Machine Learning Architecture for Broadcast and Multicast Communications

Abstract

Techniques and apparatus for a machine learning architecture for broadcast and multicast communications are described. In an embodiment, a network entity determines a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted over a wireless communication system, wherein the communications are directed to a target group of user equipment (UE). The network entity forms a network entity DNN based on the determined configuration of the DNN, and uses the network entity DNN to process the broadcast or multicast communications. In an embodiment, the network entity forms a common DNN to process the broadcast or multicast communications and / or propagate the broadcast or multicast communications to the target group of UEs.

Description

Background Art

[0001] The development of wireless communication systems is generally driven by the need for data throughput. As an example, as more and more devices gain access to wireless communication systems, the need for data increases. Evolving devices also execute data-intensive applications that utilize more data than traditional applications, such as data-intensive streaming video applications, data-intensive social media applications, data-intensive audio services, etc. Therefore, in order to accommodate the increased data usage, evolving wireless communication systems utilize increasingly complex architectures to provide more data throughput relative to older wireless communication systems.

[0002] As an example, fifth generation (5G) standards and technologies use higher frequency bands such as 6 kilohertz (GHz) bandwidth (e.g., 5G millimeter wave (mmW) technology) to transmit data to increase data capacity. However, using these higher frequency ranges to transmit and recover information brings challenges. For example, higher frequency signals are more susceptible to multipath fading, scattering, atmospheric absorption, diffraction, interference, etc. relative to lower frequency signals. When the information is recovered at the receiver, these signal distortions often lead to errors. As another example, the hardware capable of transmitting, receiving, routing and / or otherwise using these higher frequencies can be complex and expensive, which increases the processing cost of wireless networking devices. Summary of the invention

[0003] This document describes techniques and apparatus for a machine learning architecture for broadcast and multicast communications. In an embodiment, a network entity determines a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted over a wireless communication system, wherein the communications are directed to a target group of user equipment (UE). The network entity forms a network entity DNN based on the determined configuration of the DNN, and uses the network entity DNN to process the broadcast or multicast communications. In an embodiment, the network entity forms a common DNN to process the broadcast or multicast communications and / or propagate the broadcast or multicast communications to a target group of UEs.

[0004] The details of one or more embodiments of the machine learning architecture for broadcast and multicast communications are set forth in the accompanying drawings and the following description. Other features and advantages will be apparent from the description and drawings, and from the claims. This summary is provided to introduce the subject matter that is further described in the detailed description and drawings. Therefore, this summary should not be considered to describe essential features, nor should it be considered to limit the scope of the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] The following describes details of one or more aspects of a machine learning architecture for broadcast and multicast communications. The same reference numerals are used in different instances in the specification and drawings to indicate similar elements:

[0006] Figure 1 An example environment is illustrated in which various aspects of a machine learning architecture for broadcast and multicast communications can be implemented.

[0007] Figure 2 An example device diagram of a device capable of implementing various aspects of a machine learning architecture for broadcast and multicast communications is illustrated.

[0008] Figure 3 An example device diagram of a device capable of implementing various aspects of a machine learning architecture for broadcast and multicast communications is illustrated.

[0009] Figure 4 Illustrated are example machine learning modules that can implement various aspects of a machine learning architecture for broadcast and multicast communications.

[0010] Figure 5 An example block diagram of a processing chain utilized by a device to process communications transmitted over a wireless communication system is illustrated.

[0011] Figure 6 An example operating environment for utilizing multiple deep neural networks in a wireless communication system is illustrated.

[0012] Figure 7 An example transaction diagram between various devices for configuring a neural network using a neural network formation configuration is illustrated.

[0013] Figure 8 An example of generating multiple neural network formation configurations is illustrated.

[0014] Fig. 9 An example transaction diagram between various devices for communicating a neural network forming configuration is illustrated.

[0015] Fig.10 An example environment is illustrated in which various aspects of a machine learning architecture for broadcast and multicast communications can be implemented.

[0016] Fig.11 An example environment is illustrated in which various aspects of a machine learning architecture for broadcast and multicast communications can be implemented.

[0017] Fig.12 An example environment is illustrated in which various aspects of a machine learning architecture for broadcast and multicast communications can be implemented.

[0018] Fig.13An example transaction diagram between various devices according to various embodiments of a machine learning architecture for broadcast and multicast communications is illustrated.

[0019] Fig.14 An example transaction diagram between various devices according to various embodiments of a machine learning architecture for broadcast and multicast communications is illustrated.

[0020] Fig.15 An example transaction diagram between various devices according to various embodiments of a machine learning architecture for broadcast and multicast communications is illustrated.

[0021] Fig.16 An example transaction diagram between various devices according to various embodiments of a machine learning architecture for broadcast and multicast communications is illustrated.

[0022] Fig.17 Illustrated is an example method for using a machine learning framework for broadcast and multicast communications.

[0023] Fig.18 Illustrated is an example method for using a machine learning framework for broadcast and multicast communications. DETAILED DESCRIPTION

[0024] In conventional wireless communication systems, the transmitter and receiver processing chains include complex functions. For example, a channel estimation block in the processing chain estimates or predicts how a signal is distorted while propagating through the transmission environment. As another example, a channel equalizer block reverses the signal distortion identified by the channel estimation block. These complex functions typically become more complex when processing higher frequency ranges such as 5G mmW signals at or near the 6 GHz band.

[0025] DNNs provide alternative solutions for complex processing such as those used in wireless communication systems. By training DNNs on transmitter and / or receiver processing chain operations, DNNs can replace conventional functions in a variety of ways, such as by replacing some or all conventional processing blocks for processing broadcast or multicast communication signals, replacing individual processing chain blocks, etc. Dynamic reconfiguration of DNNs—such as by modifying various parameter configurations (e.g., coefficients, layer connections, kernel size)—also provides the ability to adapt to changing operating conditions.

[0026] This document describes aspects of a machine learning architecture for broadcast and multicast communications. In an embodiment, a network entity associated with a wireless communication system determines a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted to a target group of UEs via the wireless communication system. In some embodiments, a network entity such as a core network server or a base station (BS) determines the configuration based on various characteristics associated with the target group of UEs, such as an estimated location or UE capabilities. The network entity forms a network entity DNN based on the determined configuration of the DNN, and uses the network entity DNN to process broadcast or multicast communications to transmit the broadcast or multicast communications via the wireless communication system and direct them to the target group of UEs.

[0027] One or more aspects of a machine learning architecture for broadcast and multicast communications include processing broadcast or multicast communications using a DNN to direct the broadcast or multicast communications to a target group of UEs over a wireless communication system. In an embodiment, a network entity receives feedback from at least one UE in the target group of UEs and determines a modification to the DNN based on the feedback. The network entity then transmits an indication of the modification to the target group of UEs. Alternatively or additionally, the network entity updates the DNN with the modification to form a modified DNN, and uses the modified DNN to process the broadcast or multicast communications to transmit the broadcast or multicast communications over the wireless communication system and direct them to the target group of UEs.

[0028] Using DNNs to process broadcast and multicast communications allows various devices operating in a wireless communication system to correct for changes in current operating conditions, such as changes in the location of a target UE. By monitoring these changes, such as through metrics or feedback from the UE, the DNN can be adjusted to correct or address these changes. Alternatively or additionally, the DNN can be modified based on the capabilities of the target UE. These modifications improve the overall performance of how broadcast and multicast communications are transmitted and / or recovered (e.g., lower bit errors, improved signal quality, improved latency). As another example, a DNN can be trained to process complex inputs corresponding to a complex environment, such as a complex environment including a target group of UEs, where each UE has different processing power and / or estimated location. Thus, the DNN provides a flexible and modifiable solution to complex processing.

[0029] The phrases "transmitted via," "communications exchanged," and "communications associated with" include generating communications to be transmitted via a wireless communication system (e.g., processing pre-transmission communications) and / or processing communications received via a wireless communication system. Thus, "processing communications transmitted via a wireless communication system," "communications exchanged via a wireless communication system," and "communications associated with a wireless communication system" include generating transmissions (e.g., pre-transmission processing), processing received transmissions, or any combination thereof.

[0030] Example Environment

[0031] Figure 1 An example environment 100 is illustrated, which includes a user equipment 110 (UE 110) capable of communicating with a base station 120 (illustrated as base stations 121 and 122) via one or more wireless communication links 130 (wireless links 130), illustrated as wireless links 131 and 132. For simplicity, UE 110 is implemented as a smart phone, but may also be implemented as any suitable computing or electronic device, such as a mobile communication device, a modem, a cellular phone, a gaming device, a navigation device, a media device, a laptop, a desktop computer, a tablet, a smart appliance, a vehicle-based communication system, or an Internet of Things (IoT) device, such as a sensor or actuator. Base station 120 (e.g., an evolved universal terrestrial radio access network Node B, E-UTRAN Node B, evolved Node B, eNodeB, eNB, next generation Node B, gNode B, gNB, ng-eNB, etc.) may be implemented in a macro cell, a micro cell, a small cell, a pico cell, etc., or any combination thereof.

[0032] The base station 120 communicates with the user equipment 110 using wireless links 131 and 132, which may be implemented as any suitable type of wireless links. The wireless links 131 and 132 include control and data communications, such as downlinks of data and control information transmitted from the base station 120 to the user equipment 110, uplinks of other data and control information transmitted from the user equipment 110 to the base station 120, or both. The wireless link 130 may include one or more wireless links (e.g., radio links) or bearers implemented using any suitable communication protocol or standard, or a combination of communication protocols or standards, such as the Third Generation Partnership Project Long Term Evolution (3GPP LTE), the Fifth Generation New Radio (5G NR), etc. Multiple wireless links 130 may be aggregated in carrier aggregation to provide higher data rates for the UE 110. Multiple wireless links 130 from multiple base stations 120 may be configured for coordinated multi-point (CoMP) communication with the UE 110.

[0033] The base stations 120 are collectively a radio access network 140 (e.g., a RAN, an evolved universal terrestrial radio access network, E-UTRAN, a 5G NR RAN, or an NR RAN). The base stations 121 and 122 in the RAN 140 are connected to a core network 150. The base stations 121 and 122 are connected to the core network 150 at 102 and 104, respectively, via an NG2 interface for control plane signaling and an NG3 interface for user plane data communications when connected to a 5G core network, or an S1 interface for control plane signaling and user plane data communications when connected to an evolved packet core (EPC) network. The base stations 121 and 122 can communicate using an Xn application protocol (XnAP) over an Xn interface, or using an X2 application protocol (X2AP) over an X2 interface to exchange user plane and control plane data at 106. The user equipment 110 can be connected to a public network such as the Internet 160 for interacting with a remote service 170 via the core network 150. Remote services 170 represent computing, communications, and storage devices used to provide any of a variety of services, including interactive voice or video communications, file transfer, streaming voice or video, and other technical services implemented in any manner, such as voice calls, video calls, website access, messaging services (e.g., text messaging or multimedia messaging), photo file transfers, enterprise software applications, social media applications, video games, streaming video services, and podcasts.

[0034] Example Device

[0035] Figure 2 An example device diagram 200 of a user equipment 110 and one of the base stations 120 is illustrated. Figure 3 An example device diagram 300 of a core network server 302 is illustrated. For clarity, the user equipment 110, the base station 120 and / or the core network server 302 may include Figure 2 or Figure 3 Additional functions and interfaces omitted in .

[0036] The user equipment 110 includes an antenna 202 for communicating with the base station 120 in the RAN 140, a radio frequency front end 204 (RF front end 204), and a wireless transceiver (e.g., an LTE transceiver 206 and / or a 5G NR transceiver 208). The RF front end 204 of the user equipment 110 can couple or connect the LTE transceiver 206 and the 5G NR transceiver 208 to the antenna 202 to facilitate various types of wireless communications. The antenna 202 of the user equipment 110 may include an array of multiple antennas configured to be similar to each other or different from each other. The antenna 202 and the RF front end 204 can be tuned and / or tunable to one or more frequency bands defined by the 3GPP LTE and 5G NR communication standards and implemented by the LTE transceiver 206 and / or the 5G NR transceiver 208. Additionally, antenna 202, RF front end 204, LTE transceiver 206, and / or 5G NR transceiver 208 may be configured to support beamforming for transmitting and receiving communications with base station 120. By way of example and not limitation, antenna 202 and RF front end 204 may be implemented for operation in sub-gigahertz frequency bands, sub-6 GHZ frequency bands, and / or above 6 GHz frequency bands defined by 3GPP LTE and 5G NR communication standards.

[0037] The user device 110 also includes a processor 210 and a computer readable storage medium 212 (CRM 212). The processor 210 can be a single-core processor or a multi-core processor composed of a variety of materials such as silicon, polysilicon, high-K dielectrics, copper, etc. The computer-readable storage medium described herein does not include propagation signals. The CRM 212 may include any suitable memory or storage device that can be used to store device data 214 of the user device 110, such as a random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory. The device data 214 includes user data, multimedia data, beamforming codebooks, applications, neural network tables, and / or an operating system of the user device 110 that can be executed by the processor 210 to implement user plane communications, control plane signaling, and user interaction with the user of the user device 110.

[0038] In some embodiments, the computer-readable storage medium 212 includes a neural network table 216 that stores various architectures and / or parameter configurations for forming a neural network, such as, by way of example and not limitation, specifying a fully connected layer neural network architecture, a convolutional layer neural network architecture, parameters for a recurrent neural network layer, the number of connected hidden neural network layers, an input layer architecture, an output layer architecture, the number of nodes utilized by the neural network, coefficients utilized by the neural network (e.g., weights and biases), kernel parameters, the number of filters utilized by the neural network, stride / pooling configurations utilized by the neural network, activation functions for each neural network layer, interconnections between neural network layers, neural network layers to be skipped, etc. Thus, the neural network table 216 includes any combination of NN formation configuration elements (e.g., architecture and / or parameter configurations) that can be used to create a NN formation configuration (e.g., a combination of one or more NN formation configuration elements) that defines and / or forms a DNN. In some embodiments, a single index value of the neural network table 216 maps to a single NN formation configuration element (e.g., a 1:1 correspondence). Alternatively or in addition, a single index value of the neural network table 216 maps to a NN formation configuration (e.g., a combination of NN formation configuration elements). In some embodiments, the neural network table includes input characteristics for each NN formation configuration element and / or NN formation configuration, wherein the input characteristics describe properties about the training data used to generate the NN formation configuration element and / or NN formation configuration, as further described. Unless the context dictates otherwise, the terms "NN formation configuration", "neural network formation configuration", "DNN configuration", and "configuration" are used interchangeably in this specification.

[0039] In some embodiments, the CRM 212 may also include a user equipment neural network manager 218 (UE neural network manager 218). Alternatively or in addition, the UE neural network manager 218 may be implemented in whole or in part as hardware logic or circuitry that is integrated or separate from other components of the user equipment 110. The UE neural network manager 218 accesses the neural network table 216, such as by an index value, and forms a DNN using the NN formation configuration elements specified by the NN formation configuration. In an embodiment, the UE neural network manager forms multiple DNNs to process wireless communications (e.g., downlink communications and / or uplink communications exchanged with the base station 120).

[0040] Figure 2The device diagram of the base station 120 shown includes a single network node (e.g., gNode B). The functions of the base station 120 may be distributed across multiple network nodes or devices and may be distributed in any manner suitable for performing the functions described herein. The base station 120 includes an antenna 252, a radio frequency front end 254 (RF front end 254), and one or more wireless transceivers (e.g., one or more LTE transceivers 256 and / or one or more 5G NR transceivers 258) for communicating with the UE 110. The RF front end 254 of the base station 120 can couple or connect the LTE transceiver 256 and the 5G NR transceiver 258 to the antenna 252 to facilitate various types of wireless communications. The antenna 252 of the base station 120 may include an array of multiple antennas that are similarly configured to each other or configured differently from each other. The antenna 252 and the RF front end 254 can be tuned to and / or tunable to one or more frequency bands defined by the 3GPP LTE and 5G NR communication standards and implemented by the LTE transceiver 256 and / or the 5G NR transceiver 258. Additionally, the antenna 252 , the RF front end 254 , the LTE transceiver 256 , and / or the 5G NR transceiver 258 may be configured to support beamforming, such as massive multiple-input multiple-output (Massive-MIMO), for transmitting and receiving communications with the UE 110 .

