Communication apparatus and method

Abstract

A communication device and method can be used in 802.11 series protocols, such as 802.11be and 802.11bf. The device comprises an input interface, a neural network processor and an output interface, the neural network processor storing a neural network comprising an input layer and an output layer; the input interface is configured to acquire at least one of input indication information, function indication information, output indication information or parameter indication information; the neural network processor is configured to execute a function indicated by the function indication information according to the input layer and the output layer; and the output interface is configured to output a result after the neural network processor executes the function. The input layer structure and the output layer structure of the neural network can be determined based on the information acquired by the input interface. Through the device, the function implemented by the MAC layer can be implemented based on the neural network, the complexity of the neural network is simplified, and hardware acceleration is facilitated.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a communication device and method. Background Technology

[0002] With the evolution of Wi-Fi standards and networks, Wi-Fi systems have become increasingly complex. Meanwhile, the quality of the Medium Access Control (MAC) layer architecture plays a crucial role in overall network performance, such as quality of service (QoS) guarantees, fairness guarantees, and spectrum utilization improvements. Currently, MAC layer design is feature-driven, with each feature having its own corresponding algorithm, and these algorithms are designed independently. For example, in Wi-Fi, channel access uses carrier-sense multiple access with collision avoidance (CSMA / CA) and binary exponential backoff (BEB); rate adaptation uses the Intel rate adaptation algorithm (which can be called Intel Iwl); orthogonal frequency division multiple access (OFDMA) has corresponding resource unit (RU) allocation algorithms; and channel aggregation also has corresponding decision-making algorithms.

[0003] With the development of artificial intelligence (AI) technology, its application in next-generation wireless communication systems to improve network performance has become a trend. However, the performance improvement also brings various challenges, such as high complexity and poor interoperability.

[0004] Therefore, how to effectively combine AI to improve the performance of the MAC layer architecture is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a communication device and method that can implement the functions of the MAC layer based on a neural network, simplifying the complexity of the neural network and facilitating hardware acceleration.

[0006] In a first aspect, embodiments of this application provide a communication device, the device comprising: an input interface, a neural network processor, and an output interface, wherein the neural network processor stores a neural network, the neural network including an input layer and an output layer; the input interface is used to acquire at least one of input indication information, function indication information, output indication information, or parameter indication information; the neural network processor is used to execute the function indicated by the function indication information according to the input layer and the output layer, wherein the structure of the input layer is determined according to at least one of the following information: the input indication information and the function indication information, and the structure of the output layer is determined according to at least one of the following information: the output indication information and the function indication information, wherein the output indication information is used to indicate at least one of the following structures of the output layer: the number of neurons and the activation function, and the parameter indication information is used to indicate at least one of the following information: the bias of neurons in the neural network and the weight of neurons in the neural network; the output interface is used to output the result after the neural network processor executes the function.

[0007] The communication device provided in this application embodiment can implement the functions of the MAC layer based on a neural network. For example, when the function performed by the MAC layer changes, the input and output layers of the neural network are adjusted accordingly, thereby enabling different functions to be implemented based on the neural network. In other words, different functions can be achieved by adjusting the input and output layers of the neural network, thus effectively simplifying the complexity of the neural network and facilitating hardware acceleration.

[0008] In one possible implementation, the neural network further includes a hidden layer, the structure of which is fixed.

[0009] In this embodiment, the structure of the hidden layer is fixed, such that the structure of the hidden layer can be determined according to the function indicated by the function indication information. The hidden layer provided by this embodiment exhibits good performance (e.g., better algorithm performance, higher throughput, lower latency, etc.) for different functions implemented by the neural network, facilitating hardware acceleration and providing good scalability.

[0010] In one possible implementation, the neural network further includes hidden layers, the structure of which is the same for different functions.

[0011] In this embodiment, different functions correspond to the same hidden layer structure, which further simplifies the complexity of the neural network and ensures that the neural network performs well when implementing different functions.

[0012] In one possible implementation, the parameter indication information is used to indicate at least one of the following: the bias of neurons in the structure of the hidden layer, the weight of neurons in the structure of the hidden layer, the bias of neurons in the structure of the output layer, and the weight of neurons in the structure of the output layer.

[0013] In one possible implementation, the result of the neural network processor performing the function includes a decision result determined based on at least one of the function indication information and the output indication information. The decision result includes outputting the result of the neural network processor performing the function in a maximum value manner, or outputting the result of the neural network processor performing the function in a probabilistically random selection manner.

[0014] In one possible implementation, the function indication information is used to indicate any one or more of the following functions: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted resource unit (RU) allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission.

[0015] In one possible implementation, the input indication information is used to indicate at least one of the following: the input information of the neural network, the number of input information items, and the dimension of the input information items.

[0016] In one possible implementation, the number of input information items and the dimension of the input information items are used to determine the number of neurons in the input layer.

[0017] In one possible implementation, the input information includes any one or more of the following: carrier sensing results, received power, modulation and coding scheme (MCS) used by the data packet, number of multiple input multiple output (MIMO) data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, duration of carrier sensing results, or duration of channel aggregation; wherein, the channel access behavior includes access or non-access.

[0018] In one possible implementation, the input information may further include any one or more of the following: packet loss rate, throughput, the time since the last successful transmission of the communication device, the time since the last successful transmission of other communication devices on the channel, the load information of the communication device or the load information of other communication devices on the channel, wherein the other communication devices are communication devices other than the communication device.

[0019] Secondly, embodiments of this application provide a communication device, which includes an input control module, a neural network module, an output control module, and an action decision module. The neural network module includes an input layer and an output layer. The input control module is used to determine the structure of the input layer based on input indication information and / or function indication information. The output control module is used to determine the structure of the output layer based on output indication information and / or function indication information. The neural network module is used to execute the function indicated by the function indication information based on the input layer and the output layer. The action decision module is used to select an action based on the output information of the neural network and output the decision result after the neural network executes the function.

[0020] In one possible implementation, the neural network further includes a hidden layer, the structure of which is fixed.

[0021] In one possible implementation, the neural network further includes hidden layers, the structure of which is the same for different functions.

[0022] In one possible implementation, the parameter indication information is used to indicate at least one of the following: the bias of neurons in the structure of the hidden layer, the weight of neurons in the structure of the hidden layer, the bias of neurons in the structure of the output layer, and the weight of neurons in the structure of the output layer.

[0023] In one possible implementation, the decision result is determined based on at least one of the function indication information and the output indication information. The decision result includes outputting the result of the neural network processor performing the function in a maximum value manner, or outputting the result of the neural network processor performing the function in a probabilistically randomly selected manner.

[0024] In one possible implementation, the function indication information is used to indicate any one or more of the following functions: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted resource unit (RU) allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission.

[0025] In one possible implementation, the input indication information is used to indicate at least one of the following: the input information of the neural network, the number of input information items, and the dimension of the input information items.

[0026] In one possible implementation, the number of input information items and the dimension of the input information items are used to determine the number of neurons in the input layer.

[0027] In one possible implementation, the input information includes any one or more of the following: carrier sensing results, received power, modulation and coding scheme (MCS) used by the data packet, number of multiple-input multiple-output (MIMO) data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, duration of carrier sensing results, or duration of channel aggregation; wherein, the channel access behavior includes access or non-access.

[0028] In one possible implementation, the input information may further include any one or more of the following: packet loss rate, throughput, the time since the last successful transmission of the communication device, the time since the last successful transmission of other communication devices on the channel, the load information of the communication device or the load information of other communication devices on the channel, wherein the other communication devices are communication devices other than the communication device.

[0029] Thirdly, embodiments of this application provide a communication method applied to a first communication device storing a neural network, the neural network including an input layer and an output layer. The method includes: acquiring at least one of input indication information, function indication information, output indication information, or parameter indication information; executing the function indicated by the function indication information according to the input layer and the output layer, wherein the structure of the input layer is determined according to at least one of the following: the input indication information and the function indication information; the structure of the output layer is determined according to at least one of the following: the output indication information and the function indication information; the output indication information is used to indicate at least one of the following in the structure of the output layer: the number of neurons and the activation function; the parameter indication information is used to indicate at least one of the following: the bias of neurons in the neural network and the weight of neurons in the neural network; and outputting the result after the neural network executes the function.

[0030] In one possible implementation, the neural network further includes a hidden layer, the structure of which is fixed.

[0031] In one possible implementation, the neural network further includes hidden layers, the structure of which is the same for different functions.

[0032] In one possible implementation, the parameter indication information is used to indicate at least one of the following: the bias of neurons in the structure of the hidden layer, the weight of neurons in the structure of the hidden layer, the bias of neurons in the structure of the output layer, and the weight of neurons in the structure of the output layer.

[0033] In one possible implementation, the neural network further includes hidden layers with a fixed structure. The result of the neural network performing the function includes a decision result determined based on at least one of the function instruction information and output instruction information. The decision result includes outputting the result of the neural network processor performing the function at its maximum value, or outputting the result of the neural network processor performing the function in a probabilistically randomly selected manner.

[0034] In one possible implementation, the neural network further includes a hidden layer with a fixed structure. The functional indication information is used to indicate one or more of the following functions: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted resource unit (RU) allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission.

[0035] In one possible implementation, the neural network further includes hidden layers with a fixed structure. The input indication information is used to indicate at least one of the following: the input information of the neural network, the number of inputs, and the dimension of the inputs.

[0036] In one possible implementation, the neural network further includes a hidden layer with a fixed structure. The number of input information items and the dimension of the input information are used to determine the number of neurons in the input layer.

[0037] In one possible implementation, the neural network further includes a hidden layer with a fixed structure. The input information includes any one or more of the following: carrier sensing results, received power, modulation and coding scheme (MCS) used for the data packets, number of multiple-input multiple-output (MIMO) data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, duration of carrier sensing results, or duration of channel aggregation; wherein, the channel access behavior includes access or non-access.

[0038] In one possible implementation, the neural network further includes hidden layers with a fixed structure. The input information also includes any one or more of the following:

[0039] Packet loss rate, throughput, the time since the last successful transmission of the first communication device, the time since the last successful transmission of other first communication devices on the channel, the load information of the first communication device or the load information of other first communication devices on the channel, wherein the other first communication devices are first communication devices other than the first communication device.

