Artificial intelligence-based synchronization signal scanning

Abstract

Certain aspects of this disclosure provide techniques for AI-based synchronization signal scanning. A method for wireless communication by a device includes monitoring at least a first frequency bandwidth across a set of time windows. The method also includes using an artificial intelligence (AI) model to identify at least one synchronization signal block (SSB) centered on a Global Synchronization Channel Number (GSCN) within the first frequency bandwidth, the AI ​​model including a kernel size tuned to the SSB aspect ratio. The method further includes communicating with network entities at least in part based on at least one SSB.

Description

[0001] Cross-references to related applications

[0002] This patent application claims priority and benefit to U.S. Patent Application No. 18 / 535,269, filed December 11, 2023, which is hereby expressly incorporated herein by reference in its entirety. Background Technology Technical Field

[0004] Various aspects of this disclosure relate to wireless communication, and more specifically to techniques for scanning synchronization signals.

[0005] Related technical descriptions

[0006] Wireless communication systems are widely deployed to provide a variety of telecommunications services, such as telephone, video, data, messaging, broadcasting, or other similar services. These wireless communication systems may employ multiple access technologies that enable communication with several users by sharing available wireless communication system resources.

[0007] Despite significant technological advancements in wireless communication systems over the years, challenges remain. For example, complex and dynamic environments can still attenuate or block signals between wireless transmitters and receivers. Therefore, there is a continuous expectation for improving the technical performance of wireless communication systems, including, for example: improving communication speed and data carrying capacity; improving the efficiency of shared communication media; reducing the power used by transmitters and receivers during communication; improving the reliability of wireless communication; avoiding redundant transmission and / or reception and related processing; improving the coverage area of ​​wireless communication; increasing the number and types of devices that can access the wireless communication system; increasing the ability of different types of devices to communicate with each other; and increasing the number and types of available wireless communication media. Therefore, there is a need for further improvements to wireless communication systems to overcome the aforementioned technical challenges and other obstacles. Summary of the Invention

[0008] In some wireless communication systems, user equipment (UE) can scan certain broadcast signals to establish a communication link with a network entity (e.g., a base station). For example, a UE can perform a full frequency scan across the entire frequency bandwidth available for synchronization signals (e.g., one or more synchronization signal blocks (SSBs)). Full frequency scanning or other scanning techniques can consume significant time and power to detect synchronization signals. Therefore, frequency scanning and its accuracy can affect the latency and / or power consumption associated with establishing a communication link between the UE and the network entity.

[0009] The aspects described herein provide techniques for artificial intelligence (AI)-based sync signal scanning and techniques for training AI models for such sync signal scanning. AI models can be trained and / or configured to detect certain features of SSBs, including, for example, the time-frequency aspect ratio associated with the SSB and / or any other suitable features as further described herein. In some cases, the kernel of the AI ​​model can be tuned to the aspect ratio of the SSB in the time-frequency grid. AI-based sync signal scanning can achieve improved accuracy with respect to SSB detection in pre-scans. The AI-based sync signal scanning described herein can reduce scan time with respect to full-frequency scanning and / or other scanning techniques.

[0010] Some aspects provide a method for wireless communication by a device. The method includes monitoring at least a first frequency bandwidth across a set of time windows. The method also includes using an artificial intelligence (AI) model to identify at least one synchronization signal block (SSB) centered on a globally synchronized channel number (GSCN) within the first frequency bandwidth, the AI ​​model including a kernel size tuned to the aspect ratio of the SSB. The method further includes communicating with network entities at least in part based on the at least one SSB.

[0011] Some aspects provide a method. This method includes obtaining training data comprising: multiple spectral energy images and multiple labels, wherein at least one of the multiple labels includes at least one indication of whether the corresponding spectral energy image contains at least a portion of an SSB. The method also includes using the training data to train an AI model to provide one or more probabilities of detecting one or more SSBs within a frequency bandwidth spanning a set of time windows, the frequency bandwidth including multiple GSCNs.

[0012] Other aspects provide: one or more means operable to, configured to, or otherwise adapted to perform any portion of any method described herein (e.g., such that performance can be implemented by only one means or in a distributed manner across multiple means); one or more non-transitory computer-readable media comprising instructions that, when executed by one or more processors of the one or more means, cause the one or more means to perform any portion of any method described herein (e.g., such that instructions can be included in only one computer-readable medium or in a distributed manner across multiple computer-readable media, such that instructions can be executed by only one processor or by multiple processors in a distributed manner, such that in the one or more means...). Each device may include one or more processors, and / or enable execution to be performed by only one device or in a distributed manner across multiple devices; one or more computer program products embodied on one or more computer-readable storage media including code for performing any part of any method described herein (e.g., enabling the code to be stored in only one computer-readable medium or in a distributed manner across computer-readable media); and / or one or more devices including one or more components for performing any part of any method described herein (e.g., enabling execution to be performed by only one device or by multiple devices in a distributed manner). By way of example, a device may include a processing system, a device having a processing system, or a processing system cooperating via one or more networks. A device may include: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to enable the device to perform any part of any method described herein. In some examples, one or more processors may be pre-configured to perform the various functions or operations described herein without being configured by software.

