Neural network based transmission feedback in cellular networks

Abstract

Two devices (802, 804) that are in wireless communication implement a soft transmission feedback scheme. A data sending device (802) wirelessly transmits to a data receiving device (804) a data block (818) and a first transmission (822) generated using one or more neural networks (808), which the data receiving device processes using one or more neural networks (812) to attempt to recover the data block and generate a transmission feedback (824) indicating a status of the recovery attempt. A second transmission (832) is generated using one or more neural networks (810) using the feedback, which is wirelessly transmitted to the data sending device. One or more neural networks (806) process the second transmission to generate a retransmission control signal (820). One or more neural networks (808) selectively include at least a portion of the data block for retransmission to the data receiving device in a third transmission based on the retransmission control signal.

Description

Background Technology

[0001] Wireless communication networks are increasingly tasked with providing high data throughput between wireless devices, such as between a base station and its corresponding user equipment (UE). Technologies that provide sufficient data throughput, including those using relatively high carrier bands for radio frequency (RF) signals (e.g., up to 100 GHz or higher), often make RF signals susceptible to transmission errors caused by factors such as multipath fading, atmospheric absorption, bodily absorption, diffraction, or interference. A common solution to address these transmission errors is to use a transmission feedback scheme, in which the data receiving device notifies the receiving device of successful or unsuccessful reception of a data block via a return transmission signal of an acknowledgment (or "ACK") or non-acknowledgment (or "NACK"). For example, some cellular systems utilize an Automatic Repeat Request (ARQ) scheme, where the data block to be transmitted is encoded at the data transmitting device using a Cyclic Redundancy Check (CRC) value, a parity check value, or other error detection codes, allowing the data receiving device to use the error detection codes to determine whether the data block has been correctly received and recovered. The data receiving device can then send an ACK / NACK signal to the data sending device, which instructs the data receiving device whether the data block has been correctly recovered. If the ACK / NACK signal indicates that the signal representing the data block was corrupted during transmission, the data sending device retransmits the data block.

[0002] Because retransmission of each corrupted data block negatively impacts spectral efficiency, more sophisticated transmission feedback schemes have been developed to provide improved transmission efficiency while maintaining data transmission reliability. One such scheme is Hybrid ARQ (or HARQ), where the data block to be transmitted is encoded using forward error correction (FEC) codes. FEC codes allow the data receiver to detect, and in some cases correct, certain errors identified in the received data. If the FEC codes are insufficient to allow the data receiver to correct all identified errors, the data receiver can request retransmissions similar to a simpler ARQ scheme. HARQ schemes can be further enhanced by using a soft combination method, where different codes are used to transmit each successive transmission of a data block, and then each composition code is used at the data receiver to obtain a single, combined recovered data block.

[0003] While such transport feedback schemes can effectively provide sufficient data transmission reliability, their implementation can be relatively complex. For example, implementing HARQ with soft composition in the 5G New Radio (NR) specification issued by the 3GPP requires using a set of parallel processes to employ an asynchronous HARQ protocol with a corresponding set or group of code blocks for each transport block, and in some implementations, further utilizing a multidimensional HARQ acknowledgment codebook to facilitate the reuse of HARQ acknowledgments. This inherent complexity of the transport feedback scheme typically translates into a corresponding complexity in designing, testing, and implementing hard-coded implementations of the transport feedback scheme at the device. Furthermore, such hard-coded implementations often resist changes or updates to the underlying transport feedback scheme, which either requires additional effort to ensure legacy compatibility of the updated scheme or results in the abandonment of the implementation device. Summary of the Invention

[0004] Two devices, such as a base station and a UE, performing wireless communication implement one or more DNNs or other jointly trained neural networks to provide a soft transmission feedback scheme. A data transmitting device wirelessly transmits a data block and a first transmission generated using one or more neural networks to a data receiving device, which processes the first transmission using one or more neural networks to attempt to recover the data block and generate a feedback transmission signal indicating the status of the recovery attempt. A second transmission is generated by one or more neural networks using the feedback signal, and the second transmission is wirelessly transmitted to the data transmitting device. One or more neural networks process the second transmission to generate a retransmission control signal. Based on the retransmission control signal, one or more neural networks selectively include at least a portion of the data block for retransmission to the data receiving device in a third transmission.

[0005] In one aspect, a computer-implemented method in a data transmitting apparatus includes: buffering a first data block in a delay buffer; processing the first data block at a transmitter neural network of the data transmitting apparatus to generate a first transmission for wireless communication to a data receiving apparatus; and wirelessly transmitting the first transmission to the data receiving apparatus. The method further includes processing a second transmission wirelessly received from the data receiving apparatus at a receiver neural network of the data transmitting apparatus to generate a retransmission control signal. The method further includes applying a gating function to the first data block in the delay buffer using the retransmission control signal as a control input, and providing the output of the gating function to an input of the transmitter neural network to generate a third transmission for wireless communication to the data receiving apparatus; and wirelessly transmitting the third transmission to the data receiving apparatus.

[0006] In some embodiments, the retransmission control signal comprises a binary signal, and generating the third transmission comprises: generating the third transmission based on at least a portion of the first data block in response to the binary signal having a first value, and generating the third transmission independently of the first data block in response to the binary signal having a second value. In other embodiments, the retransmission control signal comprises a nonlinear activation function, and generating the third transmission based on the nonlinear activation function comprises generating a third transmission to include one or more portions of the first data block.

[0007] The above method can also include using one or more training datasets in conjunction with at least one neural network of the data receiving device to jointly train the transmitter neural network and the receiver neural network. In some embodiments, each of the transmitter neural network and the receiver neural network has multiple neural network architecture configurations, each neural network architecture configuration being associated with a different corresponding scheduling authorization type among multiple scheduling authorization types, and jointly training the transmitter neural network and the receiver neural network involves individually training many of the multiple neural network architecture configurations. In this case, the method can also include determining the current scheduling authorization type implemented for the data transmitting device, and configuring the transmitter neural network and the receiver neural network to implement one of the multiple neural network architecture configurations associated with the current scheduling authorization type. The current scheduling authorization type can include: authorization to transmit control information only, authorization to transmit user plane data only, and authorization to transmit both control information and user plane data.

[0008] In some embodiments, generating the third transmission includes further generating the third transmission at the transmitter neural network based on a second data block received after the first data block. Furthermore, in some embodiments, processing the second transmission at the receiver neural network of the data transmitting device further includes processing the second transmission at the receiver neural network of the data transmitting device to attempt to recover the second data block from the second transmission. In this case, the method may also include generating a transmission feedback signal at the receiver neural network of the data transmitting device, the transmission feedback signal representing the state of recovering the second data block from the second transmission, and wherein generating the third transmission further includes processing the transmission feedback signal at the transmitter neural network of the data transmitting device.

[0009] In some embodiments, the receiver neural network and the transmitter neural network of the data transmitting device comprise deep neural networks.

[0010] According to another aspect, a data transmission device includes a radio frequency front end, at least one processor coupled to the radio frequency front end, and a non-transitory computer-readable medium storing one or more instruction sets configured to manipulate the at least one processor to perform the above-described method.

[0011] According to another aspect, a computer-implemented method in a data receiving device includes: receiving a first transmission wirelessly transmitted from a data transmitting device at a receiver neural network of the data receiving device; processing the first transmission at the receiver neural network of the data receiving device to attempt to recover a first data block; and generating a first transmission feedback signal representing the attempt to recover the first data block. The method further includes generating a second transmission based on the first transmission feedback signal at a transmitter neural network of the data receiving device; and wirelessly transmitting the second transmission to the data transmitting device. In some embodiments, the first transmission feedback signal represents a failed attempt to recover the first data block, and the method further includes: receiving a third transmission wirelessly transmitted from the data transmitting device at the receiver neural network; processing the third transmission at the receiver neural network of the data receiving device to attempt to recover at least a portion of the first data block; generating a second transmission feedback signal representing the attempt to recover at least a portion of the first data block; and generating a fourth transmission for wireless communication to the data transmitting device at the transmitter neural network based on the second transmission feedback signal.

[0012] The method may further include jointly training the transmitter neural network and the receiver neural network using one or more training datasets in conjunction with at least one neural network of the data generating device. In some embodiments, each of the transmitter neural network and the receiver neural network has multiple neural network architecture configurations, each neural network architecture configuration being associated with a different corresponding scheduling authorization type among multiple scheduling authorization types, and jointly training the transmitter neural network and the receiver neural network includes individually training many of the multiple neural network architecture configurations. Furthermore, the method may include determining a current scheduling authorization type implemented for the data receiving device, and configuring the transmitter neural network and the receiver neural network to implement one of the multiple neural network architecture configurations associated with the current scheduling authorization type. The current scheduling authorization type may include: an authorization to transmit control information only, an authorization to transmit user plane data only, or an authorization to transmit both control information and user plane data. In some embodiments, the current scheduling authorization type includes an authorization to transmit user plane data only, and configuring the receiver neural network to implement the neural network architecture configuration includes configuring the receiver neural network to implement a neural network architecture configuration trained for processing only the transmission of user plane data. In other embodiments, the current scheduling authorization type includes authorization for both transmission control information and user plane data, and configuring the receiver neural network to implement a neural network architecture configuration includes configuring the receiver neural network to implement a neural network architecture configuration trained for processing the transmission of both representation control information and user plane data.

[0013] The first transmission feedback signal can contain either a binary signal or a differentiable signal. The receiver neural network and the transmitter neural network can both contain deep neural networks.

[0014] The current scheduling authorization type can include authorization to transmit only user plane data; and configuring the receiver neural network to implement a neural network architecture configuration can include configuring the receiver neural network to implement a neural network architecture configuration trained for processing only the transmission of user plane data. Alternatively, the current scheduling authorization type can include authorization to transmit both control information and user plane data, and configuring the receiver neural network to implement a neural network architecture configuration can include configuring the receiver neural network to implement a neural network architecture configuration trained for processing both control information and user plane data transmissions.