[0040] In one possible implementation, the neural network further includes a hidden layer with a fixed structure. Acquiring at least one of the input indication information, function indication information, output indication information, or parameter indication information includes: the STA sending a function request message to the access point AP, the function request message including the function indication information, the function request message being used to request activation of the function indicated by the function indication information; the STA receiving a function response message from the AP, the function response message being used to acknowledge the function request message; and the STA receiving indication information from the AP, the indication information including at least one of the following: the input indication information, the output indication information, and the parameter indication information.

[0041] Fourthly, embodiments of this application provide a communication method applied to a second communication device. The method includes: receiving a function request message from a first communication device, the function request message including function indication information, the function request message being used to request the activation of a function indicated by the function indication information; sending a function response message to the first communication device, the function response message being used to confirm the function request message; and sending indication information to the first communication device, the indication information including at least one of the following: input indication information, output indication information, and parameter indication information.

[0042] In one possible implementation, the method further includes: training a neural network according to the function indicated by the function indication information to obtain the parameter indication information, wherein the parameters of the neural network indicated by the parameter indication information correspond to the function indicated by the function indication information.

[0043] Fifthly, embodiments of this application provide a computer-readable storage medium for storing a computer program that, when run on a computer, causes the methods shown in the third aspect or any possible implementation thereof to be executed.

[0044] Sixthly, embodiments of this application provide a computer-readable storage medium for storing a computer program that, when run on a computer, causes the methods shown in the fourth aspect or any possible implementation thereof to be executed.

[0045] In a seventh aspect, embodiments of this application provide a computer program product comprising a computer program or computer code that, when run on a computer, causes the methods shown in the third aspect or any possible implementation thereof to be executed.

[0046] Eighthly, embodiments of this application provide a computer program product comprising a computer program or computer code that, when run on a computer, causes the methods shown in the fourth aspect or any possible implementation thereof to be executed.

[0047] Ninthly, embodiments of this application provide a computer program that, when run on a computer, executes the methods shown in the third aspect or any possible implementation thereof.

[0048] In a tenth aspect, embodiments of this application provide a computer program that, when run on a computer, executes the methods shown in the fourth aspect or any possible implementation thereof.

[0049] Eleventhly, embodiments of this application provide a wireless communication system, which includes a first communication device and a second communication device. The first communication device is used to execute the method shown in the third aspect or any possible implementation of the third aspect, and the second communication device is used to execute the method shown in the fourth aspect or any possible implementation of the fourth aspect. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the architecture of a communication system provided in an embodiment of this application;

[0051] Figure 2a This is a schematic diagram of a fully connected neural network containing three layers;

[0052] Figure 2b This is a schematic diagram of a neuron calculating its output based on the input.

[0053] Figure 3a This is a schematic diagram of the structure of an access point provided in an embodiment of this application;

[0054] Figure 3b This is a schematic diagram of another access point structure provided in an embodiment of this application;

[0055] Figure 3c This is a schematic diagram of another access point structure provided in an embodiment of this application;

[0056] Figure 4a This is a schematic diagram of the structure of a site provided in an embodiment of this application;

[0057] Figure 4b This is a schematic diagram of another site structure provided in an embodiment of this application;

[0058] Figure 4c This is a schematic diagram of the structure of another site provided in an embodiment of this application;

[0059] Figure 5 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0060] Figure 6 This is a schematic diagram of the hidden layer structure of the neural network provided in the embodiments of this application;

[0061] Figure 7a This is a schematic diagram of another communication device provided in an embodiment of this application;

[0062] Figure 7b This is a schematic diagram of the neural network structure provided in the embodiments of this application;

[0063] Figure 8 This is a flowchart illustrating a communication method provided in an embodiment of this application;

[0064] Figure 9 This is a flowchart illustrating another communication method provided in an embodiment of this application;

[0065] Figure 10 This is a flowchart illustrating another communication method provided in an embodiment of this application. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described below in conjunction with the accompanying drawings.

[0067] The terms "first" and "second," etc., used in the specification, claims, and drawings of this application are used only to distinguish different objects and not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0068] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0069] In this application, "at least one (item)" means one or more, "more than one" means two or more, "at least two (items)" means two or three or more, and "and / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items. For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c".

[0070] In the MAC layer architecture of a communication device, the operation of the MAC layer can be driven by characteristics, meaning different characteristics correspond to different algorithms. However, in this communication device, the MAC layer's algorithm architecture uses different algorithms for different functions, leading to the following problems when the MAC layer processes algorithms:

[0071] Poor adaptability: The algorithm design is optimized for specific scenarios. For example, if a problem occurs in one scenario, it is patched up. After patching up one scenario, another scenario may have a problem.

[0072] Mutually exclusive: Features influence each other. For example, packet loss can reduce MCS, increase transmit power, and delay channel access. However, when decisions occur simultaneously, they will influence each other, which may result in 1+1<2.

[0073] The development cycle is long and evolution is difficult: After the feature is specified, the performance of the corresponding feature is optimized through continuous investment in research and development (R&D). Once a new MAC feature is available, new R&D work must be carried out.

[0074] In another communication device's MAC layer architecture, the algorithm processing architecture of the MAC layer can be reconstructed using AI. For example, a certain feature can be implemented based on AI; however, this approach often involves reconstructing only a single feature. This independent reconstructing method has the following problems:

[0075] High complexity: Each feature requires the design of a neural network.

[0076] Interoperability with multiple vendors is not possible: Distributing neural network models trained by the AP to STAs from different vendors requires an additional negotiation mechanism, and standardization has not been achieved.

[0077] Hardware acceleration is difficult: each feature corresponds to a set of neural networks, which is not easy to accelerate using dedicated chips.

[0078] Poor scalability: It is essentially a rule-based architecture, with each feature being optimized and developed separately, resulting in long development cycles and slow deployment.

[0079] In view of this, this application provides a communication device and method that can implement the functions of the MAC layer based on a neural network. For example, when the function performed by the MAC layer changes, the input and output layers of the neural network in the MAC layer are adjusted accordingly, thereby enabling different functions to be implemented based on the neural network. By adjusting the input and output layers of the neural network, different functions can be achieved, effectively simplifying the complexity of the neural network and facilitating hardware acceleration. Furthermore, in the AI-based MAC (also known as AI-MAC or neural network-based MAC) algorithm architecture provided in this application, when different functions of the MAC layer need to be implemented through the neural network, the hidden layer structure of the neural network can be fixed, meaning the input and output layers can be adjusted according to different functions. Therefore, using a fixed hidden layer structure for different functions not only facilitates hardware acceleration but also provides good scalability, such as being able to extend to the MAC features of next-generation Wi-Fi. It is understood that the above-mentioned features can also be understood as functions, applications, or services, etc., and this application does not limit the names of the features. For consistency, the following will use functions as an example to describe the device and method provided in the embodiments of this application.

[0080] The following details the communication system involved in the embodiments of this application.

[0081] The method provided in this application can be applied to wireless local area network (WLAN) systems, such as Wi-Fi. It can also be applied to the Institute of Electrical and Electronics Engineers (IEEE) 802.11 series protocols, such as 802.11a / b / g, 802.11n, 802.11ac, 802.11ax, 802.11be, 802.11bf, or other future protocols, which will not be listed here. The method provided in this application can also be applied to various communication systems, such as cellular systems (including but not limited to: Long Term Evolution (LTE) systems, 5th-generation (5G) communication systems, and new communication systems emerging in future communication development (such as 6G), Internet of Things (IoT) systems, Narrow Band Internet of Things (NB-IoT) systems, and other short-range communication systems (including but not limited to: Bluetooth, Ultra Wideband (UWB)).

[0082] The method provided in this application can be applied to scenarios where one node transmits data with one or more nodes. For example, single-user uplink / downlink transmission, multi-user uplink / downlink transmission, and device-to-device (D2D) transmission. Each of the aforementioned nodes can be a communication device in a wireless communication system; that is, the method provided in this application can be implemented by a communication device in a wireless communication system. For example, the communication device can be at least one of an access point (AP) or a station (STA).

[0083] The access point in this application is a device with wireless communication capabilities and AI processing capabilities. It supports communication or sensing using the WLAN protocol and has the function of communicating or sensing with other devices (such as stations or other access points) in the WLAN network. Of course, it can also have the function of communicating or sensing with other devices. Alternatively, the access point is equivalent to a bridge connecting wired and wireless networks. Its main function is to connect various wireless network clients together and then connect the wireless network to the Ethernet. In a WLAN system, the access point can be called an access point station (AP STA). The device with wireless communication capabilities can be a complete device or a chip or processing system installed in a complete device. Devices with these chips or processing systems can implement the methods and functions of the embodiments of this application under the control of the chip or processing system. The AP in the embodiments of this application is a device that provides services to the STA and can support the 802.11 series of protocols. For example, the access point can be an access point for terminals (such as mobile phones) to enter a wired (or wireless) network. It is mainly deployed in homes, inside buildings, and inside parks, with a typical coverage radius of tens to hundreds of meters. Of course, it can also be deployed outdoors. For example, an AP can be a communication server, router, switch, bridge, or other communication entity; an AP can include various forms of macro base stations, micro base stations, relay stations, etc., and of course, an AP can also be a chip or processing system in these various forms of devices, thereby implementing the methods and functions of the embodiments of this application. The access point in this application can be an extrameally high throughput (EHT) AP, or an access point of a future Wi-Fi standard, etc.

[0084] The station in this application is a device with wireless communication capabilities and AI processing capabilities, supporting communication or sensing using the WLAN protocol, and having the ability to communicate or sense with other stations or access points in the WLAN network. In a WLAN system, a station can be called a non-access point station (non-AP STA). For example, an STA is any user communication device that allows a user to communicate or sense with an AP and thus communicate with the WLAN. This device with wireless communication capabilities can be a complete device, or it can be a chip or processing system installed in a complete device. Devices with these chips or processing systems installed can implement the methods and functions of the embodiments of this application under the control of the chip or processing system. For example, a station can be a wireless communication chip, a wireless sensor, or a wireless communication terminal, and can also be called a user. For another example, a station can be a mobile phone with Wi-Fi communication capabilities, a tablet computer with Wi-Fi communication capabilities, a set-top box with Wi-Fi communication capabilities, a smart TV with Wi-Fi communication capabilities, a smart wearable device with Wi-Fi communication capabilities, an in-vehicle communication device with Wi-Fi communication capabilities, and a computer with Wi-Fi communication capabilities, etc.

