Machine learning signaling and operation for wireless local area network (WLAN)

Abstract

An apparatus for wireless communication by a first wireless local area network (WLAN) device has a memory and one or more processors coupled to the memory. The processor is configured to transmit a first message indicating support of the first WLAN device for machine learning. The processor is also configured to receive a second message from a second WLAN device. The second message indicates support of the second WLAN device for one or more machine learning model types. The processor is configured to activate a machine learning session with the second WLAN device based at least in part on the second message. The processor is further configured to receive machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. patent application No. 17 / 953,138, filed on September 26, 2022, and entitled “MACHINE LEARNINGSIGNALING AND OPERATIONS FOR WIRELESS LOCAL AREA NETWORKS (WLANs),” the disclosure of which is expressly incorporated by reference in its entirety. Technical Field

[0003] The present disclosure relates generally to wireless communications and, more particularly, to signaling and operational aspects of a machine learning framework for wireless local area networks (WLANs). Background Art

[0004] A wireless local area network (WLAN) can be formed by one or more wireless access points (APs) that provide a shared wireless communication medium for use by multiple client devices, also known as wireless stations (STAs). The basic building block of a WLAN that complies with the Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards is the basic service set (BSS), which is managed by the AP. Each BSS is identified by a basic service set identifier (BSSID), which is advertised by the AP. The AP periodically broadcasts beacon frames to enable any STA within the range of the AP to establish or maintain a communication link with the WLAN.

[0005] Machine learning techniques include supervised learning, unsupervised learning, and reinforcement learning. However, the 802.11 specification currently does not have support for using these machine learning techniques across devices. Proprietary models can be used by individual STAs or APs, but these models may only be used to optimize features that are left to be implemented. It would be desirable to apply machine learning techniques to wireless communications to achieve greater efficiency. Summary of the invention

[0006] Some aspects of the present disclosure relate to an apparatus for wireless communication by a first wireless local area network (WLAN) device. The apparatus has a memory and one or more processors coupled to the memory. The processor is configured to send a first message indicating support for machine learning by the first WLAN device. The processor is also configured to receive a second message from a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The processor is also configured to activate a machine learning session with the second WLAN device based at least in part on the second message. The processor is configured to receive machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0007] In other aspects of the present disclosure, a method for wireless communication by a first wireless local area network (WLAN) device includes sending a first message indicating support for machine learning by the first WLAN device. The method also includes receiving a second message from a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The method also includes activating a machine learning session with the second WLAN device based at least in part on the second message. The method includes receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0008] Some aspects of the present disclosure relate to an apparatus for wireless communication by a first wireless local area network (WLAN) device. The apparatus has a memory and one or more processors coupled to the memory. The processor is configured to receive a first message indicating support for machine learning by the first WLAN device. The processor is also configured to send a second message to a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The processor is also configured to activate a machine learning session with the second WLAN device based at least in part on the second message. The processor is configured to send machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0009] In other aspects of the present disclosure, a method for wireless communication by a first wireless local area network (WLAN) device includes receiving a first message indicating support for machine learning by the first WLAN device. The method also includes sending a second message to a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The method also includes activating a machine learning session with the second WLAN device based at least in part on the second message. The method includes sending machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0010] Other aspects of the present disclosure relate to an apparatus for wireless communication by a first wireless local area network (WLAN) device. The apparatus includes means for sending a first message indicating support for machine learning by the first WLAN device. The apparatus also includes means for receiving a second message from a second WLAN device. The second message indicates support for at least one machine learning model type by the second WLAN device. The apparatus also includes means for activating a machine learning session with the second WLAN device based at least in part on the second message. The apparatus includes means for receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0011] In other aspects of the present disclosure, a non-transitory computer-readable medium is disclosed that stores program code for execution by a first wireless local area network (WLAN) device. The program code includes program code for sending a first message indicating support for machine learning by the first WLAN device. The program code also includes program code for receiving a second message from a second WLAN device. The second message indicates support for at least one machine learning model type by the second WLAN device. The program code also includes program code for activating a machine learning session with the second WLAN device based at least in part on the second message. The program code includes program code for receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0012] In general, various aspects include methods, apparatuses, systems, computer program products, non-transitory computer-readable media, access points (APs), stations (STAs), user equipment, base stations, wireless communication devices, and processing systems as substantially described with reference to the accompanying drawings and the specification and as illustrated by the accompanying drawings and the specification.

[0013] The foregoing has been fairly broadly outlined according to the features and technical advantages of the examples of the present disclosure so that the following detailed description may be better understood. Additional features and advantages will be described. The disclosed concepts and specific examples may be easily used as the basis for modifying or designing other structures for the same purpose of performing the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. When considered in conjunction with the accompanying drawings, the characteristics of the disclosed concepts (both their organization and method of operation) and the associated advantages will be better understood according to the description below. Each of the figures in the accompanying drawings is provided for the purpose of illustration and description and is not intended to be a definition of limitations to the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to fully understand the features of the present disclosure, a specific description may be obtained by referring to various aspects (some of which are shown in the accompanying drawings). However, it should be noted that the accompanying drawings only illustrate certain aspects of the present disclosure and are therefore not to be considered as limiting the scope of the present disclosure, as the description may allow for other equally effective aspects. The same reference numerals in different drawings may identify the same or similar elements.

[0015] Figure 1 is a schematic diagram illustrating an example wireless communication network.

[0016] Figure 2 is a block diagram illustrating an example protocol data unit (PDU) that may be used for communications between an access point (AP) and one or more stations (STAs).

[0017] Figure 3 It is shown Figure 2 Block diagram of example fields in a PDU.

[0018] Figure 4 is a block diagram illustrating an example implementation of designing a machine learning model using a system on a chip (SOC) including a general-purpose processor in accordance with certain aspects of the present disclosure.

[0019] Figure 5A , 5B 5 and 5C are schematic diagrams illustrating neural networks according to aspects of the present disclosure.

[0020] Figure 5D is a schematic diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0021] Figure 6 is a block diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0022] Figure 7 is a block diagram illustrating a high-level machine learning model in accordance with aspects of the present disclosure.

[0023] Figure 8 is a block diagram illustrating a reinforcement learning model in accordance with aspects of the present disclosure.

[0024] Fig. 9A is a table illustrating machine learning capabilities of wireless devices according to aspects of the present disclosure.

[0025] Fig. 9B is a call flow diagram illustrating a machine learning framework for a wireless local area network (WLAN) in accordance with aspects of the present disclosure.

[0026] Fig.10is a block diagram of an example wireless communication device in accordance with aspects of the present disclosure.

[0027] Fig.11A is a block diagram of an example access point (AP) in accordance with aspects of the present disclosure.

[0028] Fig. 11B is a block diagram of an example station (STA) in accordance with aspects of the present disclosure.

[0029] Fig.12 is a flow chart illustrating an example process, such as performed by a wireless device, in accordance with various aspects of the present disclosure.

[0030] Fig.13 is a block diagram of an example access point (AP) supporting a machine learning framework for a wireless local area network (WLAN) in accordance with various aspects of the present disclosure.

[0031] Fig.14 is a block diagram of an example station (STA) supporting a machine learning framework for a WLAN in accordance with various aspects of the present disclosure.

[0032] Fig.15 is a table illustrating WLAN stages for various machine learning operations in accordance with various aspects of the present disclosure.

[0033] Fig.16 is a state diagram illustrating states of a machine learning session in accordance with various aspects of the present disclosure.

[0034] Fig.17 is a block diagram illustrating components of information elements for machine learning information exchange in accordance with various aspects of the present disclosure.

[0035] Fig.18 is a block diagram illustrating components of an Aggregation Control (A-Control) field for machine learning information exchange in accordance with various aspects of the present disclosure.

[0036] Fig.19 is a call flow diagram illustrating a machine learning framework for WLANs in accordance with various aspects of the present disclosure.

[0037] Fig. 20 is a block diagram of an example wireless communication device supporting a machine learning framework for a WLAN in accordance with various aspects of the present disclosure.

[0038] Fig.21 is a flow chart illustrating an example process, such as performed by a wireless device, in accordance with various aspects of the present disclosure.

[0039] Fig. 22is a flow chart illustrating an example process, such as performed by a wireless device, in accordance with various aspects of the present disclosure. DETAILED DESCRIPTION

[0040] The following description refers to some specific examples for the purpose of describing the innovative aspects of the present disclosure. However, it will be readily appreciated by those skilled in the art that the teachings herein can be applied in a variety of different ways. Some or all of the described examples can be implemented in any device, system, or network capable of sending and receiving radio frequency (RF) signals in accordance with one or more of the following: the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard, the IEEE 802.15 standard, the Bluetooth Special Interest Group (SIG) standard, the Bluetooth SIG ... Standards, or Long Term Evolution (LTE), 3G, 4G or 5G (New Radio (NR)) standards promulgated by the Third Generation Partnership Project (3GPP), and other standards. The described implementations can be implemented in any device, system or network capable of sending and receiving RF signals according to one or more of the following technologies or techniques: code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), single carrier FDMA (SC-FDMA), single user (SU) multiple input multiple output (MIMO) and multi-user (MU)-MIMO. The described implementations can also be implemented using other wireless communication protocols or RF signals that are suitable for use in one or more networks in a wireless personal area network (WPAN), a wireless local area network (WLAN), a wireless wide area network (WWAN) or an Internet of Things (IOT) network.

[0041] As mentioned above, the 802.11 specification does not currently support the use of machine learning across devices. Proprietary models can be used by individual stations (STAs) or APs, but these models may only be used to optimize features that are left to be implemented. An example is rate adaptation. Wireless standards (e.g., 802.11 specifications) do not specify what algorithms clients can use. Clients can use heuristic algorithms or machine learning-based techniques. For applications such as enhanced distributed channel access (EDCA) optimization, the 802.11 specification only allows specified behaviors. Various aspects of the present disclosure create a flexible framework whereby machine learning models can be used for such use cases defined in the 802.11 specifications and determined by access points to be suitable. Therefore, access points can share the model across entities in the network.

[0042] Various aspects of the present disclosure relate to signaling details and operational aspects for implementing artificial intelligence and machine learning in a wireless local area network (WLAN). Signaling details may include container structures such as frames and elements. Operational aspects may include operational mode changes and indication of information during machine learning usage in a basic service set (BSS).

[0043] According to various aspects of the present disclosure, information is efficiently transmitted on an on-demand basis between devices in a WLAN. The information transmitted may depend on whether the device is in the discovery phase, the probe phase, the association establishment phase, or the post-association phase. During the discovery phase, a beacon message is sent. The beacon message may indicate basic information, such as whether the WLAN device supports machine learning. The information container during the discovery phase may indicate machine learning support, which enables the access point to discover machine learning support with low overhead. In some examples, additional machine learning information may be provided during the probe phase. For example, supported or expected use cases, and machine learning support levels may be indicated via messages sent in probe request and probe response frames during the probe phase. Containers exchanged during the probe phase may indicate support for specific use cases.

[0044] During the association establishment phase with the access point, complete machine learning information can be provided for each desired use case. In addition, machine learning functions can also be transmitted. Optionally, model structure, model parameters and / or model inputs and outputs can be transmitted. However, in some aspects, model structure, model parameters, and / or model inputs and outputs can be transmitted after association. The container during the association phase can indicate model structure information and model parameter information. When operating with an access point, complete information and model updates for each use case can be transmitted after association. In addition, changes to the machine learning operation mode can be transmitted. The container during the post-association phase can include model structure and parameters, as well as information required to enable and disable machine learning use cases and provide performance feedback on the model. Communication in each of the stages can be between two peer stations or between an access point and a station.

[0045] Machine learning operations in a BSS can occur during a machine learning session, which has three different states. When a client is associated with an access point and when the client is in an inactive state for machine learning, the associated machine learning inactive state can be the initial state of the client. Communication between two peer stations (such as an access point and a station, or two point-to-point (P2P) stations) can establish a machine learning session by activating a machine learning session. Once activated, the client enters a machine learning active state, during which the client can configure and use machine learning algorithms. In some examples, either peer can terminate or suspend the machine learning session, causing the device to enter an inactive state or a suspended state, respectively.

[0046] According to various aspects of the present disclosure, machine learning operations in a BSS during an active machine learning session specify that a control entity (such as an access point) has fine-grained control over how machine learning is used in the BSS. The access point can dynamically enable or disable machine learning inference in its BSS. In some aspects, the access point can dynamically enable or disable other aspects of machine learning use, such as training a model or uploading a trained model. Signaling can be introduced to indicate whether machine learning inference is allowed to be used within a specific time interval. Signaling can also be introduced to indicate what part of the data can be used to derive machine learning inference. Signaling can also indicate what part of the data can be used for machine learning training (e.g., model training). Signaling can also be introduced to indicate the time interval in which the access point coordinates with a non-access point station to verify the learned machine learning model. For signaling related to downloadable models, the access point can indicate that a new model is available for download. The access point can also provide different model parameters to different stations. When the access point belongs to an access point multi-link device and the station belongs to a non-access point multi-link device, the access point can also provide different model parameters to the same non-access point multi-link device on different links. In some aspects, an access point may request a station to indicate whether a particular downloaded machine learning model is being used by the station. The access point may solicit feedback on the performance of the downloaded model. Non-access point operations are also contemplated and described in detail below.

[0047] Aspects of the present disclosure relate to information containers for machine learning information exchange. An information element is an example of a container. An information element should be flexible enough to carry a variety of different types and degrees of information related to machine learning. One or more fields or subfields may always be included in an information element. In some examples, one or more optional fields or subfields and subelements may be included in an information element. An information element may have multiple variants, each of which may be used for a specific purpose.

[0048] Frames are another type of container for machine learning information exchange. In some examples, frames may be defined for exchanging machine learning models and their parameters, session (re)start, session suspension, session teardown, operation mode updates, and instructions related to machine learning operations. Frames enable semi-static configuration for machine learning usage in WLANs.

[0049] For faster information exchange, other types of containers can be used for machine learning information exchange. For example, an aggregation control (A-Control) or similar field can be used to quickly exchange information related to machine learning operations.

[0050] Certain aspects of the subject matter described in this disclosure may be implemented to achieve one or more of the following potential advantages. In some examples, the described techniques (such as activating a machine learning session) may be used by a controlling entity (e.g., an access point) to efficiently maintain and manage state information of a controlled entity (e.g., a station). For example, an access point may store information related to the mobility, location, battery, etc. of a station with which the access point has established a session, and determine which models to download to the station based on the information.