[0015] According to another aspect, a data receiving device includes a radio frequency front end, at least one processor coupled to the radio frequency front end, and at least one storage component storing an instruction set configured to manipulate the at least one processor to perform the above-described method. Attached Figure Description

[0016] Those skilled in the art will better understand this disclosure by referring to the accompanying drawings, and many features and advantages of this disclosure will be apparent. The same reference numerals are used in different drawings to indicate similar or identical items.

[0017] Figure 1 This is a diagram illustrating an example wireless system employing a neural network-based retransmission scheme according to some embodiments.

[0018] Figure 2 This is a diagram illustrating an example configuration of a base station and a UE implementing a neural network-based transmitter and receiver processing path according to some embodiments.

[0019] Figure 3 This illustrates training and configuration according to some embodiments. Figure 2 A diagram showing an example configuration of the base station and the UE's core network server.

[0020] Figure 4 This is a diagram illustrating a machine learning module employing at least one neural network for a neural network-based transmission feedback scheme, according to some embodiments.

[0021] Figure 5 This is a diagram showing the replacement of individual functional blocks of the transmitter and receiver processing paths according to some embodiments with corresponding processing modules employing neural networks.

[0022] Figure 6 This is a diagram illustrating an example implementation of a neural network manager for a base station and a UE according to some embodiments.

[0023] Figure 7 This is a flowchart illustrating an example method for the joint training and initial configuration of a neural network for a base station and a UE, according to some embodiments.

[0024] Figure 8 This is a diagram illustrating an embodiment of the transmission (TX) and reception (RX) processing modules of a data transmission apparatus and a data reception apparatus according to some embodiments.

[0025] Figure 9 This is a diagram illustrating the operation of a data transmitting device and a data receiving device for transmitting data blocks and soft transmission feedback generated by a neural network, according to some embodiments. Detailed Implementation

[0026] Traditional wireless communication networks employ transmitter and receiver processing paths with complex functionalities. Typically, each processing block within a path is designed, tested, and implemented relatively independently of the others. The processing paths are then integrated and further tested and tuned. As described herein, it is possible to bypass much of the design, testing, and implementation work of the transmission or reception processing paths by using neural networks (often referred to as artificial neural networks) to replace some or all of the individual blocks of the processing paths. In this approach, neural networks on the transmission and reception processing paths of a base station and one or more wirelessly connected UEs can be jointly trained to provide functionality similar to one or more conventional individual processing blocks in the corresponding paths. Furthermore, these neural networks can be dynamically reconfigured during operation to adapt to changing operating conditions by modifying parameters such as coefficients, layer size and connections, kernel size, and other configurations.

[0027] See below for reference Figure 1-9 Specifically, a functionality traditionally implemented by a separate hard-coded design is the transmission feedback scheme between the data transmitting and receiving devices, which can be entirely or partially replaced by a neural network-based configuration. In some embodiments, the base station and the UE each employ a transmitter (TX) processing module and a receiver (RX) processing module, wherein the base station's TX processing module communicates operationally with the UE's RX processing module, and the UE's TX processing module communicates operationally with the base station's RX processing module. Each processing module implements at least one neural network, such as a deep neural network (DNN) (i.e., a neural network with at least one hidden layer), and the neural networks of the base station and the UE's TX and RX processing modules are trained jointly or separately using one or more training datasets to provide data retransmission functionality, such as encoding and decoding, modulation and demodulation, in addition to various other transmission / reception functionalities.

[0028] As a general overview of data retransmission functionality, in at least one embodiment, the TX processing module of a data transmitting device (e.g., a base station for the purposes of this overview) receives a first data block as input and generates a first transmission from at least this input for communication to a data receiving device (e.g., a UE for the purposes of this overview). The RX processing module of the data receiving device processes this transmission to generate a transmission feedback signal representing a differentiable signal or other soft signal indicating the recovery status of the first data block, such as the probability or likelihood of successfully recovering the first data block. The transmission feedback signal is provided as input to the TX processing module of the data receiving device, which processes the transmission feedback signal to generate a second transmission for communication to the data transmitting device.

[0029] The RX processing module of the data transmitting device processes the second transmission to generate a retransmission control signal. This retransmission control can be implemented as a binary value (e.g., "0" for NACK and "1" for ACK) or a non-linear activation function (e.g., a sigmoid activation function or a hyperbolic tangent (tanh) activation function) or other soft signal. The TX processing module of the data transmitting device then receives this retransmission control signal, a first data block (previously transmitted data block), and a second data block (data not yet transmitted) as inputs, and processes these inputs to generate a third transmission for communication to the data receiving device. In processing these inputs, the TX processing module uses the retransmission control signal to determine whether to retransmit the first data block, and, if the retransmission input signal is an activation function, further identifies portions(s) of the first data block to be retransmitted, and includes the first data block or identified portions of the first data block in the third transmission. If the retransmission input signal reflects that the first data block has been fully recovered at the data receiving device, the TX processing module incorporates the second data block into the third transmission.

[0030] The RX processing module of the data receiving device receives and processes the third transmission. If the third transmission includes retransmissions of some or all of the first data block, this processing may include an attempt to extract a portion or all of the first data block. If the third transmission includes a new transmission containing a second data block, this processing may include an attempt to extract the second data block. In either case, the RX processing module of the data receiving device generates a transmission feedback signal, which is provided as input to the TX processing module of the data receiving device for incorporation into another transmission in the same manner as described above for processing at the data transmitting device.

[0031] Figure 1The following diagram illustrates downlink (DL) and uplink (UL) operation of an example wireless communication network 100 employing a neural network-based transmission feedback scheme according to some embodiments. As shown, the wireless communication network 100 includes a core network 102 coupled to one or more wide area networks (WANs) 104 or other packet data networks (PDNs) such as the Internet. The core network 102 is also connected to one or more edge networks 106 via one or more backhaul networks (not shown). Each edge network 106 (generally referred to as a “cell” or “radio access network” (RAN)) represents a corresponding fixed or mobile coverage area and includes at least one base station 108 that wirelessly communicates with one or more UEs, such as UE 110, via radio frequency (RF) signaling using one or more applicable radio access technologies (RATs) specified by one or more communication protocols or standards. Thus, the base station 108 serves as a wireless interface between UE 110 and various networks and services provided by the core network 102 and other networks such as packet-switched (PS) data services, circuit-switched (CS) services, etc. Traditionally, signaling communication from base station 108 to UE110 is called "downlink" or "DL", while signaling communication from UE110 to base station 108 is called "uplink" or "UL".

[0032] Base station 108 can employ any of a variety of RATs, such as NodeB (or Base Transceiver Station (BTS)) operation as a Universal Mobile Telecommunications System (UMTS) RAT (also known as "3G"), Enhanced NodeB (eNodeB) operation as a 3GPP Long Term Evolution (LTE) RAT, and 5G NodeB ("gNB") operation as a 3GPP 5th Generation (5G) New Radio (NR) RAT. UE 110 can further implement any of a variety of electronic devices operable to communicate with base station 108 via a suitable RAT, including, for example, mobile cellular phones, cellular-enabled tablets or laptops, desktop computers, cellular-enabled video game systems, servers, cellular-enabled devices, cellular-enabled car communication systems, etc.

[0033] Information communication on the air interface formed between base station 108 and UE 110 takes the form of RF signals representing both control plane signaling and user data plane signaling. However, due to the relatively high frequencies and relatively tight timing margins typically used, RF signaling is susceptible to attenuation and interference, and may therefore frequently lead to failed attempts to extract at least part of the information from the received signal. As mentioned above, conventional wireless networks attempt to mitigate the risk of signal corruption by using a hard-coded transmission feedback process, in which the data receiving device sends an ACK / NACK signal back to the data transmitting device to notify the data transmitting device whether the transmitted information has been correctly extracted, or whether some part of the transmitted information has been corrupted and therefore needs to be retransmitted. However, as mentioned above, conventional transmission feedback methods are hard-coded and fixed, requiring significant design, testing, and implementation work, and are generally difficult to modify or otherwise upgrade using components already deployed in the field.

[0034] Therefore, in some embodiments, both base station 108 and UE 110 implement transmitter (TX) and receiver (RX) processing paths that integrate one or more neural networks (NNs), which can be jointly trained and trained or otherwise configured to provide data retransmission from the data transmitting device based on soft transmission feedback from the data receiving device. For illustration, Figure 1 View 101 depicts an overview of the downlink transmission process using soft transmission feedback. Base station 108, acting as a data transmitting device, uses its NN-based TX processing path to generate a transmission 112 incorporating DL data intended for use by UE 110, acting as a data receiving device, and then wirelessly transmits transmission 112 to UE 110. UE 110's NN-based RX processing path processes transmission 112 and generates a soft transmission feedback signal, which is provided as input to UE 110's NN-based feedback path, which in turn generates a control signal 114 (referred to in the figure as "soft ACK / NACK") to be transmitted back to base station 108. Base station 108's NN-based RX processing path processes the feedback of control signal 114 and generates a retransmission control signal, which base station 108's TX processing path then uses to determine whether some or all of the DL data should be selectively incorporated into the upcoming transmission 116, i.e., whether the upcoming transmission 116 should be at least partially implemented as a retransmission of some or all of the previously transmitted DL data.

[0035] like Figure 1As shown in view 103, uplink transmissions are handled in a similar manner, but UE 110 acts as the data transmitter and base station 108 acts as the data receiver. In this case, UE 110, as the data transmitter, uses its NN-based TX processing path to generate transmission 112, which incorporates UL data intended for use by base station 108, as the data receiver, and then wirelessly transmits transmission 122 to base station 108. Base station 108's NN-based feedback path processes transmission 122 and generates a soft transmission feedback signal, which is provided to base station 108's NN-based feedback path as input, which in turn generates a control signal 124 (also referred to as "soft ACK / NACK" in the figure) transmitted back to UE 110. The NN-based RX processing path of UE 110 processes the control signal 124 and generates a retransmission control signal. The TX processing path of UE 110 then uses the retransmission information signal to determine whether to selectively incorporate some or all of the UL data into the upcoming transmission 126, that is, whether to implement the upcoming transmission 126 at least in part as a retransmission of some or all of the previously transmitted DL data.

