may be used interchangeably with “one or more.” Where only one item is intended, the phrase “only one” or similar language is used. Moreover, as used herein, the terms “have,” “contain,” “include,” etc., are intended to be open-ended terms. Additionally, the phrase “based on” is intended to mean “at least partially based on” unless otherwise explicitly stated. Moreover, as used herein, the term “or” is intended to be inclusive when used in a sequence and may be used interchangeably with “and / or” unless otherwise explicitly stated (e.g., in combination with “either of” or “only one of”).

Claims

1. A wireless communication method executed by a client, comprising: A first autoencoder is used to determine feature vectors associated with one or more features, which are related to the client's environment; A second autoencoder is used to determine a latent vector, at least in part, based on the feature vector, the latent vector including compressed channel state feedback; and Transmit the feature vector and the latent vector.

2. The method of claim 1, wherein the first autoencoder comprises: A first encoder is configured to receive an observed environment vector as input and provide the feature vector as output; as well as A first decoder is configured to receive the feature vector as input and provide the observed environment vector as output.

3. The method of claim 2, wherein the second autoencoder comprises: A second encoder is configured to receive the observed wireless communication vector and the feature vector as input and provide the latent vector as output; as well as The second decoder is configured to receive the latent vector and the feature vector as input and provide the observed wireless communication vector as output.

4. The method of claim 1, wherein determining the feature vector comprises: The observed environment vector is provided as input to the first autoencoder.

5. The method of claim 4, wherein the observed environment vector comprises one or more feature components, wherein the one or more feature components indicate: Client vendor identifier, Client antenna configuration, Large-scale channel characteristics Channel state information reference signal configuration, Images obtained through imaging equipment A portion of the estimated propagation channel, or Its combination.

6. The method of claim 5, wherein the large-scale channel characteristics indicate: Delay spread associated with the channel, Channel-related power delay profile Doppler measurements associated with the channel, The Doppler spectrum associated with the channel, Signal-to-noise ratio associated with the channel, The signal-to-noise-plus-interference ratio associated with the channel. The power received by the reference signal, Received signal strength indicator, or Its combination.

7. The method of claim 1, wherein the latent vector is associated with a wireless communication task, wherein the wireless communication task includes: Determine the Channel State Feedback (CSF). Determine the location information associated with the client. Determine the modulation associated with wireless communication. Determine the waveform associated with wireless communication, or Its combination.

8. The method of claim 7, wherein the wireless communication task includes determining the CSF, and wherein the method further comprises: Receive Channel State Information (CSI) Reference Signal (CSI-RS); The CSI is determined at least in part based on the CSI-RS; as well as The CSI is provided as input to the second autoencoder.

9. The method of claim 1, wherein transmitting the feature vector and the latent vector comprises: Use the following to transmit the feature vector and the latent vector: Physical uplink control channel, Physical uplink shared channel, or Its combination.

10. The method of claim 1, wherein at least one of the first autoencoder or the second autoencoder comprises a variational autoencoder.

11. The method of claim 1, further comprising: Train at least one of the first autoencoder or the second autoencoder.

12. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Use reparameterization.

13. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Determine the set of neural network parameters that maximizes the variational lower bound function corresponding to the first autoencoder and the second autoencoder.

14. The method of claim 13, wherein the negative variational lower bound function corresponds to the sum of a first loss function associated with the first autoencoder and a second loss function associated with the second autoencoder.

15. The method of claim 14, wherein the first loss function includes a first reconstruction loss and a first regularization term for the first autoencoder, and wherein the second loss function includes a second reconstruction loss and a second regularization term for the second autoencoder.

16. The method of claim 14, wherein the first autoencoder and the second autoencoder are regular autoencoders, and wherein the variational lower bound function does not include a regularization term.

17. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: The first autoencoder and the second autoencoder are trained to determine the format of the feature vector.

18. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Use unsupervised learning procedures.

19. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Performing a fully joint learning procedure, wherein performing the fully joint learning procedure includes: jointly training the first autoencoder and the second autoencoder.

20. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: The fully joint learning procedure is executed, wherein executing the fully joint learning procedure includes alternating between training the first autoencoder and training the second autoencoder.

21. The method of claim 20, wherein alternating between training the first autoencoder and training the second autoencoder comprises: The first multiple training iterations associated with the first autoencoder are performed according to the first training frequency; as well as A second set of training iterations associated with the second autoencoder is performed based on a second training frequency, which is higher than the first training frequency.

22. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Execute a partial joint learning procedure, wherein executing the partial joint learning procedure includes: Provide the observed environment vectors to the server; and The server receives the first autoencoder, wherein the first autoencoder is trained at least in part based on the observed environment vectors.

23. The method of claim 22, wherein the first autoencoder is trained at least in part based on at least one additional context vector associated with at least one additional client.

24. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Execute a partial joint learning procedure, wherein executing the partial joint learning procedure includes: Update the second autoencoder to determine the updated neural network parameter set; and The updated neural network parameter set is transmitted to the server.

25. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Execute a partial joint learning procedure, wherein executing the partial joint learning procedure includes: Executing a first plurality of training iterations associated with the first autoencoder according to a first training frequency, wherein executing the training iterations in the first plurality of training iterations includes: Provide the observed environment vectors to the server; and Receive an updated first autoencoder from the server, wherein the updated first autoencoder is trained at least in part based on the observed environment vectors; and A second set of training iterations associated with the second autoencoder is performed based on a second training frequency, which is higher than the first training frequency.

26. The method of claim 11, wherein training at least one of the first autoencoder or the second autoencoder comprises: Execute a partial joint learning procedure, wherein executing the partial joint learning procedure includes: Receive the neural network parameter set associated with the first autoencoder and the second autoencoder from the server; Obtain the observed environment training vectors; The observed environment training vector is input into the first encoder of the first autoencoder to determine the training feature vector; Obtain the observed wireless communication training vectors; The training feature vector and the observed wireless communication training vector are input into the second encoder of the second autoencoder to determine the training latent vector; The training feature vector and the training latent vector are input into the second decoder of the second autoencoder to determine the second training output of the second autoencoder; The loss associated with the second autoencoder is determined at least in part based on the second training output, wherein the loss is associated with the set of parameters of the neural network; Determine multiple gradients of the loss with respect to a set of autoencoder parameters, wherein the set of autoencoder parameters corresponds to the second autoencoder; The autoencoder parameter set is updated at least in part based on the plurality of gradients; The autoencoder parameter set is updated a specified number of times to determine the final updated autoencoder parameter set; and The final updated autoencoder parameter set is transmitted to the server.

27. A wireless communication method executed by a server, comprising: Receive feature vectors associated with one or more features from the client, the one or more features being associated with the client's environment; Receive from the client a latent vector at least in part based on the feature vector, the latent vector including compressed channel state feedback; The observed wireless communication vector is determined at least in part based on the eigenvector and the latent vector; and Wireless communication actions are performed, at least in part, based on determining the observed wireless communication vector.

28. The method of claim 27, wherein the feature vector is at least partially based on the observed environment vector.

29. An apparatus for wireless communication at a client, comprising: Memory; as well as One or more processors coupled to the memory, the one or more processors being configured to: A first autoencoder is used to determine feature vectors associated with one or more features, which are related to the client's environment; A second autoencoder is used to determine a latent vector, at least in part, based on the feature vector, the latent vector including compressed channel state feedback; and Transmit the feature vector and the latent vector.

30. The apparatus of claim 29, wherein the one or more processors are configured to perform any one of the methods of claims 1-26.

31. An apparatus for wireless communication at a server, comprising: Memory; as well as One or more processors coupled to the memory, the one or more processors being configured to: Receive feature vectors associated with one or more features from the client, the one or more features being associated with the client's environment; Receive from the client a latent vector at least in part based on the feature vector, the latent vector including compressed channel state feedback; The observed wireless communication vector is determined at least in part based on the eigenvector and the latent vector; and Wireless communication actions are performed, at least in part, based on determining the observed wireless communication vector.

Images