[0085] WLAN systems can provide high-speed, low-latency transmission. As WLAN application scenarios continue to evolve, WLAN systems will be applied to more scenarios and industries, such as the Internet of Things (IoT) industry, the Internet of Vehicles (IoV) industry, the banking industry, enterprise offices, stadiums, exhibition halls, concert halls, hotel rooms, dormitories, hospital wards, classrooms, supermarkets, squares, streets, production workshops, and warehouses. Of course, devices supporting WLAN communication or sensing (such as access points or sites) can be sensor nodes in smart cities (e.g., smart water meters, smart electricity meters, smart air quality monitoring nodes), smart devices in smart homes (e.g., smart cameras, projectors, displays, televisions, speakers, refrigerators, washing machines, etc.), nodes in the IoT, entertainment terminals (e.g., AR, VR wearable devices), smart devices in smart offices (e.g., printers, projectors, amplifiers, speakers, etc.), IoV devices, infrastructure in daily life scenarios (e.g., vending machines, supermarket self-service navigation kiosks, self-checkout machines, self-service ordering machines, etc.), and equipment in large sports and music venues. For example, access points and sites can be devices used in vehicle-to-everything (V2X) networks, IoT nodes and sensors in the Internet of Things (IoT), smart cameras, smart remote controls, smart water and electricity meters in smart homes, and sensors in smart cities. The specific forms of STAs and APs are not limited in this application embodiment; they are merely illustrative examples.

[0086] Although this application primarily uses the deployment of IEEE 802.11 networks as an example, those skilled in the art will readily understand that the various aspects covered in this application can be extended to other networks employing various standards or protocols, such as high-performance radio LANs (HIPERLANs) (a wireless standard similar to IEEE 802.11, primarily used in Europe), wide area networks (WANs), wireless local area networks (WLANs), personal area networks (PANs), or other networks now known or to be developed in the future.

[0087] For example, the communication system to which the method provided in this application is applied may include access points (APs) and stations (STAs), or multiple access points (APs) or multiple stations (STAs). The access point can also be understood as an access point entity, and the station can also be understood as a station entity. For example, this application can be applied to scenarios where APs and STAs communicate or sense each other in a WLAN. Optionally, the AP can communicate or sense a single STA, or the AP can communicate or sense multiple STAs simultaneously. As another example, this application can be applied to scenarios where multiple APs collaborate in a WLAN. Yet another example, this application can be applied to scenarios where STAs perform D2D communication with each other in a WLAN. Both the AP and STA can support WLAN communication protocols, which may include protocols from the IEEE 802.11 series, such as those applicable to the 802.11be standard, or standards later than 802.11be.

[0088] Figure 1 This is a schematic diagram of the architecture of a communication system provided in an embodiment of this application. The communication system may include one or more APs and one or more STAs. Figure 1 The diagram illustrates two access points, such as AP1 and AP2, and three stations, such as STA1, STA2, and STA3. It is understood that one or more APs can communicate with one or more STAs. Of course, APs can communicate with each other, and STAs can communicate with each other.

[0089] Understandable Figure 1 The use of STA (station) as a mobile phone and AP (access point) as a router in this application is merely an example and does not imply any limitation on the types of AP and STA used in this application. Furthermore, Figure 1 Two APs and three STAs are shown only as examples, but the number of APs or STAs may be more or less, and this application does not limit this.

[0090] Before introducing the communication device involved in this application, let's first introduce the terminology involved in the embodiments of this application.

[0091] 1. Neural Networks

[0092] A neural network (NN) is a machine learning technique that simulates the neural network of the human brain in an attempt to achieve artificial intelligence. A neural network typically consists of at least three layers: an input layer, a hidden layer (also called an intermediate layer), and an output layer. Deeper neural networks may contain more hidden layers between the input and output layers.

[0093] Taking the simplest neural network as an example, we will explain its internal structure and implementation. Figure 2a This is a schematic diagram of a fully connected neural network containing three layers. (Example) Figure 2a As shown, this neural network can include three layers: an input layer, a hidden layer, and an output layer. The input layer has three neurons, the hidden layer has four neurons, and the output layer has two neurons. Each neuron in each layer is fully connected to the neurons in the next layer. Each connection between neurons corresponds to a weight, which can be updated through training. Each neuron in the hidden and output layers can also correspond to a bias, which can be updated through training. Updating the neural network means updating these weights and biases. The weights and biases shown here can also be understood as the parameters of the neural network described below (such as the parameters indicated by parameter indicators). Knowing the structure of the neural network, such as the number of neurons in each layer and how the output of one neuron is input into another (i.e., the connections between neurons), as well as the parameters of the neural network, namely the weights and biases, is equivalent to knowing all the information about the neural network.

[0094] As can be understood, since weights can be interpreted as parameters corresponding to the connections between different neurons, for ease of understanding, the weights on the connections between neurons in the input layer and neurons in the hidden layer are collectively referred to as the weights of neurons in the hidden layer structure, the weights on the connections between neurons in the hidden layer are also called the weights of neurons in the hidden layer structure, and the weights on the connections between neurons in the hidden layer and neurons in the output layer are called the weights of neurons in the output layer structure.

[0095] Depend on Figure 2a It can be seen that each neuron may have multiple input connections, and each neuron calculates the output based on the input. Figure 2b This is a schematic diagram of a neuron calculating its output based on the input. For example... Figure 2b For example, if a neuron (such as a neuron in the output layer) contains 3 inputs, 1 output, and 2 computational functions, then the formula for calculating the output can satisfy the following conditions:

[0096] Output = Activation function(Input 1 * Weight 1 + Input 2 * Weight 2 + Input 3 * Weight 3 + Bias)

[0097] The symbol “*” represents the mathematical operation “multiply” or “multiply by”, which will not be elaborated further below.

[0098] As can be understood, the input layer only has output connections. Each neuron in the input layer is the value input to the neural network, and the output value of each neuron directly serves as the input to all output connections. The output layer only has input connections, and the output can be calculated using the formula described above. Optionally, the output layer may not have an activation function, meaning the aforementioned formula can be transformed into: Output = Input 1 * Weight 1 + Input 2 * Weight 2 + Input 3 * Weight 3 + Bias.

[0099] For example, the output of a k-layer neural network can satisfy the following condition:

[0100] y = f k (f k-1 (…(f1(w1*x+b1)))

[0101] Where x represents the input of the neural network, y represents the output of the neural network, and w i b represents the weights of the i-th layer of the neural network. i f represents the bias of the i-th layer of the neural network. i Let represent the activation function of the i-th layer of the neural network. i = 1, 2, ..., k.

[0102] 2. Input instruction information

[0103] Input indication information can be used to indicate at least one of the following: the input information of the neural network, the number of input information, and the dimension of the input information.

[0104] For example, input information may include basic observation information and extended observation information. Basic observation information refers to information directly obtainable by the communication device, such as carrier sensing results, received power, the MCS used by the data packet, the number of MIMO data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, and the duration of carrier sensing results or channel aggregation. The carrier sensing result can be whether the channel is busy or idle. If carrier sensing is performed on a sub-channel basis, the carrier sensing result can be understood as the busy or idle status of each sub-channel. Queuing delay refers to the time interval from when a data packet enters the transmission queue to when it is transmitted. Access delay refers to the time interval between when a data packet is successfully transmitted. Channel access behavior can refer to both access and non-access behaviors. Data packet transmission results include successful transmission and transmission failure. Channel information can refer to Channel State Information (CSI) or Signal-to-Noise Ratio (SNR). Channel aggregation can refer to at least one of the channels being aggregated or the number of channels being aggregated, or it can refer to whether each sub-channel is aggregated. Derived observation information refers to information calculated by the communication device based on basic observation information, such as packet loss rate, throughput, the time since the last successful transmission of the communication device, the time since the last successful transmission of other communication devices on the channel, the load information of the communication device, or the load information of other communication devices on the channel, where other communication devices are those other than the communication device in question.

[0105] For example, the input information can differ depending on the function implemented by the neural network. The following examples illustrate this.

[0106] Example 1: When a neural network is used to implement or support AI-assisted rate adaptation (or can be understood as when the neural network corresponds to AI-assisted rate adaptation), or when a neural network is used to implement or support AI-assisted rate adaptation combined with channel access (or when the neural network corresponds to AI-assisted rate adaptation combined with channel access), the input information of the neural network includes, but is not limited to, one or more of the following: carrier sensing results, received power, MCS used by the data packet, number of MIMO data streams, channel bandwidth, queuing delay, access delay, data packet transmission results, and channel information. Implementing AI-assisted rate adaptation through a neural network allows the communication device to obtain one or more of the following information: which MCS is used for transmission, how many MIMO data streams are used for transmission, or what channel bandwidth is used for transmission, etc. Implementing AI-assisted rate adaptation combined with channel access through a neural network allows the communication device to obtain one or more of the following information: no access, which MCS is used for access, how many MIMO data streams are used for transmission, or what channel bandwidth is used for transmission.

[0107] Example 2: When a neural network is used to implement or support AI-assisted channel access functionality (or can be understood as when the neural network corresponds to AI-assisted channel access functionality), the input information of the neural network includes, but is not limited to, one or more of the following: carrier sensing results, queuing delay, access delay, channel access behavior, and data packet transmission results. The queuing delay and access delay shown in this application can be used to reflect the traffic load of the communication device itself and / or the congestion level of the channel.

[0108] It is understood that in Examples 1 and 2 above, the input information of the neural network may also include, but is not limited to, one or more of the following: packet loss rate, throughput, the time since the last successful transmission of this communication device, and the time since the last successful transmission of other communication devices on the channel. By implementing AI-assisted channel access functionality through neural networks, communication devices can use AI to decide whether to access the channel.

[0109] Example 3: When a neural network is used to implement or support AI-assisted spatial multiplexing (or can be understood as when the neural network corresponds to AI-assisted spatial multiplexing), the input information of the neural network includes, but is not limited to, one or more of the following: packet error rate (PER), throughput, and communication device load information. Implementing AI-assisted spatial multiplexing through a neural network allows the communication device to dynamically adjust the interference threshold using AI, thereby increasing concurrency opportunities, improving throughput, and reducing latency.