[0051] Receiving machine learning model structure information and machine learning model parameters during a machine learning session enables a controlling entity (e.g., an access point) to provide a controlled entity (e.g., a station) with a model structure and parameters customized for the state information of the station. For example, based on the battery power information of the mobile station, the access point can download a more energy-efficient model structure and parameters, such as a decision tree with a smaller maximum depth for a random forest model. The maximum depth of a decision tree can provide a measure of how many times a decision tree model can be segmented before it produces a prediction. In another example, if the controlled entity (e.g., another peer station) has high computing power, the controlling entity (e.g., a peer station) can download a model with floating point parameters. The exchange of machine learning model structure information and machine learning model parameters enables the controlled entity to rebuild and use a machine learning model trained by the controlling entity, and the controlling entity can have higher processing power and limited energy constraints, thereby enabling low-cost controlled entities to enjoy the benefits of complex machine learning techniques. Depending on the machine learning use case, this can help the controlled entity achieve higher throughput, lower latency, or improve overall network efficiency.

[0052] Figure 1A block diagram of an example wireless communication network 100 is shown. According to some aspects, the wireless communication network 100 may be an example of a wireless local area network (WLAN), such as a Wi-Fi network (and will be referred to as WLAN 100 hereinafter). For example, the WLAN 100 may be a network that implements at least one standard of the IEEE 802.11 family of wireless communication protocol standards, such as standards defined by the IEEE 802.11-2016 specification or its revisions, including but not limited to 802.11ay, 802.11ax, 802.11az, 802.11ba, and 802.11be. The WLAN 100 may include a plurality of wireless communication devices, such as an access point (AP) 102 and a plurality of stations (STAs) 104. Although only one AP 102 is shown, the WLAN 100 may also include a plurality of APs 102.

[0053] Each of the STAs 104 may also be referred to as a mobile station (MS), a mobile device, a mobile phone, a wireless phone, an access terminal (AT), a user equipment (UE), a subscriber station (SS) or a subscriber unit, and other examples. The STAs 104 may represent various devices, such as mobile phones, personal digital assistants (PDAs), other handheld devices, netbooks, notebook computers, tablet computers, laptop computers, display devices (e.g., TVs, computer monitors, navigation systems, and other examples), music or other audio or stereo equipment, remote control devices (“remote controls”), printers, kitchen or other home appliances, electronic key chains (e.g., for passive keyless entry and start (PKES) systems), and other examples.

[0054] A single AP 102 and the associated set of STAs 104 may be referred to as a basic service set (BSS), which is managed by the corresponding AP 102 . Figure 1Also shown is an example coverage area 108 of the AP 102, which may represent a basic service area (BSA) of the WLAN 100. Users may identify a BSS by a service set identifier (SSID), and other devices may identify a BSS by a basic service set identifier (BSSID), which may be a medium access control (MAC) address of the AP 102. The AP 102 periodically broadcasts a beacon frame ("beacon") including the BSSID to enable any STA 104 within the wireless range of the AP 102 to "associate" or re-associate with the AP 102 to establish a corresponding communication link 106 (also referred to as a "Wi-Fi link" hereinafter) with the AP 102, or to maintain a communication link 106 with the AP 102. For example, the beacon may include an identification of a primary channel used by the corresponding AP 102, and a timing synchronization function for establishing or maintaining timing synchronization with the AP 102. The AP 102 may provide access to external networks to various STAs 104 in the WLAN 100 via corresponding communication links 106.

[0055] To establish a communication link 106 with an AP 102, each of the STAs 104 is configured to perform a passive or active scanning operation ("scan") on frequency channels in one or more frequency bands (e.g., 2.4 GHz, 5 GHz, 6 GHz, or 60 GHz bands). To perform a passive scan, the STA 104 listens for a beacon that is transmitted by the corresponding AP 102 at a periodic time interval referred to as a target beacon transmission time (TBTT) (measured in a time unit (TU) where a TU may be equal to 1024 microseconds (μs)). To perform an active scan, the STA 104 generates a probe request and sequentially transmits the probe request on each channel to be scanned, and listens for a probe response from the AP 102. Each STA 104 may be configured to identify or select an AP 102 to associate with based on the scan information obtained through the passive or active scan, and perform authentication and association operations to establish a communication link 106 with the selected AP 102. The AP 102 assigns an association identifier (AID) to the STA 104 at the end of the association operation, and the AP 102 uses the AID to track the STA 104.

[0056] Due to the increasing popularity of wireless networks, STA 104 may have the opportunity to select one of many BSSs within the range of the STA or to select among multiple APs 102, which together form an extended service set (ESS) including multiple connected BSSs. The extended network station associated with WLAN 100 may be connected to a wired or wireless distribution system that allows multiple APs 102 to be connected in such an ESS. Therefore, STA 104 may be covered by more than one AP 102 and may be associated with different APs 102 at different times for different transmissions. In addition, after associating with AP 102, STA 104 may also be configured to periodically scan the surrounding environment to find a more suitable AP 102 to associate with. For example, a STA 104 that is moving relative to the associated AP 102 may perform a "roaming" scan to find another AP 102 with more ideal network characteristics (such as a larger received signal strength indicator (RSSI) or a reduced traffic load).

[0057] In some cases, STA 104 can form a network without AP 102 or other devices other than STA 104 itself. An example of such a network is a self-organizing network (or wireless self-organizing network). A self-organizing network can be referred to as a mesh network or a point-to-point (P2P) network instead. In some cases, a self-organizing network can be implemented in a larger wireless network (such as WLAN 100). In such an implementation, although STA 104 may be able to communicate with each other using a communication link 106 through AP 102, STA 104 can also communicate with each other directly via a direct wireless link 110. In addition, two STAs 104 can communicate via a direct wireless link 110, regardless of whether the two STAs 104 are associated with the same AP 102 and served by the same AP 102. In such a self-organizing system, one or more of STA 104 can assume the role played by AP 102 in the BSS. Such STA 104 can be referred to as a group owner (GO) and can coordinate transmissions within the self-organizing network. Examples of direct wireless link 110 include a Wi-Fi direct connection, a connection established by using a Wi-Fi Tunnel Direct Link Setup (TDLS) link, and other P2P group connections.

[0058] The AP 102 and the STA 104 may operate and communicate in accordance with the IEEE 802.11 family of wireless communication protocol standards (such as those defined by the IEEE 802.11-2016 specification or its revisions, including but not limited to 802.11ay, 802.11ax, 802.11az, 802.11ba, and 802.11be) (via corresponding communication links 106). These standards define WLAN radio and baseband protocols for the physical (PHY) layer and the medium access control (MAC) layer. The AP 102 and the STA 104 send and receive wireless communications (hereinafter also referred to as "Wi-Fi communications") to each other in the form of PHY protocol data units (PPDUs) (or physical layer convergence protocol (PLCP) PDUs). The AP 102 and STA 104 in the WLAN 100 may send PPDUs on an unlicensed spectrum, which may be a portion of the spectrum including bands traditionally used by Wi-Fi technology, such as the 2.4 GHz band, the 5 GHz band, the 60 GHz band, the 3.6 GHz band, and the 900 MHz band. Some implementations of the AP 102 and STA 104 described herein may also communicate in other bands, such as the 6 GHz band, which may support both licensed and unlicensed communications. The AP 102 and STA 104 may also be configured to communicate on other bands, such as shared licensed bands, where multiple operators may have licenses to operate on the same or overlapping band or bands.

[0059] Each of the frequency bands may include multiple sub-bands or frequency channels. For example, PPDUs compliant with IEEE 802.11n, 802.11ac, 802.11ax, and 802.11be standard revisions may be sent on 2.4 GHz, 5 GHz, or 6 GHz frequency bands, each of which is divided into multiple 20 MHz channels. Thus, these PPDUs are sent on physical channels with a minimum bandwidth of 20 MHz, but larger channels may be formed by channel bonding. For example, by bonding multiple 20 MHz channels together, a PPDU may be sent on a physical channel with a bandwidth of 40 MHz, 80 MHz, 160 MHz, or CCC20 MHz.

[0060] Each PPDU is a composite structure that includes a PHY preamble and a payload in the form of a PHY service data unit (PSDU). The receiving device can use the information provided in the preamble to decode the subsequent data in the PSDU. In the case where the PPDU is sent on a bonded channel, the preamble field can be copied and sent in each subchannel in multiple subchannels. The PHY preamble may include a traditional part (or "traditional preamble") and a non-traditional part (or "non-traditional preamble"). Traditional preambles can be used for packet detection, automatic gain control and channel estimation, and other purposes. Traditional preambles can also be used to maintain compatibility with traditional devices. The format of the non-traditional part of the preamble, the encoding of the non-traditional part of the preamble, and the information provided in the non-traditional part of the preamble are based on the specific IEEE 802.11 protocol to be used to send the payload.

[0061] Figure 2 An example protocol data unit (PDU) 200 that may be used for wireless communication between an AP 102 and one or more STAs 104 is shown. For example, the PDU 200 may be configured as a PPDU. As shown, the PDU 200 includes a PHY preamble 202 and a PHY payload 204. For example, the preamble 202 may include a legacy portion that itself includes: a legacy short training field (L-STF) 206, which may be composed of two binary phase shift keying (BPSK) symbols; a legacy long training field (L-LTF) 208, which may be composed of two BPSK symbols; and a legacy signal field (L-SIG) 210, which may be composed of two BPSK symbols. The legacy portion of the preamble 202 may be configured in accordance with the IEEE 802.11a wireless communication protocol standard. The preamble 202 may also include a non-legacy portion including, for example, one or more non-legacy fields 212 that conform to an IEEE wireless communication protocol, such as IEEE 802.11ac, 802.11ax, 802.11be, or a later wireless communication protocol.

[0062] The L-STF 206 generally enables the receiving device to perform coarse timing and frequency tracking and automatic gain control (AGC). The L-LTF 208 generally enables the receiving device to perform fine timing and frequency tracking, and also performs an initial estimation of the wireless channel. The L-SIG 210 generally enables the receiving device to: determine the duration of the PDU, and use the determined duration to avoid sending outside the PDU. For example, the L-STF 206, L-LTF 208 and L-SIG 210 can be modulated according to a binary phase shift keying (BPSK) modulation scheme. The payload 204 can be modulated according to a BPSK modulation scheme, an orthogonal BPSK (Q-BPSK) modulation scheme, an orthogonal amplitude modulation (QAM) modulation scheme, or other suitable modulation schemes. The payload 204 may include a PSDU, which includes a data field (DATA) 214, which in turn may carry higher layer data, for example, in the form of a medium access control (MAC) protocol data unit (MPDU) or an aggregated MPDU (A-MPDU).

[0063] Figure 3 Show Figure 2 2. An example L-SIG 300 in a PDU 200 of FIG. 20. The L-SIG 300 includes a data rate field 302, reserved bits 304, a length field 306, parity bits 308, and a tail field 310. The data rate field 302 indicates the data rate (note that the data rate indicated in the data rate field 302 may not be the actual data rate of the data carried in the payload 204). The length field 306 indicates the length of the packet in, for example, symbols or bytes. The parity bits 308 can be used to detect bit errors. The tail field 310 includes tail bits that can be used by a receiving device to terminate the operation of a decoder (e.g., a Viterbi decoder). The receiving device can use the data rate indicated in the data rate field 302 and the length indicated in the length field 306 to determine the duration of the packet in, for example, microseconds (μs) or other time units.

[0064] In some aspects, the access point 102 and the station 104 may include means for transmitting, means for receiving, means for communicating, and means for advertising. Such means may include reference Figure 1 , 2 , 4, 10, 11A and 11B discussed one or more components of the access point 102 and the station 104.

[0065] Figure 4An example implementation of a system on chip (SOC) 400 according to certain aspects of the present disclosure is shown, and the SOC 400 may include a central processing unit (CPU) 402 or a multi-core CPU configured to generate gradients for neural network training. The SOC 400 may be included in the access point 102 and the station 104. Variables (e.g., neural signals and synaptic weights), system parameters associated with a computing device (e.g., a neural network with weights), latency, frequency band information, and task information may be stored in a memory block associated with a neural processing unit (NPU) 408, in a memory block associated with the CPU 402, in a memory block associated with a graphics processing unit (GPU) 404, in a memory block associated with a digital signal processor (DSP) 406, in a memory block 418, or may be distributed across multiple blocks. Instructions executed at the CPU 402 may be loaded from a program memory associated with the CPU 402, or may be loaded from the memory block 418.

[0066] The SOC 400 may also include additional processing blocks customized for specific functions, such as a GPU 404, a DSP 406, a connection block 410 (which may include fifth generation (5G) connections, fourth generation long term evolution (4G LTE) connections, Wi-Fi connections, USB connections, Bluetooth connections, etc.), and a multimedia processor 412 (which may, for example, detect and recognize gestures). In one implementation, the NPU is implemented in the CPU, DSP, and / or GPU. The SOC 400 may also include a sensor processor 414, an image signal processor (ISP) 416, and / or a navigation module 420 (which may include a global positioning system).

[0067] The SOC 400 may be based on the ARM instruction set. In one aspect of the present disclosure, the instructions loaded into the general processor 402 may include code for sending a first message indicating support for machine learning by the first WLAN device. The instructions loaded into the general processor 402 may also include code for receiving a second message from a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The instructions loaded into the general processor 402 may also include code for activating a machine learning session with the second WLAN device based at least in part on the second message. The instructions loaded into the general processor 402 may also include code for receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0068] In one aspect of the present disclosure, the instructions loaded into the general processor 402 may include code for receiving a first message indicating support for machine learning by the first WLAN device. The instructions loaded into the general processor 402 may also include code for sending a second message to a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device. The instructions loaded into the general processor 402 may also include code for activating a machine learning session with the second WLAN device based at least in part on the second message. The instructions loaded into the general processor 402 may also include code for sending machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0069] Deep learning architectures can perform object recognition tasks by learning to represent inputs at successively higher levels of abstraction in each layer, thereby building up useful feature representations of the input data. In this way, deep learning solves a major bottleneck of traditional machine learning. Before the advent of deep learning, machine learning methods for object recognition problems may rely heavily on human engineered features, perhaps combined with shallow classifiers. For example, a shallow classifier can be a two-class linear classifier, where the weighted sum of the feature vector components can be compared to a threshold to predict which category the input belongs to. Human-designed features can be templates or core programs customized for a specific problem domain by engineers with domain expertise. In contrast, deep learning architectures can learn to represent features that are similar to those that human engineers might design, but through training. In addition, deep networks can learn to represent and recognize new types of features that humans may not have considered.