[0013] For illustrative purposes, the following description and figures illustrate certain features. Attached Figure Description

[0014] The accompanying drawings depict certain features of the various aspects described herein and should not be considered as limiting the scope of this disclosure.

[0015] Figure 1 An example wireless communication network is depicted.

[0016] Figure 2 An example decomposed base station architecture is described.

[0017] Figure 3 Various aspects of the example base station and example user equipment (UE) are described.

[0018] Figure 4A , Figure 4B , Figure 4C and Figure 4D Various example aspects of data structures used in wireless communication networks are described.

[0019] Figure 5 An example synchronization signal block (SSB) in the time-frequency domain is illustrated.

[0020] Figure 6 An example AI architecture is shown that can be used for artificial intelligence (AI) enhanced wireless communication.

[0021] Figure 7 An example AI architecture of a first wireless device communicating with a second wireless device is illustrated.

[0022] Figure 8 An example artificial neural network is shown.

[0023] Figure 9 An example operation for performing SSB pre-scan by the UE is illustrated.

[0024] Figure 10 Examples are provided for performing actions as described in this article. Figure 9 The example operation of the described SSB pre-scan.

[0025] Figure 11 An example convolutional neural network trained to detect SSBs is shown.

[0026] Figure 12 Example operations for training a machine learning model to detect SSBs are shown.

[0027] Figure 13 Example scan times for AI-based SSB scanning and full-frequency scanning techniques are shown.

[0028] Figure 14 The process flow for communication in the system between the UE and network entities is described.

[0029] Figure 15 The method for training AI models is described.

[0030] Figure 16 A method for wireless communication is described.

[0031] Figure 17 Various aspects of the example communication device are described. Detailed Implementation

[0032] This disclosure provides apparatus, methods, processing systems, and computer-readable media for synchronous signal scanning based on artificial intelligence (AI).

[0033] In some wireless communication systems (e.g., 5G new radio systems and / or future wireless communication systems), user equipment (UE) can scan for certain broadcast signals (e.g., synchronization signals) to establish a communication link with network entities (e.g., base stations). For example, during initial cell acquisition, the UE can scan for certain frequency resources for broadcast signals carrying synchronization information (such as SSBs), as described herein. Figure 6 Further description. In addition, broadcast signals can be transmitted at specific periods (e.g., once every 5 milliseconds (ms) to 160 ms).

[0034] Technical challenges in scanning synchronization signals include, for example, the timing, accuracy, and / or power required to perform the synchronization signal scan. Since the UE may not have information about the frequency location of the broadcast signal or the transmission time of the broadcast signal, the UE may scan multiple frequency bandwidths to detect SSBs, for example, by performing a full-frequency scan across the entire frequency bandwidth available for the synchronization signal. As an example, the UE may perform a full-frequency scan when the device switches out of offline mode (such as airplane mode). Offline mode involves a disconnected state in which the UE avoids transmitting radio frequency signals. A full-frequency scan can consume significant time and power to detect SSBs. In some cases, the UE may perform spectral energy correlation techniques to reduce the latency of searching for SSBs. However, such spectral energy correlation techniques may provide false SSB detections under certain conditions, and therefore, in response to false detections, the UE may search for SSBs even when no SSBs have been transmitted. Therefore, the frequency scan and the accuracy of such scans can affect the latency and / or power consumption associated with establishing a communication link between the UE and network entities.

[0035] The aspects described herein overcome the aforementioned technical problems by providing techniques for AI-based synchronous signal scanning and techniques for training AI models for such synchronous signal scanning. More specifically, AI models (e.g., neural networks) can be trained to detect SSBs in a spectral energy image (e.g., a spectrogram or matrix indicating the value of spectral energy over time) representing the frequency bandwidth monitored at a specific duration (e.g., 20 ms), for example, as described herein. Figure 13 Further described. As an example, the UE can perform a synchronization signal pre-scan, which identifies candidate frequencies and / or timings at which an SSB can be received. The UE can compress samples of the frequency bandwidth monitored over time into a spectral energy image. The UE can provide the spectral energy image to an AI model, and the AI ​​model can output the probability of whether an SSB is detected in the spectral energy image, as described herein. Figures 10 to 12Further description: When an SSB is detected in a specific frequency bandwidth, the UE can use the information carried in the corresponding SSB to perform cell acquisition and establish a communication link with network entities.