[0110] Example 4: When a neural network is used to implement or support AI-assisted channel aggregation (or can be understood as when the neural network corresponds to AI-assisted channel aggregation), or when the neural network is used to implement or support AI-assisted channel aggregation combined with channel access (or when the neural network corresponds to AI-assisted channel aggregation combined with channel access), the input information of the neural network includes, but is not limited to, one or more of the following: carrier sensing results, channel aggregation status, data packet transmission results, duration of carrier sensing results, duration of channel aggregation, or received power. Optionally, the input information of the neural network may also include, but is not limited to, one or more of the following: packet loss rate, the time since the last successful transmission of this communication device, the time since the last successful transmission of other communication devices on the channel, the load information of this communication device, or the load information of other communication devices on the channel. By implementing AI-assisted channel aggregation through a neural network, the communication device can obtain one or more of the following information: whether channel aggregation is performed, which channels are aggregated, or how many channels are aggregated. By implementing AI-assisted channel aggregation and joint channel access functions through neural networks, communication devices can obtain one or more of the following information: whether channel aggregation is performed and whether access is granted, which channels are aggregated and whether access is granted, or how many channels are aggregated and whether access is granted.

[0111] It should be noted that the input information can differ for different functions; or even for the same function, the input information can differ when the required decision outcome is different. For example, for the same function, the required decision outcome can be directly proportional to the type of input information; the more decision outcomes required, the more input information is needed. In other words, the type of input information needs to be determined based on the function the neural network aims to achieve and / or the decision outcome of the neural network.

[0112] It is understood that the input information corresponding to the different functions shown in Examples 1 to 4 above is merely an example and should not be construed as a limitation on the embodiments of this application. In specific implementations, the input information of the neural network can be determined according to the function to be achieved by the neural network and / or the decision result, and this application will not list them all.

[0113] The number of input information can be understood as the quantity of input information to the neural network, or the quantity or length of input information to the neural network within a predetermined time period. Alternatively, the number of input information can also be understood as the historical quantity of input information contained in the neural network, the total length of the input information, or the length of each type of input information. For example, if the input information includes carrier sensing results and received power, and the number of input information is T, then the neural network requires a total of T carrier sensing results and T received power. As another example, if the input information includes carrier sensing results and received power, and the number of input information is T1 and T2 respectively, then the neural network requires T1 carrier sensing results and T2 received power. As yet another example, if the input information includes carrier sensing results and received power, and the number of input information represents the total quantity of input information, T, then the neural network requires [T / 2] carrier sensing results and [T / 2] received power, where the symbol "[T / 2]" indicates rounding down from T / 2.

[0114] The dimension of the input information can be used to determine the number of neurons corresponding to that input information. For example, when the dimension of each type of input information is the same, the number of neurons in the input layer = the dimension of the input information * the total number of input information items. For instance, if the input information includes carrier sensing results and received power, with T1 and T2 inputs respectively, the neural network requires T1 carrier sensing results and T2 received power items as input; and the dimension of the input information is the same, such as 2, then the number of neurons in the input layer = T1 * 2 + T2 * 2. For example, the number of neurons in the input layer = the number of input information 1 * the dimension of input information 1 + the number of input information 2 * the dimension of input information 2 + the number of input information 3 * the dimension of input information 3 + ... It can be understood that the input information 1, input information 2, and input information 3 shown above are only examples; the ellipsis indicates the number and dimension of other types of input information that the neural network may need for a particular function.

[0115] It should be noted that the input information, the number of input information, and the dimension of input information shown in this application may be included in the same input instruction information, or they may be included in different input instruction information. This application embodiment does not limit this.

[0116] 3. Output instruction information

[0117] Output indication information can be used to indicate at least one aspect of the structure of the output layer of a neural network: the number of neurons and the activation function. For example, the activation function can include any one or more of the following: no activation function (e.g., represented by None), rectified linear unit (ReLU), sigmoid, and softmax. For instance, each activation function can correspond to an index, so the output indication information can indicate the activation function using an index. As another example, the output information can indicate the activation function using a bitmap, etc., and so on.

[0118] 4. Functional indicator information

[0119] Functional indication information is used to indicate the functions that the neural network is to implement (supports or enables). For example, the functional indication information may be used to indicate any one or more of the following: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted RU allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission.

[0120] 5. Parameter indication information

[0121] Parameter indication information is used to indicate parameters of the neural network, such as biases and / or weights. The parameter indication information is used to indicate at least one of the following: the bias of neurons in the hidden layer structure, the weight of neurons in the hidden layer structure, the bias of neurons in the output layer structure, and the weight of neurons in the output layer structure.

[0122] It is understandable that the parameters of the neural network indicated by the parameter indication information can correspond to the functions indicated by the function indication information. For example, when the AP trains the neural network, it can train it separately according to different functions, thereby obtaining the parameters of the neural network corresponding to different functions.

[0123] The architecture of the communication device involved in this application and the MAC layer of the communication device are described in detail below.

[0124] The MAC layer is a key functional component of the 802.11 standard. Upper-layer applications invoke the MAC layer's functions by calling the interface primitives it provides. For example, the MAC layer can provide two main categories of interface primitives to the upper layers: data and management. The data portion provides the interface for sending and receiving ordinary data packets, while the management portion handles management functions such as channel management, connection management, QoS management, and function control management. Exemplary functions of the MAC layer may include, but are not limited to: rate adaptation, channel access, channel aggregation, RU allocation, spatial multiplexing, frame aggregation, beam management, or multi-link transmission. The MAC layer functions described here can be understood as the management functions provided by the MAC layer to communication devices to implement the aforementioned functions.

[0125] For example, Figure 3a This is a schematic diagram of an access point structure provided in an embodiment of this application. The AP can be multi-antenna or single-antenna. Figure 3a As shown, the AP includes a physical layer (PHY) and a medium access control (MAC) layer. The physical layer can process physical layer signals, and the MAC layer can process MAC layer signals. In this embodiment, the functionality of the MAC layer in the AP can be enhanced, such as by including AI processing functionality. That is, the functionality of the MAC layer is enhanced so that it can also implement AI-related functions or neural network-related functions. This AI processing function can be achieved through an AI processing module in the MAC layer (e.g., ...). Figure 3a Alternatively, the AI ​​processing function can be implemented through an AI processing module (not shown in the attached diagram) located outside the MAC layer. This AI processing module can be used to train neural networks and / or use the trained neural networks for inference and decision-making, etc., to realize various AI-assisted functions, such as AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted RU allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and one or more of the following AI-assisted multi-link transmission functions:

[0126] For example, at least one new MAC layer can be added to the AP, such as including two MAC layers: a first MAC layer and a second MAC layer. The first MAC layer can be used to process signals from other MAC layers, such as implementing existing MAC layer functions like association authentication, power control, channel measurement, multiple access, rate adaptation, channel aggregation, RU allocation, etc., which will not be listed here. The second MAC layer can be used to handle AI-related functions or functions related to neural networks. For example, Figure 3b This is a schematic diagram of another access point provided in an embodiment of this application. Figure 3a The AP shown is different, Figure 3b As shown, the second MAC may include an AI processing module. For example, Figure 3c This is a schematic diagram of another access point provided in an embodiment of this application. Figure 3b The difference is, Figure 3c The AI ​​processing module shown can be located outside the second MAC layer. This is understandable. Figure 3c The dashed line between the physical layer and the AI ​​processing module indicates that the AI ​​processing module can handle the functions required by the physical layer, or it may not. For an explanation of the AI ​​processing module, please refer to [link to documentation / reference]. Figure 3a The description of the second MAC layer will not be detailed here. For example, the first MAC layer can support the functions of older AP versions, while the second MAC layer can support the functions of newer AP versions. For example, when an AP detects poor performance during communication, such as a packet loss rate exceeding a preset threshold or a throughput falling below a corresponding threshold, the AP can implement related functions through the second MAC layer, such as AI-assisted channel access, AI-assisted rate adaptation, and AI-assisted channel aggregation, which will not be listed here. It is understood that this application embodiment does not limit the triggering conditions for the AP to use the first MAC layer or the second MAC layer.

[0127] It is understood that the second MAC layer can also be called the AI-MAC layer, etc., and this application does not limit the name of the second MAC.

[0128] For example, Figure 4aThis is a schematic diagram of a site structure provided in an embodiment of the application. The STA can be a single-antenna or multi-antenna device, and can have two or more antennas. The STA can include a PHY and a MAC layer. The physical layer can be used to process physical layer signals, and the MAC layer can be used to process MAC layer signals. In this embodiment, the functionality of the MAC layer in the STA can be enhanced, such as by including AI processing functionality, i.e., enhancing the functionality of the MAC layer so that it can also implement AI-related functions or neural network-related functions. This AI processing function can be achieved through an AI processing module in the MAC layer (e.g., ...). Figure 4a Alternatively, the AI ​​processing function can be implemented through an AI processing module (not shown in the attached diagram) located outside the MAC layer. This AI processing module can be used to train neural networks (powerful STAs, such as STAs with algorithms exceeding a certain threshold, can train neural networks) and / or use the trained neural network for inference and decision-making to achieve various AI-assisted functions, such as AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted RU allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and one or more of the following AI-assisted multi-link transmission functions:

[0129] For example, at least one new MAC layer can be added to the STA, such as including two MAC layer processing circuits, like a first MAC layer and a second MAC layer. For example, Figure 4b This is a schematic diagram of another site structure provided in an embodiment of this application. Figure 4a The difference between the STA shown is that, Figure 4b The second MAC shown may include an AI processing module. For example, Figure 4c This is a schematic diagram of the structure of another site provided in an embodiment of this application. Figure 4b The difference is, Figure 4c The AI ​​processing module shown can be located outside the second MAC layer. It is understood that, regarding... Figure 4b and Figure 4c The explanation can be found here. Figure 3b and Figure 3c The descriptions will not be detailed here.

[0130] It is understood that the AI ​​processing module shown above can also be called an AI processing circuit, a neural network-based processing circuit, a neural network processing circuit, a neural network processor, or an AI processing chip, etc. The specific name of the AI ​​processing module is not limited in the embodiments of this application.