[0036] Figure 2 Example hardware configurations of UE 110 and base station 108 according to some embodiments are shown. It should be noted that the depicted hardware configurations represent the processing and communication components most directly related to the neural network-based transmission feedback process described herein, and certain components that are well understood to be frequently implemented in such electronic devices such as displays, user input / output (I / O) devices, power supplies, etc., are omitted.

[0037] In the depicted configuration, UE 110 includes an array 202 of one or more antennas 203, an RF front-end 204, and one or more wireless modems that implement corresponding cellular protocols for RF-based communication with base station 108 and other base stations such as LTE modem 206 and 5G NR modem 208. The RF front-end 204 operates to conduct signals between modems 206, 208 and array 202 to facilitate various types of wireless communication. Antennas 203 can comprise an array of multiple antennas configured to be similar or different from each other and can be tuned to one or more frequency bands associated with a corresponding RAT.

[0038] UE 110 also includes one or more processors 210 and one or more non-transitory computer-readable media 212. The one or more processors 210 may include, for example, one or more central processing units (CPUs), graphics processing units (GPUs), artificial intelligence (AI) accelerators, or other application-specific integrated circuits (ASICs). For illustration, processor 210 may include an application processor (AP) used by UE 110 to execute operating systems and various user-level software applications, as well as one or more processors used by modems 206, 208. Computer-readable media 212 may include any of a variety of media used by electronic devices to store data and / or executable instructions, such as random access memory (RAM), read-only memory (ROM), cache, flash memory, solid-state drives (SSDs), or other mass storage devices. For ease of illustration and brevity, and considering the frequent use of system memory or other memory to store data and instructions for execution by processor 210, computer-readable media 212 is referred to herein as "memory 212," but it should be understood that unless otherwise stated, the reference to "memory 212" should also apply to other types of storage media.

[0039] One or more memories 212 of UE 110 are used to store one or more executable software instruction sets and associated data sets, which manipulate one or more processors 210 and other components of UE 110 to perform various functions described herein and belonging to UE 110. The executable software instruction sets include, for example, an operating system (OS) and various drivers (not shown), various software applications (not shown), and a UE neural network manager 214, which implements one or more neural networks for UE 110, such as those used in the transmitter and receiver processing paths described in detail below. The data stored in the one or more memories 212 includes, for example, UE device data 216 and one or more neural network architecture configurations 218. UE device data 216 represents, for example, user data, multimedia data, beamforming codebooks, software application configuration information, etc. The one or more neural network architecture configurations 218 include one or more data structures containing data and other information representing the corresponding architecture and / or parameter configurations used by UE neural network manager 214 to form the corresponding neural network of UE 110. The information included in neural network architecture configuration 218 includes parameters such as specifying fully connected layer neural network architecture, convolutional layer neural network architecture, recurrent neural network layers, the number of connected hidden neural network layers, input layer architecture, output layer architecture, the number of nodes utilized by the neural network, the coefficients utilized by the neural network (e.g., weights and biases), kernel parameters, the number of filters utilized by the neural network, the stride / pooling configuration utilized by the neural network, the activation function of each neural network layer, the interconnections between neural network layers, neural network layers to be skipped, etc. Therefore, neural network architecture configuration 218 includes any combination of NN forming configuration elements (e.g., architecture and / or parameter configurations) that can be used to create NN forming configurations that define and / or form DNNs (e.g., combinations of one or more NN forming configuration elements).

[0040] Turning to the hardware configuration of base station 108, it should be noted that although the figures shown represent an implementation of base station 108 as a single network node (e.g., gNB), the functionality of base station 108 and therefore its hardware components can alternatively be distributed across multiple network nodes or devices, and can be distributed in a manner that performs the functions described herein. Like UE 110, base station 108 includes an array 222 of one or more antennas 223 for communicating with UE 110, an RF front end 224, and one or more wireless modems such as LTE modem 226 and 5G NR modem 228, as well as one or more processors 230 and one or more non-transitory computer-readable storage media 232 (for brevity, computer-readable media 232 is referred to herein as "memory 232," similar to memory 212 in UE 110). These components operate in a similar manner to those described above with reference to the corresponding components of UE 110.

[0041] One or more memories 232 of base station 108 store one or more executable software instruction sets and associated data that manipulate one or more processors 210 and other components of base station 108 to perform various functions described herein and attributable to base station 108. The executable software instruction sets include, for example, an operating system (OS) and various drivers (not shown), various software applications (not shown), a base station manager 233, and a base station neural network manager 234. Base station manager 233 configures modems 226, 228 for communication with UE 110 and with a core network such as core network 102. Base station neural network manager 234 implements one or more neural networks for base station 108, such as neural networks used in transmitter and receiver processing paths, as described in detail below.

[0042] The data stored in one or more memories 232 of base station 108 includes, for example, base station data 236 and one or more neural network architecture configurations 238. Base station data 236 represents, for example, network scheduling data, radio resource management data, beamforming codebooks, software application configuration information, etc. One or more neural network architecture configurations 238 include one or more data structures containing data and other information representing the corresponding architecture and / or parameter configurations used by base station neural network manager 234 to form the corresponding neural network of base station 108. Similar to neural network architecture configuration 218 of UE 110, the information included in neural network architecture configuration 238 includes, for example, parameters specifying fully connected layer neural network architecture, convolutional layer neural network architecture, recurrent neural network layers, the number of connected hidden neural network layers, input layer architecture, output layer architecture, the number of nodes utilized by the neural network, coefficients utilized by the neural network, kernel parameters, the number of filters utilized by the neural network, the stride / pooling configuration utilized by the neural network, the activation function of each neural network layer, the interconnections between neural network layers, neural network layers to be skipped, etc. Therefore, neural network architecture configuration 238 contains any combination of NN forming configuration elements that can be used to create NN forming configurations that define and / or form DNNs or other neural networks.

[0043] In at least one embodiment, the software stored in memory 232 further includes a training module 240, which operates to train one or more neural networks implemented at base station 108 or UE 110 using one or more input datasets. Training can be performed for various purposes, such as processing communications transmitted through a wireless communication system, such as encoding downlink communications, modulating downlink communications, demodulating downlink communications, decoding downlink communications, encoding uplink communications, modulating uplink communications, demodulating uplink communications, and decoding uplink communications. Training can be included in both offline (i.e., when not actively involved in processing communications) and / or online (i.e., when actively involved in processing information).

[0044] In some embodiments, base station 108 further includes an inter-base station interface 242, such as an Xn or X2 interface, which base station manager 233 configures to exchange user plane, control plane, and other information between other base stations and manages communication between base station 108 and UE 110. Base station 108 may also include a core network interface 244, which base station manager 233 configures to exchange user plane, control plane, and other information with core network functions and / or entities.

[0045] Figure 3 It shows, for example Figure 1An example hardware configuration of the core network server 300 implemented in the core network 102 is described below. It should be noted that the depicted hardware configuration represents the processing and communication components most directly related to the neural network-based transmission feedback process described herein, and certain components that are well understood to be frequently implemented in servers within the core network are omitted. In the depicted configuration, the core network server 300 provides all or part of the functions, entities, services, and / or gateways in the core network 102. Each function, entity, and / or service of the core network 102 can be provided as a service in the core network 102, distributed across multiple servers or embodied on a dedicated server. For example, the core network server 300 can provide all or part of the services or functions such as User Plane Function (UPF), Access and Mobility Management Function (AMF), Serving Gateway (S-GW), Packet Data Network Gateway (P-GW), Mobility Management Entity (MME), Evolved Packet Data Gateway (ePDG), etc. The core network server 300 is shown as being embodied in a single server, which includes one or more processors 304 and one or more non-transitory computer-readable storage media 306 (also referred to herein as "memory 306").

[0046] Memory 306 stores one or more core network applications 310 that implement various core network functions, such as the aforementioned UPF, AMF, S-GW, P-GW, MME, and ePDG. Memory 306 also includes a core network neural network manager 312 that manages one or more neural network architecture configurations 318 for handling communications exchanged between UE 110 and base station 108. In some embodiments, the core network neural network manager 312 analyzes various parameters, such as current signal channel conditions (e.g., reported by the base station, access point, and UE), capabilities at the base station (e.g., antenna configuration, cell configuration, MIMO capability, radio capability, and processing capability), and capabilities of the UE (e.g., antenna configuration, MIMO capability, radio capability, and processing capability). For example, each base station 108 in network 100 acquires various parameters during communication with a UE camped on an associated cell and forwards these parameters to the core network neural network manager 312. The core network neural network manager 312 selects a neural network architecture configuration 318 based on these parameters to improve the accuracy of the neural network processing the communication. Improved accuracy means that the accuracy of the output generated by the neural network is improved relative to a neural network configured with another neural network architecture configuration 318, such as fewer bit errors. The core network neural network manager 312 then transmits the selected NN forming configuration to one or both of the base station 108 and the UE 110. Furthermore, the core network neural network manager 312 is capable of receiving feedback from the UE and / or BS from the base station 108 and selecting an updated neural network architecture configuration 318 based on the feedback.

[0047] In some embodiments, in addition to the training module 240 of base station 108 ( Figure 1 In addition to or instead of the training provided using the techniques described below, memory 306 also stores training modules 314 for training and retraining one or more neural network architecture configurations 318. Core network server 302 can also be a core network interface 320 for communicating user plane, control plane, and other information with other functions or entities in core network 102, base station 108, or UE 110. For illustration, core network server 302 uses core network interface 320 to transmit neural network architecture configurations 318 to base station 108.

[0048] Figure 4An example machine learning (ML) module 400 for implementing neural networks according to some embodiments is shown. As described herein, one or both of base station 108 and UE 110 implement one or more DNNs or other neural networks in one or both of the transmission processing paths or reception processing paths for processing incoming and outgoing wireless communications. Therefore, ML module 400 illustrates an example module for implementing one or more of these neural networks.