[0070] A deep learning architecture can learn a hierarchy of features. For example, if presented with visual data, the first layer can learn to recognize relatively simple features in the input stream, such as edges. In another example, if presented with auditory data, the first layer can learn to recognize spectral power in specific frequencies. A second layer that uses the output of the first layer as input can learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For example, higher layers can learn to represent complex shapes in visual data or words in auditory data. Higher layers can also learn to recognize common visual objects or spoken phrases.

[0071] Deep learning architectures can perform particularly well when applied to problems that have a natural hierarchical structure. For example, the classification of motor vehicles can benefit from first learning to recognize wheels, windshields, and other features. These features can be combined in different ways at higher levels to recognize cars, trucks, and airplanes.

[0072] Neural networks can be designed with a variety of connection patterns. In a feedforward network, information is transmitted from a lower layer to a higher layer, where each neuron in a given layer transmits to a neuron in a higher layer. As described above, hierarchical representations can be established in successive layers of a feedforward network. Neural networks can also have recursive or feedback (also known as top-down) connections. In a recursive connection, the output from a neuron in a given layer can be transmitted to another neuron in the same layer. Recursive architectures can help identify patterns that span more than one input data block in the input data blocks that are sequentially delivered to the neural network. The connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. When the recognition of high-level concepts can assist in discerning specific low-level features of the input, a network with many feedback connections can be helpful.

[0073] The connections between layers of a neural network can be fully connected or partially connected. Figure 5A An example of a fully connected neural network 502 is shown. In the fully connected neural network 502, a neuron in a first layer may transmit an output to each neuron in a second layer, such that each neuron in the second layer will receive an input from each neuron in the first layer. Figure 5B An example of a locally connected neural network 504 is shown. In the locally connected neural network 504, neurons in a first layer may be connected to a limited number of neurons in a second layer. More generally, the locally connected layers of the locally connected neural network 504 may be configured such that each neuron in a layer will have the same or similar connection pattern, but with connection strengths that may have different values ​​(e.g., 510, 512, 514, and 516). The connection patterns of the local connections may result in spatially different receptive fields in higher layers because higher layer neurons in a given area may receive inputs that are tuned through training to properties of a limited portion of the total input to the network.

[0074] An example of a locally connected neural network is a convolutional neural network. Figure 5C An example of a convolutional neural network 506 is shown. The convolutional neural network 506 can be configured such that the connection strengths associated with the inputs to each neuron in the second layer are shared (e.g., 508). Convolutional neural networks can be well suited for problems in which the spatial location of the inputs is meaningful.

[0075] One type of convolutional neural network is a deep convolutional network (DCN). Figure 5DA detailed example of a DCN 500 is shown, which is designed to recognize visual features based on an image 526 input from an image capture device 530 (such as a vehicle-mounted camera). The DCN 500 of the current example can be trained to recognize traffic signs and numbers provided on traffic signs. Of course, the DCN 500 can be trained for other tasks, such as recognizing lane markings or recognizing traffic lights.

[0076] The DCN 500 can be trained using supervised learning. During training, an image (such as an image 526 of a speed limit sign) can be presented to the DCN 500, and then a forward pass can be calculated to produce an output 522. The DCN 500 can include a feature extraction portion and a classification portion. Once the image 526 is received, a convolutional layer 532 can apply a convolutional kernel (not shown) to the image 526 to generate a first feature map set 518. As an example, the convolutional kernel used for the convolutional layer 532 can be a 5x5 kernel that generates a 28x28 feature map. In this example, because four different feature maps are generated in the first feature map set 518, four different convolutional kernels are applied to the image 526 at the convolutional layer 532. A convolutional kernel may also be referred to as a filter or a convolutional filter.

[0077] The first feature map set 518 may be downsampled by a maximum pooling layer (not shown) to generate a second feature map set 520. The maximum pooling layer reduces the size of the first feature map set 518. That is, the size of the second feature map set 520 (such as 14x14) is smaller than the size of the first feature map set 518 (such as 28x28). The reduced size provides similar information to subsequent layers while reducing memory consumption. The second feature map set 520 may be further convolved via one or more subsequent convolutional layers (not shown) to generate one or more subsequent feature map sets (not shown).

[0078] exist Figure 5D In the example of , the second feature map set 520 is convolved to generate a first feature vector 524. In addition, the first feature vector 524 is further convolved to generate a second feature vector 528. Each feature of the second feature vector 528 can include a number corresponding to a possible feature of the image 526, such as "sign", "60", and "100". A softmax function (not shown) can convert the numbers in the second feature vector 528 into probabilities. As such, the output 522 of the DCN 500 is the probability that the image 526 includes one or more features.

[0079] In this example, the probabilities for "symbol" and "60" in output 522 are higher than the probabilities for other items in output 522, such as "30", "40", "50", "70", "80", "90", and "100". Before training, output 522 produced by DCN 500 may be incorrect. Therefore, the error between output 522 and the target output can be calculated. The target output is the true value of image 526 (e.g., "symbol" and "60"). Then, the weights of DCN 500 can be adjusted so that the output 522 of DCN 500 is more closely aligned with the target output.

[0080] To adjust the weights, the learning algorithm may calculate a gradient vector for the weights. The gradient may indicate the amount by which the error will increase or decrease if the weights are adjusted. At the top layer, the gradient may correspond directly to the value of the weights connecting the activated neurons in the penultimate layer and the neurons in the output layer. In the lower layers, the gradient may depend on the value of the weights and the calculated error gradients of the higher layers. The weights may then be adjusted to reduce the error. This way of adjusting weights may be referred to as "backward propagation" because it involves "passing backwards" through the neural network.

[0081] In practice, the error gradient of the weights can be calculated over a small number of examples so that the calculated gradient is close to the true error gradient. This approximation method can be called stochastic gradient descent. Stochastic gradient descent can be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level. After learning, the DCN can be presented with new images (e.g., a speed limit sign of image 526), ​​and a forward pass through the network can produce output 522, which can be viewed as an inference or prediction of the DCN.

[0082] Deep belief network (DBN) is a probability model including multi-layer hidden nodes. DBN can be used to extract the hierarchical representation of training data set. DBN can be obtained by superimposing the layers of restricted Boltzmann machine (RBM). RBM is a type of artificial neural network that can learn probability distribution through input set. Since RBM can learn probability distribution without information about the category to which each input should be classified, RBM is often used in unsupervised learning. Using mixed unsupervised and supervised paradigms, the bottom RBM of DBN can be trained in an unsupervised manner and can act as a feature extractor, and the top RBM can be trained in a supervised manner (based on the joint distribution of the input and target category from the previous layer) and can act as a classifier.

[0083] A deep convolutional network (DCN) is a convolutional network configured with additional pooling and normalization layers. DCN has achieved state-of-the-art performance on many tasks. DCN can be trained using supervised learning, in which both the input and output targets are known for many examples and are used to modify the weights of the network using a gradient descent method.

[0084] The DCN can be a feed-forward network. In addition, as described above, the connections from neurons in the first layer of the DCN to a set of neurons in the next higher layer are shared across the neurons in the first layer. The feed-forward and shared connections of the DCN can be used for fast processing. For example, the computational burden of the DCN can be much smaller than that of a similarly sized neural network that includes recursive connections or feedback connections.

[0085] The processing of each layer of the convolutional network can be considered as a spatially invariant template or basic projection. If the input is first decomposed into multiple channels (such as the red channel, green channel, and blue channel of a color image), the convolutional network trained on the input can be considered as three-dimensional, where two spatial dimensions are along the axis of the image and the third dimension captures color information. The output of the convolutional connection can be considered to form a feature map in a subsequent layer, where each element of the feature map (e.g., 220) receives input from a series of neurons in the previous layer (e.g., feature map 218) and from each channel in the multiple channels. The values ​​in the feature map can be further processed using nonlinearities (such as correcting max(0,x)). The values ​​from adjacent neurons can be further pooled, which corresponds to downsampling and can provide additional local invariance and dimensionality reduction. Normalization (which corresponds to whitening) can also be applied through lateral inhibition between neurons in the feature map.

[0086] As more annotated data points become available or as computing power increases, the performance of deep learning architectures can also increase. Modern deep neural networks are routinely trained using thousands of times more computing resources than were available to a typical researcher just fifteen years ago. New architectures and training paradigms can further improve the performance of deep learning. Rectified linear units can reduce the training problem known as vanishing gradients. New training techniques can reduce overfitting and thereby enable larger models to achieve better generalization. Encapsulation techniques can abstract the data within a given receptive field and further improve overall performance.

[0087] Figure 6 6 is a block diagram showing a deep convolutional network 650. The deep convolutional network 650 may include multiple different types of layers based on connections and weight sharing. Figure 6As shown in FIG. 6 , the deep convolution network 650 includes convolution blocks 654A and 654B. Each of the convolution blocks 654A and 654B may be configured with a convolution layer (CONV) 656 , a normalization layer (LNorm) 658 , and a maximum pooling layer (MAX POOL) 660 .

[0088] The convolution layer 656 may include one or more convolution filters that may be applied to the input data to generate a feature map. Although only two convolution blocks 654A, 654B in the convolution block are shown, the present disclosure is not limited thereto, but on the contrary, any number of convolution blocks 654A, 654B may be included in the deep convolution network 650, depending on design preferences. The normalization layer 658 may normalize the output of the convolution filter. For example, the normalization layer 658 may provide whitening or side suppression. The maximum pooling layer 660 may provide spatial downsampling aggregation for local invariance and dimensionality reduction.

[0089] For example, the parallel filter bank of the deep convolutional network can be loaded on the CPU 402 or GPU 404 of the SOC 400 to achieve high performance and low power consumption. In an alternative embodiment, the parallel filter bank can be loaded on the DSP 406 or ISP 416 of the SOC 400. In addition, the deep convolutional network 650 can access other processing blocks that may be present on the SOC 400, such as the sensor processor 414 and the navigation module 420 dedicated to sensors and navigation, respectively.

[0090] The deep convolutional network 650 may also include one or more fully connected layers 662 (FC1 and FC2). The deep convolutional network 650 may also include a logistic regression (LR) layer 664. Between each layer 656, 658, 660, 662, 664 of the deep convolutional network 650 are weights (not shown) to be updated. The output of each of these layers (e.g., 656, 658, 660, 662, 664) may serve as an input to a subsequent layer in these layers (e.g., 656, 658, 660, 662, 664) of the deep convolutional network 650 to learn hierarchical feature representations from the input data 652 (e.g., images, audio, video, sensor data, and / or other input data) initially provided at the convolutional block 654A. The output of the deep convolutional network 650 is a classification score 666 for the input data 652. The classification score 666 may be a set of probabilities, where each probability is a probability that the input data includes a feature from a feature set.

[0091] Figure 7 is a block diagram illustrating a high-level machine learning model in accordance with aspects of the present disclosure. Figure 7A machine learning (ML) model is specified by the following terms: X-input; Y-output; and f-basis function, such that Y=f(X). For example, the input can be measurement results and preprocessing steps, and the output can be the characteristics of the Wi-Fi feature to be optimized. The machine learning model learns the mapping f:X→Y from the data set D={X,Y}. In other words, the machine learning model operates as a function approximator. In one example, when the input X is an observation vector including a received signal strength indicator (RSSI), a packet data radio (PDR), and the number of overlapping basic service sets (OBSS), the output Y can be the optimal uplink modulation and coding scheme (MCS) index.

[0092] Machine learning paradigms include supervised learning, unsupervised learning, and reinforcement learning. For supervised learning, a labeled dataset such as labeled images is given. In case of supervised learning, data collection is an offline process. The goal of supervised learning is to learn the mapping between the labeled dataset and the labels. However, collecting labeled data is a challenge for supervised learning. Examples of supervised learning models include neural networks, decision trees, support vector machines (SVM), etc. For unsupervised learning, the data is not labeled. Therefore, unsupervised learning attempts to find patterns and insights. An example of an unsupervised learning technique is clustering.

[0093] Figure 8 is a block diagram illustrating an example of a reinforcement learning model according to aspects of the present disclosure. Reinforcement learning is another machine learning paradigm. In the case of reinforcement learning, a system is given, and the system may include an agent 802 interacting with an environment 804. The system also includes a state S t 、Action A t and reward R t , where t represents the time step. The goal of reinforcement learning is to learn a strategy, such as t To Action A t The agent 802 can learn through experience by taking actions and observing rewards. Example reinforcement learning techniques include deep Q-networks, policy gradients, etc.

[0094] There are multiple levels of machine learning use and collaboration. In other words, there are hierarchies for where the machine learning model and the data for the model are stored. For example, both the model and the data can be stored locally. In another example, the model is local, but the locally stored data and the data requested from other devices in the BSS can be used for training, so that the standard can be applied to data exchange. For example, an access point (AP) can request measurement results from an associated station (STA). In another level of the hierarchy, the model is local and the model can be transmitted between the access point and the station. For example, the station can download machine learning model structure information and machine learning model parameters from the access point. Model refinement can be allowed, but the refined model remains local in this level of the hierarchy. For example, in this level of the hierarchy, the data is local and requested, and the standard can be applied to data exchange, model definition, and model transmission. In another level of the hierarchy, the model is shared, so that the station and the access point exchange model structure and parameters. In this level of the hierarchy, the model is refined and shared, and the data is both local and requested. At this hierarchical level, standards can apply to data exchange, model definition, and model transfer.

[0095] Although aspects of the present disclosure are described with respect to supervised learning models, the present disclosure is not limited thereto. For example, reinforcement learning schemes (e.g., deep Q-learning) and unsupervised models are also contemplated.

[0096] Fig. 9A is a table illustrating examples of machine learning capabilities of wireless devices according to aspects of the present disclosure. In these aspects of the present disclosure, a wireless device (such as an access point or station) can transmit machine learning capabilities. The combination of use cases, machine learning models, outputs, and inputs can form a machine learning function. Each machine learning function provided by an access point is uniquely identified by a function ID value assigned to the function by the access point.