[0131] It is understood that the 802.11 standard focuses on the PHY and MAC portions; therefore, the embodiments of this application do not elaborate on other portions included in the AP and STA. Exemplarily, the AP and STA may also include at least one of the following: memory, scheduler, controller, processor, or radio frequency circuitry.

[0132] The communication device described below focuses on the MAC layer; other descriptions of this communication device can be found above. Figures 3a to 3c as well as Figures 4a to 4c Etc. Optionally, the communication device shown below may include an AP and / or a STA, etc. Optionally, the communication device may also be understood as... Figures 3a to 3c The AI ​​processing module shown, or, as... Figures 4a to 4c The AI ​​processing module shown. Therefore, the embodiments of this application are for... Figure 5 The product shape of the communication device shown is not limited.

[0133] Figure 5 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application, such as... Figure 5 As shown, the communication device includes:

[0134] The system includes an input interface 501, a neural network processor 502, and an output interface 503. The neural network processor 502 stores a neural network, which includes an input layer and an output layer. Optionally, the input interface 501 and the output interface 503 can be understood as communication interfaces or pins, etc. Optionally, the input interface 501 and the output interface 503 can be understood as transceivers, etc. The neural network processor 502 can be understood as a neural network processing circuit, a neural network processing chip, or a neural network chip system, etc.

[0135] The input interface 501 is used to acquire at least one of input indication information, function indication information, output indication information, or parameter indication information. For explanations of input indication information, function indication information, output indication information, and parameter indication information, please refer to the related terminology introduction above; they will not be detailed here.

[0136] The neural network processor 502 is used to perform functions indicated by function instruction information based on the input layer and the output layer.

[0137] Output interface 503 is used to output the results after the neural network processor has performed its functions.

[0138] It is understandable that the functions performed by the neural network processor 502 can be understood as the functions that the MAC layer can implement, such as... Figure 3a and Figure 4a The enhanced MAC layer shown can implement functions such as... Figure 3b , Figure 3c, Figure 4b and Figure 4c The second MAC layer, as shown, implements certain functions. The functions achievable by a neural network processor will not be detailed here.

[0139] The communication device provided in this application embodiment can implement the functions of the MAC layer based on a neural network. For example, when the function performed by the MAC layer changes, the input and output layers of the neural network are adjusted accordingly, thereby enabling different functions to be implemented based on the neural network. In other words, different functions can be achieved by adjusting the input and output layers of the neural network, thus effectively simplifying the complexity of the neural network and facilitating hardware acceleration.

[0140] exist Figure 5 In the communication device shown, the input interface, used to acquire at least one of input indication information, function indication information, output indication information, or parameter indication information, can be implemented in the following ways:

[0141] Implementation Method 1

[0142] An input interface is used to acquire input indication information, function indication information, output indication information, and parameter indication information. Optionally, the input interface can acquire input indication information, function indication information, output indication information, and parameter indication information simultaneously (such as information contained in the same message). Optionally, the input interface can acquire function indication information and input indication information first (such as information contained in one message), and then acquire output indication information and parameter indication information (such as information contained in another message). Optionally, the input interface can acquire function indication information, input indication information, output indication information, and parameter indication information sequentially (such as different types of information contained in different messages). Optionally, the input interface can acquire function indication information, parameter indication information, input indication information, and parameter indication information sequentially, etc., which will not be listed here. That is to say, when the input interface needs to acquire all of the above information, the embodiments of this application do not limit the order in which the input interface acquires the above information.

[0143] Implementation Method Two

[0144] The input interface is used to obtain function indication information and parameter indication information. That is, when the function indicated by the function indication information is determined, the corresponding input and output indication information can be determined based on that function. For example, when the function to be implemented by the neural network is determined based on the function indication information, the corresponding input information, input layer structure, and output layer structure can be determined based on that function indication information. In other words, the input information, input layer structure, and output layer structure corresponding to each function can be determined by the AP (Application Processor), or defined by the protocol, etc. This can also be understood as: the input information, the number of inputs, the dimension of the input information, and the output layer structure corresponding to each function can be determined by the AP, or defined by the protocol, etc. For example, Figure 5 When the communication device shown is a STA, the input information, input layer structure, and output layer structure corresponding to each function can be broadcast from the AP to the STA. For example, Figure 5 When the communication device shown is an AP, the input information, input layer structure and output layer structure corresponding to each function can be determined by the AP itself or by the PAN coordinator, etc. This application embodiment does not limit this.

[0145] Implementation Method 3

[0146] The input interface is used to obtain function indication information. That is, when the function indicated by the function indication information is determined, the corresponding input indication information, output indication information, and parameter indication information can be determined based on the function. For example, when the function to be implemented by the neural network is determined based on the function indication information, the corresponding input information, input layer structure, output layer structure, and the weights and biases of the neural network can be broadcast by the AP, or defined by the protocol, etc. For an explanation of this implementation method, please refer to Implementation Method Two above; it will not be detailed here.

[0147] It is understood that the three implementation methods shown above are merely examples. For instance, an input interface might be used to obtain function indication information, input indication information, and parameter indication information; or, it might be used to obtain function indication information, output indication information, and parameter indication information; or, it might be used to obtain function indication information, input indication information, and output indication information, etc. These will not be listed individually here. For further explanation regarding the input interface and the methods for obtaining function indication information, input indication information, output indication information, and parameter indication information, please refer to the following section... Figure 9 and Figure 10 The description.

[0148] exist Figure 5In the communication device shown, the neural network processor determines the input layer structure and output layer structure of the neural network based on the information input through the input interface, and then obtains the result after executing the function according to the function instruction information of the input layer and the output layer.

[0149] For example, the method for determining the input layer structure of a neural network can be as follows:

[0150] Optionally, the structure of the input layer can be determined based on the function indication information. For example, each function can correspond to a different input layer structure. Alternatively, the structure of the input layer can be determined based on the input indication information. For example, if the number of input information and / or the dimensions of the input information are predetermined based on the function (e.g., by AP broadcast or protocol definition), the structure of the input layer is determined based on the input information indicated by the input indication information. Another example is when the dimensions of the input information are predetermined (e.g., by AP broadcast or protocol definition), the structure of the input layer can be determined based on the input information indicated by the input indication information and the number of input information. Yet another example is the structure of the input layer determined based on the input information indicated by the input indication information, the number of input information, and the dimensions of the input information. Optionally, the structure of the input layer can be determined based on both the function indication information and the input indication information.

[0151] For example, the method for determining the output layer structure of a neural network can be as follows:

[0152] Optionally, the structure of the output layer can be determined based on the function indication information. For example, each function corresponds to a different output layer structure. Optionally, the structure of the output layer can be determined based on the output indication information. Optionally, the structure of the output layer can be determined based on both the function indication information and the output indication information.

[0153] For example, the output of a neural network processor can be described as follows:

[0154] For example, the result of a neural network processor executing a function can be a decision result determined based on at least one of function indication information and output indication information. The decision result includes outputting the result of the neural network processor executing the function at its maximum value, or outputting the result of the neural network processor executing the function in a probabilistically randomized manner. That is, the decision result can select the result corresponding to the maximum output value, or the result randomly selected based on the output probability. In other words, the neural network processor can select a result from the output results of the neural network as the decision result, and the specific content of this decision result corresponds to the function indicated by the function indication information. Optionally, the decision result can be determined based on the function indicated by the function indication information. Optionally, the decision result can be determined based on the output indication information, such as the output indication information being used to indicate the output layer structure of the neural network and the decision result. For example, suppose the neural network corresponds to an AI-assisted rate adaptation function, and suppose the neural network has four output neurons, each neuron corresponding to a rate adaptation result, and the outputs of these four neurons as a whole are a probability distribution. The output value p_i of each neuron satisfies p_1+p_2+p_3+p_4=1. Two methods for selecting the decision result can be included: selecting the rate adaptation result corresponding to the largest p_i value, or randomly selecting the rate adaptation result according to the distribution of p_1, p_2, p_3, and p_4. Of course, the above decision result can also be determined autonomously by the communication device, etc., and this application embodiment does not limit this.

[0155] In one possible implementation, the neural network further includes a hidden layer. For example, the structure of the hidden layer can be fixed; for instance, the structure of the hidden layer corresponding to different functions can be broadcast by the AP or defined by a protocol. Alternatively, the indication information of the hidden layer structure corresponding to different functions can be included in the function indication information (or input indication information, parameter indication information, etc.). Or, the indication information of the hidden layer structure corresponding to different functions can be indicated to the STA by the AP through other information (such as information other than function indication information, input indication information, output indication information, and parameter indication information). For example, if the hidden layer structure includes two structures, the communication device can select one structure from these two structures based on the function indication information. As another example, the hidden layer structure can include three structures, and the communication device can select one structure from these three structures based on the function indication information. For example, the structure of the hidden layer can be the same for different functions. For example, the input layer structure and output layer structure of the neural network can vary according to different functions, but the hidden layer structure can be the same.

[0156] Figure 6 This is a schematic diagram of the hidden layer structure of the neural network provided in an embodiment of this application. For example... Figure 6As shown, both of the first two hidden layer structures employ a fully connected layer (FC) with a residual structure (also known as a fully connected network). For example, hidden layer structure one (such as...) Figure 6 The example opt1 shows an 8-layer fully connected network with 3 skip connections, 32 neurons per layer, and ReLU activation function for each neuron. Hidden layer structure two (as shown) Figure 6 The shown opt2) uses a 6-layer fully connected network, including 2 skip connections, with 64 neurons per layer, and each neuron has a ReLU activation function. Hidden layer structure three (as shown) Figure 6 As shown in opt3), a recurrent neural network is used. The input first passes through two fully connected layers, then one gated recurrent unit (GRU) layer, and finally another fully connected layer connecting to the output layer. Each layer has 32 neurons, and the activation function of the fully connected layers is ReLU. It should be noted that the GRU layer in hidden layer structure 3 can also be replaced with other recurrent neural networks (RNNs), such as long short-term memory (LSTM) networks or bidirectional LSTMs.

[0157] The three hidden layer structures provided in this application have low complexity. Hidden layer structure one and hidden layer structure two have inference computation complexity less than 512-point complex Fast Fourier Transform (FFT). Hidden layer three uses a GRU structure, which can better extract information from historical observations and improve prediction accuracy. Furthermore, regardless of the function the neural network is meant to perform, the above hidden layer structures can be used. This low-complexity neural network core structure with unified multi-feature characteristics facilitates hardware acceleration.