[0049] In the depicted example, the ML module 400 implements at least one deep neural network (DNN) 402, which has multiple sets of connected nodes (e.g., neurons and meta- or perceptrons) organized into three or more layers. The nodes between layers can be configured in various ways, such as partially connected configurations where a first subset of nodes in the first layer is connected to a second subset of nodes in the second layer; fully connected configurations where each node in the first layer is connected to every node in the second layer, etc. 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. Both neuron and perceptron nodes are capable of generating output information based on adaptive learning using various algorithms. Using the DNN 402, the ML module 400 performs various types of analyses, including unilinear regression, multiple linear regression, logistic regression, stepwise regression, binary classification, multi-class classification, multivariate adaptive regression splines, local estimation scatter plot smoothing, etc.

[0050] In some implementations, the ML module 400 learns adaptively based on supervised learning. In supervised learning, the ML module receives various types of input data as training data. The ML module 400 processes the training data to learn how to map the inputs to the desired output. As an example, the ML module 400 receives digital samples of a signal as input data and learns how to map the signal samples to binary data reflecting information embedded within the signal. As another example, the ML module 400 receives binary data as input data and learns how to map the binary data to digital samples of a signal having binary data embedded within the signal. Furthermore, as another example and described in more detail below, the ML module 400 receives signal samples as input data and learns how to generate non-binary or "soft" transmission feedback signals, or conversely, receives data representing soft transmission feedback signals, and learns from this input whether data needs to be retransmitted, and in some cases, which parts of the data need to be transmitted. During the training process, the ML module 400 uses labeled or known data as input to the DNN 402. The DNN 402 uses nodes to analyze the input and generate corresponding outputs. The ML module 400 compares the corresponding output with the ground truth data and adjusts the algorithm implemented by the nodes to improve the accuracy of the output data. Subsequently, the DNN 402 applies the adjusted algorithm to the unlabeled input data to generate the corresponding output data.

[0051] The ML module 400 uses one or both of statistical analysis and adaptive learning to map inputs to outputs. For example, the ML module 400 uses characteristics learned from training data to associate unknown inputs with outputs that are statistically likely within a threshold range or value. This allows the ML module 400 to receive complex inputs and identify corresponding outputs. Some implementations train the ML module 400 on communication characteristics transmitted over a wireless communication system (e.g., time / frequency interleaving, time / frequency deinterleaving, convolutional coding, convolutional decoding, power levels, channel equalization, inter-symbol interference, quadrature amplitude modulation / demodulation, frequency division multiplexing / demultiplexing, and transport channel characteristics). This allows the trained ML module 400 to receive samples of signals as input, such as samples of downlink signals received at the UE, and to recover information from the downlink signals, such as binary data embedded in the downlink signals.

[0052] Furthermore, in some embodiments, the communication characteristics used to train the ML module 400 include aspects related to providing transmission feedback, such as ACK / NACK or HARQ type feedback, such that the data receiving device(s) ML module 400 is trained to provide probabilistic or other non-binary transmission feedback to the data transmitting device, indicating that the transmission of data was successfully recovered in the transmission from the data transmitting device to the data receiving device, and further, the data transmitting device(s) ML module 400 is trained to process such transmission feedback when deciding whether to include some or all of the previously transmitted data in the next transmission.

[0053] In the depicted example, DNN 402 comprises 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. Each layer has an arbitrary number of nodes, wherein the number of nodes between layers can be the same or different. That is, the input layer 404 can have the same and / or different number of nodes as the output layer 406, and the output layer 406 can have the same and / or different number of nodes as the one or more hidden layers 408, and so on.

[0054] Node 410 corresponds to one of several nodes included in input layer 404, where these nodes perform individual, independent computations. As further described, a node receives input data and processes the input data using one or more algorithms to produce output data. Typically, the algorithm includes weights and / or coefficients that change based on adaptive learning. Thus, the weights and / or coefficients reflect information learned by the neural network. In some cases, each node is able to determine whether to pass the processed input data to one or more next nodes. For illustration, after processing the input data, node 410 is able to determine whether to pass the processed input data to one or both of nodes 412 and 414 in hidden layer 408. Alternatively or additionally, node 410 passes the processed input data to nodes based on the layer connection architecture. This process can be repeated across multiple layers until DNN 402 generates output using nodes of output layer 406 (e.g., node 416).

[0055] Neural networks can also employ various architectures that determine which nodes are connected within the network, how data is advanced and / or retained within the network, what weights and coefficients are used to process the input data, and how the data is processed. These various factors collectively describe neural network architecture configurations, such as the neural network architecture configurations 218, 238, and 318 briefly described above. For illustration, recurrent neural networks, such as Long Short-Term Memory (LSTM) neural networks, form loops between node connections to retain information from the first part of the input data sequence. The recurrent neural network then uses the retained information for subsequent parts of the input data sequence. As another example, feedforward neural networks pass information to the forward connections without forming loops to retain information. Although described in the context of node connections, it should be understood that neural network architecture configurations can encompass a variety of parameter configurations that influence how a DNN 402 or other neural networks process input data.

[0056] The neural network architecture configuration can be characterized by various architecture and / or parameter configurations. For illustration, consider an example of a DNN 402 implementing a convolutional neural network (CNN). Typically, a convolutional neural network corresponds to a DNN where each layer uses convolution operations to process data to filter the input data. Therefore, the CNN architecture configuration can be characterized by, for example, pooling parameters, kernel parameters, weights, and / or layer parameters.

[0057] Pooling parameters correspond to the parameters of the pooling layer within a specified convolutional neural network that reduces the dimensionality of the input data. For illustration, a pooling layer can merge the outputs of nodes in a first layer into the inputs of nodes in a second layer. Alternatively or additionally, pooling parameters specify how and where the neural network pools data in the data processing layer. For example, a pooling parameter indicating "max pooling" configures the neural network to pool by selecting the maximum value from the data packets generated by the nodes in the first layer, and using that maximum value as the input to a single node in the second layer. A pooling parameter indicating "average pooling" configures the neural network to generate an average value from the data packets generated by the nodes in the first layer, and uses that average value as the input to a single node in the second layer.

[0058] Kernel parameters indicate the size of the filter (e.g., width and height) used to process the input data. Alternatively or additionally, kernel parameters specify a kernel method used to filter and process the input data. For example, a support vector machine corresponds to a kernel method that uses 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 filter size and / or the type of kernel method to be applied to a neural network.

[0059] The weight parameters specify the weights and biases used by the algorithm within a node to classify the input data. In some implementations, the weights and biases are learned parameter configurations, such as those generated from the training data.

[0060] Layer parameters specify layer connections and / or layer types, such as fully connected layer types indicating which nodes in the first layer (e.g., output layer 406) are connected to each node in the second layer (e.g., hidden layer 408), partially connected layer types indicating which nodes in the first layer are disconnected from the second layer, activation layer types indicating which filters and / or layers are activated within the neural network, etc. Alternatively or additionally, layer parameters specify the type of node layer, such as normalization layer type, convolutional layer type, pooling layer type, etc.

[0061] Although described in the context of pooling parameters, kernel parameters, weight parameters, and layer parameters, it should be understood that other parameter configurations can be used to form DNNs consistent with the guidelines presented in this paper. Therefore, neural network architecture configurations can include any suitable type of configuration parameters that can be applied to a DNN, influencing how the DNN processes input data to generate output data.

[0062] In some embodiments, the configuration of the ML module 400 is based on the current operating environment. For illustration, consider an ML module trained to generate binary data from digital samples of a signal. The transmission environment frequently modifies the characteristics of the signal traveling through it. The transmission environment often changes, affecting how the environment modifies the signal. For example, a first transmission environment modifies the signal in a first manner, while a second transmission environment modifies the signal in a different manner than the first. These differences affect the accuracy of the output generated by the ML module 400. For example, a DNN 402 configured to process communication transmitted in the first transmission environment may produce errors (e.g., bit errors exceeding a threshold) when processing communication transmitted in the second transmission environment.

[0063] Therefore, in some embodiments, the apparatus implementing ML module 400 generates and stores different neural network architecture configurations for different transmission environments or configurations. For this purpose, one or both of base station 108 or core network server 302 can train ML module 400 using any combination of base station neural network manager 234, training module 240, core network neural network manager 312, and / or training module 314. Training can be performed offline without active communication exchange or online during active communication exchange. For example, base station 108 or core network server 302 can mathematically generate training data, access files storing training data, obtain real-world communication data, etc. Then, base station 108 or core network server 302 extracts and stores various learned neural network architecture configurations for subsequent use. Some implementations store input characteristics along with each neural network architecture configuration, whereby the input characteristics describe various characteristics of the transmission environment corresponding to the respective neural network architecture configuration. In embodiments, the neural network manager selects a neural network architecture configuration by matching the current transmission environment and / or the current operating environment with the input characteristics.

[0064] Now go to Figure 5 , Figure 5 Example methods are illustrated for using ML modules, particularly DNNs or other neural networks, in base station 108 and UE 110 to replace traditional hard-coded or fixed signal processing paths, according to some embodiments. As described above, wireless communication systems typically include a variety of complex components and functions, including converting digital data into RF signals at a data transmitting device, recovering digital data from received RF signals at a data receiving device, and providing transmission feedback to correct errors in data transmission and recovery processes by retransmitting at least a portion of the original data. This functionality can typically be divided into two processing paths at each of the data transmitting and receiving devices: a transmission (TX) processing path for converting data and other inputs into RF signals for output, and a reception (RX) processing path for converting RF signals into corresponding data and other outputs, some of which can then be used as inputs to a co-location control signal path (e.g., as transmission feedback).

[0065] For demonstration purposes, Figure 5 Example figures 500 and 502 are depicted, each of which represents a representation by... Figure 1The example processing path used by the data transmitting and receiving devices in network 100 is as follows: For downlink transmission, base station 108 is the data transmitting device, and UE 110 is the data receiving device. Conversely, for uplink transmission, UE 110 is the data transmitting device, and base station 108 is the data receiving device. Therefore, which of the two devices constitutes the data transmitting device and which constitutes the data receiving device depends on which of the two devices is the originating source of the data being transmitted.