[0036] The techniques for AI-based synchronization signal scanning and training described herein offer various beneficial effects and / or advantages. AI-based synchronization signal scanning, as described herein, can, for example, reduce the error rate associated with SSB detection in pre-scanning with correlation-based energy scanning techniques. For example, AI models can be trained and / or configured to detect certain features of SSBs to achieve improved accuracy in SSB detection. This improved accuracy in SSB detection can lead to reduced latency and / or power consumption for synchronization and cell acquisition. AI-based synchronization signal scanning, as described herein, can reduce scan time with respect to full-frequency scanning. This reduction in scan time can reduce latency and / or power consumption for synchronization and cell acquisition. The AI ​​model training techniques described herein enable AI models that can adapt to various channel conditions and / or communication scenarios. For example, online training allows AI models to be trained under various channel conditions and / or communication scenarios. AI models can be trained to detect SSBs under various channel conditions and / or communication scenarios, including, for example, line-of-sight conditions, non-line-of-sight conditions, various UE mobility states, various transmission ranges, various frequency bands, multipath conditions, fading, scattering, interference, noise, etc. Therefore, AI models can detect SSBs across various channel conditions and / or communication scenarios with improved accuracy, reduced latency, and / or reduced power consumption.

[0037] An introduction to wireless communication networks

[0038] The techniques and methods described herein can be used in a variety of wireless communication networks. Although aspects herein may be described using terms commonly associated with 3G, 4G, 5G, 6G, and / or other generations of wireless technologies, aspects of this disclosure are equally applicable to other communication systems and standards not explicitly mentioned herein.

[0039] Figure 1 An example of a wireless communication network 100 in which the aspects described herein can be implemented is depicted.

[0040] Generally, wireless communication network 100 includes various network entities (alternatively, network elements or network nodes). Network entities are typically communication devices and / or communication functions performed by communication devices (e.g., user equipment (UE), base station (BS), components of a BS, servers, etc.). Because such communication devices are part of wireless communication network 100 and facilitate wireless communication, they may be referred to as wireless communication devices. For example, various functions of the network and various devices associated with and interacting with the network may be considered network entities. Furthermore, wireless communication network 100 includes terrestrial and non-terrestrial aspects (also referred to herein as non-terrestrial network entities). Terrestrial aspects include terrestrial network entities such as terrestrial network entities (e.g., BS 102), and non-terrestrial aspects include satellite 140. These non-terrestrial aspects may include onboard network entities (e.g., one or more BSs) capable of communicating with other network elements (e.g., terrestrial BSs) and UEs.

[0041] In the depicted example, wireless communication network 100 includes BS 102, UE 104 and one or more core networks (such as Evolved Packet Core (EPC) 160 and 5G Core (5GC) network 190) that interoperate to provide communication services over various communication links, including wired and wireless links.

[0042] Figure 1 Various example UE 104s are described, which may more generally include: cellular phones, smartphones, Session Initiation Protocol (SIP) phones, laptops, personal digital assistants (PDAs), satellite radios, GPS devices, multimedia devices, video devices, digital audio players, cameras, game consoles, tablet computers, smart devices, wearable devices, vehicles, electricity meters, air pumps, large or small kitchen appliances, healthcare devices, implants, sensors / actuators, displays, Internet of Things (IoT) devices, always-on (AON) devices, edge processing devices, data centers, or other similar devices. UE 104 may also be more generally referred to as mobile devices, wireless devices, stations, mobile stations, subscriber stations, mobile subscriber stations, mobile units, subscriber units, wireless units, remote units, remote devices, access terminals, mobile terminals, wireless terminals, remote terminals, mobile phones, and others.

[0043] BS 102 communicates wirelessly with UE 104 via communication link 120 (e.g., sending or receiving signals to or from UE 104). Communication link 120 between BS 102 and UE 104 may include uplink (UL) transmission (also referred to as reverse link) from UE 104 to BS 102 and / or downlink (DL) transmission (also referred to as forward link) transmission from BS 102 to UE 104. In various aspects, communication link 120 may utilize multiple-input multiple-output (MIMO) antenna technologies, including spatial multiplexing, beamforming, and / or transmit diversity.