[0097] As an example of conveying machine learning capabilities, the access point may advertise use cases for which the access point supports machine learning. Exemplary use cases include enhanced distributed channel access (EDCA) optimization, interference estimation, rate adaptation, channel state information (CSI) enhancement, and traffic classification. Fig. 9A In the example shown in , the access point supports Enhanced Distributed Channel Access (EDCA) optimization and traffic classification (as seen in the second column of the table). Each use case can be indicated using a binary representation. Fig. 9AIn the example of , EDCA optimization is represented by 0001 and traffic classification is represented by 1001, but these values ​​are exemplary only. The fields and values ​​of each use case may be standardized. In some aspects, the access point shares the use case with the station during discovery. The use case may also be referred to as a descriptor or feature ID. Announcement may refer to broadcasting or otherwise transmitting a message.

[0098] The access point can advertise one or more machine learning models for each use case. The advertised machine learning model is a type of learning technique and structure. For example, Fig. 9A The table of shows a random forest technique for the first business classification use case with 20 decision trees, each with a maximum depth of seven. Another example is a deep neural network (DNN) with three hidden layers, each with 50 neurons. Fig. 9A In the example of , the access point advertises the DNN model for the second service classification use case. Fig. 9A In the EDCA optimization shown in , the access point advertises a decision tree with a maximum depth of seven. Similar to the use cases, the standard can define fields and values ​​for each type of machine learning model. Fig. 9A In the example of , the decision tree is represented by 01001, the random forest is represented by 01000, and the DNN is represented by 10000, but these values ​​are just example values. In some aspects, the structure of the machine learning model can have a standardized encoding.

[0099] Access points can advertise the output of machine learning models. For example, Fig. 9A For the EDCA optimization use case shown in , the model can output the maximum contention window CW_max and the minimum contention window CW_min for video and voice access categories (AC_VI and AC_VO). For the random forest service classification model, Fig. 9A In the example of , the access point advertises six dimensions and probability values ​​for six differentiated service code points (DSCPs). For DNN, the access point advertises outputs for ten dimensions and probability values ​​for ten DSCPs. For the output, the fields and values ​​can be standardized.

[0100] The access point can advertise the input features of the machine learning model. For example, for Fig. 9A In the EDCA optimization use case shown in , the access point advertises fifteen inputs along with the encoding of the characteristics. Example inputs may include measured loss rate and measured throughput. In the traffic classification example, the access point advertises five and eight input characteristics, respectively, along with the encoding of the characteristics. Example inputs may include maximum packet size and minimum inter-arrival time. For the input characteristics, the fields and values ​​may be standardized.

[0101] Fig. 9A The first column on shows the function ID. According to aspects of the present disclosure, the function ID is a representation of a four-element tuple: <descriptor, ML model, input, output>. The function ID is a unique identifier for the machine learning function supported by the access point. This field can be standardized, where the value is assigned by the access point. The access point can support a limited number of functions, each with a unique function ID. Fig. 9A In the example of , the EDCA use case has a functional ID of zero, while the traffic classification use case has values ​​of two and three, respectively.

[0102] As mentioned above, the wireless specification (e.g., 802.11) may standardize some of the machine learning information. For example, each use case may have a unique descriptor that may be standardized in the specification (e.g., traffic prediction = 0001, rate adaptation = 0010, etc.). During discovery, access points may share descriptors with non-access point stations.

[0103] In some aspects, access points may share partial information of machine learning models during discovery. Full information may be shared during or after establishment. A full description of a machine learning model includes: (i) model name (e.g., type), (ii) model structure, and (iii) model parameters. Options for normalizing these components are now discussed.

[0104] In some aspects, the specification may define a set of machine learning models for each use case. For example, for a particular use case, the specification may define a random forest model with a fixed structure. In these aspects, the model parameters may be exchanged over an air interface (such as an 802.11 air interface).

[0105] In other aspects, the specification may define encodings for separate components of a machine learning algorithm. In other words, different components of a machine learning model may be standardized. For example, a standard may define encodings for the name / type of a machine learning model, the structure of a machine learning model, and the parameters of a machine learning model. A model type may be a candidate set of machine learning models (e.g., random forest (RF), deep neural network (DNN), convolutional neural network (CNN), recurrent neural network (RNN)). Each type may be assigned an identifier (e.g., RF=0001, DNN=0010, etc.). The specification may also define a model structure, including a machine learning algorithm and model parameters. Each of these components may be exchanged over an air interface (e.g., an 802.11 air interface). Parameters may be carried in elements and frames. An example of a parameter is a weight for a trained neural network.

[0106] In other aspects of the present disclosure, external standardized interfaces define machine learning algorithm components, including machine learning models, structures, and parameters. For example, components can be defined using Predictive Model Markup Language (PMML) or Open Neural Network Exchange (ONNX). In these aspects, the exchange of model types, structures, and parameters can occur over an air interface (e.g., 802.11), or an access point can provide a uniform resource locator (URL) of an external server to enable a station to download information.

[0107] According to various aspects of the present disclosure, the specification can standardize the input. In some aspects, a standard set of input features can be defined, such as maximum loss rate, average RSSI, etc. As described above, these aspects may be appropriate when the machine learning model (including the structure) is also standardized.

[0108] In other aspects, the specification may define a set of measurements (e.g., RSSI, throughput, loss rate) and a set of operations (e.g., maximum, average, last known value, etc.). In these aspects, input features to the machine learning model may be represented as one or more operations on the measurements (e.g., average RSSI, maximum loss rate, last known MCS). As described above, these aspects may be appropriate when encoding of machine learning models and structures is used (in 802.11 or external standards).

[0109] In other aspects, the specification may define a combination of a standard set of input features and a standard set of measurements. In these aspects, the specification may define a standard set of inputs, and the access point may use, for example, as described above when the standard defines a set of measurements and operations to select additional inputs. Information related to the inputs to the machine learning model may be shared after it is established.

[0110] The specification may define a set of candidate outputs for a use case. In some aspects, an access point may only support a subset of these outputs. For example, for EDCA optimization, the standard may define CW_min, CW_max, arbitration interframe spacing (AIFS), transmission opportunity limit (TXOPLimit) for access categories video, voice, background traffic, and best effort traffic (AC_VI, AC_VO, AC_BK, AC_BE) as sixteen potential outputs. Fig. 9A In the example of , an access point supporting EDCA optimization only provides CW_min and CW_max for AC_VI and AC_VO as outputs. In other aspects, the specification may define a set of output variables and a set of operations on the output variables (similar to as described above with respect to inputs). Information about the output of the machine learning model can be shared after it is established.

[0111] Fig. 9B is a call flow diagram illustrating a machine learning framework for a wireless local area network (WLAN) in accordance with aspects of the present disclosure. The capability exchange between an access point and a non-access point station will now be described in more detail. In accordance with aspects of the present disclosure, an access point may advertise use cases for machine learning support (e.g., EDCA optimization, interference estimation, traffic classification). For example, Fig. 9B As seen in FIG. 1 , at 910, access point 102 sends a message indicating its machine learning capabilities. Each access point may advertise multiple machine learning models for the same use case (e.g., Fig. 9A For each machine learning function (which can be identified by a function ID), the access point advertises the level of machine learning support. For example, the access point can advertise support for proprietary models by non-access point STAs. That is, non-access point stations can use proprietary models to optimize WLAN features that are otherwise specified in the standard. The access point can also provide downloadable trained models. Return to see Fig. 9A , these support levels can include downloaded models that non-access points cannot retrain, as seen with EDCA optimization. This support level can be selected when fairness with respect to legacy stations that do not use machine learning is a consideration. Other levels of support can be seen with respect to traffic classification. In the first level, non-access points can retrain downloaded models, but uploads of updated models and aggregations are not supported (e.g., non-federated learning). In the second level, non-access points can retrain downloaded models, and updated models can be uploaded to access points (e.g., federated learning). This level can be seen in the last row of the traffic classification use case (leveraging DNN).

[0112] For a multi-link device (MLD) access point, the supported use cases, machine learning models or parameters may be different on different links. Access point advertisements may be sent by an access point before, during or after association.

[0113] Non-access point stations can also send announcements. For example, Fig. 9B As seen in FIG. 1 , at 920, the non-access point station 104 may send a message indicating the machine learning capabilities of the station 104. The non-access point station may indicate a use case of interest, such as using a traffic classification model. The non-access point station may advertise support for specific machine learning models, measurements, and operations. In one example, the non-access point station may support a random forest model, but may not support a DNN. As another example, the non-access point station may support basic operations, but may not support a fast Fourier transform (FFT).

[0114] Examples of operations include basic operations, statistical operations, and signal processing operations. Basic operations may include, for example, sampling, such as sampling with a specified observation window and sampling interval; logarithm, such as logarithm with a specified base; last known value; and / or count attribute, such as count attribute with a specified observation window. Statistical operations may include, for example, averaging, such as averaging with a specified observation window; and / or quantile operations, such as quantile operations with a specified observation window and quantiles. Examples of signal processing operations include Fast Fourier Transform (FFT), such as an FFT with a specified size.

[0115] The non-access point station may indicate computing capabilities (e.g., hardware acceleration and million instructions per second (MIPS) limit) and location relative to the access point. The non-access point station may also indicate mobility information such as mobility level (e.g., walking or stationary) and the environment of the station (e.g., residential, outdoor, or stadium).

[0116] The non-access point station may select from the same four support levels as the access point. For example, for one of these levels, the non-access point station may indicate that the non-access point supports using the downloaded model but does not support retraining the downloaded model. For MLD non-access points, the supported use cases and machine learning models may be different on different links. These announcements by the non-access point may be sent during the association phase or in the post-association phase.

[0117] See again Fig. 9B At 930, the access point may transmit information associated with a machine learning model for use between the access point 102 and the station 104 based on the exchanged capability messages. In some aspects, the access point may share partial information of the machine learning model during discovery. The complete information may be shared during or after establishment. The complete description of the machine learning model includes: (i) model name (e.g., type), (ii) model structure, and (iii) model parameters. In various aspects of the present disclosure, the exchange of model types, structures, and parameters may occur via an air interface (e.g., 802.11), or the access point may provide a uniform resource locator (URL) of an external server to enable the station to download the information.

[0118] At 940, the access point 102 and the station 104 may communicate based on the machine learning model. For example, the machine learning model may be used for enhanced distributed channel access (EDCA) optimization, interference estimation, rate adaptation, channel state information (CSI) enhancement, and traffic classification. The use of the machine learning model may be enabled by the BSS of the access point, including by the station 104 that may not otherwise have the ability to train complex models. Machine learning techniques may be used to (jointly) optimize one or more 802.11 features that may be difficult to optimize using traditional (e.g., non-machine learning) techniques. For example, rate adaptation is a complex problem that involves the interaction of multiple 802.11 parameters (bandwidth, number of spatial streams, MCS, etc.), and machine learning may be used to jointly optimize these parameters.

[0119] According to aspects of the present disclosure, inputs to a machine learning model may be described as building blocks, which include a set of measurements or observations and a set of operations. Examples of measurements include radio statistics such as received signal strength indication (RSSI), number of spatial streams, modulation and coding scheme (MCS), number of channels, number of retransmissions, handover / association failures (with cause codes if available), number of successful triggered transmissions (e.g., based on number of triggered PPDUs) versus total packets sent (number of single user (SU) PPDUs), number of request to send / clear to send (RTS / CTS) exchanges, RSSI of neighboring access points, multi-user (MU) order, number of MU multiple-input multiple-output (MIMO) PPDUs and packet error rate (PER), number of MU orthogonal frequency division multiplexing access (OFDMA) PPDUs and PER, and power control parameters (e.g., allowed transmit power).

[0120] Other measurements are related to medium statistics such as congestion statistics (e.g., busy status, number of collisions), number of neighboring access points, (SU / MU) EDCA value, whether spatial reuse (SR) is enabled, and overlapping basic service set / preamble detection (OBSS-PD) threshold.

[0121] Other measurements are also related to application and usage statistics such as delay statistics, packet size distribution, user activity per hour (for power saving), user speed (if available), residential or enterprise environment (if known), and requested Quality of Service (QoS) values / Admission Control (AC) / latency.

[0122] Some measurements may be related to the device version, capabilities, and status. Examples include which revision (e.g., specification name), Wi-Fi capabilities, remaining battery status, and authentication type.

[0123] Using these building blocks, any arbitrary input feature can be described. For any given measurement result, zero or more operations can be performed. In a first example, where the feature used for the machine learning model is the downlink aggregate packet size, the measurement result can have the downlink packet size, and the operation can be a summation operation using a specified observation window. In a second example, where the feature used for the machine learning model is the maximum packet size, the measurement result can have the downlink packet size, and the operation can be a quantile operation using a specified observation window and quantile (e.g., 100). In a third example, where the feature used for the machine learning model is a 512-point FFT, the measurement result can have a downlink packet size. In this third example, the three operations can be: sampling using a specified observation window and sampling interval, summing using a specified observation window, and performing an FFT of a specified size.

[0124] By enabling machine learning, communication between access points (APs) and stations (STAs) can be improved. In some examples, the described techniques (such as communication between access points and non-access point stations based on machine learning models) can be used to (jointly) optimize one or more 802.11 features. These features may be difficult to optimize using traditional (e.g., non-machine learning) techniques. In addition, the downloadable model can help establish fairness between users of the machine learning model.

[0125] Fig.10 1 is a block diagram of an example wireless communication device 1000 according to aspects of the present disclosure. In some implementations, the wireless communication device 1000 may be a wireless communication device for use in a STA (such as the one described above with reference to FIG. Figure 1 In some implementations, the wireless communication device 1000 may be an example of a device used in an AP (such as one of the STAs 104 described above). Figure 1 The wireless communication device 1000 is an example of a device used in the AP 102 described above. The wireless communication device 1000 is capable of sending and receiving wireless communications in the form of, for example, wireless packets. For example, the wireless communication device 1000 can be configured to send and receive packets in the form of physical layer convergence protocol (PLCP) protocol data units (PPDUs) and medium access control (MAC) protocol data units (MPDUs) that comply with IEEE 802.11 wireless communication protocol standards (such as standards defined by the IEEE 802.11-2016 specification or its revisions, including but not limited to 802.11ay, 802.11ax, 802.11az, 802.11ba, and 802.11be).