[0066] Figure 500 illustrates the transmitter block 504 and receiver block 506 conventionally implemented at the data transmitting device and data receiving device, respectively. The transmitter block 504 contains a transmitter processing path that proceeds from top to bottom. The transmitter processing path begins with the input data entering the coding stage, followed by the modulation stage, and then the RF analog transmission (TX) stage.

[0067] A coding level can encompass any type and number of coding levels used by a device to transmit data through a wireless communication system. For illustration, an example coding level receives binary data as input and processes the binary data using various coding algorithms to append information such as frame information to the binary data. Alternatively or additionally, the coding level may transform the binary data, for example, by applying forward error correction, which adds redundancy to aid in information recovery at the receiver. As another example, the coding level converts binary data into symbols.

[0068] The modulation stage receives the output generated by the coding stage as input and embeds the input into the signal. For example, the modulation stage generates digital samples of the signal(s) embedded together with the input from the coding stage. Therefore, in transmitter block 504, the coding stage and modulation stage represent higher-level transmitter processing paths, which typically contain lower-level complex functions such as convolutional coding, 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 receiver block 506.

[0069] Receiver block 506 implements a receiver processing path that performs complementary processing relative to the transmitter processing path of transmitter block 504. The receiver processing path shown 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. The RF analog RX stage receives the signal transmitted by transmitter block 504 and generates the input 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 the data embedded in the signal (e.g., the data embedded by the modulation stage of transmitter block 504). The demodulation stage then recovers the symbol and / or binary data.

[0070] The decoder 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 decoder stage extracts payload data from frames and / or time slots, etc. Therefore, the decoder stage generates the recovered information.

[0071] Since transmitter block 504 transmits data and other information to receiver block 506 via RF signals, and these RF signals are subject to interference and other disturbances that may significantly modify the RF signals, the "information" recovered by receiver block 506 from the modified RF signals may not accurately reflect the transmitted information. The transmitter processing path and receiver processing path together implement a transmission feedback scheme, wherein the encoding stage of the transmitter processing path encodes the data to be transmitted using forward error correction (FEC), parity bits, or other error detection and / or correction codes; the decoding stage of the receiver processing path attempts to decode the encoded data recovered from the received RF signaling; and the retransmission stage of the receiver processing path provides transmission feedback, such as ACK / NACK signals or HARQ feedback signals, for wireless transmission back to transmitter block 504 at the data receiving device using the corresponding control signal processing path. The receiver processing path at the data transmission device recovers the control signal feedback, thereby the retransmission stage of the transmission processing path of the data transmission device uses the transmission feedback to guide the input, encoding, modulation and / or RF analog stages to retransmit at least a portion of the previous data when the transmission feedback indicates that the previous transmission was not correctly received at the receiver block 506.

[0072] As noted, for clarity, the transmitter and receiver processing paths illustrated by transmitter block 504 and receiver block 506 have been simplified, and these processing paths can contain multiple complex modules. These complex modules are specific to particular functions or conditions. For example, consider a receiver processing path for processing Orthogonal Frequency Division Modulation (OFDM) transmissions. To recover information from OFDM transmissions, a receiver block typically contains multiple processing blocks, each dedicated to a specific function, such as an equalization block to correct distortion in the received signal, a channel estimation block to estimate transmission channel properties to identify effects such as scattering, power attenuation, etc., on the transmission. At high frequencies such as 6 GHz or terahertz (THz) bands, these blocks can be resource-intensive and require complex hardware implementations. Furthermore, implementing blocks that generate outputs with accuracy within desired thresholds typically requires more specific and less flexible components. To illustrate, an equalization block that works for signals in the 6 GHz band may not perform with the same accuracy in other bands, thus requiring different equalization blocks for different bands and increasing the complexity of the corresponding devices.

[0073] In at least some embodiments, the transmitter and receiver processing paths employ one or more DNNs or other neural networks (using...). Figure 4Instead of implementing the transmitter and receiver processing paths as chains of independent and complex functional blocks requiring separate design, testing, and implementation work, an instance of the ML module 400 is used. Block diagram 502 illustrates this approach, where transmitter block 508 includes one or more DNNs 510 in the transmitter processing path, and receiver block 512 includes one or more DNNs 514 in the receiver processing path, where each DNN 510, 514 provides functionality for at least a portion of at least one conventionally separate block of the corresponding processing path. For illustration, in the depicted example, one or more DNNs 510 in transmitter block 508 correspond to the coding, modulation, and retransmission levels of transmitter block 504. However, it should be understood that DNNs 510 are capable of performing any high-level or low-level operations found within the transmitter processing chain. As an example, a first DNN 510 performs low-level transmitter-side forward error correction, a second DNN 510 performs low-level transmitter-side convolutional coding, and so on. Alternatively or additionally, DNN 510 performs high-level processing, such as full transmitter chain processing corresponding to the coding and modulation stages of transmitter block 508. Similarly, one or more DNNs 514 in receiver block 512 perform functions, for example, those belonging to the demodulation, decoding, and transmission feedback stages. One or more DNNs 514 are capable of performing any high-level or low-level operations found within the receiver processing path, such as low-level receiver-side bit error correction, low-level receiver-side symbol recovery, high-level full receiver chain demodulation and decoding, etc. Thus, DNNs in a wireless communication network can be configured to replace one or both of the high-level or low-level operations in the transmitter and receiver processing paths. As described below, these DNNs can be reconfigured through retraining, which provides greater flexibility and adaptability to the processing path compared to the more specific and less flexible components found in conventional implementations.

[0074] As described above, network devices such as UE 110 and base station 108 in wireless communication can be configured to use one or more DNNs at each network device to process wireless communication switching, wherein each DNN replaces one or more functions (e.g., uplink processing, downlink processing, uplink coding processing, downlink decoding processing, transmission feedback generation, transmission feedback processing, etc.) conventionally implemented by one or more hard-coded or fixed design blocks. For this purpose, Figure 6An example operating environment 600 for implementing a DNN at UE 110 and base station 108 is shown. In the depicted example, the UE neural network manager 214 of UE 110 implements a downlink (DL) receiver (RX) processing module 602 and an uplink (UL) transmitter processing module 604. Similarly, the base station neural network manager 234 implements a DL TX processing module 606 and an UL RX processing module 608. Each of the processing modules 602, 604, 606, and 608 implements one or more DNNs via the implementation of a corresponding ML module, such as those referenced above. Figure 4 The ML module 400 is described by one or more DNNs 402.

[0075] The DL TX processing module 606 and DL RX processing module 602 of base station 108 interoperate to support the DL wireless communication path 610 between base station 108 as a data transmitter and UE 110 as a data receiver, while the UL TX processing module 604 of the UE and the UL RX processing module 608 of base station 108 interoperate to support the UL wireless communication path 612 between UE 110 as a data transmitter and base station 108 as a data receiver. Therefore, one or more DNNs of DL TX processing module 606 are trained to receive DL data and DL control plane information as inputs and generate corresponding outputs to be transmitted via the RF analog stage of base station 108 (see...). Figure 5 The transmitter block 508 transmits the signal as an RF signal. Conversely, one or more DNNs of the DL RX processing module 602 are trained to receive signals from the RF analog stage (see [link to DNN]). Figure 5 The receiver block 512 extracts the output from the transmitted RF signal as input and generates recovered DL data and DL control plane information as output for further processing at UE 110 (or, in some cases, directly applied to the DL control plane information). Similarly, one or more DNNs of the UL TX processing module 604 are trained to receive UL data and UL control plane information as input and generate corresponding outputs for transmission as RF signals via the RF analog stage of UE 110. Conversely, one or more DNNs of the UL RX processing module 608 are trained to receive the output extracted from the transmitted RF signal by the RF analog stage as input and generate recovered UL user plane data and UL control plane information as output for further processing at base station 108.

[0076] During the execution of these uplink and downlink communications, one or more errors may be introduced in the transmission of the RF signal itself, in the formation of the RF signal at the data transmitting device, or in the interpretation of the RF signal at the data receiving device. To facilitate robust recovery from such errors, processing modules 602, 604, 606, and 608 together implement DL and UL transmission feedback loops to allow UE 110 to provide base station 108 with some form of ACK / NACK signal, HARQ signal, or another DL transmission feedback signal indicating whether the most recently transmitted DL data block has been successfully received and recovered, enabling base station 108 to selectively retransmit at least a portion of this DL data block based on the DL transmission feedback signal. Similarly, base station 108 is also allowed to provide UE 110 with some form of ACK / NACK, HARQ, or another UL transmission feedback signal indicating whether the most recently transmitted UL data block has been successfully received and recovered, enabling UE 110 to selectively retransmit at least a portion of this UL data block based on the UL transmission feedback signal.

[0077] To facilitate this robust retransmission process, the base station neural network manager 234 implements a transmission recovery feedback path 614 between the UL RX processing module 608 and the DL TX processing module 606, while the UE neural network manager 214 implements a transmission recovery feedback path 616 between the DL RX processing module 602 and the UL TX processing module 604. When processing transmissions wirelessly transmitted from the corresponding TX processing modules 604, 606, the DNN(s) of the RX processing modules 602, 608 are trained or otherwise configured to generate differentiable or “soft” transmission feedback signals, probabilities, or other possibility representations that enable the RX processing modules to fully recover the data blocks represented in the transmission and provide soft transmission feedback signals to the TX processing modules of the same device via the transmission recovery feedback paths connecting the two processing modules. One or more DNNs of the TX processing modules are then trained or otherwise configured to incorporate the soft transmission feedback signals into control plane transmissions wirelessly transmitted to another device. The RX processing module at another device is trained or otherwise configured to process control plane transmissions to generate retransmission control signals provided to the RX processing module of the same device via a transmission recovery feedback path. When preparing to generate a wireless communication transmission to the corresponding RX processing module of the other device, the DNN(s) of the TX processing module is trained or otherwise configured to process the retransmission control signals to selectively retransmit at least a portion of previously transmitted data blocks by selectively forming outgoing data plane transmissions containing one or more portions of data blocks based on the retransmission control signals. (See below for reference.) Figure 8 and Figure 9The uplink and downlink transmission feedback loops implemented by the DNNs of processing modules 602, 604, 606 and 608 of UE 110 and base station 108 are described in more detail.