[0158] The communication device provided in this application embodiment can realize AI-assisted functions based on neural networks. The input layer structure and output layer structure for different functions can be adjusted according to the information obtained from the input interface, effectively simplifying the complexity of the neural network. Meanwhile, the hidden layer structure of the neural network can be as follows: Figure 6 Any of the hidden layer structures shown can be achieved through... Figure 6 The hidden layer structure shown has good performance for different functions (such as good algorithm performance, high throughput, low latency, etc.), which not only facilitates hardware acceleration, but also has good scalability.

[0159] The structure of the communication device shown above is only an example. The communication device provided in this application will be described below in terms of module division. Figure 7a This is a schematic diagram of another communication device provided in an embodiment of this application, such as... Figure 7a As shown, the communication device includes:

[0160] The system comprises an input control module, a neural network module, an output control module, and an action decision module. The neural network module includes an input layer, hidden layers, and an output layer. The hidden layer structure can be referenced as follows: Figure 6 The three structures shown are illustrated. The methods for determining the input layer structure and output layer structure can be found in [reference needed]. Figure 5 The descriptions shown will not be elaborated upon here.

[0161] For example, the input control module is used to determine the structure of the neural network input layer based on input indication information and / or function indication information, and the output control module is used to determine the structure of the neural network output layer based on output indication information and / or function indication information. The neural network module can execute the function indicated by the function indication information according to the input layer, hidden layer, and output layer. The action decision module can select an action based on the output information of the neural network, such as selecting the action corresponding to the maximum output value, or randomly selecting an action based on the output probability.

[0162] For example, the above-mentioned input control module, neural network module, output control module, and action decision module can correspond to Figure 5 The neural network processor shown. That is, the input control module, neural network module, output control module, and action decision module can be used... Figure 5 The neural network processor implementation shown.

[0163] Understandable, regarding Figure 7a For a detailed description of the communication device shown, please refer to [link / reference]. Figure 5 and Figure 6 These will not be detailed here.

[0164] It should be noted that, Figure 5 and Figure 6 The communication devices shown can be understood as different ways of dividing the same communication device. The above division methods are illustrative, and other division methods may be used in actual implementation. As mentioned above... Figure 6 This can be understood as dividing a communication device into functional modules. Besides dividing each function into its own modules, two or more functions can be integrated into a single module. These modules can be implemented in hardware or as software functional modules. In specific implementations, the communication device may also include... Figure 5 and Figure 6 This application does not limit the number of other devices or modules.

[0165] It should be understood that, as long as the above conditions are met... Figure 5 or Figure 6 Any form of product that incorporates the functionality of the aforementioned communication device falls within the protection scope of the embodiments of this application.

[0166] To further understand the communication device shown in the embodiments of this application, the following description uses examples of functions implemented by neural networks.

[0167] Example 1

[0168] For example, such as Figure 7b As shown, the input information includes carrier sensing results, action information, and PER. The input information has a dimension of 2, and each input information has 1 value. Therefore, the input layer structure includes 4 neurons. The output indication information indicates 2 neurons, the activation function is softmax, and the action decision method is to select the action corresponding to the maximum value of the neural network output (e.g., ...). Figure 7b The value of argmax shown indicates that the output layer contains two neurons, and the decision result is selected from the output of the neural network with the maximum value. The hidden layer structure can be as follows: Figure 6 Any of the structures shown. Figure 7a Taking the communication device shown as an example, when the communication device receives input instruction information, the input control module can be in an active state, controlling the number of neurons in the input layer. When the communication device receives output instruction information, the output control module can be in an active state, controlling the number of neurons and activation function in the output layer, and the action decision module can select the action corresponding to the maximum value of the neural network output.

[0169] Example 2

[0170] For example, the function indication information indicates either AI-assisted rate adaptation or MCS selection. The input information includes the carrier sense result (RSSI) and the received power (P), and the number of each input item is T, meaning the input information can be represented as (RSSI... t ,P t The value of t is a positive integer less than or equal to T (i.e., 1, 2, 3, 4, ..., T). If each input information corresponds to one neuron (i.e., the dimension of the input information is 1), then the input layer structure of the neural network includes 2T neurons.

[0171] The hidden layer structure of the neural network is a 32*8 residual fully connected network, that is, the hidden layer contains a total of 8 fully connected layers, 3 skip connections, 32 neurons in each layer, and the activation function of each neuron is ReLU.

[0172] Based on output indication information and / or function indication information, it is determined that the output layer structure includes N neurons, and the activation function of each neuron is a linear function. If an output layer neuron outputs the expected reward of one level MCS, then the output of the neural network is N levels MCS (e.g., MCS0 to MCS). N-1 The expected reward is calculated as follows. Assume the decision-making method involves selecting the MCS corresponding to the maximum output layer value. Therefore, after obtaining parameter indication information, the communication device can infer from the historical T RSSIs and T Ps detected by the neural network, and select the MCS corresponding to the maximum output layer value for transmission.

[0173] It is understood that the input information shown above is only an example. If the communication device expects more decision results, the types of input information may include other information, which will not be detailed here.

[0174] Currently, most commercially available rate adaptive algorithms for Wi-Fi are sampling-based, including the Minstrel algorithm and the Iwl-mvm-rs algorithm. Their basic principle is to sample different MCS (Mean Cross Section) levels and statistically analyze the characteristics of each MCS, such as throughput and packet loss rate, ultimately determining the MCS based on these statistical features. However, this type of method has at least the following drawbacks: 1. Because the sampling period is long (if the sampling period is too short, it cannot reflect the channel conditions), the obtained statistical values ​​cannot reflect the current wireless environment (because the wireless environment changes over time), resulting in poor transmission performance; 2. After changes in the wireless environment, long-term resampling is required to adjust to a suitable MCS, leading to poor adaptability.

[0175] However, the embodiments of this application utilize the predictive power of neural networks for rate-adaptive MCS selection, which is highly adaptable, low in complexity, and has good scalability, and can achieve multi-vendor interoperability (vendor interoperability can be understood as the hidden layer structure being fixed, allowing different vendors to use it, such as...). Figure 6 The hidden layers shown, as well as the methods for determining the input and output layer structures, have been negotiated.

[0176] Example 3

[0177] For example, the function indicated by the function indication information is AI-assisted channel aggregation. Assume the input information is divided into two categories: basic observation information and derived observation information. The number of basic observation information is T', and the number of derived observation information is 1. Each type of input information corresponds to 1 neuron (i.e., the dimension of the input information is 1). Assume the number of channels corresponding to the input information is M. Then the input information of the neural network can be understood as the historical T' basic observation information and the historical 1 derived observation information of M channels. Further assume that at time t, the basic observation information of the communication device for the m-th channel ∈ {1,…,M} is o, a, I, and the derived observation information is load. 0 and load 1 Where o represents the carrier sensing result, a represents the channel aggregation case, I represents the duration of a (o,a) pair (i.e., the time during which o and a remain unchanged), and load 0 This indicates the load of the device itself (such as the load of a communication device). 1 This represents the load of other nodes on this channel (i.e., the load information of other communication devices on the channel). Therefore, the input layer of the neural network contains (3*T′+2*1)*M neurons.

[0178] The hidden layer structure of the neural network is a 32*8 residual fully connected network, that is, the hidden layer contains a total of 8 fully connected layers, 3 skip connections, 32 neurons in each layer, and the activation function of each neuron is ReLU.

[0179] Based on the output indication information and / or function indication information, it is determined that the output layer structure includes N neurons, and the activation function of each neuron is a linear function. Then, the output of the neural network can be the expected reward for N channel aggregation methods, i.e., Q(Ch0) to Q(Ch... N-1 ), where Ch n This represents a channel aggregation method. Assume the decision is made by selecting the channel aggregation method corresponding to the maximum output layer value.

[0180] Therefore, after obtaining the parameter indication information, the communication device can infer from the historical dataset of the M channels it has monitored based on the neural network, and select the channel aggregation method corresponding to the maximum value of the output layer for transmission.

[0181] The communication device in this application embodiment can perform inference and decision-making through a neural network to achieve AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, or AI-assisted channel access, thereby improving performance. Optionally, when the parameter indication information is obtained through training on other communication devices, the communication device does not need to train its own neural network, which can also reduce the power consumption of the communication device. In addition, compared with traditional channel aggregation methods (such as channel aggregation methods that do not use neural networks), the embodiments of this application also have advantages such as strong adaptability, low complexity, good scalability, and the ability to achieve interoperability among multiple vendors.

[0182] Example 4

[0183] The function indicated by the functional indication information is AI-assisted channel access, that is, using AI to decide whether the communication device should access the network. For example, the input information may include access behavior, carrier sensing result, duration of the carrier sensing result, the time since the last successful transmission of this communication device, and the time since the last successful transmission of other communication devices. Assuming each input information corresponds to one neuron, the neural network input layer includes n historical pieces of information (access behavior, carrier sensing result, duration of the carrier sensing result), and the time since the last successful transmission of this communication device and the time since the last successful transmission of other communication devices. Therefore, the input layer includes 3n+2 neurons, such as n=10 or 20, etc., this is just an example. The hidden layer structure can be as follows: Figure 6 The opt1 or opt2 shown.

[0184] Based on the functional indication information and / or output indication information, the output layer of the neural network has 2 neurons, and the activation function is a linear activation function. Therefore, the output of the neural network can include the expected reward for accessing the channel and the expected reward for not accessing the channel. The decision result can be to select the action with the largest expected reward, i.e., the action corresponding to the maximum output value.

[0185] Example 5

[0186] The function indicated by the functional indication information is AI-assisted spatial multiplexing. For example, AI can be used to dynamically adjust the interference threshold, thereby increasing concurrency opportunities, improving throughput, and reducing latency. For instance, input information may include the number of consecutive packet losses, throughput, PER, and the load information of the communication device. Assuming each input corresponds to one neuron, the input layer of the neural network includes four neurons. The hidden layer structure can be as follows: Figure 6 As shown in opt3, since GRU is a recurrent neural network that makes predictions based on information from a historical period, the four neurons in the input layer refer to the input dimension of GRU at each time step.