[0126] The wireless communication device 1000 may be or may include a chip, a system on chip (SoC), a chipset, a package, or a device that includes one or more modems 1002, such as a Wi-Fi (IEEE 802.11 compliant) modem. In some implementations, the one or more modems 1002 (collectively referred to as "modems 1002") further include a wireless wide area network (WWAN) modem (e.g., a 3GPP 4G LTE or 5G compliant modem). In some implementations, the wireless communication device 1000 further includes one or more processors, processing blocks, or processing elements 1004 (collectively referred to as "processors 1004") coupled to the modem 1002. In some implementations, the wireless communication device 1000 further includes one or more radio units 1006 (collectively referred to as "radio units 1006") coupled to the modem 1002. In some implementations, the wireless communication device 1000 further includes one or more memory blocks or elements 1008 (collectively referred to as "memory 1008") coupled to the processor 1004 or the modem 1002.

[0127] The modem 1002 may include an intelligent hardware block or device, such as, for example, an application specific integrated circuit (ASIC), etc. The modem 1002 is typically configured to implement a PHY layer, and in some implementations, also implements a portion of a MAC layer (e.g., a hardware portion of a MAC layer). For example, the modem 1002 is configured to modulate packets and output the modulated packets to the radio unit 1006 for transmission on a wireless medium. Similarly, the modem 1002 is configured to obtain modulated packets received by the radio unit 1006 and demodulate the packets to provide demodulated packets. In addition to the modulator and demodulator, the modem 1002 may also include a digital signal processing (DSP) circuit, an automatic gain control (AGC) circuit, an encoder, a decoder, a multiplexer, and a demultiplexer. For example, when in a transmit mode, data obtained from the processor 1004 may be provided to an encoder, which encodes the data to provide encoded bits. The coded bits may then be mapped to a number of spatial stream network slice subnets (NSS) for spatial multiplexing or a number of space-time streams for space-time block coding (STBC). The coded bits in these streams may then be mapped to points in a modulation constellation (using the selected MCS) to provide modulated symbols. The modulated symbols in the corresponding spatial streams or space-time streams may be multiplexed, transformed via an inverse fast Fourier transform (IFFT) block, and then provided to a DSP circuit (e.g., for Tx windowing and filtering). The digital signal may then be provided to a digital-to-analog converter (DAC). The resulting analog signal may then be provided to an upconverter and ultimately to a radio unit 1006. In an implementation involving beamforming, the modulated symbols in the corresponding spatial streams are precoded via a steering matrix before they are provided to the IFFT block.

[0128] When in receive mode, the DSP circuit is configured to acquire a signal including modulated symbols received from the radio unit 1006, for example, by detecting the presence of the signal and estimating the initial timing and frequency offset. The DSP circuit is also configured to digitally condition the signal, for example, using channel (narrowband) filtering and analog impairment adjustment (such as correcting in-phase / quadrature (I / Q) imbalance) and by applying digital gain to ultimately obtain a narrowband signal. The output of the DSP circuit can then be fed to the AGC, which is configured to use information extracted from the digital signal (for example, in one or more received training fields) to determine the appropriate gain. The output of the DSP circuit is also coupled to a demultiplexer, which demultiplexes the modulated symbols when multiple spatial streams or space-time streams are received. The demultiplexed symbols can be provided to a demodulator, which is configured to extract the symbols from the signal and, for example, calculate a log-likelihood ratio (LLR) for each bit position in each spatial stream for each subcarrier. The demodulator is coupled to a decoder, which can be configured to process the LLRs to provide decoded bits. The decoded bits may then be descrambled and provided to the MAC layer (eg, processor 1004) for processing, evaluation, or interpretation.

[0129] The radio unit 1006 typically includes at least one radio frequency (RF) transmitter (or "transmitter chain") and at least one RF receiver (or "receiver chain"), which can be combined into one or more transceivers. For example, each of the RF transmitter and receiver can include various analog circuits, including at least one power amplifier (PA) and at least one low noise amplifier (LNA), respectively. The RF transmitter and receiver can then be coupled to one or more antennas. For example, in some implementations, the wireless communication device 1000 can include or be coupled to the following: multiple transmit antennas (each transmit antenna has a corresponding transmit chain) and multiple receive antennas (each receive antenna has a corresponding receive chain). The symbols output from the modem 1002 are provided to the radio unit 1006, and the radio unit 1006 then transmits the symbols via the coupled antennas. Similarly, the symbols received via the antennas are obtained by the radio unit 1006, and the radio unit 1006 then provides the symbols to the modem 1002.

[0130] The processor 1004 may include an intelligent hardware block or device designed to perform the functions described herein, such as, for example, a processing core, a processing block, a central processing unit (CPU), a microprocessor, a microcontroller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD) (such as a field programmable gate array (FPGA)), discrete gate or transistor logic, discrete hardware components, or any combination thereof. The processor 1004 processes information received through the radio unit 1006 and the modem 1002, and processes information to be output by the modem 1002 and the radio unit 1006 for transmission over a wireless medium. For example, the processor 1004 may implement at least a portion of the control plane and the MAC layer, which is configured to perform various operations related to the generation, transmission, reception, and processing of MPDUs, frames, or packets. In some implementations, the MAC layer is configured to generate an MPDU to provide to the PHY layer for encoding, and receive decoded information bits from the PHY layer for processing as an MPDU. The MAC layer may also be configured to allocate time and frequency resources, for example, for OFDMA and other operations or technologies. In some implementations, the processor 1004 may generally control the modem 1002 to cause the modem 1002 to perform the various operations described above.

[0131] The memory 1008 may include a tangible storage medium, such as a random access memory (RAM) or a read-only memory (ROM), or a combination thereof. The memory 1008 may also store non-transitory processor or computer executable software (SW) code, which contains instructions that, when executed by the processor 1004, cause the processor to perform various operations described herein for wireless communication, including generation, transmission, reception, and interpretation of MPDUs, frames, or packets. For example, various functions of the components disclosed herein, or various blocks or steps of the methods, operations, processes, or algorithms disclosed herein may be implemented as one or more modules of one or more computer programs.

[0132] Fig.11A is a block diagram of an example AP 1102 according to aspects of the present disclosure. For example, the AP 1102 may be a reference Figure 1 102. The AP 1102 includes a wireless communication device (WCD) 1110 (although the AP 1102 itself may also be generally referred to as a wireless communication device, as used herein). For example, the wireless communication device 1110 may be a wireless communication device (WCD) 1110. Fig.101000 is an example implementation of the wireless communication device 1000 described herein. The AP 1102 also includes a plurality of antennas 1120 coupled to the wireless communication device 1110 for sending and receiving wireless communications. In some implementations, the AP 1102 further includes an application processor 1130 coupled to the wireless communication device 1110, and a memory 1140 coupled to the application processor 1130. The AP 1102 also includes at least one external network interface 1150 that enables the AP 1102 to communicate with a core network or a backhaul network to obtain access to an external network including the Internet. For example, the external network interface 1150 may include one or both of a wired (e.g., Ethernet) network interface and a wireless network interface (such as a WWAN interface). One or more of the above components may communicate directly or indirectly with other components via at least one bus. The AP 1102 also includes a housing that surrounds the wireless communication device 1110, the application processor 1130, the memory 1140, and at least a portion of the antenna 1120 and the external network interface 1150.

[0133] Fig. 11B 1 is a block diagram of an example STA 1104 according to aspects of the present disclosure. For example, STA 1104 may be a reference Figure 1 104. STA 1104 includes a wireless communication device 1115 (although STA 1104 itself may also be generally referred to as a wireless communication device, as used herein). For example, wireless communication device 1115 may be a wireless communication device 1115. Fig.10 1000 is an example implementation of a wireless communication device 1000 described herein. STA 1104 also includes one or more antennas 1125 coupled to the wireless communication device 1115 for sending and receiving wireless communications. STA 1104 further includes an application processor 1135 coupled to the wireless communication device 1115, and a memory 1145 coupled to the application processor 1135. In some implementations, STA 1104 also includes a user interface (UI) 1155 (such as a touch screen or keyboard) and a display 1165, which can be integrated with UI 1155 to form a touch screen display. In some implementations, STA 1104 can also include one or more sensors 1175, such as, for example, one or more inertial sensors, accelerometers, temperature sensors, pressure sensors, or altitude sensors. One or more of the above components can communicate directly or indirectly with other components via at least one bus. STA 1104 also includes a housing that encloses the wireless communication device 1115 , the application processor 1135 , the memory 1145 , and at least portions of the following: antenna 1125 , UI 1155 , and display 1165 .

[0134] The access point 102 and the station 104 may each include a machine learning (ML) module in the Figure 4 , 10 , CPU 402, GPU 404, DSP 406, NPU 408, memory 418, processor 1004, memory 1008, memory 1140, application processor 1130, memory 1145 and / or application processor 1135 of 11A and 11B. The machine learning module can send a first message indicating support for machine learning by a first wireless local area network (WLAN) device. The machine learning module can also receive second information from a second WLAN device. The second message indicates support for at least one machine learning model type by the second WLAN device. The machine learning module also activates a machine learning session with the second WLAN device based at least in part on the second message. The machine learning module can also receive machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0135] Fig.12 is a flow chart illustrating an example process 1200, performed, for example, by a wireless device, in accordance with various aspects of the present disclosure. The example process 1200 is an example of a machine learning framework for a wireless local area network (WLAN).

[0136] like Fig.12 As shown in , in some aspects, process 1200 may include sending a first message indicating a first machine learning capability of a first wireless device (block 1202). For example, the first machine learning capability may indicate a first use case and at least one corresponding machine learning model. The first machine learning capability may also indicate support for proprietary models. In some aspects, the first wireless device may advertise a machine learning support level indicating at least one of the following: the machine learning model cannot be retrained by the second wireless device, the machine learning model can be retrained by the second wireless device but upload of updated model parameters is not supported, or the machine learning model can be retrained by the second wireless device and upload of updated model parameters is supported.

[0137] In some aspects, process 1200 may include receiving a second message from the second wireless device indicating second machine learning capabilities of the second wireless device (block 1204). For example, the second machine learning capabilities may indicate a second use case and at least one supported second machine learning model. The second machine learning capabilities may also indicate a location of the second wireless device relative to the first wireless device, a mobility level of the second device, and a computing capability of the second wireless device.

[0138] In some aspects, process 1200 may include transmitting information associated with a machine learning model for use between the first wireless device and the second wireless device based on the second machine learning capability and the first machine learning capability (block 1206). For example, the first wireless device may transmit the information by sending a function identifier (ID) identifying the machine learning function. The function ID may indicate the selected use case, machine learning model, set of inputs, and outputs. In some aspects, transmitting the information may include sending a link to an external server for downloading the machine learning model. In other aspects, the transmission of the information may include sending initial information for supported machine learning functions during discovery and sending additional information for supported machine learning functions during or after establishment.

[0139] In some aspects, process 1200 may include communicating with a second wireless device based at least in part on a machine learning model (block 1208). For example, the machine learning model may be a model or a standardized set of components in a set of standardized models for the selected use case. In some aspects, parameters and structure of the machine learning model are defined by an external interface. The set of inputs to the machine learning model may be a set of standardized input features and / or input features defined by the first wireless device and including a set of standardized measurements and a set of standardized operations applied to the measurements. The set of outputs from the machine learning model may be a subset of the standardized candidate outputs and / or the set of standardized outputs.

[0140] Fig.13 is a block diagram of an example wireless communication device 1300 supporting a machine learning framework for WLAN according to various aspects of the present disclosure. In some implementations, the wireless communication device 1300 is configured to perform the above-referenced Fig.12 The wireless communication device 1300 may be a wireless communication device as described above with reference to Fig.10 1000 . For example, the wireless communication device 1300 may be a chip, an SoC, a chipset, a package, or a device that includes at least one modem (e.g., a Wi-Fi (IEEE 802.11) modem or a cellular modem, such as modem 1002), at least one processor (such as processor 1004), at least one radio unit (such as radio unit 1006), and at least one memory (such as memory 1008). In some implementations, the wireless communication device 1300 may be a processor for communicating with an AP (such as those described above, respectively). Figure 1 and Fig.11AIn some other implementations, the wireless communication device 1300 may be an AP that includes such a chip, SoC, chipset, package or device and at least one antenna (such as antenna 1120).

[0141] The wireless communication device 1300 includes a transmitting component 1302, a receiving component 1304, a transmitting information component 1306, and a communication component 1308. Portions of one or more of the components 1302, 1304, 1306, and 1308 may be implemented at least in part using hardware or firmware. For example, the receiving component 1304 may be implemented at least in part by a modem (such as modem 1002). In some implementations, at least some of the components 1302, 1304, 1306, and 1308 are at least partially implemented as software stored in a memory (such as memory 1008). For example, portions of one or more of the components 1302, 1304, 1306, or 1308 may be implemented as non-transitory instructions (or "codes") that are executable by a processor (such as processor 1004) to perform the functions or operations of the corresponding module.

[0142] The sending component 1302 is configured to send a first message indicating a first machine learning capability of a first wireless device.

[0143] The receiving component 1304 is configured to receive, from the second wireless device, a second message indicating second machine learning capabilities of the second wireless device.

[0144] The transmitting information component 1306 is configured to transmit information associated with a machine learning model for use between the first wireless device and the second wireless device based at least in part on the second machine learning capability and the first machine learning capability.

[0145] The communication component 1308 is configured to communicate with a second wireless device based at least in part on the machine learning model.

[0146] Fig.14 is a block diagram of an example wireless communication device 1400 supporting a machine learning framework for WLAN according to various aspects of the present disclosure. In some implementations, the wireless communication device 1400 is configured to perform the above-referenced Fig.12 The wireless communication device 1400 may be a wireless communication device as described above with reference to Fig.101000. For example, the wireless communication device 1400 may be a chip, an SoC, a chipset, a package, or a device that includes at least one modem (e.g., a Wi-Fi (IEEE 802.11) modem or a cellular modem, such as modem 1002), at least one processor (such as processor 1004), at least one radio unit (such as radio unit 1006), and at least one memory (such as memory 1008). In some implementations, the wireless communication device 1400 may be a processor for communicating between STAs (such as those described above, respectively). Figure 1 and Fig. 11B In some other implementations, the wireless communication device 1400 may be a station (STA) that includes such a chip, SoC, chipset, package or device and at least one antenna (such as antenna 1125).