[0041] The base station 120 also includes a processor 260 and a computer readable storage medium 262 (CRM 262). The processor 260 may be a single core processor or a multi-core processor composed of a variety of materials such as silicon, polysilicon, high-K dielectrics, copper, etc. The CRM 262 may include any suitable memory or storage device that can be used to store device data 264 of the base station 120, such as a random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory. The device data 264 includes network scheduling data, radio resource management data, beamforming codebooks, applications, and / or an operating system of the base station 120 that can be executed by the processor 260 to implement communication with the user equipment 110.

[0042] CRM 262 also includes a base station manager 266. Alternatively or additionally, base station manager 266 can be implemented in whole or in part as hardware logic or circuitry that is integrated or separate from other components of base station 120. In at least some aspects, base station manager 266 configures LTE transceiver 256 and 5G NR transceiver 258 for communication with user equipment 110 and with a core network such as core network 150.

[0043] The CRM 262 also includes a base station neural network manager 268 (BS neural network manager 268). Alternatively or in addition, the BS neural network manager 268 can be implemented in whole or in part as hardware logic or circuitry that is integrated or separate from other components of the base station 120. In at least some aspects, the BS neural network manager 268 selects a NN formation configuration to be utilized by the base station 120 and / or the UE 110 to configure a deep neural network for processing wireless communications, such as by selecting a combination of NN formation configuration elements. In some embodiments, the BS neural network manager receives feedback from the UE 110 and selects the neural network formation configuration based on the feedback. Alternatively or in addition, the BS neural network manager 268 receives a neural network formation configuration indication from a core network 150 element via a core network interface 276 or an inter-base station interface 274 and forwards the neural network formation configuration indication to the UE 110.

[0044] The CRM 262 includes a training module 270 and a neural network table 272. In an embodiment, the base station 120 manages and deploys the NN formation configuration to the UE 110. Alternatively or in addition, the base station 120 maintains the neural network table 272. The training module 270 teaches and / or trains the DNN using known input data. For example, the training module 270 trains the DNN for different purposes, such as by processing communications transmitted by a wireless communication system (e.g., encoding downlink communications, modulating downlink communications, demodulating downlink communications, decoding downlink communications, encoding uplink communications, modulating uplink communications, demodulating uplink communications, decoding uplink communications). This includes training the DNN offline (e.g., when the DNN is not actively involved in processing communications) and / or online (e.g., when the DNN is actively involved in processing communications).

[0045] In an embodiment, the training module 270 extracts learned parameter configurations from the DNN to identify NN formation configuration elements and / or NN formation configurations, and then adds and / or updates the NN formation configuration elements and / or NN formation configurations in the neural network table 272. The extracted parameter configurations include any combination of information that defines the behavior of the neural network, such as node connections, coefficients, active layers, weights, biases, pooling, etc.

[0046] The neural network table 272 stores a plurality of different NN formation configuration elements and / or NN formation configurations generated using the training module 270. In some embodiments, the neural network table includes input characteristics for each NN formation configuration element and / or NN formation configuration, wherein the input characteristics describe properties of the training data used to generate the NN formation configuration element and / or NN formation configuration. For example, by way of example and not limitation, the input characteristics include power information, signal to interference plus noise ratio (SINR) information, channel quality indicator (CQI) information, channel state information (CSI), Doppler feedback, frequency band, block error rate (BLER), quality of service (QoS), hybrid automatic repeat request (HARQ) information (e.g., first transmission error rate, second transmission error rate, maximum retransmission), latency, radio link control (RLC), automatic repeat request (ARQ) metrics, received signal strength (RSS), uplink SINR, timing measurements, error metrics, UE capabilities, base station capabilities (BS capabilities), power modes, Internet Protocol (IP) layer throughput, end-to-end latency, end-to-end packet loss rate, etc. Thus, the input features sometimes include layer 1, layer 2, and / or layer 3 metrics. In some embodiments, a single index value of the neural network table 272 maps to a single NN formation configuration element (e.g., a 1:1 correspondence). Alternatively or in addition, a single index value of the neural network table 272 maps to a NN formation configuration (e.g., a combination of NN formation configuration elements).

[0047] In an embodiment, the base station 120 synchronizes the neural network table 272 with the neural network table 216 so that the NN formation configuration elements and / or input characteristics stored in one neural network table are replicated in the second neural network table. Alternatively or in addition, the base station 120 synchronizes the neural network table 272 with the neural network table 216 so that the NN formation configuration elements and / or input characteristics stored in one neural network table represent complementary functions in the second neural network table (e.g., the NN formation configuration elements in the first neural network table are used for transmitter path processing, and the NN formation configuration elements in the second neural network table are used for receiver path processing).

[0048] The base station 120 also includes an inter-base station interface 274, such as an Xn and / or X2 interface, which the base station manager 266 configures to exchange user plane, control plane and other information between other base stations 120 to manage the communication of the base station 120 with the user equipment 110. The base station 120 includes a core network interface 276, which the base station manager 266 configures to exchange user plane, control plane and other information with core network functions and / or entities.

[0049] exist Figure 3, the core network server 302 may provide all or part of the functions, entities, services, and / or gateways in the core network 150. Each function, entity, service, and / or gateway in the core network 150 may be provided as a service in the service core network 150 distributed on multiple servers or embodied on a dedicated server. For example, the core network server 302 may provide all or part of the services or functions of a user plane function (UPF), an access and mobility management function (AMF), a serving gateway (S-GW), a packet data network gateway (P-GW), a mobility management entity (MME), an evolved packet data gateway (ePDG), etc. The core network server 302 is illustrated as being embodied on a single server including a processor 304 and a computer readable storage medium 306 (CRM 306). The processor 304 may be a single-core processor or a multi-core processor composed of a variety of materials such as silicon, polysilicon, high-K dielectrics, copper, etc. CRM 306 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), a hard drive, or flash memory, that can be used to store device data 308 for core network server 302. Device data 308 includes data that supports core network functions or entities that can be executed by processor 304, and / or an operating system of core network server 302.

[0050] The CRM 306 also includes one or more core network applications 310, which in one embodiment are embodied on the CRM 306 (as shown). The one or more core network applications 310 can implement functions such as UPF, AMF, S-GW, P-GW, MME, ePDG, etc. Alternatively or in addition, the one or more core network applications 310 can be implemented in whole or in part as hardware logic or circuitry integrated or separated from other components of the core network server 302.

[0051] The CRM 306 also includes a core network neural network manager 312 that manages NN formation configurations for processing communications exchanged between the UE 110 and the base station 120. In some embodiments, the core network neural network manager 312 analyzes various parameters, such as current signal channel conditions (e.g., as reported by the base station 120, as reported by other wireless access points, as reported by the UE 110 (via the base station or other wireless access points)), capabilities at the base station 120 (e.g., antenna configuration, cell configuration, multiple-input multiple-output (MIMO) capabilities, radio capabilities, processing capabilities), capabilities of the UE 110 (e.g., antenna configuration, MIMO capabilities, radio capabilities, processing capabilities), etc. For example, the base station 120 obtains the various parameters during communications with the UE and forwards the parameters to the core network neural network manager 312. Based on the parameters, the core network neural network manager selects a NN formation configuration that improves the accuracy of the DNN processing the communications. Improved accuracy indicates improved accuracy, such as lower bit errors, of the output generated by the neural network relative to a neural network configured with another NN formation configuration. The core network neural network manager 312 then transmits the selected NN formation configuration to the base station 120 and / or the UE 110. In an embodiment, the core network neural network manager 312 receives UE and / or BS feedback from the base station 120 and selects an updated NN formation configuration based on the feedback.

[0052] The CRM 306 includes a training module 314 and a neural network table 316. In an embodiment, the core network server 302 manages and deploys NN formation configurations to multiple devices in the wireless communication system, such as UE 110 and base station 120. Alternatively or in addition, the core network server maintains the neural network table 316 external to the CRM 306. The training module 314 teaches and / or trains the DNN using known input data. For example, the training module 314 trains the DNN to process different types of pilot communications transmitted through the wireless communication system. This includes training the DNN offline and / or online. In an embodiment, the training module 314 extracts learned NN formation configurations and / or learned NN formation configuration elements from the DNN and stores the learned NN formation configuration elements in the neural network table 316. Therefore, the NN formation configuration includes any combination of architectural configurations (e.g., node connections, layer connections) and / or parameter configurations (e.g., weights, biases, pooling) that define or affect the behavior of the DNN. In some embodiments, a single index value of the neural network table 316 maps to a single NN formation configuration element (e.g., 1:1 correspondence). Alternatively or in addition, a single index value of the neural network table 316 maps to a NN formation configuration (eg, a combination of NN formation configuration elements).

[0053] In some embodiments, the training module 314 of the core network neural network manager 312 generates NN formation configurations and / or NN formation configuration elements that are complementary to those stored in the neural network table 216 at the UE 110 and / or the neural network table 272 at the base station 121. As an example, the training module 314 generates the neural network table 316 using NN formation configurations and / or NN formation configuration elements that have large changes in architecture and / or parameter configuration relative to medium and / or low changes used to generate the neural network table 272 and / or the neural network table 216. For example, the NN formation configurations and / or NN formation configuration elements generated by the training module 314 correspond to fully connected layers, full kernel sizes, frequent sampling and / or pooling, high weighted accuracy, etc. Therefore, the neural network table 316 sometimes includes a high accuracy neural network at the tradeoff of increased processing complexity and / or time.

[0054] The NN formation configurations and / or NN formation configuration elements generated by the training module 270 sometimes have more fixed architectures and / or parameter configurations (e.g., fixed connection layers, fixed kernel size, etc.) and fewer variations relative to those generated by the training module 314. The training module 270 generates, for example, streamlined NN formation configurations (e.g., faster computation time, less data processing) relative to those generated by the training module 314 to optimize or improve the performance of end-to-end network communications at the base station 121 and / or the UE 110. Alternatively or in addition, the NN formation configurations and / or NN formation configuration elements stored at the neural network table 216 at the UE 110 include more fixed architectures and / or parameter configurations relative to those stored in the neural network table 316 and / or the neural network table 272 that reduce requirements at the UE 110 relative to the base station 121 and / or the core network server 302 (e.g., computation speed, fewer data processing points, fewer computations, less power consumption, etc.). In an embodiment, the fixed (or flexible) architecture and / or parameter configuration at each neural network is varied based on the processing resources of the device for the purpose of forming the corresponding DNN (e.g., processing power, memory constraints, quantization constraints (e.g., 8-bit vs. 16-bit), fixed-point computations vs. floating-point computations, floating-point operations per second (FLOPS), power availability). Thus, a UE or access point with fewer processing resources relative to a core network server or base station receives a NN formation configuration optimized for the available processing resources.

[0055] The neural network table 316 stores a plurality of different NN formation configuration elements generated using the training module 314. In some embodiments, the neural network table includes input characteristics for each NN formation configuration element and / or NN formation configuration, wherein the input characteristics describe properties about the training data used to generate the NN formation configuration. For example, the input characteristics can include power information, SINR information, CQI, CSI, Doppler feedback, RSS, error metric, minimum end-to-end (E2E) delay, expected E2E delay, E2E QoS, E2E throughput, E2E packet loss rate, service cost, etc.

[0056] The CRM 306 also includes an end-to-end machine learning controller 318 (E2E ML controller 318). The E2E ML controller 318 determines an end-to-end machine learning configuration (E2E ML configuration) for processing information exchanged via E2E communications such as QoS flows. In an embodiment, the E2E ML controller analyzes any combination of ML capabilities (e.g., supported ML architectures, number of layers supported, available processing power, memory limitations, available power budgets, fixed-point processing vs. floating-point processing, maximum kernel size capabilities, compute capabilities) of devices participating in the E2E communication. Alternatively or in addition, the E2E ML controller analyzes any combination of QoS requirements, QoS parameters, and / or QoS characteristics to determine an E2E ML configuration that meets the associated requirements, parameters, and / or characteristics. In some embodiments, the E2E ML controller obtains metrics characterizing the current operating environment and analyzes the current operating environment to determine an E2EML configuration. This includes determining an E2EML configuration that includes an architecture configuration combined with a parameter configuration that defines a DNN, or includes determining an E2E ML configuration that includes only a parameter configuration for updating a DNN.

[0057] In determining an E2E ML configuration, the E2E ML controller sometimes determines a partitioned E2E ML configuration that distributes processing functionality associated with the E2E ML configuration across multiple devices. For clarity, Figure 3 The end-to-end machine learning controller 318 is illustrated as being separate from the core network neural network manager 312, but in alternative or additional embodiments, the core network neural network manager 312 includes functionality performed by the end-to-end machine learning controller 318, and vice versa. Figure 3 The core network server 302 is illustrated as implementing the E2E ML controller 318 , although alternative or additional devices can implement the E2E ML controller, such as the base station 120 and / or other network elements.

[0058] The core network server 302 also includes a network slice manager 320. Generally, the network slice manager 190 uses network slices to provide different quality of service flows through the wireless communication network (for example, providing different quality of service flows between at least one UE 110, at least one base station 120, and the core network 150). Sometimes, the network slice manager 320 works in conjunction with the E2E ML controller 318 to partition network resources to provide communication exchanges that meet or exceed the quality of service level. For example, the quality of service level can be specified by one or more quality of service parameters such as latency, throughput (for example, bandwidth or data rate), reliability, or error rate (for example, bit error rate). Other example quality of service parameters include availability, packet loss, or jitter. In addition to the quality of service level, the network slice can also provide a specific level of security through encryption. In some embodiments, the network slice manager 320 associates each network slice with one or more end-to-end machine learning architectures to provide a quality of service level. For clarity, Figure 3 The network slice manager 320 is illustrated as being separate from the core network neural network manager 312 and the E2E ML controller 318, but in alternative or additional embodiments, the core network neural network manager 312 includes functionality performed by the network slice manager 320, or vice versa. Figure 3 A core network server 302 is illustrated that implements the network slice manager 320, but alternative or additional devices can implement the network slice manager 320, such as a base station 120 and / or other network elements.

[0059] The core network server 302 also includes a core network interface 322 for communicating user plane, control plane and other information with other functions or entities in the core network 150, the base station 120 or the UE 110. In some embodiments, the core network server 302 transmits the NN formation configuration to the base station 120 using the core network interface 322. The core network server 302 uses the core network interface 322 to receive feedback from the base station 120 and / or from the UE 110 through the base station 120, alternatively or additionally.

[0060] Having described an example environment and example devices that can be used for neural network formation configuration feedback in wireless communications, consider now a discussion of configurable machine learning modules in accordance with one or more implementations.

[0061] Configurable machine learning modules

[0062] Figure 4 An example machine learning module 400 is illustrated. The machine learning module 400 implements an adaptive set of algorithms that learn and recognize patterns within data. The machine learning module 400 can be implemented using any combination of software, hardware, and / or firmware.

[0063] exist Figure 4 In the embodiment of the present invention, the machine learning module 400 includes a deep neural network 402 (DNN 402) having a group of connected nodes (e.g., neurons and / or perceptrons) organized into three or more layers. In various aspects, the DNN 402 includes at least two hidden layers, but in alternative embodiments, a neural network with a single hidden layer provides the same functionality as described with respect to the DNN 402. The nodes between the layers can be configured in a variety of ways, such as a partially connected configuration, in which a first subset of nodes in a first layer is connected to a second subset of nodes in a second layer; a fully connected configuration, in which each node in a first layer is connected to each node in a second layer, and so on. Neurons process input data to produce continuous output values, such as any real number between 0 and 1. In some cases, the output value indicates how close the input data is to the desired category. Perceptrons perform linear classification on the input data, such as binary classification. Whether neurons or perceptrons, these nodes are capable of generating output information based on adaptive learning using various algorithms. Using DNN, the machine learning module 400 performs various different types of analysis, including single linear regression, multiple linear regression, logistic regression, stepwise regression, binary classification, multi-class classification, multivariate adaptive regression splines, local estimation scatter plot smoothing, etc.

[0064] In some embodiments, the machine learning module 400 learns adaptively based on supervised learning. In supervised learning, the machine learning module 400 receives various types of input data as training data. The machine learning module 400 processes the training data to learn how to map the input to the desired output. As an example, the machine learning module 400 receives a digital sample of a signal as input data and learns how to map the signal sample to binary data reflecting the information embedded in the signal. As another example, the machine learning module 400 receives binary data as input data and learns how to map the binary data to a digital sample of a signal, wherein the binary data is embedded in the signal. During the training process, the machine learning module 400 uses labeled or known data as input to the DNN. The DNN uses nodes to analyze the input and generate corresponding outputs. The machine learning module 400 compares the corresponding output with the real data and adjusts the algorithm implemented by the node to improve the accuracy of the output data. Afterwards, the DNN applies the adjusted algorithm to the unlabeled input data to generate corresponding output data.