[0187] The number of neurons in the output layer of the neural network, determined based on functional indication information and / or output indication information, is n, and the activation function is softmax. The output of the neural network can then include the probabilities of n interference thresholds. The decision result can be the interference threshold corresponding to the maximum output value. Therefore, if the carrier sensing result is greater than the interference threshold, the communication device considers the channel busy; if the carrier sensing result is less than the interference threshold, the communication device considers the channel idle.

[0188] It is understood that the five examples described above are merely examples. The communication device provided according to the embodiments of this application can also realize more functions, and the input layer structure and output layer structure corresponding to different functions will also be different, which will not be listed one by one here.

[0189] Based on the communication devices described above, the method provided in this application will be described below using the first and second communication devices as examples. It is understood that the description of the first and second communication devices can be found above regarding... Figures 5 to 7a These will not be detailed here.

[0190] In the communication apparatus described below in this application, the first communication device can be a resource-constrained node in the network (e.g., a node that cannot complete neural network training but can use neural networks for inference and decision-making), such as a STA in an AP-STA communication scenario, or a weaker AP (e.g., a traditional access point) in a multi-AP collaboration scenario, or a weaker STA (e.g., a traditional site) in a D2D communication scenario. Correspondingly, the second communication device can be a resource-unconstrained or more powerful node in the network (e.g., a node that can both complete neural network training and use neural networks for inference and decision-making), such as an AP in an AP-STA communication scenario, or a more powerful AP (e.g., an access point supporting the latest 802.11 standard) in a multi-AP collaboration scenario, or a more powerful STA (e.g., a site supporting the latest 802.11 standard) in a D2D communication scenario. Here, "latest 802.11 standard" can refer to future Wi-Fi standards, such as Wi-Fi 8 (the next generation of 802.11be), Wi-Fi 9 (the next generation of Wi-Fi 8), and Wi-Fi 10 (the next generation of Wi-Fi 9). "Traditional access point" can refer to an access point that only supports standards prior to the latest 802.11 standard, and "traditional site" can refer to a site that only supports standards prior to the latest 802.11 standard.

[0191] Optionally, the communication device in this application includes a module capable of performing reasoning and decision-making using a neural network, such as an AI processing module. For example, the AI ​​processing module in the second communication device can not only perform reasoning and decision-making using a neural network, but also train the neural network; while the AI ​​processing module in the first communication device can perform reasoning and decision-making using a neural network trained by the second communication device. Of course, in some optional embodiments, the AI ​​processing module in the first communication device can also train the neural network.

[0192] It is understood that the differences between the first communication device and the second communication device include, but are not limited to: the strength of their capabilities, whether their resources are limited, or whether they have the ability to train neural networks.

[0193] The communication method provided in this application embodiment enables resource-constrained nodes in a network (which may refer to the first communication device in this application) to obtain relevant information (such as the function, structure, parameters, input, and output of the neural network) from a more powerful node in the network (which may refer to the second communication device in this application). This lays the foundation for resource-constrained nodes to achieve AI-assisted rate adaptation, AI-assisted rate adaptation joint channel access, and AI-assisted channel aggregation. In general, because neural network training requires significant computing power and energy consumption, usually only some nodes (typically more powerful nodes) can independently train and use the neural network; while other nodes (typically resource-constrained nodes) cannot complete neural network training or the cost of neural network training is too high (e.g., very high energy consumption), making it impossible to use AI-based rate adaptation algorithms to improve performance. However, using the method provided in this application embodiment, (after the first communication device requests the activation of AI-assisted functions, such as AI-assisted rate adaptation, AI-assisted channel access, or AI-assisted rate adaptation combined with channel access, etc.) the second communication device can inform the first communication device of relevant information about its trained neural network (such as the neural network's function, structure, parameters, input and output information, etc.). This allows the first communication device to recover the neural network trained by the second communication device based on this information, and to use its observed data and the recovered neural network to perform inference and decision-making, obtaining corresponding decision results for transmission; without the first communication device needing to train the neural network itself. This not only lays the foundation for resource-constrained nodes to implement AI-assisted rate adaptation or AI-assisted rate adaptation combined with channel access, but also reduces the energy consumption of the first communication device.

[0194] Figure 8This is a flowchart illustrating a communication method provided in an embodiment of this application. The method can be applied to a first communication device (such as a STA or a functionally limited AP), which stores a neural network, including an input layer and an output layer. Figure 8 As shown, the method includes:

[0195] 801. The first communication device acquires at least one of input indication information, function indication information, output indication information, or parameter indication information.

[0196] For example, the first communication device can simultaneously acquire input indication information, function indication information, output indication information, and parameter indication information. Alternatively, the method by which the first communication device acquires the above information can be referred to Figure 9 Steps 901 and 902 are shown.

[0197] 802. The first communication device performs the function indicated by the function instruction information based on the input layer and the output layer.

[0198] 803. The first communication device outputs the result after the neural network performs its function.

[0199] Understandably, for detailed explanations of steps 801 to 803, please refer to [link / reference]. Figures 5 to 7a These will not be detailed here.

[0200] The following will explain further. Figure 8 The method shown.

[0201] Figure 9 This is a flowchart illustrating another communication method provided in an embodiment of this application, such as... Figure 9 As shown, the method includes:

[0202] 901. The first communication device sends a function request message, which includes function indication information. The function request message is used to request the activation of the function indicated by the function indication information. Correspondingly, the second communication device receives the function request message.

[0203] For example, when the first communication device detects poor performance during communication, such as a packet loss rate exceeding a preset threshold, throughput falling below a corresponding threshold, rate variation exceeding a corresponding threshold, or the number of consecutive transmission failures exceeding a threshold, it can send a function request message to the second communication device, such as requesting the activation of AI-assisted rate adaptation or AI-assisted rate adaptation combined with channel access. For example, when the first communication device wants to transmit data for a certain service (which is a latency-sensitive service or a service with high latency requirements), it can send a function request message to the second communication device, such as requesting the activation of AI-assisted channel access or AI-assisted spatial multiplexing. The specific AI-assisted functions to be activated can be determined by the first communication device according to its own strategy; this embodiment does not limit this.

[0204] For example, the function requested by the first communication device to be enabled may exist in the function request message in the form of a field, or in the form of a bitmap. This application embodiment does not limit this.

[0205] For example, function indication information exists in the form of a field in the function request message, such as the AI ​​function request field. For instance, when the AI ​​function request field is set to a first value, it requests to enable the AI-assisted rate adaptive function; when it is set to a second value, it requests to enable the AI-assisted rate adaptive combined channel access function; when it is set to a third value, it requests to enable AI-assisted channel aggregation; and when it is set to a fourth value, it requests to enable AI-assisted channel aggregation combined channel access, etc., and so on. This application does not limit the bit length occupied by the AI ​​function request field. For example, the bit length occupied by the AI ​​function request field can be determined based on the type of function requested in the function request message.

[0206] For example, function indication information exists in the form of a bitmap within the function request message. This bitmap could be called an AI function request bitmap, where each bit corresponds to a function. When a bit in the AI ​​function request bitmap is set to 1, the corresponding function is requested to be enabled. Alternatively, a bit in the AI ​​function request bitmap can be set to 0, also requesting the enabling of the corresponding function. The length of the AI ​​function request bitmap can be determined based on the type of function requested in the function request message.

[0207] 902. The second communication device sends a function response message in response to the function request message. Correspondingly, the first communication device receives the function response message.

[0208] Optionally, after receiving the aforementioned function request message, the second communication device may send a function response message to the first communication device to respond to (i.e., agree or refuse) the request in the function request message. The function response message may exist in the form of a field. For example, the function response message may be 1 bit long; when the bit is set to 1, it indicates agreement to the function request message; when the bit is set to 0, it indicates rejection of the function request message. Alternatively, when the bit is set to 0, it indicates agreement to the function request message; when the bit is set to 1, it indicates rejection of the function request message. This embodiment of the application does not limit whether setting the function response message to 1 or 0 indicates agreement to the function request message.

[0209] Optionally, when the second communication device agrees to the request in the aforementioned function request message, the second communication device may send the parameters of the neural network to the first communication device. In one possible implementation, the second communication device may carry the parameters of the neural network in the aforementioned function response message. It is understood that if the function response message carries the parameters of the neural network (which may be all or some of the parameters), it indicates that the second communication device agrees to the request in the function request message, and in this case, it is not necessary to use additional fields or bits (in the function response message) to indicate agreement to the function request message. Of course, even if the function response message carries the parameters of the neural network (which may be all or some of the parameters), additional fields or bits may be used in the function response message to indicate agreement to the function request message. This application embodiment does not impose any limitations on this.

[0210] In another possible implementation, the second communication device can send the parameters of the neural network separately (i.e., send the parameters of the neural network without using the function response message), as described below regarding step 903.

[0211] 903. The second communication device sends indication information, which includes at least one of the following: input indication information, output indication information, and parameter indication information. Correspondingly, the first communication device receives the indication information.

[0212] Optionally, step 903 above can also be replaced by: the second communication device sending indication information, which includes input indication information and / or output indication information. Correspondingly, the first communication device receives the indication information. And the second communication device sending parameter indication information, which the first communication device receives.

[0213] It is understood that the parameter indication information sent by the second communication device is the parameter of the neural network corresponding to the function indicated by the function indication information. In other words, the second communication device can send the parameters of its trained neural network to the first communication device.

[0214] The same step 903 described above can enable the first communication device and the second communication device to reach an agreement on the input and output information of the neural network, and at the same time, to reach an agreement on the input layer structure and output layer structure of the neural network.

[0215] In one example, the hidden layer structure of this neural network is a variety of predefined hidden layer structures (such as...). Figure 6 The hidden layer structure can be one of the three types shown. Each hidden layer structure can be characterized by one or more of the following information: the number of hidden layers, the number of neurons per layer, the activation function used by each neuron in the layer, and the connection method between neurons in the layers. These predefined multiple hidden layer structures can be specified in a standard protocol, or they can be communicated to the first communication device in advance by the second communication device. For example, the first and second communication devices can exchange the predefined multiple hidden layer structures during the connection establishment process (if the first and second communication devices are STA and AP respectively, then the connection establishment process can refer to the association process). For example, the second communication device can indicate which of the predefined multiple hidden layer structures the neural network's hidden layer structure is through the hidden layer structure indication information. For example, three hidden layer structures are predefined, identified by selection 1 (opt1), selection 2 (opt 2), and selection 3 (opt 3); if the hidden layer indicated by the hidden layer structure indication information is opt 2, then it means that the hidden layer structure of the neural network is the hidden layer structure identified by opt 2 (i.e., one or more of the following: the number of hidden layers, the number of neurons per layer, the activation function used by each neuron in the layer, and the connection method between neurons in the layers). It is understood that the embodiments of this application can save bits and reduce signaling overhead by predefining a variety of hidden layer structures and indicating which of the predefined hidden layer structures to use through indication information.