[0147] The wireless communication device 1400 includes a transmitting component 1402, a receiving component 1404, a transmitting information component 1406, and a communication component 1408. Portions of one or more of the components 1402, 1404, 1406, and 1408 may be implemented at least in part using hardware or firmware. For example, the receiving component 1404 may be implemented at least in part by a modem (such as modem 1002). In some implementations, at least some of the components 1402, 1404, 1406, and 1408 are at least partially implemented as software stored in a memory (such as memory 1008). For example, portions of one or more of the components 1402, 1404, 1406, or 1408 may be implemented as non-transitory instructions (or "codes") that are executable by a processor (such as processor 1004) to perform the functions or operations of the corresponding module.

[0148] The sending component 1402 is configured to send a first message indicating a first machine learning capability of a first wireless device.

[0149] Receiving component 1404 is configured to receive, from a second wireless device, a second message indicating second machine learning capabilities of the second wireless device.

[0150] The transmitting information component 1406 is configured to transmit information associated with a machine learning model for use between the first wireless device and the second wireless device based at least in part on the second machine learning capability and the first machine learning capability.

[0151] The communication component 1408 is configured to communicate with a second wireless device based at least in part on the machine learning model.

[0152] Various aspects of the present disclosure relate to signaling details and operational aspects for implementing artificial intelligence and machine learning in WLANs. Signaling details may include container structures such as frames and information elements. Operational aspects may include operational mode changes and indication of information during machine learning usage in a basic service set (BSS).

[0153] According to various aspects of the present disclosure, information is efficiently transmitted between devices in a WLAN on an on-demand basis. Which information to transmit depends on whether the device is in a discovery state or an association state.

[0154] Fig.15 is a table showing WLAN phases for various machine learning operations according to various aspects of the present disclosure. During the discovery phase, beacon messages are sent. For example, basic information such as whether the WLAN device supports machine learning can be provided. Relative to providing all machine learning information in the beacon message, the information container during the discovery phase can indicate machine learning support with low overhead. During the exploration phase, additional machine learning information is provided. For example, supported or expected use cases, and supported machine learning levels (as described above) can be indicated. Containers exchanged during the exploration phase can indicate support for specific use cases.

[0155] During the association phase with the access point, complete machine learning information may be provided for each desired use case. Machine learning functionality may be transmitted. Optionally, model structure, model parameters, and / or model inputs and outputs may be transmitted. However, in some aspects, when operating with an access point, machine learning model structure information, machine learning model parameters, and / or model inputs and outputs may be transmitted during a post-association phase. The container during the association phase should be able to carry machine learning model structure information and model parameter information.

[0156] The machine learning model parameters may indicate algorithm-specific information, such as weights of a neural network-based model, decision variables at nodes of a decision tree or random forest model, and / or decision boundaries at nodes of a decision tree or random forest model. In some aspects, the machine learning model structure information may indicate the number of convolutional layers of a convolutional neural network-based model, the number of pooling layers of a convolutional neural network-based model, the number of fully connected layers of a neural network-based model, and / or the number of input and output features for a neural network-based model. The machine learning model structure information may also indicate the number of neurons in a convolutional layer of a convolutional neural network-based model, the number of neurons in a fully connected layer of a neural network-based model, and / or the activation function for a hidden layer of a neural network-based model. The machine learning model structure information may indicate a loss function for a neural network model, discard information for a neural network model, a maximum depth of a decision tree for a decision tree model, the number of decision trees for a random forest model, and / or a maximum depth of a decision tree for a random forest model.

[0157] When operating with an access point, after association, complete information and model updates for each use case can be transmitted. In addition, changes to the machine learning mode of operation can be transmitted. The container during the post-association phase can indicate machine learning model structure information and machine learning parameters, enable and disable use cases, and provide feedback on the model. The communication in each of the phases can be between two peer stations or between an access point and a station.

[0158] As mentioned above, information is efficiently transmitted between devices in a WLAN on an on-demand basis. For example, a beacon may include only limited information. In some implementations, a beacon may indicate the access point's (AP) support for machine learning. The indication may be binary: supported or not supported. Alternatively, a beacon may advertise a list of use cases, such as descriptors or feature identifiers that the access point supports.

[0159] Additional partial information may be requested through probing. A non-access point wireless station (STA) may request a set of use cases and machine learning models supported by an access point. In response, the access point may provide basic information of the supported machine learning models, such as function ID, model type, structure, and support level.

[0160] In some implementations, access points and non-access point wireless devices may exchange information using Generic Advertisement Service (GAS) frames instead of beacons and probes.

[0161] The complete information can be provided during the association phase or after the association phase. In this context, the complete information includes model inputs and outputs, model structure, and model parameters. By sharing this information during or after association, only legitimate clients receive access to the machine learning model.

[0162] Fig.16 1602 is a state diagram illustrating states of machine learning sessions according to various aspects of the present disclosure. Machine learning operations in a BSS may occur throughout a machine learning session, which has multiple states. When a client is in an inactive state for machine learning, an associated machine learning inactive state 1602 is an initial state, where the client is associated with an access point. In another example, the client may be in a machine learning active state immediately after association. Communication between two peer stations (such as an access point and a station, or two point-to-point (P2P) stations) may establish a machine learning session by activating a machine learning session. A station initiates session activation by sending an activation request to a control station. The control station sends an activation response to the initiating station. Session activation may occur after association, after session termination, or after session suspension. Once activated, the client enters an associated machine learning active state 1604, during which the client may configure and use machine learning algorithms. While in the associated machine learning active state 1604, either station may update the operating mode (OM). A machine learning session may be established for each machine learning function identified by a function ID, or a machine learning session may be established across all machine learning functions between peer stations.

[0163] Either peer may terminate or suspend a machine learning session. The first station initiates a session termination or suspension by sending a termination or suspension request to the second station. The second station may send a session termination or suspension response to confirm the termination or suspension. As a result, the client enters an associated machine learning suspended state 1606 or an associated machine learning inactive state 1602. State information is associated with and stored for an active session. By pausing the session, the information remains stored and reactivation can occur quickly. By tearing down the session, state information is discarded. If a peer station is temporarily not using machine learning functionality (e.g., due to low battery power or high processor load), the peer station may suspend the machine learning session. If a peer station does not intend to use machine learning functionality in the future (e.g., because of poor machine learning model performance), the peer station may terminate the machine learning session. In another example, if a station intends to disassociate from an associated access point, the station may terminate the machine learning session.

[0164] According to aspects of the present disclosure, during an active machine learning session, machine learning operations in a BSS provide for fine-grained control by a controlling entity (such as an access point) over how machine learning is used in the BSS. This fine-grained control enables the controlling entity to be fair to traditional stations while enhancing the performance of stations that support machine learning, preventing abuse, and allowing the controlling entity to revert to fallback options (such as non-machine learning methods). For example, channel access should be fair so that stations with machine learning capabilities and stations without machine learning capabilities have equal access. In order to provide fair access, the controlling entity can disable machine learning for all stations. Although the controlling entity may be described as an access point, the present disclosure is not limited to this. The controlling entity may also be a non-access point station for a point-to-point session.

[0165] The access point operations will now be described. The following operations apply to all model support levels (eg, proprietary models, federated learning models, downloaded and trained models that have not been uploaded, and downloaded models that cannot be retrained).

[0166] An access point may dynamically enable or disable machine learning inference in its BSS. In some aspects, an access point may dynamically enable or disable other aspects of machine learning use, such as training or uploading of trained models. Signaling may be introduced to indicate whether the use of machine learning inference is allowed within a specific time interval. For example, an access point may configure time division multiplexing so that stations with machine learning capabilities operate during some service periods, while stations without machine learning capabilities operate during other service periods. This may allow access points to create intervals in which stations that do not support machine learning remain unaffected by the operation of stations that support machine learning. Signaling may also indicate whether the use of machine learning inference is recommended within a specific time interval, and whether the use of machine learning inference within a specific time interval is mandatory. If a station does not support machine learning inference or has temporarily disabled machine learning, the station should not operate during intervals in which machine learning inference is mandatory.

[0167] Signaling may also be introduced to indicate what portion of the data may be used to derive machine learning inferences. Each portion of the data may refer to a subset in the time domain or a subset of features such as the frequency domain. For example, measurements during a particular time period may be excluded from the input to the machine learning model because those measurements occurred during a time period that experienced abnormal interference. By including those measurements, the model output may be erroneous. Therefore, these measurements may be excluded to improve model performance.

[0168] The signaling may also indicate which portion of the data may be used for machine learning training (e.g., model training), where the portion may again refer to a subset in the time domain or a subset of features such as the frequency domain. Signaling may also be introduced to indicate an interval in which the access point coordinates with non-access point stations to support validation of the learned machine learning model. During this interval, the access point may send synthesized or artificial data to assist the non-access point stations in model validation.

[0169] In some aspects, signaling is on a per-use case basis. For example, during a particular interval, enhanced distributed channel access (EDCA) parameter optimization may not be allowed, but interference estimation may be allowed. Indications may be addressed to individual stations, or may be broadcast. For example, if an access point observes a station accessing a channel aggressively after downloading a machine learning function for an EDCA optimization use case, the access point may send an individual addressed signal to the station to disable use of the EDCA optimization machine learning function.

[0170] Some access point operations may only apply to downloadable models. These operations may require signaling on a per-use case basis, where the signaling is addressed individually or broadcast. For example, an access point may indicate that a new model is available for download. The indication may be per-use case. The indication may be implicit. For example, an access point may announce a period of updates during machine learning session establishment or during association with a station. An access point may also provide different model parameters to different stations. Different model parameters may be used for the same functional ID. The differences may arise from the distance between the station and the access point, the mobility of the station, the capabilities of the station, or different links in the case where the access point and the station support multi-link operation.

[0171] In some aspects, an access point may request a station to indicate whether a particular downloaded machine learning model is in use by the station. For example, even though the model has been downloaded, it may not be in use due to battery or CPU load considerations of the station. The request clarifies whether the model is in use. The access point may request a non-access point (e.g., a station) to provide a reason code for not using the downloaded model. In some aspects, if the station is not using the downloaded model, the access point may request the station to pause or terminate the machine learning session. As mentioned above, the access point maintains the state of the machine learning session. The access point may no longer wish to maintain this state and may therefore request to terminate or pause the session.

[0172] The access point may solicit feedback on the performance of the downloaded model. In some implementations, the access point may fall back to non-machine learning techniques based on the feedback (such as when the model performs poorly).

[0173] Some access point operations apply to retrainable models. These operations may require signaling on a per-use case basis, where the signaling is addressed individually or broadcast. These operations may specify signaling that the access point requests stations to indicate whether they should retrain the model. When stations use downloaded machine learning models, this information may be useful when the access point is troubleshooting different performance across stations.

[0174] Non-access point operations are now described. These operations may be applicable to devices that support downloadable models. In some aspects, a non-access point may not use a downloaded machine learning model within a specific interval. For example, the model may not be used due to poor model performance, battery limitations, or CPU load limitations. If the non-access point is not using the downloaded machine learning model, the non-access point may choose to suspend or remove the machine learning session. The non-access point may also indicate whether the non-access point uses the downloaded machine learning model for inference. The indication may be requested or not requested. In some aspects, the signaling may include an optional reason code. The non-access point may provide (e.g., requested or not requested) feedback on the performance of the model to the access point. For a retrainable model in which retraining occurs, the non-access point may indicate (e.g., requested or not requested) whether the non-access point has retrained the downloaded model.

[0175] For federated learning models, non-access points can upload downloaded models only if they retrain the downloaded models. This restriction can reduce overhead. In some cases, non-access points may not choose to retrain downloaded models. In other cases, non-access points may choose to retrain but not upload the retrained models.

[0176] If the access point's support level is different from that of the non-access point, the lower of the two support levels will be used. For example, if the access point supports federated learning for a use case (e.g., a downloadable model can be retrained and uploaded), and the associated station supports a downloaded model that cannot be retrained, then the downloaded model without retraining is used between the access point and the station.

[0177] A container for machine learning information exchange will now be described. An information element is an example of a container. According to aspects of the present disclosure, an information element should be flexible enough to carry a variety of different types of information related to machine learning. For example, an information element should be able to carry information indicating the following items: support for machine learning, support for machine learning use cases, machine learning model structure, machine learning model parameters, machine learning model inputs and outputs, and / or machine learning operation modes. An information element should be flexible enough to carry a partial set of information or a complete set of information. One or more fields or subfields may always be included in an information element. For example, fields or subfields for general information or public information may be included. An information element may also include one or more optional fields or subfields and subelements (e.g., functional information subelements).

[0178] The information element can have multiple variants. Each variant can be used for a specific purpose. For example, one variant can be used for notifications, while another variant can be used for information exchange. A third variant can be used for operating modes and indications (such as indicating whether the client is using the downloaded model, model performance feedback, etc.).

[0179] Fig.17 is a block diagram illustrating components of information elements for machine learning information exchange in accordance with various aspects of the present disclosure. Fig.17 , the information element includes a control field 1702, a general information field 1704, an optional information field 1706, and a functional information field 1708. The element ID, length, and element ID extension fields are not discussed because they are conventional 802.11 fields. The control field 1702 indicates an element variant, and may also include presence indicators for the general information field and the optional information field and subfields therein. For example, the variant may indicate a type, such as whether the element is used for notification, information exchange, etc. The presence indicator for the general information may include a descriptor bitmap that indicates which subfields are present. The presence indicator for the optional information may indicate whether the optional information field 1706 is present. In Fig.17 In the example of , the optional information field 1706 is indicated as not present using a bitmap of 000000, but other signaling may be used to indicate the absence of the optional information field.

[0180] The general information field 1704 may include a descriptor bitmap subfield 1710, an artificial intelligence / machine learning (AI / ML) capability subfield 1712, and other subfields 1714. The presence indicator for the general information may indicate which subfields are present. Fig.17 In the example of , bitmap 010000 indicates that the descriptor bitmap subfield 1710 does not exist, but the AI / ML capability subfield 1712 exists.

[0181] Descriptor Bitmap subfield 1710 indicates which use cases are supported. Fig.17 In the example of , EDCA optimization and rate adaptation are supported, but CSI enhancement is not supported. The AI / ML capability subfield 1712 indicates the supported capabilities. Fig.17 In the example of , hardware acceleration, random forest model, fast Fourier transform calculation and received signal strength indication (RSSI) measurement are supported. Neural network and federated learning are not supported. Subfield 1714 can have multiple uses. For example, subfield 1714 can be used to indicate whether machine learning is supported.