[0065] The machine learning module 400 uses statistical analysis and / or adaptive learning to map inputs to outputs. For example, the machine learning module 400 uses characteristics learned from training data to correlate unknown inputs with outputs that are statistically likely to be within a threshold range or threshold. This allows the machine learning module 400 to receive complex inputs and identify corresponding outputs. Some embodiments train the machine learning module 400 on characteristics of communications transmitted on a wireless communication system (e.g., time / frequency interleaving, time / frequency deinterleaving, convolutional encoding, convolutional decoding, power levels, channel equalization, intersymbol interference, orthogonal amplitude modulation / demodulation, frequency division multiplexing / demultiplexing, transmission channel characteristics). This allows the trained machine learning module 400 to receive samples of a signal as input, such as samples of a downlink signal received at a user device, and to recover information from the downlink signal, such as binary data embedded in the downlink signal.

[0066] exist Figure 4 4, the DNN includes an input layer 404, an output layer 406, and one or more hidden layers 408 located between the input layer 404 and the output layer 406. In various aspects, the DNN includes at least two hidden layers, but in alternative embodiments, a neural network with a single hidden layer provides the same functionality as described with respect to the DNN 402. Each layer has an arbitrary number of nodes, wherein the number of nodes between layers can be the same or different. In other words, the input layer 404 can have the same number and / or a different number of nodes as the output layer 406, the output layer 406 can have the same number and / or a different number of nodes as the hidden layer 408, and so on.

[0067] Node 410 corresponds to one of several nodes included in input layer 404, wherein the nodes perform independent calculations. As further described, the node receives input data and processes the input data using an algorithm to generate output data. Sometimes, the algorithm includes weights and / or coefficients that change based on adaptive learning. Therefore, the weights and / or coefficients reflect the information learned by the neural network. In some cases, each node can determine whether to pass the processed input data to the next node. For illustration, after processing the input data, node 410 can determine whether to pass the processed input data to node 412 and / or node 414 of hidden layer 408. Alternatively or in addition, node 410 passes the processed input data to a node based on a layer connection architecture. This process can be repeated throughout multiple layers until the DNN generates an output using the nodes of the output layer 406.

[0068] Neural networks can also employ various architectures that determine which nodes within the neural network are connected, how data is advanced and / or retained in the neural network, what weights and coefficients are used to process input data, how data is processed, and so on. These various factors collectively describe a NN formation configuration. To illustrate, a recurrent neural network, such as a long short-term memory (LSTM) neural network, forms loops between node connections in order to retain information from a previous portion of an input data sequence. The recurrent neural network then uses the retained information for subsequent portions of the input data sequence. As another example, a feedforward neural network passes information to a forward connection without forming a loop to retain the information. Although described in the context of node connections, it is to be understood that a NN formation configuration can include various parameter configurations that affect how the neural network processes input data.

[0069] The NN formation configuration of a neural network can be characterized by various architectures and / or parameter configurations. For illustration, consider an example of a DNN implementing a convolutional neural network. Typically, a convolutional neural network corresponds to a DNN in which layers process data using convolution operations to filter input data. Accordingly, by way of example and not limitation, a convolutional NN formation configuration can be characterized by pooling parameters, kernel parameters, weights, and / or layer parameters.

[0070] Pooling parameters correspond to parameters that specify a pooling layer within a convolutional neural network, which reduce the dimensionality of the input data. For illustration, a pooling layer is capable of combining node outputs at a first layer into node inputs at a second layer. Alternatively or in addition, pooling parameters specify how and where in a data processing layer the neural network pools data. Pooling parameters indicating "max pooling" configure the neural network, for example, to perform pooling by selecting a maximum value from a group of data generated by nodes of the first layer, and use the maximum value as an input to a single node of the second layer. Pooling parameters indicating "average pooling" configure the neural network to generate an average value from a group of data generated by nodes of the first layer, and use the average value as an input to a single node of the second layer.

[0071] Kernel parameters indicate the size of the filter (e.g., width and height) used to process the input data. Alternatively or in addition, kernel parameters specify the type of kernel method used to filter and process the input data. Support vector machines, for example, correspond to kernel methods that use regression analysis to identify and / or classify data. Other types of kernel methods include Gaussian processes, canonical correlation analysis, spectral clustering methods, etc. Therefore, kernel parameters can indicate the type of filter size and / or kernel method applied in a neural network.

[0072] The weight parameters specify the weights and biases within a node used by the algorithm to classify input data. In an embodiment, the weights and biases are learned parameter configurations, such as those generated from training data.

[0073] The layer parameters specify layer connections and / or layer types, such as a fully connected layer type indicating that each node in a first layer (e.g., output layer 406) is connected to each node in a second layer (e.g., hidden layer 408), a partially connected layer type indicating which nodes in the first layer are disconnected from the second layer, an activation layer type indicating which filters and / or layers within a neural network are to be activated, etc. Alternatively or additionally, the layer parameters specify a type of node layer, such as a normalization layer type, a convolutional layer type, a pooling layer type, etc.

[0074] Although described in the context of pooling parameters, kernel parameters, weight parameters, and layer parameters, it is understood that other parameter configurations can be used to form a DNN without departing from the scope of the claimed subject matter. Thus, a NN formation configuration can include any other type of parameter that can be applied to a DNN that affects how the DNN processes input data to generate output data.

[0075] Some embodiments configure the machine learning module 400 based on the current operating environment. To illustrate, consider a machine learning module trained to generate binary data from digital samples of a signal. The transmission environment often modifies the characteristics of the signal propagating through the environment. The transmission environment often changes, which affects how the environment modifies the signal. A first transmission environment, for example, modifies the signal in a first manner, while a second transmission environment modifies the signal in a manner different from the first transmission environment. These differences affect the accuracy of the output results generated by the machine learning module. For example, a neural network configured to process communications transmitted through a first transmission environment may generate errors (e.g., bit errors exceeding a threshold) when processing communications transmitted through a second transmission environment.

[0076] Various embodiments generate and store NN formation configurations and / or NN formation configuration elements (e.g., various architectures and / or parameter configurations) for different transmission environments. For example, the base station 120 and / or the core network server 302 trains the machine learning module 400 using any combination of the BS neural network manager 268, the training module 270, the core network neural network manager 312, and / or the training module 314. The training can occur offline when no active communication exchanges are occurring, or online during active communication exchanges. For example, the base station 120 and / or the core network server 302 can mathematically generate training data, access files storing training data, obtain real-world communication data, etc. The base station 120 and / or the core network server 302 then extracts the various learned NN formation configurations and stores them in a neural network table. Some embodiments store input characteristics with each NN formation configuration, wherein the input characteristics describe various attributes of the transmission environment corresponding to the corresponding NN formation configuration. In an embodiment, the neural network manager selects the NN formation configuration and / or NN formation configuration elements by matching the current transmission environment and / or the current operating environment with the input characteristics.

[0077] Having described a configurable machine learning module, consider now a discussion of deep neural networks in a wireless communication system in accordance with one or more embodiments.

[0078] Deep Neural Networks in Wireless Communication Systems

[0079] Wireless communication systems include various complex components and / or functions, such as reference Figure 1 Example environment 100, Figure 2 An example device diagram 200 and Figure 3 The various devices and modules described in the exemplary device diagram 300 of FIG. In some embodiments, devices participating in a wireless communication system link a series of functions together to enable information to be exchanged over a wireless connection.

[0080] To demonstrate, Figure 5 An example block diagram 500 and an example block diagram 502 are illustrated, each depicting an example processing chain utilized by a device in a wireless communication system. For simplicity, the block diagrams illustrate high-level functionality, and it is to be understood that for clarity, the block diagrams may include Figure 5 Additional features omitted.

[0081] exist Figure 5In the upper half of the block diagram 500, the block diagram 500 includes a transmitter block 504 and a receiver block 506. The transmitter block 504 includes a transmitter processing chain that proceeds from top to bottom. The transmitter processing chain starts with input data, which proceeds to a coding stage, then a modulation stage, followed by a radio frequency (RF) analog transmission (Tx) stage. The coding stage can include any type and number of coding stages used by a device to transmit data over a wireless communication system.

[0082] To illustrate, the encoding stage receives binary data as input and processes the binary data using various encoding algorithms to append information to the binary data, such as frame information. Alternatively or in addition, the encoding stage transforms the binary data, such as by applying forward error correction, which adds redundancy to aid in information recovery at the receiver. As another example, the encoding stage converts the binary data into symbols.

[0083] The example modulation stage receives as input the output generated by the coding stage and embeds the input onto the signal. For example, the modulation stage generates digital samples of the signal embedded with the input from the coding stage. Thus, in the transmitter block 504, the coding stage and the modulation stage represent a high-level transmitter processing chain that often includes low-level complex functions such as convolutional encoding, serial to parallel conversion, cyclic prefix insertion, channel coding, time / frequency interleaving, etc. The RF analog Tx stage receives the output from the modulation stage, generates an analog RF signal based on the modulation stage output, and transmits the analog RF signal to the receiver block 506.

[0084] Receiver block 506 uses a receiver processing chain to perform complementary processing relative to transmitter block 504. The receiver processing chain illustrated in receiver block 506 proceeds from top to bottom and includes an RF analog receive (Rx) stage followed by a demodulation stage and a decoding stage.

[0085] The RF analog RX stage receives the signal transmitted by the transmitter block 504 and generates a signal used by the demodulation stage. As an example, the RF analog Rx stage includes a down-conversion component and / or an analog-to-digital converter (ADC) to generate samples of the received signal. The demodulation stage processes the input from the RF analog RX stage to extract data embedded on the signal (e.g., data embedded by the modulation stage of the transmitter block 504). For example, the demodulation stage recovers symbol and / or binary data.

[0086] The decoding stage receives input from the demodulation stage, such as recovered symbols and / or binary data, and processes the input to recover the transmitted information. For illustration, the decoding stage corrects data errors based on forward error correction applied at the transmitter block, extracts payload data from frames and / or time slots, etc. Thus, the decoding stage generates recovered information.

[0087] As noted, the transmitter and receiver processing chains illustrated by the transmitter block 504 and the receiver block 506 have been simplified for clarity and can include multiple complex modules. Sometimes, these modules are specific to specific functions and / or conditions. For example, consider a receiver processing chain that processes orthogonal frequency division modulation (OFDM) transmissions. In order to recover information from OFDM transmissions, the receiver block 506 typically includes multiple processing blocks, each of which is dedicated to a specific function, such as an equalization block that corrects distortion in the received signal, a channel estimation block that estimates transmission channel properties to identify the effects on the transmission due to scattering, power attenuation, etc. At high frequencies such as 5G mmW signals in the 6 GHz band, these blocks can be computationally and / or monetaryly expensive (e.g., requiring a large amount of processing power, requiring expensive hardware). In addition, implementing blocks that generate outputs with accuracy within a desired threshold often requires more specific and less flexible components. To illustrate, an equalization block for a 5G mmW signal in the 6 GHz band may not be able to perform with the same accuracy in other bands, so different equalization blocks are required for different bands and complexity is added to the corresponding device.

[0088] Some embodiments include DNNs in the transmit and / or receiver processing chains. In block diagram 502, transmitter block 508 includes one or more deep neural networks 510 (DNN 510) in the transmitter processing chain, while receiver block 512 includes one or more deep neural networks 514 (DNN 514) in the receiver processing chain.

[0089] For simplicity, the DNN 510 in the transmitter block 508 corresponds to the encoding and modulation stages of the transmitter block 504. It is to be understood, however, that the DNN 510 is capable of performing any high-level and / or low-level operations present in the transmitter processing chain. For example, a first DNN performs low-level transmitter-side forward error correction, a second DNN performs low-level transmitter-side convolutional encoding, and so on. Alternatively or in addition, the DNN 510 performs high-level processing, such as end-to-end processing corresponding to the encoding and modulation stages of the transmitter block 508.

[0090] In a similar manner, the DNN 514 in the receiver block 512 performs receiver processing chain functions (e.g., demodulation stage, decoding stage). The DNN 514 is capable of performing any high-level and / or low-level operations present within the receiver processing chain, such as low-level receiver-side error correction, low-level receiver-side symbol recovery, high-level end-to-end demodulation and decoding, etc. Therefore, the DNN 514 in the wireless communication system can be configured to replace high-level operations and / or low-level operations in the transmitter and receiver processing chains. Sometimes, as further described, the DNN 514 that performs high-level operations and / or low-level operations can be configured and / or reconfigured based on the current operating environment. Relative to more specific and less flexible components, this DNN reconfigurability, together with DNN coefficient updates, provides greater flexibility and adaptability to the processing chain.

[0091] Some embodiments use multiple DNNs to process communication exchanges through a wireless communication system, where each DNN has a corresponding purpose (e.g., uplink processing, downlink processing, uplink encoding processing, downlink decoding processing, etc.). For demonstration purposes, now consider Figure 6 An example operating environment 600 including UE 110 and base station 121 is illustrated. Figure 6 In an embodiment, UE 110 and base station 121 exchange communications with each other through a wireless communication system by processing the communications using a plurality of DNNs.

[0092] exist Figure 6 , base station neural network manager 268 of base station 121 includes a downlink processing module 602 for processing downlink communications, such as for generating downlink communications for transmission to UE 110. For illustration, as further described, base station neural network manager 268 forms a deep neural network 604 (DNN 604) in downlink processing module 602 using a NN formation configuration. In some examples, DNN 604 corresponds to Figure 5 's DNN 510. In other words, DNN 604 performs some or all of the transmitter processing functions for generating downlink communications.

[0093] Similarly, the UE neural network manager 218 of the UE 110 includes a downlink processing module 606, wherein the downlink processing module 606 includes a deep neural network 608 (DNN 608) for processing (received) downlink communications. In various embodiments, the UE neural network manager 218 uses the NN formation configuration to form the DNN 608. Figure 6 In the example, DNN 608 corresponds to Figure 5514 of UE 110, wherein the deep neural network 606 of UE 110 performs some or all of the receiver processing functions for the (received) downlink communication. Therefore, DNN 604 and DNN 608 perform complementary processing (e.g., encoding / decoding, modulation / demodulation) to each other.

[0094] DNN 604 and / or DNN 608 can include multiple deep neural networks, where each DNN is dedicated to a corresponding channel, a corresponding purpose, etc. As an example, base station 120 uses a first DNN of DNN 604 to process downlink control channel information, uses a second DNN of DNN 604 to process downlink data channel information, etc. As another example, UE 110 uses a first DNN of DNN 608 to process downlink control channel information, uses a second DNN of DNN 608 to process downlink data channel information, etc.

[0095] The base station 120 and / or the UE 110 also use the DNN to process uplink communications. In the environment 600, the UE neural network manager 218 includes an uplink processing module 610, wherein the uplink processing module 610 includes a deep neural network 612 (DNN 612) for generating and / or processing uplink communications (e.g., encoding, modulation). In other words, the uplink processing module 610 processes pre-transmission communications as part of processing uplink communications. The UE neural network manager 218 forms the DNN 612, for example, using the NN formation configuration. Sometimes, the DNN 612 corresponds to Figure 5 Thus, DNN 612 performs some or all of the transmitter processing functions for generating uplink communications for transmission from UE 110 to base station 120.

[0096] Similarly, the uplink processing module 614 of the base station 120 includes a deep neural network 616 (DNN 616) for processing (received) uplink communications, wherein the base station neural network manager 268 forms the DNN 616 using the NN formation configuration, as further described. In the example, the DNN 616 of the base station 120 corresponds to Figure 5 514 and performs some or all of the receiver processing functions for (received) uplink communications, such as uplink communications received from UE 110. Sometimes, DNN 612 and DNN 616 perform complementary functions to each other. Alternatively or in addition, uplink processing module 610 and / or uplink processing module 614 include multiple DNNs, each of which has a dedicated purpose (e.g., processing a corresponding channel, performing a corresponding uplink function, etc.). Figure 6DNNs 604, 608, 612, and 616 are illustrated as residing within respective neural network managers to indicate that the neural network managers form the DNNs, and it is understood that the DNNs can be formed external to the neural network managers (e.g., UE neural network manager 218 and base station neural network manager 268) within different components, processing chains, modules, etc.

[0097] Having described deep neural networks in wireless communication systems, consider now a discussion of signaling and control transactions on a wireless communication system according to one or more embodiments, which are used to configure a deep neural network for downlink and uplink communications.