[0044] BS 102 may typically include: NodeB, enhanced NodeB (eNB), next-generation enhanced NodeB (ng-eNB), next-generation NodeB (gNB or gNodeB), access point, transceiver base station, radio base station, radio transceiver, transceiver functionality, transmit / receive point, and / or others. Each of BS 102 provides communication coverage for a corresponding coverage area 110, which may sometimes be referred to as a cell, and in some cases may overlap (e.g., a small cell 102' may have a coverage area 110' that overlaps with the coverage area 110 of a macro cell). For example, BS may provide communication coverage for macro cells (covering a relatively large geographic area), pico cells (covering a relatively small geographic area, such as a stadium), femtocells (covering a relatively small geographic area (e.g., a home)), and / or other types of cells.

[0045] Generally, a cell can refer to a portion, partition, or segment of wireless communication coverage served by network entities within a wireless communication network. A cell can have geographical characteristics (such as a geographical coverage area) and radio frequency characteristics (such as time and / or frequency resources dedicated to the cell). For example, multiple cells employing different frequency resources (e.g., bandwidth portions) and / or different time resources can cover a specific geographical coverage area. As another example, a single cell can cover a specific geographical coverage area. In some contexts (e.g., carrier aggregation scenarios and / or multi-connectivity scenarios), the terms "cell" or "serving cell" can refer to or correspond to a specific carrier frequency (e.g., component carrier) used for wireless communication, and "cell group" can refer to or correspond to multiple carriers used for wireless communication. As an example, in a carrier aggregation scenario, a UE can communicate on multiple component carriers corresponding to multiple (serving) cells in the same cell group, and in a multi-connectivity (e.g., dual-connectivity) scenario, a UE can communicate on multiple component carriers corresponding to multiple cell groups.

[0046] Although BS 102 is described as a single communication device in various aspects, it can be implemented in a variety of configurations. For example, to give a few examples, one or more components of the base station can be decomposed, including a central unit (CU), one or more distributed units (DU), one or more radio units (RU), a near real-time (near RT) RAN intelligent controller (RIC), or a non-real-time (non-RT) RIC. In another example, various aspects of the base station can be virtualized. More generally, a base station (e.g., BS 102) can include components located at a single physical location or components located at various physical locations. In examples where the base station includes components located at various physical locations, the various components can each perform functions, such that the various components collectively achieve functionality similar to a base station located at a single physical location. In some aspects, a base station including components located at various physical locations can be referred to as a decomposed radio access network architecture (such as an open RAN (O-RAN) or virtualized RAN (VRAN) architecture). Figure 2 An example decomposed base station architecture is depicted and described.

[0047] Different BSs 102 within the wireless communication network 100 can also be configured to support different radio access technologies (such as 3G, 4G, and / or 5G). For example, a BS 102 configured for 4G LTE (collectively referred to as Evolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (E-UTRAN)) can interface with EPC 160 via a first backhaul link 132 (e.g., S1 interface). A BS 102 configured for 5G (e.g., 5G NR or Next Generation RAN (NG-RAN)) can interface with 5GC 190 via a second backhaul link 184. BSs 102 can communicate directly or indirectly (e.g., via EPC 160 or 5GC 190) on a third backhaul link 134 (e.g., X2 interface), which can be wired or wireless.

[0048] Wireless communication network 100 can subdivide the electromagnetic spectrum into various categories, bands, channels, or other characteristics. In some aspects, subdivision is provided based on wavelength and frequency, where frequency may also be referred to as carrier, subcarrier, channel, tone, or subband. For example, 3GPP currently defines frequency range 1 (FR1) as including 410MHz to 7125MHz, which is often (interchangeably) referred to as “sub-6GHz”. Similarly, 3GPP currently defines frequency range 2 (FR2) as including 24,250MHz to 71,000MHz, which is sometimes (interchangeably) referred to as “millimeter wave” (“mmW” or “mmWave”). In some cases, FR2 can be further defined according to subranges (such as a first subrange FR2-1 including 24,250MHz to 52,600MHz and a second subrange FR2-2 including 52,600MHz to 71,000MHz). Base stations configured to communicate using mmWave / near mmWave radio bands (e.g., mmWave base stations such as BS 180) can utilize beamforming (e.g., 182) with UEs (e.g., 104) to improve path loss and range.

[0049] The communication link 120 between BS 102 and, for example, UE 104 can be via one or more carriers, which may have different bandwidths (e.g., 5MHz, 10MHz, 15MHz, 20MHz, 100MHz, 400MHz and / or other MHz) and may be aggregated in various ways. The carriers may or may not be adjacent to each other. The allocation of carriers may be asymmetric with respect to DL and UL (e.g., more or fewer carriers may be allocated to DL compared to UL).