[0078] As described above, in various embodiments, DNNs or other neural networks are used to implement certain functions of the transmitter processing path and receiver processing path in each of base station 108 and UE 110. This neural network-based approach provides design flexibility and facilitates efficient updates relative to traditional block-by-block design and textual methods. However, before a DNN can be deployed and put into operation, it is typically trained or otherwise configured to provide suitable outputs for a given set of one or more inputs. For this purpose, Figure 7 An example method 700, according to some embodiments, for initializing a DNN-based processing path for UEs 110 and 108 prior to active operation of wireless communication is illustrated. Note that references... Figure 7 The described order of operations is for illustrative purposes only, and different orders of operations may be performed. Furthermore, one or more operations may be omitted, or one or more additional operations may be included in the method shown.

[0079] Method 700 begins at block 702, wherein the DNNs of processing modules 602, 604 of UE 110 and processing modules 606, 608 of base station 108 are jointly trained. In some embodiments, training module 240 of base station 108 manages the joint training, but in other embodiments, training module 314 of core network server 300 manages the joint training. In any method, such joint training typically includes initializing the bias weights and coefficients of various DNNs with initial values ​​typically chosen pseudo-randomly; inputting a training dataset (e.g., representing known user plane data and control plane information) at the TX processing module of a first device (e.g., base station 108); wirelessly transmitting the resulting output as a transmission to the RX module of a second device (e.g., UE 110); processing the transmission at the RX processing module of the second device; determining the error between the actual and expected output; and backpropagating the error in the DNNs of both the TX processing module of the first device and the RX processing module of the second device, and repeating this process for the next input dataset. This process is repeated until a certain number of training iterations are performed or until a certain minimum error rate is reached. The same process is performed between the TX processing module of the second device and the RX processing module of the first device. Furthermore, as described herein, the RX processing module of one device generates transmission feedback, which is incorporated into a control signal wirelessly transmitted to the data transmitting device, and the RX processing module of the data transmitting device processes this control signal to provide a retransmission control signal to the TX processing module of the data transmitting device. Therefore, the joint training process also includes training the DNNs of the respective processing modules to implement this feedback loop. For example, joint training can include jointly training one or more DNNs of the RX and TX processing modules of one device (e.g., base station 108) with one or more DNNs of the RX and TX processing modules of another device (e.g., UE 110).

[0080] As described above, the joint training process can be performed when the processing modules of base station 108 and UE 110 are offline (i.e., when there is no active communication of control information or user plane data) or when the processing modules of base station 108 and UE 110 are online (i.e., when there is active communication of control information or user plane data). Furthermore, in some embodiments, in certain cases, a subset of DNNs can be trained or retrained while other DNNs remain static, rather than jointly training all DNNs. For illustration, base station neural network manager 234 can detect that a particular processing module, such as UE DL TX processing module 606, is operating inefficiently or incorrectly due to, for example, the presence of an interference source near UE 110, and therefore base station neural network manager 234 can schedule individual retraining of the DNN(multiple) of UE DL TX processing module 606 while keeping the other DNNs of other processing modules 602, 604, and 608 in their current configuration.

[0081] As the result of joint or individual training of the DNNs of base station 108 and UE 110, each DNN has a specific neural network architecture configuration, or a DNN architecture configuration when the implemented neural network is a DNN. This characterizes the architecture and parameters of the corresponding DNN, such as the number of hidden layers, the number of nodes at each layer, the connections between each layer, weights, coefficients, and other bias values ​​implemented at each node.

[0082] Relative to user plane data and control plane transmission feedback information, radio transmissions from base station 108 to UE 110 or from UE 110 to base station 108 can represent, for example, any of three scheduling grant types: radio transmissions containing user plane data but not transmission feedback information (i.e., a grant of transmitting only user plane data as shown in box 705); radio transmissions containing transmission feedback information but not user plane data (i.e., a grant of transmitting only control information as shown in box 707); and radio transmissions containing both user plane data and transmission feedback information (i.e., a grant of transmitting both control information and user plane information as shown in box 709). It should be understood that the resulting configuration of the DNN trained jointly or individually can be at least partially based on which of these three scheduling grant types was implemented during training. Therefore, to better ensure that the DNN-based transmitter and receiver processing paths can operate under any of these three scheduling grant types, in some embodiments, the joint training process of box 702 is repeated for each represented scheduling grant. In other words, as shown in box 704, the DNN is jointly trained using training data representing user data transmission only to generate a DNN architecture configuration specific to user data-only scheduling authorization configuration; the DNN is jointly trained using training data representing feedback information transmission only to generate a DNN architecture configuration specific to transmission feedback-only scheduling authorization configuration; and the DNN is further jointly trained using training data representing transmissions with both user data and transmission feedback to generate a DNN architecture configuration specific to scheduling authorizations provided for both data and transmission feedback.

[0083] With the DNNs trained and one or more DNN architecture configurations determined for each DNN, base station 108 and UE 110 are ready to use the trained DNNs to initiate DL and UL radio transmissions. In embodiments where different DNN architecture configurations have been developed for different operating environments or configurations, such as for the different scheduling license types described above, at block 708, base station neural network manager 234 determines the operating environment that will exist or be configured for one or more upcoming transmissions between base station 108 and UE 110. For example, base station neural network manager 234 is able to determine which scheduling license type will be configured for communication between base station 108 and UE 110. At block 710, base station neural network manager 234 guides the implementation of DNN architecture configurations for the DNNs at base station 108 and UE 110, which are associated with the indicated operating environment, such as the indicated scheduling license type. This can include signaling to UE 110 via, for example, Layer 1, Layer 2, or Layer 3 control signaling between base station 108 and UE 110 an identifier of scheduling authorization type or other operating environment identifier or identifier of specific DNN architecture configuration. Otherwise, if an operating environment-specific DNN architecture configuration is not implemented, the procedures in boxes 706 and 708 can be skipped or otherwise omitted.

[0084] When configuring the DNN using its corresponding DNN architecture configuration, at block 710, base station 108 and UE 110 perform wireless communication via the generation and communication of wireless transmissions using the DNN of the data transmitting device and the corresponding reception and processing of wireless transmissions using the DNN of the data receiving device. See below for reference. Figure 8 and Figure 9 This process, which includes incorporating soft transmission feedback, will be described in more detail.

[0085] Figure 8 An example operating environment 800 employing a DNN-based transmission feedback data transmitting apparatus 802 and a data receiving apparatus 804 according to some embodiments is illustrated. For downlink transmission, the data transmitting apparatus 802 represents base station 108 and the data receiving apparatus 804 represents UE 110. Conversely, for uplink transmission, the data transmitting apparatus 802 represents UE 110 and the data receiving apparatus 804 represents base station 108. Therefore, it should be understood that base station 108 implements both the components of data transmitting apparatus 802 for downlink transmission and the components of data receiving apparatus 804 for uplink transmission, while UE 110 implements both the components of data receiving apparatus 804 for downlink transmission and the components of data transmitting apparatus 802 for uplink transmission.

[0086] As referenced above Figure 6As described, the data transmitting device 802 includes an RX processing module 806 and a TX processing module 808, and the data receiving device 804 includes a TX processing module 810 and an RX processing module 812. Each of the processing modules 806, 808, 810, and 812 includes one or more DNNs (using, for example...) Figure 4 The ML module 400 contains DNNs that implement one or more functions of a corresponding transmitter processing path or receiver processing path, and these DNNs are trained individually or jointly as described above. The data transmission device 802 also includes a delay buffer 814 (e.g., a HARQ buffer) implemented in one or more registers, caches, memories, or other storage elements, and a gating function 816 implemented as one or more processors executing a corresponding instruction set, implemented as hard-coded or programmable logic, or a combination thereof. The delay buffer 814 has an input for receiving a block of data to be transmitted (e.g., input data block 818) and an output for providing the buffered data block as input to the gating function 816. The gating function 816 also includes another input for receiving a retransmission control signal 820 output by the RX processing module 806, and an output for the input layer of at least one DNN connected to the TX processing module 808. Although illustrated separately from the TX processing module 808, in some embodiments, the gating function 816 may be used as part of the TX processing module 806. As described in more detail below, the gating function 816 operates to selectively output a buffered data block or a portion thereof or a representation thereof in the delay buffer 814 as input to one or more DNNs of the TX processing module 808 based on the retransmission control signal 820.

[0087] As a general operational overview, when each data block (input data block 818) is received for transmission, the data block is buffered at delay buffer 814. TX processing module 808 operates to generate transmission 822 for wireless communication to data receiving device 804 based on one or more inputs, including outputs from gating function 816 (in some embodiments, this output depends on retransmission control signal 820 indicating the absence, partial or complete presence of the previous data block currently buffered in delay buffer 814), the current data block to be transmitted, etc. One or more DNNs of TX processing module 808 are configured to process these inputs according to their currently implemented DNN architecture to generate one or more outputs, such as one or more outputs for control plane transmissions, such as Physical Uplink Control Channel (PUCCH) transmissions or Physical Downlink Control Channel (PDCCH) transmissions, and one or more outputs for user data plane transmissions, such as Physical Uplink Shared Channel (PUSCH) or Physical Downlink Shared Channel (PDSCH). These outputs are provided together as a transmission 822 for conversion into RF signals by an RF analog stage (not shown) and for wireless transmission of these RF signals to a data receiving device 804.

[0088] At data receiving device 804, an RF analog stage (not shown) converts the RF signal into one or more digital inputs to the input layer of one or more DNNs of RX processing module 812. The one or more DNNs of RX processing module 812 are configured according to their DNN architecture to process these inputs to generate an output set representing the recovered user plane data and control plane information represented by transmission 822, or representing an attempt to recover this information. However, RF signal corruption or errors may have been introduced at the RF analog stage at data receiving device 804 or in subsequent input processing. Therefore, one or more DNNs of RX processing module 812 are trained or otherwise configured to include a transmission feedback signal 824 in the output set, indicating whether RX processing module 812 was able to successfully recover (or may have successfully recovered) the information represented by transmission 822, or whether an attempt to recover at least a portion of this information failed.