[0098] Configuring signaling and control transactions for deep neural networks

[0099] Figure 7 An example signaling and control transaction diagram 700 between a base station and a user equipment according to one or more aspects of a neural network formation configuration in wireless communications is illustrated. Alternative or additional embodiments include transactions involving a core network server. For example, in some embodiments, the core network server 302 performs various signaling and control actions performed by the base station 120, such as Figure 7 Signaling and control transactions can be handled by Figure 1 The base station 120 and the UE 110 use Figure 1-6 For clarity, Figure 7 The core network server 302 is omitted, but alternative embodiments include a core network server as further described.

[0100] As shown, at 705, UE 110 optionally indicates UE capabilities (e.g., capabilities supported by the UE) to a network entity such as base station 120. In some embodiments, the UE capabilities include ML-related capabilities, such as maximum kernel size capability, memory limit, computational capability, supported ML architectures, number of layers supported, available processing power, memory limit, available power budget, and fixed-point processing versus floating-point processing. Sometimes, the base station forwards the UE capabilities to a core network server (e.g., core network server 302).

[0101] At 710, the base station 120 determines a neural network formation configuration. In determining the neural network formation configuration, the base station analyzes any combination of information, such as the channel type (e.g., downlink, uplink, data, control, etc.) processed by the deep neural network, transmission medium properties (e.g., power measurements, signal to interference plus noise ratio (SINR) measurements, channel quality indicator (CQI) measurements), coding schemes, UE capabilities, BS capabilities, etc. In some embodiments, the base station 120 determines the neural network formation configuration based on the UE capabilities indicated at 705. Alternatively or in addition, the base station 120 obtains the UE capabilities from a networked storage device such as a server. In some embodiments, the core network server 302 determines the neural network formation configuration in a manner similar to that described with respect to the base station, and transmits the determined neural network formation configuration to the base station.

[0102] Base station 120 receives, for example, from UE 110 (not shown) a message indicating one or more capabilities of the UE, such as, for example and not limitation, connectivity information, dual connectivity information, carrier aggregation capability, downlink physical parameter values, uplink physical parameter values, supported downlink / uplink categories, inter-frequency handover, and ML capabilities (e.g., maximum kernel size capability, memory limit, computational capability, supported ML architectures, number of layers supported, available processing power, memory limit, available power budget, fixed point processing vs. floating point processing). Base station 120 (and / or core network server 302) identifies from the message the UE capabilities that affect how the UE processes communications and / or how the base station processes communications from the UE, and selects a neural network formation configuration that has an improved output accuracy relative to other neural network formation configurations.

[0103] In some embodiments, the base station 120 (and / or the core network server 302) selects a neural network formation configuration from a plurality of neural network formation configurations. Alternatively or in addition, the base station 120 (and / or the core network server 302) selects the neural network formation configuration by selecting a subset of neural network architecture formation elements in a neural network table. Sometimes, the base station 120 (and / or the core network server 302) analyzes a plurality of neural network formation configurations and / or a plurality of neural network formation configuration elements included in the neural network table and determines a neural network formation configuration by selecting and / or creating a neural network formation configuration that is aligned with current channel conditions, such as by matching channel type, transmission medium properties, etc. with input characteristics, as further described. Alternatively or in addition, the base station 120 (and / or the core network server 302) selects the neural network formation configuration based on network parameters, such as scheduling parameters (e.g., scheduled multi-user multiple input multiple output (MU-MIMO) for downlink communications, scheduled MU-MIMO for uplink communications).

[0104] At 715, the base station 120 transmits the neural network formation configuration to the UE 110. Alternatively or in addition, the core network server 302 transmits the neural network formation configuration to the base station 120, and the base station 120 forwards the neural network formation configuration to the UE 110. In some embodiments, the base station transmits a message specifying the neural network formation configuration, such as by transmitting a message including an index value that maps to a value such as Figure 2 216 of the neural network table. Alternatively or in addition, the base station transmits a message including a neural network parameter configuration (e.g., weight values, coefficient values, number of filters). In some cases, the base station 120 specifies a purpose and / or a processing assignment in the message, wherein the processing assignment indicates which channels and / or which position in the processing chain the configured neural network is applied to, such as downlink control channel processing, uplink data channel processing, downlink decoding processing, uplink encoding processing, etc. Thus, the base station can transmit the processing assignment together with the neural network formation configuration.

[0105] In some embodiments, the base station 120 transmits multiple neural network formation configurations to the UE 110. For example, the base station transmits a first message directing the UE to use a first neural network formation configuration for uplink encoding, and a second message directing the UE to use a second neural network formation configuration for downlink decoding. In some cases, the base station 120 transmits multiple neural network formation configurations and corresponding processing assignments in a single message. As another example, the base station transmits multiple neural network formation configurations using different radio access technologies (RATs). The base station can, for example, transmit a first neural network formation configuration for downlink communication processing to the UE 110 using a first RAT and / or carrier, and can transmit a second neural network formation configuration for uplink communication processing to the UE 110 using a second RAT and / or carrier.

[0106] At 720, UE 110 forms a first neural network based on the neural network formation configuration. For example, UE 110 uses the index value transmitted by the base station to access a neural network table to obtain the neural network formation configuration and / or neural network formation configuration elements. Alternatively or in addition, UE 110 extracts the neural network architecture and / or parameter configuration from the message. UE 110 then forms a neural network using the neural network formation configuration, the extracted architecture and / or parameter configuration, etc. In some embodiments, the UE processes all communications using the first neural network, while in other embodiments, the UE processes select communications using the first neural network based on the processing assignment.

[0107] At 725, the base station 120 transmits information based on the neural network formation configuration. Figure 6, base station 120 processes downlink communications using a second neural network configured with functions complementary to the first neural network. In other words, the second neural network uses a second neural network formation configuration complementary to the neural network formation configuration. Then, at 730, UE 110 recovers the information using the first neural network.

[0108] Having described signaling and control transactions that can be used to configure a neural network for handling communications, consider now a discussion of generating and transmitting a neural network forming configuration in accordance with one or more embodiments.

[0109] Generate and transmit neural network formation configuration

[0110] In supervised learning, a machine learning module processes labeled training data to generate an output. The machine learning module receives feedback about the accuracy of the generated output and modifies processing parameters to improve the accuracy of the output. Figure 8 An example 800 is illustrated that describes aspects of generating multiple NN formation configurations. Sometimes, various aspects of example 800 are described by Figure 2 and Figure 3 This may be implemented by any combination of the training module 270, the base station neural network manager 268, the core network neural network manager 312 and / or the training module 314.

[0111] Figure 8 The upper part includes Figure 4 The machine learning module 400 of FIG. 400 is a machine learning module. In an embodiment, the neural network manager determines to generate different NN formation configurations. To illustrate, consider a case where the base station neural network manager 268 determines to generate a NN formation configuration by selecting a combination of NN formation configuration elements from a neural network table, wherein the NN formation configuration corresponds to a UE decoding and / or demodulating downlink communications. In other words, the NN formation configuration (through a combination of NN formation configuration elements) forms a DNN that processes downlink communications received by the UE. However, transmission channel conditions often change, which in turn affects the characteristics of downlink communications. For example, a first transmission channel distorts downlink communications by introducing a frequency offset, a second transmission channel distorts downlink communications by introducing a Doppler effect, a third transmission channel distorts downlink communications by introducing a multipath channel effect, and so on. In order to accurately process downlink communications (e.g., reduce bit errors), various embodiments select multiple NN formation configurations, wherein each NN formation configuration (and associated combination of NN formation configuration elements) corresponds to a corresponding input condition, such as a first transmission channel, a second transmission channel, and the like.

[0112] Training data 802 represents example inputs to machine learning module 400. Figure 8, training data 802 represents data corresponding to downlink communications. Training data 802 can include, for example, digital samples of downlink communication signals, recovered symbols, recovered frame data, etc. In some embodiments, the training module generates training data mathematically or accesses a file storing training data. At other times, the training module obtains real-world communication data. Therefore, the training module can use mathematically generated data, static data, and / or real-world data to train the machine learning module 400. Some embodiments generate input characteristics 804 that describe various qualities of the training data, such as transmission channel metrics, UE capabilities, UE speed, etc.

[0113] The machine learning module 400 analyzes the training data and generates an output 806, which is represented here as binary data. Some embodiments use the same training data set and / or additional training data with the same input characteristics 804 to iteratively train the machine learning module 400 to improve the accuracy of the machine learning module 400. During training, the machine learning module 400 modifies some or all of the architecture and / or parameter configurations of the neural network included in the machine learning module 400, such as node connections, coefficients, kernel size, etc. At certain moments in training, such as when the training module determines that the accuracy meets or exceeds the expected threshold, the training process meets or exceeds the number of iterations, etc., the training module determines to extract the architecture and / or parameter configuration 808 of the neural network (e.g., pooling parameters, kernel parameters, layer parameters, weights). Then, the training module extracts the architecture and / or parameter configuration from the machine learning module 400 to be used as a NN formation configuration and / or NN formation configuration element. The architecture and / or parameter configuration can include any combination of fixed architecture and / or parameter configuration and / or variable architecture and / or parameter configuration.

[0114] Figure 8 The lower portion of includes a neural network table 810 representing a collection of NN formation configuration elements, such as Figure 2 and Figure 3 804. The neural network table 810 stores various combinations of architecture configurations, parameter configurations 808, and input characteristics 804, although alternative embodiments exclude input characteristics 804 from the table. As the machine learning module 400 learns additional information, various embodiments update and / or maintain the NN formation configuration elements and / or input characteristics 804. For example, at index 812, the neural network manager and / or training module updates the neural network table 810 to include the architecture and / or parameter configurations 808 generated by the machine learning module 400 when analyzing the training data 802.

[0115] The neural network manager and / or training module alternatively or additionally adds the input characteristics 804 to the neural network table 810 and links the input characteristics 804 to the architecture and / or parameter configuration 808. This allows the input characteristics 804 to be obtained simultaneously with the architecture and / or parameter configuration 808, such as by using an index value (e.g., reference to a NN formation configuration, reference to a NN formation configuration element) that references the neural network table 810. In some embodiments, the neural network manager selects a NN formation configuration by matching the input characteristics to the current operating environment, such as by matching the input characteristics to current channel conditions, UE capabilities, UE characteristics (e.g., speed, location, etc.), etc.

[0116] Having described generating and transmitting a neural network formation configuration, consider now a discussion of signaling and control transactions over a wireless communication system that may be used to transmit a neural network formation configuration in accordance with one or more embodiments.

[0117] Signaling and control transactions used to convey the configuration of neural network formation

[0118] Fig. 9 An example signaling and control transaction diagram 900 between a base station and a user equipment according to one or more aspects of a transmit neural network formation configuration, such as a transmit NN formation configuration, is illustrated. In an embodiment, the signaling and control transaction may be performed by Figure 1 Any one of the base station 120 and the UE 110 uses Figure 1-8 In alternative or additional embodiments, signaling and control transactions may include Figure 3 The interaction of the core network server 302 ( Fig. 9 In an embodiment, part or all of the signaling and control transactions described in reference signaling and control transaction diagram 900 correspond to those in reference Figure 7 Describes the signaling and control transactions.

[0119] As shown, at 905, the base station 120 maintains a neural network table. Alternatively or additionally, the core network server 302 maintains the neural network table. For example, a neural network manager (base station neural network manager 268, core network neural network manager 312) and / or a training module (training module 270, training module 314) generates and / or maintains a neural network table (e.g., neural network table 272, neural network table 316) using any combination of mathematically generated training data, data extracted from real-world communications, files, etc. In various embodiments, the base station 120 (and / or the core network server 302) maintains a plurality of neural network tables, wherein each neural network table includes a plurality of neural network formation configurations and / or neural network formation configuration elements for a specified purpose, such as a first neural network table designated for data channel communications, a second neural network table designated for control channel communications, etc.

[0120] At 910, the base station 120 transmits a neural network table to the UE 110. In some embodiments, the base station 120 first receives the neural network table from the core network server 302. As an example, the base station transmits the neural network table using a layer 3 message (e.g., a radio resource control (RRC) message). When transmitting the neural network table, the base station transmits any combination of an architecture and / or parameter configuration that can be used to form a deep neural network, examples of which are provided in the present disclosure. Alternatively or additionally, the base station transmits an indication along with the neural network table, the indication specifying a processing assignment for the neural network table. Thus, the base station transmits multiple neural network tables to the UE, specifying a corresponding processing assignment for each neural network table. In some embodiments, the base station 120 broadcasts the neural network table to a group of UEs. At other times, the base station 120 transmits a UE-specific neural network table to the UE 110.

[0121] At 915, base station 120 identifies a neural network formation configuration for use in processing communications. Alternatively or additionally, core network server 302 identifies the neural network formation configuration and transmits the neural network formation configuration to base station 120. For example, the base station selects a combination of neural network formation architecture elements (such as in Figure 7 ), determining a neural network formation configuration to use in processing communications by analyzing any combination of information, such as the type of channel processed by the deep neural network (e.g., downlink, uplink, data, control, etc.), transmission medium properties (e.g., power measurements, signal to interference plus noise ratio (SINR) measurements, channel quality indicator (CQI) measurements), coding schemes, UE capabilities, BS capabilities, etc. In some embodiments, base station 120 identifies a default neural network formation configuration to use as the neural network formation configuration. Thus, identifying the neural network formation configuration may include identifying a default neural network formation configuration and / or an update to the neural network formation configuration.

[0122] When identifying a neural network formation configuration, the base station 120 (and / or the core network server 302) finds a neural network formation configuration in the neural network table that corresponds to the determined neural network formation configuration. In other words, the base station 120 identifies the neural network formation configuration that is aligned with the determined neural network formation configuration, such as by correlating and / or matching the input characteristics. Figure 2 The base station may further include a neural network formation configuration and / or a neural network formation configuration element in the neural network table 272 and / or the neural network table 216. When identifying the neural network formation configuration and / or the neural network formation configuration element in the neural network table, the base station may identify the index value of the neural network formation configuration and / or the neural network formation configuration element.

[0123] At 920, base station 120 transmits an instruction directing UE 110 to form a deep neural network using a neural network formation configuration from the neural network table. Figure 7 As described at 715 of , the base station 120 transmits the index value to the UE 110 and instructs the UE 110 to form a deep neural network using the neural network formation configuration indicated by the index value. The base station may transmit the indication to the UE in any appropriate manner. As an example, the base station uses a layer 2 message (e.g., a radio link control (RLC) message, a medium access control (MAC) control element) to transmit an index value corresponding to the neural network formation configuration. In some embodiments, the base station compares the current operating environment (e.g., one or more of channel conditions, UE capabilities, BS capabilities, metrics) with the input characteristics stored in the neural network table, and identifies the stored input characteristics that align with the current operating environment. The base station then obtains the index value of the stored input characteristic, which in turn provides an index value of the neural network formation configuration and / or a neural network formation configuration element. The base station 120 then transmits the index value as an indication. Sometimes, the base station 120 includes a processing assignment indicating the location of the application deep neural network in the processing chain. In some embodiments, the base station uses a downlink control channel to transmit the index value and / or the processing assignment.

[0124] Sometimes, the base station transmits to the UE a rule specifying an operating parameter related to applying a neural network formation configuration. In one example, the rule includes a time instance indicating when to process communications with a deep neural network formed using the neural network formation configuration. Alternatively or additionally, the rule specifies a time threshold that directs the UE to use a default neural network formation configuration instead of a specified neural network formation configuration when the data channel and the control channel are within the time threshold. Additionally or alternatively, when the data channel and the control channel are within the time threshold, the rule may direct the user equipment to use the same neural network formation configuration for data channel communications and control channel communications. To illustrate, consider an example in which the UE uses a first deep neural network (formed using a first neural network formation configuration) to process data channel communications and a second deep neural network (formed using a second neural network formation configuration) to process control channel communications. If the data channel communications and the control channel communications fall within the time threshold specified by the time instance, the UE uses the default deep neural network (formed using the default neural network formation configuration) and / or the same deep neural network to process both channels because there may not be enough time to switch between the first deep neural network and the second deep neural network.

[0125] Sometimes, the base station transmits a default neural network formation configuration using a downlink control channel to specify a default neural network formation configuration to the UE, wherein the default neural network formation configuration forms a deep neural network that processes various input data. In some embodiments, the default neural network formation configuration forms a deep neural network that processes various input data with an accuracy within a threshold range. The default neural network formation configuration may include a general neural network formation configuration.

[0126] To illustrate, some embodiments generate or select neural network formation configurations for specific operating conditions, such as a first neural network formation configuration specific to UE downlink control channel processing (e.g., demodulation and / or decision making) having a current operating environment “X”, a second neural network formation configuration specific to UE downlink control channel processing having a current operating environment “Y”, etc. For example, the first neural network formation configuration may be associated with a current operating environment in which a detected interference level is high, the second neural network formation configuration may be associated with a current operating environment in which a detected interference level is low, the third neural network formation configuration may be associated with a current operating environment in which a connected UE appears stationary, the fourth neural network formation configuration may be associated with a current operating environment in which a connected UE appears to be moving and has a specific speed, and so on.