[0216] In another example, the hidden layer structure of the neural network can be broadcast by the AP, and the broadcast message can include the correspondence between functions and hidden layer structures. Binding functions to hidden layer structures ensures that when the function indicated by the function indication information is determined, the corresponding hidden layer structure is also determined, further reducing signaling overhead.

[0217] In another example, the hidden layer structures corresponding to different functions are the same. That is to say, the different functions implemented by the neural network can all correspond to... Figure 6 Any of the hidden layers shown.

[0218] 904. The first communication device performs the function indicated by the function instruction information based on the input layer and the output layer.

[0219] The explanation of how the first communication device determines the structure of the neural network can be found above, and will not be elaborated here.

[0220] Figure 10 This is a flowchart illustrating another communication method provided in an embodiment of this application, such as... Figure 10 As shown, the method includes:

[0221] 1001. The second communication device sends a function request message, which includes function indication information. This function indication information is used to notify the first communication device to enable the function indicated by the function indication information. Correspondingly, the first communication device receives the function request message.

[0222] For example, the function request message is used to request or notify the activation of any of the following functions: AI-assisted rate adaptation function, AI-assisted rate adaptation combined with channel access function, AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, etc.

[0223] 1002. The first communication device sends a function response message in response to the function request message. Correspondingly, the second communication device receives the function response message.

[0224] This feature response message can be used to agree to or reject the request message for this feature.

[0225] Optionally, the aforementioned function request message may also include parameter indication information. That is, while the second communication device is sending the function indication information, it can also send parameter indication information to the first communication device.

[0226] In this embodiment, a more capable node (such as an AP or a second communication device) can send a function request message to request or notify one or more resource-constrained nodes (such as a STA or a first communication device) to enable AI-assisted functions. This not only lays the foundation for resource-constrained nodes to achieve AI-assisted rate adaptation or AI-assisted rate adaptation joint channel access, but also enables AI-assisted functions for multiple resource-constrained nodes through a single interaction, saving air interface resources.

[0227] Of course, function indication information and parameter indication information can also be located in different messages.

[0228] 1003. The second communication device sends instruction information, which includes at least one of the following: input instruction information and output instruction information.

[0229] It is understood that step 1003 is illustrated by the example of input indication information and output indication information being contained in the same message. The input indication information and output indication information can also be contained in different messages, which will not be described in detail here.

[0230] 1004. The first communication device performs the function indicated by the function instruction information based on the input layer and the output layer.

[0231] The explanation of how the first communication device determines the structure of the neural network can be found above, and will not be elaborated here.

[0232] This application also provides a wireless communication system, which includes a first communication device and a second communication device. The first communication device and the second communication device can be used to execute any of the foregoing embodiments (such as...). Figures 8 to 10 The method in ).

[0233] In addition, this application also provides a computer program for implementing the operations and / or processes performed by the first communication device in the method provided in this application.

[0234] This application also provides a computer program for implementing the operations and / or processes performed by the second communication device in the method provided in this application.

[0235] This application also provides a computer-readable storage medium storing computer code that, when executed on a computer, causes the computer to perform the operations and / or processes performed by the first communication device in the method provided in this application.

[0236] This application also provides a computer-readable storage medium storing computer code that, when executed on a computer, causes the computer to perform the operations and / or processes performed by the second communication device in the method provided in this application.

[0237] This application also provides a computer program product, which includes computer code or a computer program that, when run on a computer, causes the operations and / or processes performed by the first communication device in the method provided in this application to be executed.

[0238] This application also provides a computer program product, which includes computer code or a computer program that, when run on a computer, causes the operations and / or processes performed by the second communication device in the method provided in this application to be executed.

[0239] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, or it may be an electrical, mechanical, or other form of connection.

[0240] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the technical effects of the solutions provided in the embodiments of this application.

[0241] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0242] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0243] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A communication device, characterized in that, The device includes: The system includes an input interface, a neural network processor, and an output interface. The neural network processor stores a neural network, which includes an input layer, an output layer, and at least one hidden layer. The structure of the hidden layer is the same for different Media Access Control (MAC) functions. The input interface is used to acquire input indication information, function indication information, and output indication information; The neural network processor is configured to execute the MAC function indicated by the function indication information according to the input layer and the output layer. The function indication information indicates one or more of the following MAC functions: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted resource unit (RU) allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission. The structure of the input layer is determined based on at least one of the following: the input indication information and the function indication information; the structure of the output layer is determined based on at least one of the following: the output indication information and the function indication information. The input indication information indicates at least one of the following: the input information of the neural network, the number of input information items, and the dimension of the input information. The output indication information indicates at least one of the following structures of the output layer: the number of neurons and the activation function. The output interface is used to output the result after the neural network processor executes the MAC function.

2. The apparatus according to claim 1, characterized in that, The input interface is also used to input parameter indication information, which is used to indicate at least one of the following: the bias of the neuron in the neural network, the weight of the neuron in the neural network; the parameter indication information is also used to indicate at least one of the following: the bias of the neuron in the structure of the hidden layer, the weight of the neuron in the structure of the hidden layer, the bias of the neuron in the structure of the output layer, and the weight of the neuron in the structure of the output layer.

3. The apparatus according to any one of claims 1-2, characterized in that, The result of the neural network processor executing the MAC function includes a decision result determined based on at least one of the function indication information and the output indication information. The decision result includes outputting the result of the neural network processor executing the MAC function in a maximum value manner, or outputting the result of the neural network processor executing the MAC function in a probabilistic random selection manner.

4. The apparatus according to claim 1, characterized in that, The number of input information items and the dimension of the input information items are used to determine the number of neurons in the input layer.

5. The apparatus according to claim 1 or 4, characterized in that, The input information includes one or more of the following: Carrier sensing results, received power, modulation and coding scheme (MCS) used in the data packets, number of multiple-input multiple-output (MIMO) data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, duration of carrier sensing results or duration of channel aggregation. The channel access behavior includes access or non-access.

6. The apparatus according to claim 5, characterized in that, The input information also includes any one or more of the following: Packet loss rate, throughput, the time since the last successful transmission of the communication device, the time since the last successful transmission of other communication devices on the channel, the load information of the communication device or the load information of other communication devices on the channel, wherein the other communication devices are communication devices other than the communication device.

7. A communication method based on implementing Media Access Control (MAC) function using a neural network, characterized in that, The method is applied to a communication device that stores a neural network, the neural network including an input layer, an output layer, and at least one hidden layer, the structure of the hidden layer being the same for different MAC functions, the method comprising: Acquire input indication information, function indication information, and output indication information; The MAC function indicated by the function indication information is executed according to the input layer and the output layer. The function indication information is used to indicate any one or more of the following MAC functions: AI-assisted channel access, AI-assisted channel aggregation, AI-assisted channel aggregation combined with channel access, AI-assisted rate adaptation, AI-assisted rate adaptation combined with channel access, AI-assisted resource unit (RU) allocation, AI-assisted spatial multiplexing, AI-assisted frame aggregation, AI-assisted beam management, and AI-assisted multi-link transmission. The structure of the input layer is determined according to at least one of the following: the input indication information and the function indication information; the structure of the output layer is determined according to at least one of the following: the output indication information and the function indication information; the input indication information is used to indicate at least one of the following: the input information of the neural network, the number of input information, and the dimension of the input information; the output indication information is used to indicate at least one of the following in the structure of the output layer: the number of neurons and the activation function. Output the result after performing the MAC function based on the neural network.

8. The method according to claim 7, characterized in that, The method further includes: Obtain parameter indication information, which is used to indicate at least one of the following: the bias of the neuron in the neural network, and the weight of the neuron in the neural network; The parameter indication information is also used to indicate at least one of the following: the bias of neurons in the structure of the hidden layer, the weight of neurons in the structure of the hidden layer, the bias of neurons in the structure of the output layer, and the weight of neurons in the structure of the output layer.

9. The method according to any one of claims 7-8, characterized in that, The result of the neural network executing the MAC function includes a decision result determined based on at least one of the function indication information and the output indication information. The decision result includes outputting the result of the neural network processor executing the MAC function in a maximum value manner, or outputting the result of the neural network processor executing the MAC function in a probabilistic random selection manner.

10. The method according to claim 7, characterized in that, The number of input information items and the dimension of the input information items are used to determine the number of neurons in the input layer.

11. The method according to claim 7 or 10, characterized in that, The input information includes one or more of the following: Carrier sensing results, received power, modulation and coding scheme (MCS) used in the data packets, number of multiple-input multiple-output (MIMO) data streams, channel bandwidth, queuing delay, access delay, channel access behavior, data packet transmission results, channel information, channel aggregation status, duration of carrier sensing results or duration of channel aggregation. The channel access behavior includes access or non-access.

12. The method according to claim 11, characterized in that, The input information also includes any one or more of the following: Packet loss rate, throughput, the time since the last successful transmission of the communication device, the time since the last successful transmission of other communication devices on the channel, the load information of the communication device or the load information of other communication devices on the channel, wherein the other communication devices are communication devices other than the communication device.

13. The method according to any one of claims 7-12, characterized in that, The acquisition of at least one of the following: input indication information, function indication information, output indication information, or parameter indication information includes: The communication device sends a function request message to the access point (AP). The function request message includes the function indication information and is used to request the activation of the function indicated by the function indication information. The communication device receives a function response message from the AP, the function response message being used to confirm the function request message; The communication device receives indication information from the AP, the indication information including at least one of the following: the input indication information, the output indication information, and the parameter indication information.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed, performs the method according to any one of claims 7-13.

15. A computer program product, characterized in that, The computer program product includes a computer program that, when executed, performs the method according to any one of claims 7-13.

Images