[0089] To illustrate, similar to conventional HARQ-type transmission feedback methods, network 100 jointly trains the transmitter DNN and receiver DNN using forward error correction (FEC) or other error detection / error correction coding. Therefore, the input data block is actually FEC-coded or otherwise error-coded by the DNN(s) of the TX processing module 808, and the result is included in transmission 822 in some form. The DNN(s) of the RX processing module 812 actually attempts to decode the error-coded data block to recover the input data block. If processing indicates the presence of an error in the recovered data (and which is uncorrected or uncorrectable as part of the processing), the DNN(s) of the RX processing module 812 formats the transmission feedback signal 824 to indicate the detection of an uncorrected / uncorrectable error (i.e., indicating some form of NACK). If processing indicates the data block has been fully recovered, the DNN(s) of the RX processing module 812 formats the transmission feedback signal to indicate successful recovery of the data block (i.e., indicating some form of ACK).

[0090] If the DNN of the RX processing module 812 is trained individually or jointly to enable the DNN to detect whether a recovered data block has an error without detecting the location of the error, similar to using parity bits, then the transmission feedback signal 824 can be formatted as a simple ACK / NACK feedback signal or another binary signal (as shown in box 826) indicating whether the data block was successfully recovered or not. If the DNN of the RX processing module 812 is trained to not only detect whether a recovered data block has an error but also, in some cases, identify which parts(s) of the data block have an error, then the transmission feedback signal 824 can be formatted as a soft probabilistic representation or other non-binary (or soft) signal (as shown in box 828) that identifies or suggests which parts of the data block may be unrecoverable and therefore require retransmission. This soft probabilistic representation can include, for example, a sigmoid function, a hyperbolic tangent (tanh) activation function, or other non-linear activation functions.

[0091] The TX processing module 810 receives the transmission feedback signal 824 and provides this signal as input to the input layer of one or more DNNs. The one or more DNNs of the TX processing module 810 are configured to process this input, along with any other suitable inputs, according to their currently implemented DNN architecture, to generate one or more outputs. These one or more outputs are provided as a transmission 832 for conversion into RF signals by an RF analog stage (not shown), wirelessly transmitting the RF signals to the data transmission device 802.

[0092] Returning to the data transmission device 802, the RF signal representing transmission 832 is received and processed by the RF analog stage (not shown) to convert the RF signal into one or more digital inputs to the input layer of one or more DNNs of the RX processing module 806. The one or more DNNs of the RX processing module 806 process these inputs according to their DNN architecture configuration (which may be selected based on the implemented scheduling authorization type or other operating environment characteristics) to generate an output set representing the recovered user plane data and control plane information represented by transmission 832. In addition to recovering (or attempting to recover) any user plane data, the DNN(s) of the RX processing module 806 operate to configure the retransmission control signal 820 based on the representation of the transmission feedback signal 824 included in transmission 832 by the TX processing module 810. That is, in practice, the RX processing module 806 extracts the representation of the transmission feedback signal 824 to provide as the retransmission control signal 820.

[0093] Therefore, the retransmission control signal 820 can adopt a format compatible with that of the transmission feedback signal 824. For example, if the transmission feedback signal 824 is formatted as an ACK / NACK signal or another binary signal, the retransmission control signal 820 can also be formatted as a binary signal (as shown in block 834), where a first value indicates that the previously transmitted data block was successfully recovered and therefore does not need to be retransmitted, and a second value indicates that the previously transmitted data block was not successfully recovered and therefore needs to be retransmitted in its entirety. However, if the transmission feedback signal 824 is formatted as a soft probability representation or another differentiable signal, the retransmission control signal 820 can also be formatted as a soft signal or a differentiable signal. For illustration, in some embodiments, the DNN of the RX processing module 806 that generates the retransmission control signal 820 is trained or otherwise configured to generate and provide a nonlinear activation function (as shown in block 836) as the retransmission control signal 820. As mentioned above, such nonlinear activation functions can include, for example, the Sigmoid function, the tanh activation function, etc.

[0094] For the next iteration of the data / control transmission by the data transmitting device 802, the gating function 816 uses the retransmission control signal 820 as a gating signal or other control signal to selectively control whether the TX processing module 808 includes a representation of some or all of the previously buffered data blocks in the delay buffer 814 used for the next transmission of wireless communication to the data receiving device 804; that is, the TX processing module 808 selectively retransmits a portion or all of the most recently transmitted data blocks based on the retransmission control signal 820, which indicates whether the most recently transmitted data blocks were successfully recovered at the data receiving device 804, and in some cases, which portions (if any) were not successfully recovered. When the retransmission control signal 820 is a binary signal, the resulting gating signal and corresponding gating function can selectively pass the most recently transmitted data blocks buffered in the delay buffer 814 to the input of one or more DNNs of the TX processing module 808. When the retransmission control signal 820 is a nonlinear activation function or another differentiable signal, applying this gating signal to the buffered data block enables, for example, the output to represent which parts of the data block will be retransmitted. This vector is then provided to one or more DNNs in the TX processing module 808 to generate the output(s) constituting the next iteration of transmission 822. Returning to the processing of transmission 832 at the RX processing module 806, it is noted that the transmission from the data receiving device 804 to the data transmitting device 802 can also use the above-described DNN-based soft transmission feedback process, and thus a similar transmission feedback process can be performed, such as generating a transmission feedback signal by processing transmission 832 at the RX processing module 806, generating a transmission based on using this transmission feedback signal as input to the TX processing module 808, "extracting" the representation of this transmission feedback signal as a retransmission control signal by the RX processing module 812, and then selectively retransmitting some or all of the data represented by transmission 832 by the TX processing module 810 based on this transmission feedback signal.

[0095] Furthermore, as mentioned above, in some embodiments, depending on different operating environments, such as the scheduling authorization type for the transmission between the data transmitting device 802 and the data receiving device 804 (… Figure 7The blocks 706 and 708 allow for different DNN architecture configurations at the TX and RX processing modules. Therefore, if the current scheduling mandate provides the transmission of user plane data along with transmission feedback or other control data, a corresponding DNN architecture configuration is selected and deployed at the DNN of the TX processing module 810, and the input data block can be multiplexed with the transmission feedback signal 824, or otherwise provided as input to one or more DNNs of the TX processing module 810. Similarly, the data transmitting device 802 selects a corresponding DNN architecture configuration trained to process both incoming user plane data and transmission feedback as input to one or more transmissions received and processed by the data receiving device 804 to extract the relevant transmission feedback and user plane data represented in these transmissions.

[0096] Figure 9 The above description is shown. Figure 8 The data transmitting device 802 and the data receiving device 804 describe a transaction graph of multiple iterations of the DNN-facilitated transmission / retransmission feedback loop. For the purpose of this description, it is assumed that the data transmitting device 802 provides transmission feedback and selected data transmission in a HARQ-like process, such that (multiple) transmitter DNNs operate to encode the transmitted data blocks using FEC or some other error correction code, and (multiple) receiver DNNs therefore operate to decode the received transmission based on the error correction code to detect and, in some cases, correct errors, and thus use soft signaling to represent transmission feedback signals and retransmission control signals. Therefore, for the purpose of Figure 9 The purpose of this example is to transmit the feedback signal and the retransmission control signal, which are respectively called the HARQ feedback signal and the HARQ control signal.

[0097] At block 902, the data transmitting device 802 receives a first data block of the data stream and buffers it at delay buffer 814. At block 904, the TX processing module 808 generates a first transmission based on the first data block, and the data transmitting device wirelessly transmits this first transmission to the data receiving device 804. Since the first data block is the initial transmission in this exchange, there is no need to consider transmission feedback from previous transmissions.

[0098] At block 906, RX processing module 812 processes the first transmission to attempt to recover the first data block and any control information represented by the first transmission. As part of this process, RX processing module 812 generates HARQ feedback signal 908. Figure 8(One embodiment of the transmission feedback signal 824) indicates whether a data block has been successfully recovered, and if not, which parts were not recovered. For the purposes of this example, it is assumed that the first byte and the last byte of the first data block were not recovered, and the HARQ feedback signal 908 is configured accordingly.

[0099] At block 910, the TX processing module 810 generates a second transmission based on the HARQ feedback signal 908, and the data receiving device 804 wirelessly transmits the second transmission to the data transmitting device 802. At block 912, the RX processing module 806 of the data transmitting device 802 processes the second transmission to extract any user plane data and control plane information represented by the received second transmission. As part of this process, the RX processing module 806 generates a HARQ control signal 914 from the second transmission. Figure 8 (An embodiment of the retransmission control signal 820), which represents the HARQ feedback signal 908, and therefore can contain a nonlinear activation function or another soft signal.

[0100] At block 916, the HARQ feedback signal 908 is used as a gating signal at gating function 816 for the first data block buffered at delay buffer 814, so that a vector representing the first and last bytes of the initial data block buffered is input to TX processing module 808. Based on this vector (and therefore based on HARQ feedback signal 908), at block 916, TX processing module 808 provides a retransmission of the indicated first and last bytes of the initial data block via generating a third transmission. Meanwhile, at block 918, a second data block is received and then buffered at delay buffer 814. Because the first and last bytes of the first data block are being retransmitted, the second data block remains in delay buffer 814 for subsequent transmissions.

[0101] The third transmission is wirelessly transmitted to the RX processing module 812, and at block 920, the RX processing module 812 processes the third transmission to attempt to recover the retransmission portion of the first data block and any control information represented by the first transmission. As part of this process, the RX processing module 812 generates a HARQ feedback signal 922, which indicates whether the retransmission portion of the first data block was successfully recovered, or if not, which portions were not recovered. For the purposes of this example, it is assumed that the first byte and the last byte of the first data block were successfully recovered, and the HARQ feedback signal 922 is configured accordingly.