[0127] Forming a deep neural network using a neural network formation configuration for a particular operating condition improves (relative to forming a deep neural network using other neural network formation configurations) the accuracy of outputs generated by the deep neural network when processing input data corresponding to the particular operating condition. However, this introduces a tradeoff in that the deep neural network formed using the neural network formation configuration for the particular operating condition generates outputs with lower accuracy when processing inputs associated with other operating conditions. In contrast, the default neural network formation configuration corresponds to a neural network formation configuration that handles a wider variety of inputs, such as a variety of inputs spanning more operating conditions. In other words, a deep neural network configured using the default neural network formation configuration handles a greater variety of communications relative to the neural network formation configuration for the particular operating condition.

[0128] At 925, UE 110 forms a deep neural network using the neural network formation configuration. As an example, the UE extracts the index value transmitted by base station 120 and obtains the neural network formation configuration and / or the neural network formation configuration element by accessing the neural network table using the index value. Alternatively or additionally, UE 110 extracts the processing assignment and forms the deep neural network in the processing chain as specified by the processing assignment.

[0129] At 930, base station 120 transmits a communication, such as a downlink data channel communication, to UE 110. At 935, UE 110 processes the communication using the deep neural network. For example, UE 110 processes the downlink data channel communication using the deep neural network to recover the data. As another example, processing the communication includes processing a reply to the communication, wherein UE 110 processes an uplink communication in reply to the downlink communication using the deep neural network.

[0130] Having described signaling and control transactions that may be used to communicate a neural network formation configuration, consider now a discussion of E2E ML for wireless networks in accordance with one or more embodiments.

[0131] E2E ML for Wireless Networks

[0132] Aspects of end-to-end communication (E2E communication) involve two endpoints exchanging information over a communication path, such as over a wireless network. Sometimes, E2E communication performs a one-way exchange of information, where a first endpoint sends information and a second endpoint receives information. At other times, E2E communication performs a two-way exchange of information, where both endpoints send and receive information. Endpoints of E2E communication can include any entity that can consume and / or generate information, such as a computing device, application, service, etc. To illustrate, consider an example in which an application executing at a UE exchanges information with a remote service over a wireless network. For this example, the E2E communication corresponds to a communication path between the application and the remote service, where the application and the remote service act as endpoints.

[0133] Although E2E communication involves endpoints that exchange information, E2E communication may alternatively or additionally include intermediate entities (e.g., devices, applications, services) that participate in the information exchange. To illustrate, consider again the example of E2E communication established over a wireless network, where an application at a UE serves as a first endpoint and a remote service serves as a second endpoint. When establishing E2E communication between endpoints, the wireless network utilizes any combination of UEs, base stations, core network servers, remote networks, remote services, etc., such as reference Figure 1 As described in the environment 100 of FIG. 1 , intermediate entities such as the base station 120 and the core network 150 are involved in establishing E2E communications and / or in participating in E2E communications to enable information exchange between endpoints.

[0134] Various factors affect the operational efficiency of E2E communications and how network elements process information exchanged via E2E communications. For example, with reference to E2E communications established using a wireless network, the current operating environment (e.g., current channel conditions, UE location, UE mobility, UE capabilities) affects how accurately the receiving endpoint recovers the information (e.g., bit error rate, packet loss). As an example, E2E communications implemented using 5G mmW technology become susceptible to more signal distortion relative to lower frequency sub-6 GHz signals, as further described.

[0135] As another example, various embodiments partition wireless network resources differently based on an end-to-end analysis of E2E communications, where wireless network resources include, by way of example and not limitation, any combination of physical hardware, physical spectrum, logical channels, network functions, services provided, quality of service, latency, and the like. Wireless network resource partitioning allows the wireless network to dynamically allocate wireless network resources based on expected usage to improve the efficiency of how wireless network resources are used (e.g., reducing the occurrence of unused and / or wasted resources). To illustrate, consider various devices connected to a wireless network, where the devices have different performance requirements relative to each other (e.g., a first device has a secure data transmission requirement, a second device has a high priority / low latency data transmission requirement, and a third device has a high data rate requirement). For at least some devices, a fixed and / or static distribution of wireless network resources (e.g., a fixed configuration of wireless network resources for implementing E2E communications) can result in unused resources and / or failure to meet performance requirements for some services. Therefore, partitioning wireless network resources can improve the overall efficiency of how wireless network resources are utilized. However, the partitioning results in changes in how one E2E endpoint pair exchanges information relative to a second E2E endpoint pair.

[0136] For further demonstration, consider a Quality of Service flow (QoS flow) corresponding to information exchanged in a wireless network. In various embodiments, the E2E communication includes and / or corresponds to a QoS flow. Some wireless networks configure QoS flows with operational rules, priority levels, classifications, etc. that affect how information is exchanged through QoS flows. For example, a QoS profile indicates to the wireless network QoS parameters and / or QoS characteristics for a particular QoS flow, such as a Guaranteed Flow Bit Rate (GFBR) parameter for indicating an uplink and / or downlink guaranteed bit rate for the QoS flow, a Maximum Flow Bit Rate (MFBR) parameter for indicating a maximum uplink and / or downlink bit rate for the QoS flow, an Allocation and Retention Priority (ARP) parameter indicating a priority level, preemption capability, and / or preemption vulnerability for the QoS flow, a Reflected QoS Attribute (RQA) indicating that the type of traffic carried on the QoS flow is subject to reflected QoS (e.g., implicit update), a Notification Control parameter indicating whether notification is requested when a guaranteed flow bit rate cannot be guaranteed, or an expected aggregation of collective non-guaranteed bit rate (Non-GBR) flows associated with a particular UE for QoS flow recovery. Aggregate bit rate parameter for bit rate, default parameters for 5QI and ARP priority level, maximum packet loss rate (MPLR) for uplink and / or downlink indicating the maximum rate of lost packets for QoS flow, resource type characteristic indicating the type of resource that can be used by QoS flow (e.g., GBR resource type, delay critical GBR resource type, non-GBR resource type), scheduling priority level characteristic to distinguish multiple QoS flows of the same UE, packet delay budget characteristic providing an upper limit on how long packets can be delayed, packet error rate characteristic indicating an upper limit on the rate at which PDUs are not successfully received, average window characteristic indicating the data window over which GFBR and / or MFBR are calculated, maximum data burst characteristic indicating the maximum amount of data that needs to be served over a predefined time period, etc. In some embodiments, the parameters and / or characteristics of the configuration of a specified QoS flow can be preconfigured (e.g., by default) and / or dynamically transmitted, such as through a QoS profile. These changes affect how the wireless network partitions various wireless network resources to support QoS flow configurations.

[0137] For example, a UE can include three applications, each of which has different performance requirements (e.g., resource type, priority level, packet delay budget, packet error rate, maximum data burst, average window, security level). These different performance requirements cause the wireless network to partition the wireless network resources assigned to the corresponding QoS flows assigned to each application that is different from each other.

[0138] For demonstration, consider a scenario where the UE includes a game application, an augmented reality application, and a social media application. In some instances, the game application interacts with a remote service (via a data network) to connect to another game application to exchange audio, video, commands, views, etc. in real time, so that the game application has high data volume and low latency performance requirements. The augmented reality application also interacts with the remote service via a data network to transmit location information and then receives image data overlaid on the camera image generated at the UE. Relative to the game application, the augmented reality application utilizes less data, but has some time sensitivity to maintain synchronization between the current position and the corresponding image overlay. Finally, the social media application interacts with the remote service via a data network to receive feed information, wherein the feed information has less data volume and time criticality relative to the data consumed by the augmented reality application and / or the game application.

[0139] Based on these performance requirements, the wireless network establishes QoS flows between the application and the data network, wherein the wireless network constructs each QoS flow based on QoS requirements, QoS parameters, and / or QoS characteristics (e.g., resource type, priority level, packet delay budget, packet error rate, maximum data burst, average window, security level) indicating high data volume performance requirements and time-sensitive performance requirements. In an embodiment, the QoS requirements, QoS parameters, and / or QoS characteristics included in the QoS profile correspond to the performance requirements of the QoS flow. As an example, the wireless network processes a QoS profile associated with a first QoS flow, the QoS profile configuring any combination of GFBR parameters, maximum data burst characteristics, ARP parameters, etc. Then, the wireless network constructs the QoS flow by partitioning wireless network resources based on the QoS parameters and / or characteristics.

[0140] While the configurability of QoS flows provides wireless networks with the flexibility to dynamically modify how wireless network resources are allocated, the configurability adds complexity in how the wireless network processes information exchanged between endpoints. Some embodiments train DNNs to perform some or all of the complex processing associated with exchanging information using E2E communications with various configurations. By training the DNNs on different processing chain operations and / or wireless network resource partitions, the DNNs are able to replace traditional complex functions, as further described. The use of DNNs in E2E communications also allows network entities to adapt the DNNs to changing operating conditions, such as by modifying various parameter configurations (e.g., coefficients, layer connections, kernel size).

[0141] One or more embodiments determine an E2E ML configuration for processing information exchanged via E2E communication. In some cases, an end-to-end machine learning controller (E2E ML controller) obtains capabilities of devices associated with end-to-end communication in a wireless network, such as machine learning (ML) capabilities of devices participating in the E2E communication, and determines the E2E ML configuration based on the ML capabilities of the devices (e.g., supported ML architectures, number of layers supported, available processing power, memory limitations, available power budgets, fixed-point processing vs. floating-point processing, maximum kernel size capabilities, compute capabilities). Alternatively or in addition, the E2E ML controller identifies a current operating environment and determines the E2E ML configuration based on the current operating environment. Some embodiments of the E2E ML controller communicate with a network slice manager to determine an E2E ML configuration corresponding to a network slice (e.g., a partitioning of wireless network resources). In determining the E2E ML configuration, some embodiments of the E2E ML controller partition the E2E ML configuration based on the devices participating in the E2E communication, and transmit the corresponding partitioning of the E2E ML configuration to each corresponding device.

[0142] To demonstrate, consider Fig.10 , which illustrates an example environment 1000 in which an example E2E ML configuration for E2E communication includes a DNN operating at multiple devices. In an embodiment, the example environment 1000 illustrates various aspects of a machine learning architecture for broadcast and multicast communications. The environment 1000 includes Figure 1 UE 110, base station 120 and remote service 170 and Figure 3 1004. In an embodiment, the UE 110 and the remote service 170 exchange information with each other using E2E communications 1002 and E2E communications 1004. For clarity, the E2E communications 1002 and 1004 are illustrated as separate and unidirectional E2E communications, such that the E2E exchange of information through each communication corresponds to one direction (e.g., E2E communication 1002 includes uplink transmissions and E2E communication 1004 includes downlink transmissions), but it is understood that in alternative or additional embodiments, the E2E communications 1002 and E2E communications 1004 correspond to a single bidirectional E2E communication including both, which is indicated in the environment 1000 using a dashed line 1006. In an embodiment, the E2E communications 1002 and E2E communications 1004 (in combination) correspond to a single QoS flow.

[0143] The environment 1000 also includes an E2E ML controller 318 implemented by the core network server 302, wherein the E2E ML controller 318 determines an E2E ML configuration for the E2E communication 1002 and / or the E2E communication 1004. In some embodiments, such as when each E2E communication corresponds to a unidirectional information exchange, the E2E ML controller determines a first E2E ML configuration for the E2E communication 1002 and a second E2E ML configuration for the E2E communication 1004. In other embodiments, the E2E ML controller determines an E2E ML configuration for a bidirectional E2E communication that includes both the E2E communications 1002 and 1004. For example, in response to UE 110 requesting a connection with remote server 170, such as through a call by an application, the E2E ML controller determines an E2E ML configuration for the corresponding connection based on any combination of ML capabilities of UE 110 (e.g., supported ML architectures, number of layers supported, processing power available for ML processing, memory constraints applied to ML processing, power budget available for ML processing, fixed-point processing versus floating-point processing), performance requirements associated with the requested connection (e.g., resource type, priority level, packet delay budget, packet error rate, maximum data burst size, average window, security level), available wireless network resources, ML capabilities of intermediate devices (e.g., base station 120, core network server 302), current operating environment (e.g., channel conditions, UE location), etc. As an example, referring to Fig. 9 as well as Fig.10 , the E2E ML controller 318 receives base station metrics and / or UE metrics describing the current operating environment through the core network server 302. As another example, the E2E ML controller receives base station ML capabilities and / or UE capabilities through the core network server 302. Alternatively or additionally, the E2E ML controller communicates with the network slice manager 320 to identify network slices that partition wireless network resources in a manner that supports QoS requirements.

[0144] In one or more embodiments, the E2E ML controller 318 analyzes the neural network table to determine the E2E ML configuration based on any combination of device capabilities, wireless network resource segmentation, operating parameters, current operating environment, ML capabilities, etc. Although described as being implemented by the core network server 302, in alternative or additional embodiments, the E2E ML controller 318 may be implemented by another network entity such as the base station 120.

[0145] For illustration and reference Figure 8, the training module 270 and / or the training module 314 trains the machine learning module 400 with variations of the training data 802 that reflect different combinations of capabilities, wireless network resource partitions, operating parameters, current operating environments, etc. The training module extracts the architecture and / or parameter configuration (e.g., the architecture and / or parameter configuration 808) and stores it in the neural network table, so that at a later point in time, the E2E ML controller 318 accesses the neural network table to obtain and / or identify the neural network formation configuration corresponding to the determined E2E ML configuration. The E2E ML controller 318 then transmits the neural network formation configuration to various devices and directs the respective devices to form respective DNNs, as further described.

[0146] When determining the E2E ML configuration, the E2E ML controller 318 sometimes partitions the E2E ML configuration based on the devices participating in the corresponding E2E communication. For example, the E2E ML controller 318 determines a first partition of the E2E ML configuration corresponding to the processing information at the UE 110, a second partition of the E2E ML configuration corresponding to the processing information at the base station 120, and a third partition of the E2E ML configuration corresponding to the processing information at the core network server 302, wherein the determination of the partition can be based on any combination of capabilities, wireless network resource partitions, operating parameters, current operating environment, etc.

[0147] As an example, consider E2E communications corresponding to voice transmission over a wireless network, such as E2E communications 1002, E2E communications 1004, and / or a combination of both E2E communications. In determining an E2E ML configuration for the E2E communication, the E2EML controller 318 alternatively or additionally identifies that the performance requirements of the E2E communication indicate a large amount of data transmission with low latency requirements. Based on the performance requirements, the E2E ML controller identifies an E2E ML configuration that, when formed by a corresponding DNN, performs end-to-end functions of exchanging voice communications and meeting the performance requirements. For illustration, the E2E ML controller determines an E2EML configuration that performs end-to-end functions for transmitting voice from a UE to a core network server, such as signal processing, voice coding, channel coding and / or channel modulation on the UE side, channel decoding, demodulation and / or signal processing on the base station side, decoding voice on the core network server side, etc., and selects a configuration designed to meet the performance requirements.

[0148] Some embodiments partition the E2E ML configuration based on the ML capabilities and / or performance requirements of the devices participating in the E2E communication. The UE, for example, may have fewer processing resources (e.g., processing power, memory constraints, quantization constraints, fixed-point computations versus floating-point computations, FLOPS, power availability relative to the base station and / or core network server, which can be indicated by the ML capabilities. In response to identifying different processing resources through analysis of the ML capabilities, the E2E ML controller partitions the E2E ML configuration so that a first partition (e.g., at the UE 110) forms a DNN that performs less processing than a DNN formed by a second or third partition (e.g., at the base station, at the core network server). Alternatively or additionally, the E2E ML controller partitions the E2E ML configuration to produce a neural network designed to not exceed the device capabilities, for example, based on the analysis capabilities, the E2E ML controller directs the UE to form a DNN with fewer layers and a smaller kernel size relative to a DNN formed by the base station and / or core network server based on the UE's processing constraints. Alternatively or additionally, the E2E ML controller partitions the E2E ML configuration to produce a neural network designed to not exceed the device capabilities, for example, based on the analysis capabilities, the E2E ML controller directs the UE to form a DNN with fewer layers and a smaller kernel size relative to a DNN formed by the base station and / or core network server based on the UE's processing constraints. Alternatively or additionally, the E2E ML controller partitions the E2E The E2E ML controller calculates whether the amount of computation performed at each device collectively meets a performance requirement corresponding to a latency budget and determines an E2E ML configuration designed to meet the performance requirement.