[0182] The function information field 1708 includes function profile subfields 1716 (only one subfield is labeled for ease of illustration). Each function profile subfield 1716 includes a function control subfield 1718 and a model information subfield 1720. The function control subfield 1718 indicates the function ID, descriptor, and model information type. Fig.17 In the example of , the function ID is zero, the descriptor is EDCA optimization, and the model information type is URL (as described above, the machine learning model is downloaded from an external server). The model information subfield 1720 indicates a link (e.g., URL) for downloading the machine learning model structure information and the machine learning model parameters. In other implementations, the model information subfield 1720 indicates the model type, the machine learning model structure information, and the machine learning model parameters. As described above, additional fields may also be provided to the server for various purposes.

[0183] Frames are another type of container for machine learning information exchange. In some examples, frames can be defined for exchanging machine learning models and their parameters, session initiation and teardown, operating mode updates, and instructions related to machine learning operations. Frames can implement semi-static configuration.

[0184] The frames used to exchange machine learning models and their parameters may be new action frames, which are used to download and upload partial or complete information of machine learning models. These frames may be sent during machine learning session establishment, such as during machine learning session activation. In addition, frames may be sent before activation or after activation. In other implementations, existing management frames may be allowed to carry this information. For example, during association, (re)association request / response frames may be used.

[0185] The frames used for session initiation and teardown can be action frames. These action frames can activate a machine learning session, or pause or tear down a machine learning session. Frames can also update machine learning modes of operation. These frames can enable or disable machine learning inference, change the machine learning model being used, etc. Frames can also provide some type of indication. These frames can indicate the availability of a new or retrained model, solicit or provide feedback on the performance of a machine learning model, etc.

[0186] For faster information exchange, other types of containers can be used for machine learning information exchange. For example, an aggregation control (A-Control) or similar field can be used to quickly exchange information related to machine learning operations.

[0187] Fig.18 is a block diagram illustrating components of an A-Control field for machine learning information exchange in accordance with various aspects of the present disclosure. Fig.18 In the example of , the control subfield of the A-control field includes a control ID subfield 1802 and a control information subfield 1804. The control ID subfield 1802 indicates a new A-Control type for artificial intelligence machine learning (AI / ML) operations.

[0188] The control information subfield 1804 includes a type subfield 1806. The type subfield 1806 may indicate the function of the A-Control field. One type may indicate activation, suspension, or teardown of a machine learning session. Another type may indicate a change in a machine learning model in use. A third type may indicate feedback for a machine learning model in use. The remaining bits in the control information subfield 1804 may carry information related to the function in the type-specific information subfield 1808. The type-specific information subfield 1808 may include a function ID that is used to identify the machine learning function associated with the information carried.

[0189] Four non-limiting examples of the use of the A-Control field are now described. In the first example related to session information, the control information subfield 1804 may indicate whether to terminate, activate, or pause a session for function ID five. In the second example related to the operating mode, the control information subfield 1804 may request to change the machine learning model from function ID four to function ID seven. In this second example, updated machine learning model structure information and machine learning model parameters are exchanged. In the third example related to indications, the control information subfield 1804 may indicate that the model performance for function ID one is unsatisfactory. In the fourth example related to rewards in a reinforcement learning scenario, the control information subfield 1804 indicates reward signaling for the reinforcement learning model, or provides information related to the updated state of the agent in the environment.

[0190] Fig.19 1 is a call flow diagram illustrating a machine learning framework for a wireless local area network (WLAN) according to aspects of the present disclosure. According to aspects of the present disclosure, an access point (AP) 102 may establish a machine learning session with a station (STA) 104. For example, Fig.19 As seen in , at 1910, the access point 102 sends a message indicating support for machine learning. In some implementations, the message may be a beacon message. At 1920, the station 104 may send a message indicating support for at least one machine learning model type. At 1930, the station 104 may initiate a machine learning session by sending an activation request to the access point 102. At 1940, the access point 102 may respond with an activation response. Session activation may occur immediately after association, after session termination, or after session suspension. At 1950, the station 104 sends a machine learning model download request to the access point 102. At 1960, the access point 102 may send machine learning model structure information and machine learning model parameters during the machine learning session. Although Fig.19 Various messages between an access point and a station are shown, but the disclosure is not limited thereto. Messages may also be exchanged between two peer stations.

[0191] Fig. 20 is a block diagram of an example wireless communication device 2000 supporting a machine learning framework for a wireless local area network (WLAN) in accordance with various aspects of the present disclosure. In some implementations, the wireless communication device 2000 is configured to perform a reference Fig.21 The wireless communication device 2000 may be the one or more steps of the process 2100 described above. Fig.101000. For example, the wireless communication device 2000 may be a chip, SoC, chipset, package, or device that includes at least one modem (e.g., a Wi-Fi (IEEE 802.11) modem or a cellular modem, such as modem 1002), at least one processor (such as processor 1004), at least one radio unit (such as radio unit 1006), and at least one memory (such as memory 1008). In some implementations, the wireless communication device 2000 may be a processor for use in an AP (such as the above-referenced respective Figure 1 and Fig.11A In some other implementations, the wireless communication device 2000 may be an AP that includes such a chip, SoC, chipset, package or device and at least one antenna (such as antenna 1120).

[0192] The wireless communication device 2000 includes a transmitting component 2002, a receiving component 2004, an activating component 2006, and a receiving component 2008. Portions of one or more of the components 2002, 2004, 2006, and 2008 may be implemented at least in part using hardware or firmware. For example, the receiving component 2004 may be implemented at least in part by a modem (such as the modem 1002). In some implementations, at least some of the components 2002, 2004, 2006, and 2008 are at least partially implemented as software stored in a memory (such as the memory 1008). For example, portions of one or more of the components 2002, 2004, 2006, or 2008 may be implemented as non-transitory instructions (or "codes") that are executable by a processor (such as the processor 1004) to perform the functions or operations of the corresponding modules.

[0193] The sending component 2002 is configured to send a first message indicating support of machine learning by the first WLAN device. In other aspects, the sending component 2002 is configured to send a second message to the second WLAN device indicating support of one or more machine learning model types by the second WLAN device. The sending component 2002 may also be configured to send machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0194] The receiving component 2004 is configured to receive a second message from the second WLAN device. The second message indicates support of the second WLAN device for one or more machine learning model types. In other aspects, the receiving component 2004 is configured to receive a first message indicating support of the first WLAN device for machine learning.

[0195] The activation component 2006 is configured to activate a machine learning session with a second WLAN device based at least in part on the second message.

[0196] The receiving component 2008 is configured to receive machine learning model structure information and machine learning model parameters from a second WLAN device during a machine learning session.

[0197] Fig.21 is a flow chart illustrating an example process performed, for example, by a wireless local area network (WLAN) device according to various aspects of the present disclosure. Fig.21 As shown in , in some aspects, the process 2100 may send a first message indicating support for machine learning by a first WLAN device (block 2102). In some aspects, the process 2100 may receive a second message from a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device (block 2104). For example, the first WLAN device may send and receive at least one information element, each of which includes at least one of the following items: control information, general information, optional information, and functional information including multiple functional profiles. Each functional profile in the functional profile includes information related to machine learning functions, machine learning model structure information, and machine learning model parameters. In some aspects, the information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, machine learning model structure information, machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

[0198] In other aspects, the first WLAN device sends and receives at least one frame for at least one of: exchanging machine learning model structure information and machine learning model parameters, activating a machine learning session, pausing a machine learning session, tearing down a machine learning session, updating a machine learning operation mode, and providing instructions related to machine learning operations.

[0199] In other aspects, the first WLAN device may send and receive at least one aggregation control (A-control) field, the A-control field including a control subfield indicating the use of A-control for machine learning purposes, the control subfield including a type indicator and a type-specific information indicator. The first message may be a beacon message, and the second message may be a probe message. In other implementations, the first message is a first general announcement service (GAS) message, and the second message is a second GAS message.

[0200] In some aspects, the process 2100 may activate a machine learning session with the second WLAN device based at least in part on the second message (block 2106). For example, the first WLAN device may be configured to activate the machine learning session after association, after termination of a previous machine learning session, or after suspension of the machine learning session. The first WLAN device may be configured to update an operating mode during the machine learning session. The machine learning session may be established for a separate machine learning function or for each machine learning function.

[0201] In some aspects, the process 2100 may receive machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session (block 2108). For example, the machine learning model parameters may indicate at least one of the following: weights of a neural network-based model, decision variables at nodes of a decision tree or random forest model, or decision boundaries at nodes of a decision tree or random forest model. The machine learning model structure information may indicate at least one of the following: the number of convolutional layers of a convolutional neural network-based model, the number of pooling layers of a convolutional neural network-based model, the number of fully connected layers of a neural network-based model, the number of input and output features for a neural network-based model, the number of neurons in a convolutional layer of a convolutional neural network-based model, the number of neurons in a fully connected layer of a neural network-based model, an activation function for a hidden layer in a neural network-based model, a loss function for a neural network-based model, discarding information for a neural network-based model, a maximum depth of a decision tree for a decision tree model, the number of decision trees for a random forest model, or a maximum depth of a decision tree for a random forest model.

[0202] Fig. 22 is a flow chart illustrating an example process performed, for example, by a wireless local area network (WLAN) device according to various aspects of the present disclosure. Fig. 22As shown in , in some aspects, the process 2200 may receive a first message indicating support for machine learning by a first WLAN device (block 2202). In some aspects, the process 2200 may send a second message to a second WLAN device. The second message indicates support for one or more machine learning model types by the second WLAN device (block 2204). For example, the first WLAN device may send and receive at least one information element, each of which includes at least one of the following items: control information, general information, optional information, and functional information including multiple functional profiles. Each functional profile in the functional profile includes information related to machine learning functions, machine learning model structure information, and machine learning model parameters. In some aspects, the information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, machine learning model structure information, machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

[0203] In other aspects, the first WLAN device sends and receives at least one frame for at least one of: exchanging machine learning model structure information and machine learning model parameters, activating a machine learning session, pausing a machine learning session, tearing down a machine learning session, updating a machine learning operation mode, and providing instructions related to machine learning operations.

[0204] In other aspects, the first WLAN device may send and receive at least one aggregation control (A-control) field, the A-control field including a control subfield indicating the use of A-control for machine learning purposes, the control subfield including a type indicator and a type-specific information indicator. The first message may be a beacon message, and the second message may be a probe message. In other implementations, the first message is a first general announcement service (GAS) message, and the second message is a second GAS message.

[0205] In some aspects, process 2200 may activate a machine learning session with a second WLAN device based at least in part on the second message (block 2206). For example, the first WLAN device may be configured to activate the machine learning session after association, after termination of a previous machine learning session, or after suspension of the machine learning session. The first WLAN device may be configured to update an operating mode during the machine learning session. The machine learning session may be established for a separate machine learning function or for each machine learning function.

[0206] In some aspects, the process 2200 may send the machine learning model structure information and the machine learning model parameters to the second WLAN device during the machine learning session (block 2208). For example, the machine learning model parameters may indicate at least one of the following: weights of a neural network-based model, decision variables at nodes of a decision tree or random forest model, or decision boundaries at nodes of a decision tree or random forest model. The machine learning model structure information may indicate at least one of the following: the number of convolutional layers of a convolutional neural network-based model, the number of pooling layers of a convolutional neural network-based model, the number of fully connected layers of a neural network-based model, the number of input and output features for a neural network-based model, the number of neurons in a convolutional layer of a convolutional neural network-based model, the number of neurons in a fully connected layer of a neural network-based model, an activation function for a hidden layer in a neural network-based model, a loss function for a neural network-based model, discarding information for a neural network-based model, a maximum depth of a decision tree for a decision tree model, the number of decision trees for a random forest model, or a maximum depth of a decision tree for a random forest model.

[0207] Example aspects

[0208] Aspect 1: An apparatus for wireless communication by a first wireless local area network (WLAN) device, comprising: a memory; and at least one processor coupled to the memory, the at least one processor being configured to: send a first message indicating support for machine learning by the first WLAN device; receive a second message from a second WLAN device indicating support for at least one machine learning model type by the second WLAN device; activate a machine learning session with the second WLAN device based at least in part on the second message; and receive machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0209] Aspect 2: An apparatus according to Aspect 1, wherein the at least one processor is configured to send and receive at least one information element, each of the at least one information element comprising at least one of the following items: control information, general information, optional information, and functional information comprising multiple functional profiles, each of the multiple functional profiles comprising information related to the machine learning function, the machine learning model structure information, and the machine learning model parameters.

[0210] Aspect 3: An apparatus according to Aspect 1 or 2, wherein the at least one information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, the machine learning model structure information, the machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

[0211] Aspect 4: The apparatus according to any of the preceding aspects, wherein each of the at least one information element has multiple variants.

[0212] Aspect 5: An apparatus according to any one of the preceding aspects, wherein the machine learning model parameters indicate at least one of the following items: weights of a neural network-based model, decision variables at nodes of a decision tree or random forest model, or decision boundaries at nodes of the decision tree or random forest model.

[0213] Aspect 6: An apparatus according to any one of the preceding aspects, wherein the machine learning model structure information indicates at least one of the following items: the number of convolutional layers of a convolutional neural network-based model, the number of pooling layers of the convolutional neural network-based model, the number of fully connected layers of the neural network-based model, the number of input and output features for the neural network-based model, the number of neurons in the convolutional layers of the convolutional neural network-based model, the number of neurons in the fully connected layers of the neural network-based model, an activation function for a hidden layer in the neural network-based model, a loss function for the neural network-based model, discarding information for the neural network-based model, a maximum depth of a decision tree for a decision tree model, the number of decision trees for a random forest model, or the maximum depth of a decision tree for the random forest model.

[0214] Aspect 7: An apparatus according to any one of the preceding aspects, wherein the at least one processor is configured to send and receive at least one frame for at least one of the following items: exchanging the machine learning model structure information and the machine learning model parameters, activating the machine learning session, pausing the machine learning session, tearing down the machine learning session, updating the machine learning operation mode, and providing instructions related to the machine learning operation.

[0215] Aspect 8: An apparatus according to any one of Aspects 1-6, wherein the at least one processor is configured to send and receive at least one aggregation control (A-control) field, the A-control field including a control subfield indicating the use of A-control for machine learning purposes, the control subfield including a type indicator and a type-specific information indicator.