[0102] At block 924, the TX processing module 810 generates a fourth transmission based on the HARQ feedback signal 922, and the data receiving device 804 wirelessly transmits the fourth transmission to the data transmitting device 802. At block 926, the RX processing module 806 of the data transmitting device 802 processes the fourth transmission to extract any user plane data and control plane information represented by the received fourth transmission. As part of this process, the RX processing module 806 generates a HARQ control signal 928 from the fourth transmission, which represents the HARQ feedback signal 922, and is therefore capable of including a non-linear activation function or another soft signal.

[0103] At block 930, the HARQ feedback signal 928 is used as a gating signal at gating function 816 for the first data block buffered at delay buffer 814, so that a vector indicating no retransmission of the first data block is input to TX processing module 808. Based on the vector (and therefore based on HARQ feedback signal 922), TX processing module 808 provides transmission of the buffered second data block via generating a fifth transmission. Meanwhile, at block 932, a third data block is received and then buffered at delay buffer 814. The fifth transmission is wirelessly transmitted to data receiving device 804, and another iteration of the transmit-receive-feedback-selective retransmission loop as described above is performed. This process is repeated for each data block of the stream until the last data block of the stream is fully recovered at data receiving device 804 or until communication is otherwise terminated.

[0104] In some embodiments, certain aspects of the techniques described above may be implemented by one or more processors of a processing system executing the software. The software includes one or more sets of executable instructions stored or otherwise tangibly embodied on a non-transitory computer-readable storage medium. The software may contain instructions and certain data that, when executed by one or more processors, manipulate the one or more processors to perform one or more aspects of the techniques described above. The non-transitory computer-readable storage medium may include, for example, disk or optical disk storage devices, solid-state storage devices such as flash memory, cache, random access memory (RAM), or one or more other non-volatile storage devices. The executable instructions stored on the non-transitory computer-readable storage medium may be source code, assembly language code, object code, or another instruction format interpreted or otherwise executed by one or more processors.

[0105] Computer-readable storage media can include any storage medium or combination of storage media that can be accessed by a computer system during use to provide instructions and / or data to the computer system. Such storage media can include, but are not limited to, optical media (e.g., optical discs (CDs), digital versatile discs (DVDs), Blu-ray discs), magnetic media (e.g., floppy disks, magnetic tapes, or magnetic-hard disk drives), volatile memory (e.g., random access memory (RAM) or cache), non-volatile memory (e.g., read-only memory (ROM) or flash memory), or microelectromechanical systems (MEMS) based storage media. Computer-readable storage media can be embedded in a computing system (e.g., system RAM or ROM), fixedly attached to a computing system (e.g., magnetic hard disks), removably attached to a computing system (e.g., optical discs or flash memory based on Universal Serial Bus (USB)), or coupled to a computer system via a wired or wireless network (e.g., network accessible storage device (NAS)).

[0106] It should be noted that not all activities or elements described in the general description are necessary; a particular activity or device may not be required as part of a specific activity or element; and one or more further activities or elements may be performed in addition to the described activities or elements. Furthermore, the order in which the activities are listed is not necessarily the order in which they are performed. These concepts have been described with reference to specific embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of this disclosure as set forth in the appended claims. Therefore, the specification and drawings are to be regarded as illustrative rather than restrictive, and all such modifications are to be included within the scope of this disclosure.

[0107] The benefits, other advantages, and solutions to problems have been described above with reference to specific embodiments. However, these benefits, advantages, solutions to problems, and any features that may lead to or make more apparent any benefit, advantage, or solution should not be construed as key, essential, or fundamental features of any or all claims. Furthermore, the specific embodiments disclosed above are merely illustrative, as the disclosed subject matter can be modified and practiced in different but equivalent ways that would be apparent to those skilled in the art benefiting from the teachings herein. The construction or design details shown herein are not limited in any way except as described in the appended claims. Therefore, it is apparent that the specific embodiments disclosed above can be changed or modified, and all such changes are considered to be within the scope of the disclosed subject matter. Therefore, the protection sought herein is as set forth in the appended claims.

Claims

1. A method implemented by a computer in a data transmitting apparatus (802), comprising: The first data block (902) is buffered in the delay buffer (814); The first data block is processed at the transmitter neural network (808) of the data transmitting device to generate a first transmission (904) for wireless communication to the data receiving device (804). The first transmission is wirelessly transmitted to the data receiving device; The second transmission (912) wirelessly received from the data receiving device is processed at the receiver neural network (806) of the data transmitting device to generate a retransmission control signal (914). By using the retransmission control signal as a control input, the gating function (816) is applied to the first data block in the delay buffer, and the output of the gating function is provided to the input of the transmitter neural network, a third transmission (916) for wireless communication to the data receiving device is generated. as well as The third transmission is wirelessly transmitted to the data receiving device; The transmitter neural network's neural network architecture configuration is selected from multiple neural network architecture configurations. Each of these configurations is associated with a different transmission environment and input characteristics describing the attributes of that environment. The neural network architecture configuration of the transmitter neural network is selected by matching the current transmission environment with the input characteristics.

2. The method according to claim 1, wherein, The second transmission represents a Hybrid Automatic Repeat Request (HARQ) signal (908) from the data receiving device.

3. The method according to claim 1, wherein: The retransmission control signal includes a binary signal (834); and Generating the third transmission includes: In response to the binary signal having a first value, the third transmission is generated based on at least a portion of the first data block; and The third transmission is generated independently of the first data block in response to the binary signal having a second value.

4. The method according to claim 1, wherein: The retransmission control signal includes a nonlinear activation function (836); and The third transmission is generated based on the nonlinear activation function to include one or more portions of the first data block.

5. The method according to claim 1, wherein, Each of the transmitter neural network and the receiver neural network has multiple neural network architecture configurations (218, 238), and each neural network architecture configuration is associated with a different corresponding scheduling authorization type among multiple scheduling authorization types; and The method further includes: The transmitter neural network and the receiver neural network are jointly trained (702) using one or more training datasets in combination with at least one neural network of the data receiving device, wherein jointly training the transmitter neural network and the receiver neural network includes training multiple neural network architecture configurations of the multiple neural network architecture configurations individually.

6. The method according to claim 5, further comprising: Determine the current scheduling authorization type (705, 707, 709) implemented for the data transmission device, wherein the current scheduling authorization type includes one of the following: authorization to transmit control information only (707), authorization to transmit user plane data only (705), or authorization to transmit both control information and user plane data (709); and Configure the transmitter neural network and the receiver neural network to implement the neural network architecture configuration among the multiple neural network architecture configurations associated with the current scheduling authorization type.

7. The method according to claim 1, wherein: Generating the third transmission includes further generating the third transmission at the transmitter neural network based on a second data block (918) received after the first data block.

8. The method according to claim 1, wherein: Processing the second transmission at the receiver neural network of the data transmitting device further includes: processing the second transmission at the receiver neural network of the data transmitting device to attempt to recover a second data block from the second transmission.

9. The method according to claim 8, further comprising: A transmission feedback signal (914) is generated at the receiver neural network of the data transmitting device, the transmission feedback signal indicating the state of recovering the second data block from the second transmission; and The generation of the third transmission further includes processing the transmission feedback signal at the transmitter neural network of the data transmission device.

10. The method according to claim 9, wherein, The transmission feedback signal includes a Hybrid Automatic Repeat Request (HARQ) control signal (914).

11. A data transmission device (108, 110, 802), comprising: RF front end (204, 224); At least one processor (210, 230) coupled to the radio frequency front end. as well as A non-transitory computer-readable medium (212, 323) storing one or more instruction sets configured to manipulate the at least one processor to perform the method according to any one of claims 1 to 10.

12. A method implemented by a computer in a data receiving device (804), comprising: The receiver neural network (812) of the data receiving device receives the first transmission (822, 906) wirelessly transmitted from the data transmitting device (802). The first transmission is processed at the receiver neural network of the data receiving device to attempt to recover the first data block (818, 902) and a first transmission feedback signal (824, 910) representing the attempt to recover the first data block is generated. The second transmission (832, 910) is generated at the transmitter neural network (810) of the data receiving device based on the first transmission feedback signal. as well as The second transmission is wirelessly transmitted to the data transmitting device; The transmitter neural network's neural network architecture configuration is selected from multiple neural network architecture configurations. Each of these configurations is associated with a different transmission environment and input characteristics describing the attributes of that environment. The neural network architecture configuration of the transmitter neural network is selected by matching the current transmission environment with the input characteristics.

13. The method according to claim 12, wherein, The second transmission represents a mixed automatic repeat request (HARQ) signal (908).

14. The method according to claim 12, wherein: The first transmission feedback signal indicates a failed attempt to recover the first data block; and The method further includes: The receiver neural network receives a third transmission (920) wirelessly transmitted from the data transmitting device. The third transmission is processed at the receiver neural network of the data receiving device to attempt to recover at least a portion of the first data block and a second transmission feedback signal (922) representing the attempt to recover at least a portion of the first data block is generated; and Based on the second transmission feedback signal, a fourth transmission (924) for wireless communication to the data transmission device is generated at the transmitter neural network.

15. The method of claim 14, wherein: Each of the transmitter neural network and the receiver neural network has multiple neural network architecture configurations (218, 238), and each neural network architecture configuration is associated with a different corresponding scheduling authorization type among multiple scheduling authorization types (705, 707, 709); and The method further includes: The transmitter neural network and the receiver neural network are jointly trained using one or more training datasets in conjunction with at least one neural network of the data transmitting device, wherein jointly training the transmitter neural network and the receiver neural network includes individually training multiple neural network architecture configurations among the multiple neural network architecture configurations.

16. The method of claim 15, further comprising: Determine the current scheduling authorization type implemented for the data receiving device, the current scheduling authorization type including one of the following: authorization to transmit control information only (707), authorization to transmit user plane data only (705), or authorization to transmit both control information and user plane data (709); and Configure the transmitter neural network and the receiver neural network to implement the neural network architecture configuration among the multiple neural network architecture configurations associated with the current scheduling authorization type.

17. A data receiving device (108, 110, 804), comprising: RF front end (204, 224); At least one processor (210, 230) coupled to the radio frequency front end. as well as At least one storage component (212, 232) of a storage instruction set configured to manipulate the at least one processor to perform the method according to any one of claims 12 to 16.

Images