[0149] In the environment 1000, the E2E ML controller 318 determines a first E2E ML configuration for processing information exchanged via the E2E communication 1002, and determines to partition the first E2E ML configuration across multiple devices, such as by partitioning the first E2E ML configuration across the UE 110, the base station 120, and the core network server 302 based on device capabilities. In other words, some embodiments determine an E2E ML configuration corresponding to a distributed DNN in which multiple devices implement and / or form parts of the DNN. To transmit the partition, the E2E ML controller 318 identifies a first neural network formation configuration (NN formation configuration) corresponding to the first partition of the E2E ML configuration, and transmits the first NN formation configuration to the UE 110 by using the core network server 302. The E2E ML controller 318 and / or the core network server 302 then directs the UE to form a user device-side deep neural network 1008 (UE-side DNN 1008) for processing the information exchanged via the E2E communication 1002. Similarly, the E2E ML controller 318 identifies a second NN formation configuration corresponding to a second partition of the E2E ML configuration, and transmits the second NN formation configuration to the base station 120. The E2E ML controller 318 and / or the core network server 302 then directs the base station 120 to use the second NN formation configuration to form a base station-side deep neural network 1010 (BS-side DNN 1010) for processing information exchanged via the E2E communication 1002. The E2E ML controller 318 also identifies a third NN formation configuration and transmits it to the core network server 302 for use in forming a core network server-side deep neural network 1012 (CNS-side DNN 1012) for processing information exchanged via the E2E communication 1002.

[0150] In an embodiment, the E2E ML controller 318 partitions the E2E ML configuration to distribute the processing computations performed on the E2E communication so that the UE-side DNN 1008 performs less processing relative to the BS-side DNN 1010 (e.g., the UE-side DNN 1008 uses fewer layers, fewer data processing points, etc. relative to the BS-side DNN 1010). Alternatively or in addition, the E2E ML controller 318 partitions the E2E ML configuration so that the BS-side DNN 1010 performs less processing relative to the CNS-side DNN 1012. In combination, the processing performed by the UE-side DNN 1008, the BS-side DNN 1010, and the CNS-side DNN 1012 exchange information across the E2E communication 1002.

[0151] In a similar manner, the E2E ML controller 318 determines a second E2E ML configuration for processing information exchanged via the E2E communication 1004, wherein the E2E ML controller partitions and / or distributes the second E2E ML configuration across multiple devices. In the environment 1000, the partition corresponds to a core network server-side deep neural network 1014 (CNS-side DNN 1014), a base station-side deep neural network 1016 (BS-side DNN 1016), and a user device-side deep neural network 1018 (UE-side DNN 1018). In combination, the processing performed by the CNS-side DNN 1014, the BS-side DNN 1016, and the UE-side DNN 1018 corresponds to exchanging information using the E2E communication 1004. Although the E2E ML controller separately determines the first and second E2E ML configurations for unidirectional E2E communications (e.g., E2E communications 1002 and 1004) in the environment 1000, it is understood that in alternative or additional embodiments, the E2E ML controller 318 determines a single E2E ML configuration corresponding to exchanging bidirectional information using E2E communications. Thus, with respect to the E2E communications 1002 and / or the E2E communications 1004, the E2E ML controller 318 determines the partitioned E2E ML configuration and transmits the corresponding portions of the partitioned E2E ML configuration to the devices participating in the E2E communications 1002 and / or the E2E communications 1004.

[0152] In an embodiment, the E2E ML controller 318 periodically re-evaluates metrics, performance requirements, radio link performance, processing capabilities of the device, or other aspects that affect or provide an indication of the current operating environment and / or current performance (e.g., bit errors, BLER) to determine whether to update the E2E ML configuration. For example, the E2E ML controller 318 determines modifications (e.g., parameter changes) to an existing DNN to better accommodate the performance requirements of devices, applications, and / or transmissions in the wireless network. A UE changing location may affect radio link performance, or a user opening an application at the UE may reduce the processing power that the user device can provide for machine learning. By re-evaluating dynamically changing conditions (e.g., changes in the operating environment, changes in the device), the E2E ML controller can modify or update the E2E ML configuration to improve the overall efficiency of how wireless network resources are utilized.

[0153] Having described E2E ML for wireless networks, consider now a discussion of a machine learning architecture for broadcast and multicast communications in accordance with one or more embodiments.

[0154] Machine Learning Architecture for Broadcast and Multicast Communications

[0155] Wireless communication systems use broadcast and multicast communications to propagate information from one device to multiple devices. When propagating information, wireless communication systems may repeat, duplicate, or share the same information among multiple devices, such as by transmitting the repetition of information on multiple beams or enabling multiple devices to access the information from the same transmission. As an example, a base station transmits information on a broadcast channel to provide multiple receiving devices with access to a corresponding network. As another example, a base station may broadcast information associated with a service to multiple devices, such as a traffic broadcast service that delivers traffic updates to multiple vehicles.

[0156] Typically, any capable receiving device within the operating range of a base station has access to the information included in a broadcast communication. However, where broadcast communications disseminate information in a one-to-all manner, multicast communications selectively disseminate information in a one-to-many manner. Multicast communications target specific devices (e.g., a subset of possible receiving devices) rather than providing information to any device capable of receiving.

[0157] Transmitting information to multiple devices, whether for broadcast or multicast communications, presents several challenges. As an example, transmitting information to multiple devices creates a more complex transmission environment relative to transmitting information to a single device. To illustrate, a mobile device communicating with a base station changes the transmission environment by moving from a first location to a second location. This becomes even more complicated when multiple mobile devices receiving broadcast or multicast communications from a base station move in different directions from each other. As another example, multiple devices sometimes have different processing capabilities from each other. For example, a first device receiving a broadcast or multicast communication may have less processing power than a second device receiving the broadcast or multicast communication. Different processing capabilities between different UEs can affect how a base station transmits information. These differences in device capabilities and changing channel conditions sometimes lead to inefficient use of network resources.

[0158] Various embodiments utilize a machine learning architecture for broadcast and multicast communications. In an embodiment, a network entity determines a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted over a wireless communication system, wherein the communications are directed to a target group of user equipment (UE). The network entity forms a network entity DNN based on the determined configuration of the DNN, and uses the network entity DNN to process the broadcast or multicast communications. In an embodiment, the network entity forms a common DNN to process the broadcast or multicast communications and / or propagate the broadcast or multicast communications to the target group of UEs.

[0159] Fig.11 An example environment 1100 of a machine learning architecture utilizing broadcast and multicast communications according to one or more implementations is illustrated. The environment 1100 includes Figure 11104, 1106, and 1108. In various embodiments, UE 1102, 1104, 1106, and 1108 represent Figure 1 UE 1102, 1104, 1106, and 1108 collectively form a target group 1110 of UEs for broadcast communications from base station 120, where the target group includes multiple UEs (eg, at least two UEs).

[0160] Base station 120 includes a base station side deep neural network 1112 (BS side DNN 1112) that processes broadcast communications 1114 for transmission over the wireless communication system. Figure 5 and Figure 6 , the BS-side DNN 1112 performs transmitter chain operations to generate broadcast communications 1114 and / or propagate the broadcast communications 1114 to a target group 1110 of UEs (e.g., UE 1102, UE 1104, UE 1106, and UE 1108) over a wireless interface. Alternatively or additionally, reference Fig.10 , BS-side DNN 1112 represents a DNN formed by a portion of the partitioned E2E ML configuration. Similarly, each UE includes a user equipment side deep neural network (UE-side DNN), which is labeled as UE-side DNN 1116, UE-side DNN 1118, UE-side DNN 1120 and UE-side DNN 1122, where the UE-side DNN performs receiver chain operations. In some embodiments, the UE-side DNN represents a device-specific DNN and / or a DNN based on the partitioned E2E ML configuration. In an embodiment, the neural network manager, E2E ML controller and / or network slice manager determines the configuration of the BS-side DNN 1112 and / or the UE-side DNN 1116, UE-side DNN 1118, UE-side DNN 1120 and UE-side DNN 1122, as further described.

[0161] The BS-side DNN 1112 processes the broadcast communication 1114 to transmit information over the wireless communication system, where the processing may include pre-transmission operations. The BS-side DNN 1112 may include any combination of ML architectures, such as a convolutional neural network (CNN) architecture, a recurrent neural network (RNN) architecture, an LSTM architecture, a fully connected layer architecture, or a partially connected layer architecture. Sometimes, the base station 120 uses a common DNN architecture to form a common DNN (e.g., BS-side DNN 1112) for communicating with each target UE, such as all capable UEs in a cell coverage area associated with the base station 120. Therefore, each corresponding UE-side DNN (e.g., UE-side DNN 1116, UE-side DNN 1118, UE-side DNN 1120, and UE-side DNN 1122) performs complementary processing to the common DNN. In other words, in some embodiments, the UE-side DNNs 1116, 1118, 1120, and 1122 use similar DNN architectures to receive and process the broadcast communication 1114. However, in alternative embodiments, the UE-side DNNs may use different DNN architectures to receive and process broadcast communications, such as using a respective DNN architecture by each respective UE based on the respective processing capabilities of the UEs. In some embodiments, the core network server may alternatively or additionally use a DNN architecture such as reference Fig.10 The described core network server-side deep neural network (CNS-side DNN) handles broadcast communications (not shown here).

[0162] Fig.12 An example environment 1200 utilizing a machine learning architecture for broadcast and multicast communications is illustrated in accordance with one or more implementations. The environment 1200 includes Figure 1 The base station 120 and Fig.11 1104, UE 1106, and UE 1108, all of which are within the coverage area of ​​​​the base station 120. In the environment 1200, the base station 120 includes a base station-side deep neural network 1202 (BS-side DNN 1202), which is configured to transmit a multicast communication 1204 to a target group 1206 of UEs, and the target group 1206 of UEs includes two UEs: UE 1104 and UE 1108. For example, referring to Figure 5 and Figure 6 , the BS-side DNN 1202 performs transmitter chain operations to process and / or generate multicast communications 1204. Sometimes, reference Fig.10 , the base station 120 forms a BS-side DNN 1202 based on a portion of the segmented E2E ML configuration as further described. In the environment 1200, the target group 1206 of UEs corresponds to a subset of UEs included in a coverage area associated with the base station 120.

[0163] Each UE in the target UE 1206 group (e.g., UE 1104, UE 1108) includes a corresponding UE-side DNN that processes and recovers information from the multicast communication 1204. For illustration, UE 1104 includes UE-side DNN 1208, and UE 1108 includes UE-side DNN 1210, wherein the UE-side DNN performs receiver chain operations and / or complementary operations to BS-side DNN 1202 to recover the multicast information. In some embodiments, UEs 1104 and 1108 form UE-side DNNs 1208 and 1210 based on a portion of a split E2E ML configuration (such as a split E2DE ML configuration used to form BS-side DNN 1202). Although in Fig.12 Not shown in the figure, but some embodiments include a CNS-side DNN at the core network server to handle multicast communications, such as reference Fig.10 Further described.

[0164] By using a DNN to process broadcast and / or multicast communications, the wireless communication system can adapt the processing to improve communication exchanges in the system. As an example, the wireless communication system configures the DNN (e.g., BS-side DNN, UE-side DNN, CNS-side DNN) based on current operating conditions or different UE capabilities to improve the overall performance of broadcast and / or multicast communications exchanged between devices (e.g., lower bit errors, improved signal quality, improved latency). Alternatively or in addition, the wireless communication system configures the DNN based on the lowest common configuration supported by each UE in the target group of UEs (e.g., using a configuration supportable by a UE with the lowest processing power).

[0165] In an embodiment, a network entity (e.g., a core network server 302, a base station 120) uses any combination of a neural network manager, an E2E ML controller, and / or a network slicer to determine one or more ML configurations for processing broadcast or multicast communications. The ML configurations may be based on content requirements for the broadcast and / or multicast communications, a current operating environment, changes in the current operating environment, UE capabilities, UE characteristics, and the like. For example, the content requirements specify quality requirements, resolution requirements, or frames per second requirements for specific UEs in a target group of UEs that receive broadcast and / or multicast communications. Alternatively or in addition, the content requirements specify quality requirements, resolution requirements, or frames per second requirements for each UE in a target group of UEs. The network entity analyzes these content requirements and determines device-specific ML configurations or E2E ML configurations designed to meet the content requirements for each UE and / or specific UE, such as references. Fig.10Sometimes, when the content requirement of the broadcast or multicast communication requirement changes, the network entity determines to change the ML configuration. Alternatively or additionally, the network entity determines the ML configuration based on a network slice designed to meet the performance and / or content requirements. In some embodiments, the network entity determines the ML configuration for each device in the device based on UE characteristics (such as an estimated location of each device, capabilities of each device, UE speed, etc.).

[0166] Configuring signaling and control transactions for deep neural networks

[0167] Fig.13 , 14 , 15 and 16 illustrate example signaling and control transaction diagrams between base stations, user equipment and / or core network servers according to one or more aspects of using a machine learning architecture for broadcast and multicast communications in wireless communications. The signaling and control transactions may be performed by Figure 1 The base station 120 and the UE 110 or Figure 3 The core web server 302 uses Figure 1-12 elements to execute.

[0168] Fig.13 The signaling and control transaction diagram 1300 illustrates a first example of signaling and control transactions using a machine learning architecture for broadcast and multicast communications in wireless communications. At 1305, the base station 120 optionally receives metrics and / or UE capabilities from the UE 110, wherein Fig.13 The UEs in the target group correspond to the UEs, such as Fig.11 The target group 1110 of UEs shown in FIG. Fig.12 The target group of UEs 1206 is shown. For example, the base station 120 receives UE capabilities from one or more UEs in the target group of UEs in response to sending a request for UE capabilities. In some embodiments, the UE capabilities include ML-related capabilities, such as maximum kernel size capabilities, memory limitations, computational capabilities, supported ML architectures, number of layers supported, available processing power, memory limitations, available power budget, and fixed-point processing versus floating-point processing. As another example, the base station 120 receives UE metrics from one or more UEs in the UE 110, such as power measurements (e.g., RSS), error metrics, timing metrics, QoS, latency, reference signal received power (RSRP), SINR information, CQI, CSI, Doppler feedback, QoS, latency, etc. Alternatively or additionally, the base station 120 generates BS metrics based on communications with the UEs, such as reference Figure 8 Those described.

[0169] At 1310, the base station 120 determines to transmit a broadcast or multicast communication. For example, the base station 120 determines to transmit a machine type communication (MTC) to a subset of IoT devices using a multicast communication. As another example, the base station 120 determines to transmit a paging message to all devices in a cell coverage area using a broadcast communication.

[0170] In response to determining to transmit a broadcast or multicast communication, at 1315, the base station 120 (via the BS neural network manager 268, an E2E ML controller implemented by the base station, and / or a network slice manager implemented by the base station) determines a configuration of the DNN for processing the communication. The configuration of the DNN may include a (partitioned) E2E ML configuration and / or multiple DNN configurations for multiple devices. In some embodiments, the base station neural network manager 268 determines to use a default configuration for the DNN. As another example, the base station neural network manager 268 determines the configuration by communicating with an E2E ML controller and / or a network slice manager (such as an E2E ML controller 318 and / or a network slice manager 320 at a core network server 302 or an E2E ML controller and / or a network slice manager implemented at the base station 120). To illustrate, the base station neural network manager 268 communicates content requirements and / or performance requirements (such as quality requirements, resolution requirements, frames per second requirements, latency requirements, or bandwidth requirements) to the E2E ML controller and / or network slice manager to determine the configuration of the DNN to be directed to meet these requirements.

[0171] In some embodiments, the base station 120 determines the DNN configuration based on one or more (estimated) locations of the UE, such as the estimated location of each UE in the target group of UEs, the single estimated location of a single UE in the target group of UEs, or the corresponding estimated locations of a subset of UEs in the target group of UEs. The target group of UEs may include all UEs within the cell coverage area or a subset of UEs within the cell coverage area, wherein, in various embodiments, the target group of UEs includes at least two UEs. For example, in response to identifying that at least one of the target UEs is located at the edge of the cell coverage area, the base station 120 determines the DNN configuration through any combination of the base station neural network manager 268, the network slice manager and / or the E2EML controller so as to reliably transmit broadcast or multicast communications to UEs located at the edge of the cell coverage area in addition to other UEs in the target group of UEs. In some embodiments, this includes analyzing the neural network table based on the estimated location, UE capabilities, radio network resource partitioning, operating parameters, metrics received at 1305, current operating environment, UE ML capabilities, etc. to determine the configuration of the DNN.