[0216] Aspect 9: The apparatus according to any one of the preceding aspects, wherein the first message comprises a beacon message and the second message comprises a probe message.

[0217] Aspect 10: The apparatus according to any one of aspects 1-8, wherein the first message comprises a first general advertisement service (GAS) message, and the second message comprises a second GAS message.

[0218] Aspect 11: An apparatus according to any of the preceding aspects, wherein the at least one processor is configured to activate the machine learning session after association, after a previous machine learning session is terminated, or after the machine learning session is suspended.

[0219] Aspect 12: An apparatus according to any of the preceding aspects, wherein the at least one processor is configured to update an operating mode during the machine learning session.

[0220] Aspect 13: An apparatus according to any of the preceding aspects, wherein the machine learning session is established for a separate machine learning function.

[0221] Aspect 14: An apparatus according to any one of Aspects 1-12, wherein the machine learning session is established for all machine learning functions.

[0222] Aspect 15: An apparatus according to any of the preceding aspects, wherein the at least one processor is configured to dynamically enable or disable machine learning inference.

[0223] Aspect 16: An apparatus according to any of the preceding aspects, wherein the at least one processor is configured to send information indicating at least one of the following items: a first time interval within which machine learning inference is allowed, a second time interval within which machine learning inference is recommended, and a third time interval within which machine learning inference is mandatory.

[0224] Aspect 17: An apparatus according to any one of Aspects 1-15, wherein the at least one processor is configured to send information indicating at least one of the following items: a first portion of data to be used to generate machine learning inferences, a second portion of data to be used for machine learning training, and a third portion of data to be used to verify the machine learning model.

[0225] Aspect 18: An apparatus according to any one of Aspects 1-15, wherein the at least one processor is configured to send information indicating at least one of the following items: whether a new machine learning model is available for download, different model parameters for different WLAN devices, a request to identify whether the second WLAN device is using the downloaded machine learning model, a request to pause or terminate the machine learning session when the second WLAN device is not using the downloaded machine learning model, a request for performance feedback on the downloaded machine learning model, and an inquiry as to whether the second WLAN device has retrained the model.

[0226] Aspect 19: An apparatus according to any of the preceding aspects, wherein the at least one processor is configured to receive a message for at least one of: when the downloaded machine learning model will not be used, and a request to suspend or terminate the machine learning session when the first WLAN device is not using the downloaded machine learning model.

[0227] Aspect 20: The apparatus of claim 1, wherein the at least one processor is configured to send information indicating at least one of: whether the first WLAN device is using a downloaded machine learning model, feedback on model performance, and whether the first WLAN device has retrained the downloaded model.

[0228] Aspect 21: The apparatus according to any one of Aspects 1-15, wherein the at least one processor is configured to upload the downloaded model in response to training the downloaded model.

[0229] Aspect 22: A method for wireless communication by a first wireless local area network (WLAN) device, comprising: sending a first message indicating support for machine learning by the first WLAN device; receiving a second message from a second WLAN device indicating support for at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; and receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0230] Aspect 23: The method according to Aspect 22 also includes: sending and receiving at least one information element, each of the at least one information element includes at least one of the following items: control information, general information, optional information, and functional information including multiple functional profiles, each of the multiple functional profiles includes information related to the machine learning function, the machine learning model structure information and the machine learning model parameters.

[0231] Aspect 24: A method according to Aspect 22 or 23, wherein the at least one information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, the machine learning model structure information, the machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

[0232] Aspect 25: The method according to any one of Aspects 22-24 further includes: sending and receiving at least one aggregation control (A-control) field, wherein the A-control field includes a control subfield indicating the use of A-control for machine learning purposes, and the control subfield includes a type indicator and a type-specific information indicator.

[0233] Aspect 26: The method according to any one of aspects 22-25, wherein the first message comprises a beacon message and the second message comprises a probe message.

[0234] Aspect 27: The method according to any one of aspects 22-25, wherein the first message includes a first general announcement service (GAS) message, and the second message includes a second GAS message.

[0235] Aspect 28: The method according to any one of Aspects 22-27 further includes dynamically enabling or disabling machine learning inference.

[0236] Aspect 29: An apparatus for wireless communication by a first wireless local area network (WLAN) device, comprising: a memory; and at least one processor coupled to the memory, the at least one processor being configured to: receive a first message indicating support for machine learning by the first WLAN device; send a second message to a second WLAN device indicating support for at least one machine learning model type by the second WLAN device; activate a machine learning session with the second WLAN device based at least in part on the second message; and send machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0237] Aspect 30: A method for wireless communication by a first wireless local area network (WLAN) device, comprising: receiving a first message indicating support for machine learning by the first WLAN device; sending a second message to a second WLAN device indicating support for at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; and sending machine learning model structure information and machine learning model parameters to the second WLAN device during the machine learning session.

[0238] Aspect 31: An apparatus for wireless communication by a first wireless local area network (WLAN) device, comprising: a unit for sending a first message indicating support of machine learning by the first WLAN device; a unit for receiving a second message from a second WLAN device indicating support of at least one machine learning model type by the second WLAN device; a unit for activating a machine learning session with the second WLAN device based at least in part on the second message; and a unit for receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0239] Aspect 32: A non-transitory computer-readable medium storing program code for execution by a first wireless local area network (WLAN) device, the program code comprising: program code for sending a first message indicating support for machine learning by the first WLAN device; program code for receiving a second message from a second WLAN device indicating support for at least one machine learning model type by the second WLAN device; program code for activating a machine learning session with the second WLAN device based at least in part on the second message; and program code for receiving machine learning model structure information and machine learning model parameters from the second WLAN device during the machine learning session.

[0240] The above disclosure provides illustration and description, but is not intended to be exhaustive or to limit the aspects to the precise forms disclosed. Modifications and variations are possible in light of the above disclosure or may be acquired from practice of the various aspects.

[0241] As used, the term "component" is intended to be broadly interpreted as hardware, firmware, and / or a combination of hardware and software. As used, a processor is implemented using hardware, firmware, and / or a combination of hardware and software.

[0242] Some aspects are described in conjunction with a threshold. As used, satisfying a threshold may refer to a value being greater than a threshold, greater than or equal to a threshold, less than a threshold, less than or equal to a threshold, equal to a threshold, not equal to a threshold, etc., depending on the context.

[0243] It will be apparent that the described systems and / or methods can be implemented in various forms of hardware, firmware, and combinations of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods does not limit the aspects. Therefore, the operation and behavior of the systems and / or methods are described without reference to specific software code, it being understood that software and hardware can be designed to implement the systems and / or methods based at least in part on the description.

[0244] Although the specific combination of features is recorded in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of various aspects. In fact, many of these features can be combined in a manner not specifically recorded in the claims and / or disclosed in the specification. Although each dependent claim listed below may directly only be subordinate to a claim, the disclosure of various aspects includes the combination of each dependent claim and each other claim in the claim set. The phrase of "at least one of" referring to the list of items refers to any combination of these items, including single members. For example, "at least one of a, b or c" is intended to cover a, b, c, ab, ac, bc and abc, and any combination of identical elements with multiples (for example, aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc and ccc or any other sorting of a, b and c).

[0245] Any element, action or instruction used should not be interpreted as critical or necessary unless explicitly described as such. In addition, as used, the articles "a" and "an" are intended to include one or more items, and can be used interchangeably with "one or more". In addition, as used, the terms "set" and "group" are intended to include one or more items (e.g., related items, unrelated items, combinations of related items and unrelated items, and / or similar items), and can be used interchangeably with "one or more". In the case of only meaning one item, the phrase "only one" or similar terms are used. In addition, as used, the terms "has", "have", "having" and / or similar terms are intended to be open terms. In addition, the phrase "based on" is intended to mean "based at least in part on", unless otherwise explicitly stated.

Claims

1. An apparatus for wireless communication by a first wireless local area network (WLAN) device, comprising: Memory; as well as at least one processor coupled to the memory, the at least one processor configured to: sending a first message indicating support of machine learning by the first WLAN device; receiving, from a second WLAN device, a second message indicating support of at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; as well as During the machine learning session, machine learning model structure information and machine learning model parameters are received from the second WLAN device.

2. The device according to claim 1, wherein: The at least one processor is configured to send and receive at least one information element, each of the at least one information element comprising at least one of the following items: control information, general information, optional information, and functional information comprising multiple functional profiles, each of the multiple functional profiles comprising information related to the machine learning function, the machine learning model structure information, and the machine learning model parameters.

3. The device according to claim 2, wherein: The at least one information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, the machine learning model structure information, the machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

4. The device according to claim 2, wherein: Each of the at least one information element has a plurality of variations.

5. The device according to claim 1, wherein: The machine learning model parameters indicate at least one of the following: weights of a neural network-based model, decision variables at nodes of a decision tree or random forest model, or decision boundaries at nodes of the decision tree or random forest model.

6. The device according to claim 1, wherein: The machine learning model structure information indicates at least one of the following items: the number of convolutional layers of the convolutional neural network based model, the number of pooling layers of the convolutional neural network based model, the number of fully connected layers of the neural network based model, the number of input and output features for the neural network based model, the number of neurons in the convolutional layers of the convolutional neural network based model, the number of neurons in the fully connected layers of the neural network based model, the activation function for the hidden layer in the neural network based model, the loss function for the neural network based model, the discarding information for the neural network based model, the maximum depth of a decision tree for a decision tree model, the number of decision trees for a random forest model, or the maximum depth of a decision tree for the random forest model.

7. The device according to claim 1, wherein: The at least one processor is configured to send and receive at least one frame for at least one of: exchanging the machine learning model structure information and the machine learning model parameters, activating the machine learning session, pausing the machine learning session, tearing down the machine learning session, updating the machine learning operation mode, and providing instructions related to the machine learning operation.

8. The device according to claim 1, wherein: The at least one processor is configured to send and receive at least one aggregation control (A-control) field, the A-control field comprising a control subfield indicating use of the A-control for machine learning purposes, the control subfield comprising a type indicator and a type-specific information indicator.

9. The device according to claim 1, wherein: The first message comprises a beacon message, and the second message comprises a probe message.

10. The device according to claim 1, wherein: The first message comprises a first Generic Advertisement Service (GAS) message, and the second message comprises a second GAS message.

11. The device according to claim 1, wherein: The at least one processor is configured to activate the machine learning session after association, after a previous machine learning session is terminated, or after the machine learning session is suspended.

12. The device according to claim 1, wherein: The at least one processor is configured to update an operating mode during the machine learning session.

13. The device according to claim 1, wherein: The machine learning session is established for a separate machine learning function.

14. The device according to claim 1, wherein: The machine learning session is established for all machine learning functions.

15. The device according to claim 1, wherein: The at least one processor is configured to dynamically enable or disable machine learning inference.

16. The device according to claim 15, wherein: The at least one processor is configured to send information indicating at least one of: a first time interval within which machine learning inference is allowed, a second time interval within which machine learning inference is recommended, and a third time interval within which machine learning inference is mandatory.

17. The device according to claim 1, wherein: The at least one processor is configured to send information indicating at least one of: a first portion of data to be used to generate machine learning inferences, a second portion of data to be used for machine learning training, and a third portion of data to be used to validate a machine learning model.

18. The device according to claim 1, wherein: The at least one processor is configured to send information indicating at least one of: whether a new machine learning model is available for download, different model parameters for different WLAN devices, a request to identify whether the second WLAN device is using the downloaded machine learning model, a request to pause or terminate the machine learning session when the second WLAN device is not using the downloaded machine learning model, a request for performance feedback for the downloaded machine learning model, and an inquiry as to whether the second WLAN device has retrained the model.

19. The device according to claim 1, wherein: The at least one processor is configured to receive a message for at least one of: when the downloaded machine learning model will not be used, and a request to suspend or terminate the machine learning session when the first WLAN device is not using the downloaded machine learning model.

20. The device according to claim 1, wherein The at least one processor is configured to send information indicating at least one of whether the first WLAN device is using the downloaded machine learning model, feedback on model performance, and whether the first WLAN device has retrained the downloaded model.

21. The device according to claim 1, wherein The at least one processor is configured to upload the downloaded model in response to training the downloaded model.

22. A method for wireless communication by a first wireless local area network (WLAN) device, comprising: sending a first message indicating support of machine learning by the first WLAN device; receiving, from a second WLAN device, a second message indicating support of at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; as well as The machine learning model structure information and the machine learning model parameters are received from the second WLAN device during the machine learning session.

23. The method according to claim 22, further comprising: Send and receive at least one information element, each of the at least one information element includes at least one of the following items: control information, general information, optional information, and functional information including multiple functional profiles, each of the multiple functional profiles includes information related to the machine learning function, the machine learning model structure information and the machine learning model parameters.

24. The method according to claim 23, wherein: The at least one information element includes information related to at least one of the following items: support for machine learning, support for machine learning use cases, the machine learning model structure information, the machine learning model parameters, machine learning model input, machine learning model output, and machine learning operation mode.

25. The method of claim 22, further comprising: At least one aggregation control (A-control) field is sent and received, wherein the A-control field includes a control subfield indicating the use of the A-control for machine learning purposes, and the control subfield includes a type indicator and a type-specific information indicator.

26. The method of claim 22, wherein: The first message comprises a beacon message, and the second message comprises a probe message.

27. The method of claim 22, wherein: The first message comprises a first Generic Advertisement Service (GAS) message, and the second message comprises a second GAS message.

28. The method of claim 22, further comprising dynamically enabling or disabling machine learning inference.

29. An apparatus for wireless communication by a first wireless local area network (WLAN) device, comprising: Memory; as well as at least one processor coupled to the memory, the at least one processor configured to: receiving a first message indicating support of machine learning by the first WLAN device; sending a second message to a second WLAN device indicating support of the at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; as well as The machine learning model structure information and the machine learning model parameters are sent to the second WLAN device during the machine learning session.

30. A method for wireless communication by a first wireless local area network (WLAN) device, comprising: receiving a first message indicating support of machine learning by the first WLAN device; sending a second message to a second WLAN device indicating support of the at least one machine learning model type by the second WLAN device; activating a machine learning session with the second WLAN device based at least in part on the second message; as well as The machine learning model structure information and the machine learning model parameters are sent to the second WLAN device during the machine learning session.

Images