[0172] When determining the configuration of the DNN, the base station 120 sometimes determines a gradient or scaled version of the configuration and / or ML architecture. For example, consider a scenario in which a first UE in a target group of UEs has less processing power relative to a second UE in the target group of UEs. Some embodiments identify a simplified version of the determined configuration and transmit the simplified version to the first UE with less processing power. The simplified version of the determined configuration represents a scaled version of the ML configuration with lower complexity relative to the determined configuration, such as a reduced number of layer connections, fewer processing points, etc. More broadly, a gradient version of an ML or DNN configuration is a variation of an architecture or configuration that performs different levels of processing, such as by having variations with different layer connections to reduce or increase the number of processing nodes, by different filtering to reduce or increase the number of data points processed, and so on. Therefore, when determining the configuration of the DNN, the base station 120 can determine a gradient or scaled version of the configuration based on UE capabilities.

[0173] At 1320, the base station 120 optionally transmits a configuration of the DNN to the UE 110 at 1325. As an example, when the base station determines to use a default DNN configuration, the UE 110 implicitly determines to use the default DNN configuration in response to not receiving the DNN configuration from the base station. Therefore, the base station does not explicitly transmit the DNN configuration. As another example, the base station 120 transmits a segmentation of the E2E ML configuration or a UE-specific DNN configuration to the UE 110. The configuration of the DNN can be included in any suitable manner, such as by using a broadcast or unicast message.

[0174] At 1330, the base station 120 forms a DNN based on the configuration of the DNN determined at 1315. In an embodiment, the DNN formed by the base station (e.g., the BS-side DNN 1112) performs at least some processing for transmitting broadcast or multicast communications over the wireless communication system. Similarly, at 1335, the UE 110 forms a DNN based on the configuration determined at 1315. For example, the UE 110 accesses a neural network table to obtain one or more parameters and uses the parameters to form a DNN, as described in reference to FIG. Figure 8 In an embodiment, the UE forms a DNN using a default neural network formation configuration, a portion of an E2E ML configuration, or a UE-specific neural network formation configuration, wherein the DNN formed by the UE 110 (e.g., UE-side DNNs 1116, 1118, 1120, and 1122) performs at least some processing associated with consuming broadcast or multicast communications.

[0175] Then, at 1340, the base station 120 and the UE 110 use the DNN to process the broadcast or multicast communication, such as reference Fig.11 and 12In an embodiment, the base station 120 and / or the UE 110 iteratively performs the signaling and control transactions described in the signaling and control transaction diagram 1300, as indicated by the dashed line 1345. These iterations allow the base station 120 and / or the UE 110 to dynamically modify the DNN that processes the broadcast or multicast communication based on the changing operating conditions, as further described.

[0176] Fig.14 The signaling and control transaction diagram 1400 illustrates a second example of signaling and control transactions using a machine learning architecture for broadcast and multicast communications in wireless communications. The example diagram 1400 performs transactions similar to those described with respect to 1300, where the signaling and control transactions include interactions with the core network server 302.

[0177] At 1405, the core network server 302 optionally receives metrics and / or capabilities from the base station 120 and / or the UE 110 (via the base station 120), wherein Fig.14 The UEs in the target group correspond to the UEs, such as Fig.11 The target group 1110 of UEs shown in FIG. Fig.12 In an embodiment, the core network server 302 receives UE capabilities or UE metrics, such as ML-related capabilities or processing capabilities, from one or more UEs in the target group of UEs via the base station 120. Alternatively or additionally, the core network server 302 receives BS metrics based on communications exchanged between the base station and the UE from the base station 120, as shown in FIG. Figure 8 Alternatively or additionally, the core network server 302 receives BS capabilities from the base station 120, such as processing power, power status, capabilities (eg, the number of connections that can be supported), operating range, etc.

[0178] At 1410, the core network server 302 determines to transmit a broadcast or multicast communication, and determines a configuration of a DNN for processing the communication at 1415. In an embodiment, the core network neural network manager 312 determines the configuration by communicating with the E2E ML controller 318 and / or the network slice manager 320. Sometimes, the configuration can be based on any combination of content requirements, performance requirements, UE capabilities, UE characteristics, or current operating conditions.

[0179] In some embodiments, the core network server 302 determines a configuration of the DNN based on one or more (estimated) locations of the UEs, such as estimated locations of a target group of UEs including all UEs within a cell coverage area or a subset of UEs within a cell coverage area. For illustration, the core network server 302 analyzes the neural network table based on any combination of estimated locations to determine the configuration. Alternatively or in addition, the core network server 302 analyzes the neural network table based on UE capabilities, wireless network resource segmentation, content requirements, performance requirements, metrics received at 1305, the current operating environment, UE ML capabilities, one or more UE speeds, etc. to determine the configuration of the DNN. Any type of configuration for the DNN may be determined, such as a CNS-specific DNN configuration, a BS-side DNN configuration, a UE-side DNN configuration, and / or a segmented E2E ML configuration, as further described.

[0180] In determining the configuration of the DNN, the core network server 302 sometimes determines a simplified version of the configuration and / or ML architecture, where the determined simplified version of the configuration represents a scaled version of the ML configuration with less or more processing complexity relative to the determined configuration, such as a reduced number of layer connections, fewer processing points, etc. Thus, as part of determining the configuration of the DNN, the core network server 302 sometimes determines a variation or scaled version of the configuration based on UE capabilities.

[0181] At 1420, the core network server optionally transmits the configuration of the DNN to the base station 120 and / or to the UE 110 at 1425. As an example, when the core network server 302 determines to use the default DNN configuration, the base station 120 and / or the UE 110 implicitly determines to use the default DNN configuration in response to not receiving the DNN configuration from the core network server 302. As another example, the core network server 302 transmits a segmentation of the E2E ML configuration or a device-specific DNN configuration to the base station 120 and / or the UE 110.

[0182] At 1430, the core network server forms a core network server-side deep neural network (CNS-side DNN) based on the configuration of the DNN determined at 1415. In an embodiment, the DNN formed by the core network server performs at least some processing for transmitting broadcast or multicast communications through a wireless communication system, which may include pre-transmission processing. At 1335, the base station 120 forms a DNN based on the configuration of the DNN determined at 1415. In an embodiment, the DNN formed by the base station performs at least some processing for transmitting broadcast or multicast communications through a wireless communication system, which includes pre-transmission processing. Similarly, at 1440, the UE 110 forms a DNN based on the configuration determined at 1415. For example, the UE 110 uses the information received at 1425 to access the neural network table to obtain a reference Figure 8 One or more parameters described in the description. This may include forming the DNN using a default neural network formation configuration, a portion of an E2EML configuration, or a UE-specific neural network formation configuration. In an embodiment, the DNN formed by the UE 110 performs at least some processing for transmitting broadcast or multicast communications over the wireless communication system.

[0183] Then, at 1445, the core network server 302, the base station 120, and the UE 110 use the DNN to process the broadcast or multicast communication, such as reference Fig.11 and 12 In an embodiment, the core network server 302, the base station 120 and / or the UE 110 iteratively perform the signaling and control transactions described in the signaling and control transaction diagram 1400 (indicated by the dashed line 1450). These iterations allow the core network server 302, the base station 120 and / or the UE 110 to dynamically modify the DNN that handles the broadcast or multicast communication based on the changing operating conditions, as further described.

[0184] Changing operating conditions affects the performance of how well each DNN processes information. For illustration, and with reference to Figure 8 , training module 270 and / or training module 314 generates a neural network formation configuration based on operating conditions as described by the input characteristics. This includes operating conditions corresponding to broadcast and multicast communications, such as operating conditions including estimated UE locations, number of target UEs, channel conditions, content requirements, or network slice configurations. As current operating conditions deviate from the operating conditions used to train the DNN, the performance of the DNN begins to deteriorate.

[0185] Various embodiments measure the performance of a DNN for exchanging broadcast or multicast communications, and determine to make modifications to the DNN when the performance fails to meet a threshold. This includes determining to make architectural changes and / or parameter changes to the DNN, such as by determining to form a new neural network configuration corresponding to the change. Modifications to the DNN may include small changes involving updating coefficient parameters of an existing architectural configuration, or larger changes involving modifying the architectural configuration. For example, a cost function measures error performance within a system, such as by comparing predicted values ​​generated by the DNN to expected values ​​or true values. When the cost function indicates that the error performance within the system meets a desired threshold, various embodiments determine to make small modifications to the system (e.g., parameter changes). When the cost function indicates that the error performance within the system fails to meet the desired threshold, various embodiments determine to make large modifications to the system (e.g., architectural configuration changes).

[0186] For demonstration purposes, now consider the signaling and control transaction diagram 1500 in Fig.15 A third example of signaling and control transactions using a machine learning architecture for broadcast and multicast communications in wireless communications is shown in FIG. In some embodiments, the signaling and control transaction diagram 1500 represents Fig.13 A continuation of signaling and control transaction diagram 1300 .

[0187] refer to Fig.11 and 12 At 1505, the base station 120 and the UE 110 use the DNN to process the broadcast or multicast communication, where the UE 110 represents a target group of UEs, which may include all UEs in the cell coverage area or a subset of UEs in the cell coverage area. In some embodiments, the processing performed at 1505 corresponds to Fig.13 The processing performed at 1340.

[0188] At 1510, base station 120 receives feedback from at least one UE in a target group of UEs. For example, UE 110 transmits one or more metrics, such as BLER, SINR, CQI feedback, or packet loss rate. Alternatively or additionally, base station 120 generates one or more metrics, such as a round trip time (RTT) delay metric.

[0189] At 1515, the base station 120 analyzes the feedback. For example, the base station 120 (via the BS neural network manager 268, the E2E ML controller implemented by the base station, and / or the network slice manager implemented by the base station) analyzes the feedback to determine whether each UE in the target group of UEs meets the performance threshold and / or cost function threshold. Sometimes, the UEs in the target group of UEs have different performance thresholds and / or cost function thresholds from each other. For illustration, the first UE in the target group may have less processing power relative to the second UE in the target group. Therefore, different UEs may have different performance requirements based on relative UE capabilities. Therefore, the base station 120 compares the metric describing the performance of the UE-side DNN at the first UE with a performance threshold / cost function value that is different from the metric describing the performance of the UE-side DNN at the second UE. For clarity, the above examples discuss analyzing metrics describing UE-side DNN performance, but it should be understood that the metrics may describe the performance of a DNN chain (e.g., a CNS-side DNN, a BS-side DNN, a UE-side DNN), a segmented DNN based on an E2E ML configuration, and the like.

[0190] At 1520, the base station 120 (via the BS neural network manager 268, the E2E ML controller implemented by the base station, and / or the network slice manager implemented by the base station) determines a modification to the DNN based on the feedback. In some embodiments, the base station 120 determines a large modification that changes the architectural configuration of the DNN. For example, in response to determining that "X" UEs in the target group fail to meet the corresponding performance and / or cost function thresholds, the base station 120 determines a large modification corresponding to changing the architectural configuration of one or more DNNs, where "X" represents an arbitrary number. Alternatively or additionally, when more than "X" UEs meet the performance and / or cost function thresholds, the base station 120 determines a small modification corresponding to changing the parameter configuration without changing the architectural configuration, such as changing coefficient values, weights, or kernel sizes. The modification may correspond to a DNN formed from the segmented E2E ML configuration and / or may correspond to a device-specific DNN (e.g., a CNS-side DNN, a BS-side DNN, a UE-side DNN). In various embodiments, the base station 120 determines a gradual modification based on the corresponding UE capabilities, as further described.

[0191] At 1525, the base station 120 transmits the modification to the UE 110. In some embodiments, the base station transmits the same modification to each UE in the target group of UEs. In other embodiments, the base station transmits a gradual modification to different UEs based on the respective UE capabilities (e.g., UEs with less processing power receive a simplified version of the modification relative to UEs with greater processing power).

[0192] At 1530, the base station 120 updates the BS-side DNN based on the modification. Similarly, at 1535, the UE 110 updates the UE-side DNN based on the modification, respectively, wherein the BS-side DNN and the UE-side DNN may correspond to a device-specific DNN or a DNN configured based on the partitioned E2E ML. In an embodiment, the base station 120 and / or the UE 110 iteratively perform the signaling and control transactions described in the signaling and control transaction diagram 1500, indicated by the dashed line 1540. These iterations allow the base station 120 and / or the UE 110 to dynamically modify the DNN that handles broadcast or multicast communications based on changing operating conditions, as further described.

[0193] Fig.16 The signaling and control transaction diagram 1600 illustrates a fourth example of signaling and control transactions using a machine learning architecture for broadcast and multicast communications in wireless communications. The example diagram 1600 performs transaction processing similar to those described with respect to 1500, where the transaction includes interaction with a core network server 302. In some embodiments, the signaling and control transaction diagram 1600 represents Fig.14 Continuation of signaling and control transaction diagram 1400.

[0194] At 1605, the core network server 302, the base station 120, and the UE 110 use the DNN to process the broadcast or multicast communication, wherein the UE 110 represents a target group of UEs, which may include all UEs in the cell coverage area or a subset of UEs in the cell coverage area. In some embodiments, the processing performed at 1605 corresponds to Fig.14 The processing of 1445 executions.

[0195] At 1610, the core network server 302 receives feedback from at least one UE in the target group of UEs through the base station 120. For example, the UE 110 transmits one or more metrics (such as BLER, SINR, CQI feedback, or packet loss rate) as feedback to the core network server 302. Alternatively or additionally, the base station 120 generates one or more metrics such as a round trip time (RTT) delay metric and sends the metric to the core network server 302 as feedback.

[0196] At 1615, the core network server 302 analyzes the feedback. For example, the core network server 302 (via the core network neural network manager 312, the E2E ML controller 318, and / or the network slice manager 320) analyzes the feedback to determine whether each UE in the target group of UEs meets the performance threshold and / or cost function threshold. In some embodiments, the core network server 302 uses the same performance threshold and / or cost function value when analyzing the performance of each UE in the target group of UEs. At other times, the core network server 302 uses a varying performance and / or cost function threshold based on the relative UE capabilities of each UE. Thus, the core network server 302 compares a metric describing the performance of the UE-side DNN at the first UE with a performance threshold / cost function value that is different from a metric describing the performance of the UE-side DNN at the second UE. For clarity, the above examples discuss analyzing metrics describing UE-side DNN performance, but it should be understood that the metrics can describe the performance of a DNN chain (e.g., a CNS-side DNN, a BS-side DNN, a UE-side DNN), a segmented DNN based on an E2E ML configuration, and the like.

[0197] At 1620, the core network server 302 (via the core network neural network manager 312, the E2E ML controller 318, and / or the network slice manager 320) determines a modification to the DNN based on the feedback. In some embodiments, the core network server 302 determines a large modification that changes the architectural configuration of the DNN. Alternatively or additionally, the core network server 302 determines a small modification that changes a parameter configuration such as a coefficient value, a weight, or a kernel size. The modification may correspond to a DNN formed from the segmented E2EML configuration and / or a device-specific DNN (e.g., a CNS-side DNN, a BS-side DNN, a UE-side DNN). In various embodiments, the core network server 302 determines a gradient version of the modification based on the corresponding UE capabilities, as further described.

[0198] At 1625, core network server 302 transmits the modification to base station 120 and / or UE 110 (via base station 120). In some embodiments, core network server 302 transmits the same modification to each UE in the target group of UEs. In other embodiments, core network server 302 transmits gradual modifications to different UEs based on the respective UE capabilities (e.g., UEs with less processing power receive a simplified version of the modification relative to UEs with more processing power).

[0199] At 1630, the core network server updates the CNS-side DNN based on the modification. Similarly, at 1635 and 1640, respectively, the base station 120 updates the BS-side DNN based on the modification, and the UE 110 updates the corresponding UE-side DNN based on the modification. The CNS-side DNN, the BS-side DNN, and / or the UE-side DNN may correspond to a device-specific DNN or a DNN configured based on a partitioned E2E ML. In an embodiment, the core network server 302, the base station 120, and the UE 110 iteratively perform the signaling and control transactions described in the signaling and control transaction diagram 1600, indicated by the dashed line 1645. These iterations allow the core network server 302, the base station 120, and / or the UE 110 to dynamically modify the DNN that handles broadcast or multicast communications based on changing operating conditions, as further described.

[0200] Example Method

[0201] Reference to one or more aspects of a machine learning architecture for broadcast and multicast communications Fig.17 and Fig.18 Example methods 1700 and 1800 are described. The order in which the method boxes are described is not intended to be interpreted as limiting, and any number of the described method boxes can be skipped or combined in any order to implement a method or an alternative method. Generally, any of the components, modules, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuits), manual processing, or any combination thereof. Some operations of the example methods can be described in the general context of executable instructions stored on a computer-readable memory locally and / or remotely of a computer processing system, and implementations can include software applications, programs, functions, and the like. Alternatively or in addition, any function described herein can be performed at least in part by one or more hardware logic components, such as, but not limited to, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an application specific standard product (ASSP), a system on a chip (SoC), a complex programmable logic device (CPLD), and the like.

[0202] Fig.17 An example method 1700 for using a machine learning framework for processing broadcast or multicast communications is illustrated. In some implementations, the operations of the method 1700 are performed by a network entity such as any of the base station 120 or the core network server 302.

[0203] At 1705, the network entity determines a configuration of a deep neural network (DNN) for processing broadcast or multicast communications directed to a target group of UEs using a wireless communication system, such as Fig.13 1315 or Fig.141415. In one or more embodiments, a network entity (e.g., a core network server 302) determines an E2E ML configuration for transmitting broadcast or multicast communications to a target group of UEs (e.g., target group 1110 of UEs, target group 1206 of UEs). As another example, a network entity (e.g., a base station 120) determines a configuration of one or more DNNs (e.g., at 710, at 915), such as a first configuration for a BS-side DNN and a second configuration for a UE-side DNN that perform complementary operations to each other. The target group of UEs may include all UEs in a cell coverage area associated with the network entity or a subset of UEs in a cell coverage area.

[0204] In some embodiments, the network entity determines to use a default configuration. In other embodiments, the network entity determines the configuration based on one or more metrics (e.g., at 1305, at 1315, at 1405, at 1415). Sometimes, the network entity determines a gradient version of the DNN configuration based on the UE capabilities of one or more UEs in the target group of UEs. Alternatively or in addition, the network entity determines the configuration of the DNN based on one or more characteristics associated with the target group of UEs, such as characteristics describing the current operating environment, the estimated location of each UE in the target group of UEs (e.g., at least two estimated locations), UE ML capabilities, etc. In some embodiments, the network entity determines the configuration of the DNN based on one or more content requirements (e.g., quality requirements, resolution requirements, frames per second requirements) of broadcast or multicast communications.

[0205] At 1710, the network entity forms a network entity DNN based on the determined configuration of the DNN, such as Fig.13 1330 or Fig.14 1430. In other words, the network entity 120 instantiates (or otherwise creates) a DNN having an architecture and / or parameters defined at least in part by the DNN configuration determined at 1705. In an embodiment, the network entity (core network server 302) forms a CNS-side DNN, where the CNS-side DNN can be based on a partitioned or device-specific DNN configuration of an E2E ML configuration. As another example, a network entity (e.g., base station 120) forms a BS-side DNN 1202. In an embodiment, the network entity accesses a neural network table to obtain one or more architecture and / or parameter configurations. In response to obtaining the configuration, the network entity uses the architecture and / or parameter configuration to form a network entity DNN. In some embodiments, the network entity optionally transmits the configuration of the DNN to a target group of UEs, such as in Fig.13 1325 and Fig.14Alternatively or additionally, the network entity implicitly indicates to the target group of UEs to use a default configuration of the DNN, such as by not transmitting the configuration to the target group of UEs.

[0206] At 1715, the network entity uses the network entity DNN to process the broadcast or multicast communication to direct the broadcast or multicast communication to the target group of UEs using the wireless communication system. For example, the CNS-side DNN (1012, 1014) at the core network server 302 processes the pre-transmission communication, such as reference Figure 6 As another example, the BS-side DNN (1112, 1202) at the base station 120 processes the broadcast or multicast communication including the pre-transmission communication to transmit the broadcast or multicast communication to the target group of UEs. In some embodiments, the BS-side DNN corresponds to a common DNN that uses the same configuration rather than a UE-specific DNN to transmit and direct the broadcast or multicast communication to each UE in the target group of UEs.

[0207] Fig.18 An example method 1800 for using a machine learning framework for processing broadcast or multicast communications is illustrated. In some implementations, the operations of the method 1800 are performed by a network entity such as any of the base station 120 or the core network server 302.

[0208] At 1805, a network entity processes a broadcast or multicast communication using a deep neural network (DNN) to direct the broadcast or multicast communication to a target group of user equipment (UE) using a wireless communication network. For example, a network entity (e.g., core network server 302, base station 120) processes a broadcast or multicast communication such as Fig.13 1340 Fig.14 1445 Fig.15 1505 and Fig.16 1605. This may include processing broadcast or multicast communications using a DNN formed using a segmentation of the E2E ML configuration or a device-specific DNN. In some embodiments, the DNN performs transmitter chain operations including pre-transmission processing to direct transmissions to a target group of UEs.

[0209] At 1810, a network entity receives feedback from at least one user equipment in a target group of UEs. For example, a network entity (e.g., core network server 302, base station 120) receives one or more metrics from a UE in a target group of UEs, such as Fig.15 and Fig.161510 and 1610 of . In an embodiment, the metric characterizes the performance of one or more processing chains, such as power measurement (e.g., RSS), error metric, timing metric, QoS, latency, reference signal received power (RSRP), SINR information, CQI, CSI, error metric, or Doppler feedback. Alternatively or additionally, the network entity receives information characterizing the current operating environment, such as UE capabilities, base station type (e.g., eNB, gNB, or NG-eNB), or power mode. In some scenarios, the network entity receives feedback related to the mobility of the UE, such as location change information.

[0210] In response to receiving the feedback, at 1815, the network entity determines modifications to the DNN based on the feedback. For example, the network entity (e.g., core network server 302, base station 120) may Fig.15 1515 or Fig.16 1615 to determine that "X" UEs in the target group of UEs meet the performance and / or cost function thresholds. This may include the network entity determining to change a large modification of the architecture configuration or a small modification of the parameter configuration, such as in Fig.15 1520 or Fig.16 As described in 1620. As another example, the network entity determines the modification of the DNN based on the location change of the UE.

[0211] At 1820, the network entity transmits an indication of the modification to the target group of UEs. For example, the core network server 302 transmits the indication to the target group of UEs via the base station 120. In some embodiments, the indication includes an indication of one or more items in the neural network formation configuration, such as in Fig. 9 920. Sometimes, a network entity (e.g., a core network server 302, a base station 120) transmits a modified variant version. To illustrate, the network entity transmits a first indication of a first neural network forming configuration to a first UE in a target group of UEs, and transmits a second indication of a second neural network forming configuration to a second UE, wherein the first and second neural network forming configurations are based on UE capabilities and form a variant version of a DNN architecture. For example, a second UE-side DNN formed using the second neural network forming configuration performs less processing than a first UE-side DNN formed using the first neural network forming configuration. In other words, the DNN architecture of the second UE-side DNN is a variant version of the DNN architecture of the first UE-side DNN.

[0212] At 1825, the network entity updates the DNN with the modification to form a modified DNN. For example, the core network server 302 updates the CNS-side DNN based on the modification, and the base station 120 updates the BS-side DNN based on the modification. In response to updating the DNN, at 1830, the network entity uses the modified DNN to process the broadcast or multicast communication to direct the broadcast or multicast communication to the target group of UEs.

[0213] Generally, any of the components, modules, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Some operations of the example methods may be described in the general context of executable instructions stored on a computer-readable memory local and / or remote to a computer processing system, and implementations may include software applications, programs, functions, and the like. Alternatively or in addition, any functionality described herein can be performed, at least in part, by one or more hardware logic components, such as, but not limited to, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an application specific standard product (ASSP), a system on a chip (SoC), a complex programmable logic device (CPLD), and the like.

[0214] Although techniques and apparatus for a machine learning architecture for broadcast and multicast communications have been described in language specific to features and / or methods, it is to be understood that the subject matter of the appended claims is not necessarily limited to the specific features or methods described. Rather, the specific features and methods are disclosed as example implementations of a machine learning architecture for broadcast and multicast communications.

[0215] Several examples are described below:

[0216] Example 1: A method performed by a network entity associated with a wireless communication system, the method comprising: determining a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted to a target group of user equipment (UE) via the wireless communication system; forming a network entity DNN at the network entity based on the determined configuration of the DNN; and using the network entity DNN to process the broadcast or multicast communications to direct the broadcast or multicast communications to the target group of UEs using the wireless communication system.

[0217] Example 2: The method of Example 1, wherein determining the configuration comprises determining the configuration of the DNN based at least in part on at least one characteristic of the target group of UEs.

[0218] Example 3: A method as described in Example 2, wherein at least one characteristic includes at least one of: an estimated location of at least one UE in a target group of user equipment (UE); or at least one UE capability of at least one UE in the target group of UEs.

[0219] Example 4: A method as described in any of the preceding examples, wherein determining the configuration further comprises: determining a gradient version of the configuration of at least one UE based on a processing capability of at least one UE in a target group of user equipment (UE).

[0220] Example 5: A method as described in any of the preceding examples, wherein using a network entity DNN to process broadcast or multicast communications further comprises: using a public DNN as the network entity DNN to process broadcast or multicast communications to direct the broadcast or multicast communications to each UE in a target group of UEs.

[0221] Example 6: A method as described in any of the preceding examples, wherein the target group of UEs is a subset of UEs in a cell coverage area of ​​the wireless communication system.

[0222] Example 7: A method as in any of the preceding examples, wherein determining the configuration is based at least in part on one or more content requirements to enable broadcast or multicast communications.

[0223] Example 8: The method of Example 7, wherein the one or more content requirements include at least one of: a quality requirement; a resolution requirement; or a frames per second requirement.

[0224] Example 9: A method as described in any of the preceding examples, wherein determining the configuration includes determining the end-to-end machine learning configuration as the configuration of the DNN.

[0225] Example 10: A method as described in Example 1, wherein the target group of UEs is all UEs in a cell coverage area of ​​a base station of the wireless communication system.

[0226] Example 11: A method performed by a network entity associated with a wireless communication system, the method comprising: using a deep neural network (DNN) to process broadcast or multicast communications to direct one or more broadcast or multicast communications to a target group of user equipment (UE) using the wireless communication system; receiving feedback from at least one UE in the target group of user equipment (UE); determining modifications to the DNN based on the feedback; transmitting an indication of the modifications to the target group of UEs; updating the DNN using the modifications to form a modified DNN; and using the modified DNN to process the broadcast or multicast communications to direct the broadcast or multicast communications to the target group of UEs using the wireless communication system.

[0227] Example 12: A method as described in Example 11, wherein the indication of the transmission modification also includes: transmitting a first indication of a first neural network forming a configuration to a first UE in a target group of UEs, the first neural network forming a configuration corresponding to an updated parameter configuration of a first corresponding UE-side DNN; and transmitting a second indication of a second neural network forming a configuration to a second UE in the target group of UEs, the second neural network forming a configuration corresponding to an updated parameter configuration of a second corresponding UE-side DNN, wherein: the first corresponding UE-side DNN uses a first DNN architecture, the second corresponding UE-side DNN uses a second DNN architecture, and the second DNN architecture is configured as a gradient version of the first DNN architecture, which performs less processing relative to the first DNN architecture.

[0228] Example 13: The method as described in Example 12 also includes: determining a gradient version of the first DNN architecture based at least in part on one or more capabilities associated with the second UE.

[0229] Example 14: A method as described in any of Examples 11 to 13, wherein the indication of the transmission modification also includes: transmitting a neural network formation configuration including architectural configuration changes to each corresponding UE-side DNN to each UE in the target group of UEs.

[0230] Example 15: A method as described in any one of Examples 11 to 14, wherein determining the modification to the DNN further includes: identifying a corresponding cost function threshold for each UE in a target group of UEs; determining based on feedback that at least one UE in the target group of UEs fails to meet the corresponding cost function threshold; and determining an architecture configuration change as a modification to the DNN.

[0231] Example 16: A method as described in Example 15, wherein determining an architecture configuration change further includes: determining at least one architecture configuration change to a corresponding UE-side DNN at at least one UE for at least one UE in a target group of UEs, and wherein updating the DNN using the modification further includes: transmitting an indication of the at least one architecture configuration change to the at least one UE.

[0232] Example 17: A network entity device comprising: a wireless transceiver; a processor; and a computer-readable storage medium comprising instructions, wherein the instructions direct the network entity device to perform operations including: determining a configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted to a target group of user equipment (UE) via a wireless communication system; forming a network entity DNN at the network entity based on the determined configuration of the DNN; and using the network entity DNN to process the broadcast or multicast communications to direct the broadcast or multicast communications to the target group of UEs using the wireless communication system.

[0233] Example 18: The network entity device as described in Example 17, the operation also includes: receiving feedback from at least one UE of a target group of user equipment (UE); determining modifications to the DNN based on the feedback; updating the DNN using the modifications to form a modified DNN; and using the modified DNN to process broadcast or multicast communications to direct the broadcast or multicast communications to the target group of UEs.

[0234] Example 19: A network entity device as described in Example 17 or Example 18, wherein determining the configuration of the DNN also includes: determining the configuration of the DNN based at least in part on transmitting broadcast or multicast communications based on a network slice configuration.

[0235] Example 20: A network entity device as described in any one of Examples 17 to 19, wherein determining the configuration of the DNN also includes: determining an end-to-end machine learning configuration (E2E ML configuration) as the configuration of the DNN, wherein determining the E2E ML configuration includes determining a segmentation of the E2E ML configuration to distribute the E2E ML configuration across multiple devices.

[0236] Example 21: A network entity apparatus as described in any of Examples 16 to 20, wherein determining the configuration of the DNN is based at least in part on one or more quality of service (QoS) requirements.

[0237] Example 22: A network entity device, comprising: a processor; and a computer-readable storage medium comprising instructions, wherein the instructions direct the network entity device to execute the method described in any one of Examples 1 to 16.

[0238] Example 23: A computer-readable storage medium comprising instructions, which, when executed by a processor, directs a network entity device comprising the processor to execute the method described in any one of Examples 1 to 16.

Claims

1. A method performed by a network entity associated with a wireless communication system, the method comprising: receiving UE capabilities of a first UE and a second UE in a target group of user equipment UEs; determining, based on the UE capability, that the second UE has a smaller processing capability than the first UE; Determining a first configuration and a second configuration of a deep neural network (DNN) for processing broadcast or multicast communications transmitted to the first UE and the second UE, respectively, over the wireless communication system, based on determining that the second UE has less processing power relative to the first UE, the second configuration being a gradient version of the first configuration of the DNN, and a first DNN formed using the second configuration performing less processing relative to a second DNN formed using the first configuration; transmitting the first configuration and the second configuration to the first UE and the second UE respectively; forming, at the network entity, a network entity DNN based on the determined first and second configurations of the DNN; as well as The broadcast or multicast communication is processed using the network entity DNN to direct the broadcast or multicast communication to a target group of UEs using the wireless communication system.

2. The method according to claim 1, wherein: Determining the first configuration and the second configuration includes: The first and second configurations of the DNN are determined based at least in part on at least one characteristic of the target group of UEs.

3. The method according to claim 2, wherein: The at least one characteristic includes at least one of the following: An estimated location of at least one UE in the target group of UEs.

4. The method according to claim 1, wherein: The target group of UEs is a subset of UEs in a cell coverage area of ​​the wireless communication system.

5. The method according to claim 1, wherein: Determining the first configuration and the second configuration is based at least in part on one or more content requirements to enable the broadcast or multicast communication.

6. The method according to claim 1, wherein: Using the network entity DNN to process the broadcast or multicast communication further includes: The broadcast or multicast communication is processed using a common DNN as the network entity DNN to direct the broadcast or multicast communication to a target group of the UEs.

7. The method according to any one of claims 1 to 6, wherein: Determining the first configuration and the second configuration includes: An end-to-end machine learning configuration is determined as a type of the first configuration and the second configuration of the DNN.

8. The method according to claim 1, wherein: The UE capability indicates processing capability.

9. The method according to claim 5, wherein: The one or more content requirements include at least one of the following: Quality requirements; Resolution requirements; or Frames per second requirement.

10. The method according to claim 1, wherein: The second configuration performs less processing relative to the first configuration based on the UE capabilities.

11. A network entity device, comprising: processor; as well as A computer-readable storage medium comprising instructions, wherein the instructions direct the network entity device to execute the method according to any one of claims 1 to 10.

12. A computer-readable storage medium comprising instructions, which, when executed by a processor, direct a network entity device comprising the processor to perform the method according to any one of claims 1 to 10.

Images