[0050] Compared to lower-frequency communication, communication using higher frequency bands may have higher path loss and shorter range. Therefore, some base stations (e.g., Figure 1The beamforming 182 of the BS 180 (180) with the UE 104 can be used to improve path loss and range. For example, the BS 180 and UE 104 may each include multiple antennas, such as antenna elements, antenna panels, and / or antenna arrays, to facilitate beamforming. In some cases, the BS 180 may transmit beamformed signals to the UE 104 in one or more transmit directions 182''. The UE 104 may receive beamformed signals from the BS 180 in one or more receive directions 182''. The UE 104 may also transmit beamformed signals to the BS 180 in one or more transmit directions 182''. The BS 180 may also receive beamformed signals from the UE 104 in one or more receive directions 182''. The BS 180 and UE 104 may then perform beamforming training to determine the optimal receive and transmit directions for each of the BS 180 and UE 104. It is worth noting that the transmit and receive directions of the BS 180 may or may not be the same. Similarly, the transmission and reception directions of UE 104 may or may not be the same.

[0051] The wireless communication network 100 further includes a Wi-Fi AP 150 that communicates with a Wi-Fi station (STA) 152 via a communication link 154 in, for example, unlicensed spectrum in 2.4 GHz and / or 5 GHz.

[0052] Some UEs 104 may use device-to-device (D2D) communication link 158 to communicate with each other. The D2D communication link 158 may use one or more sidelink channels, such as physical sidelink broadcast channel (PSBCH), physical sidelink discovery channel (PSDCH), physical sidelink shared channel (PSSCH), physical sidelink control channel (PSCCH), and / or physical sidelink feedback channel (PSFCH).

[0053] EPC 160 may include various functional components, including: Mobility Management Entity (MME) 162, other MMEs 164, Serving Gateway 166, Multimedia Broadcast Multicast Service (MBMS) Gateway 168, Broadcast Multicast Service Center (BM-SC) 170, and / or Packet Data Network (PDN) Gateway 172, as in the illustrated example. MME 162 may communicate with Home Subscriber Server (HSS) 174. MME 162 is the control node that handles signaling between UE 104 and EPC 160. Generally, MME 162 provides bearer and connectivity management.

[0054] Generally, user Internet Protocol (IP) packets are transmitted through Serving Gateway 166, which is itself connected to PDN Gateway 172. PDN Gateway 172 provides UE IP address allocation and other functions. PDN Gateway 172 and BM-SC 170 are connected to IP services 176, which may include, for example, the Internet, intranets, IP Multimedia Subsystem (IMS), packet-switched (PS) streaming services, and / or other IP services.

[0055] The BM-SC 170 provides functions for MBMS user service dispatch and delivery. The BM-SC 170 can serve as an entry point for content provider MBMS transmissions, authorize and initiate MBMS bearer services within a Public Land Mobile Network (PLMN), and / or schedule MBMS transmissions. The MBMS Gateway 168 can distribute MBMS services to BS 102 within a Broadcast-Specific Service Single Frequency Network (MBSFN) area, and / or be responsible for session management (start / stop) and collecting eMBMS-related billing information.

[0056] 5GC 190 may include various functional components, including: Access and Mobility Management Function (AMF) 192, other AMFs 193, Session Management Function (SMF) 194, and User Plane Function (UPF) 195. AMF 192 may communicate with Unified Data Management (UDM) 196.

[0057] AMF 192 is the control node that handles signaling between UE 104 and 5GC 190. AMF 192 provides services such as Quality of Service (QoS) flow and session management.

[0058] Internet Protocol (IP) packets are transmitted via UPF 195, which connects to IP service 197 and provides UE IP address allocation and other functions for 5GC 190. IP service 197 may include, for example, the Internet, intranet, IMS, PS streaming service, and / or other IP services.

[0059] In various aspects, to give a few examples, network entities or network nodes can be implemented as aggregated base stations, decomposed base stations, components of base stations, integrated access and backhaul (IAB) nodes, relay nodes, and sidelink nodes.

[0060] Figure 2An example decomposed base station 200 architecture is depicted. The decomposed base station 200 architecture may include one or more central units (CUs) 210, which may communicate directly with the core network 220 via a backhaul link, or indirectly with the core network 220 through one or more decomposed base station units, such as a near real-time (near-RT) RAN Intelligent Controller (RIC) 225 via an E2 link, or a non-real-time (non-RT) RIC 215 associated with a Service Management and Orchestration (SMO) framework 205, or both. CUs 210 may communicate with one or more distributed units (DUs) 230 via corresponding midhaul links (such as F1 interfaces). DUs 230 may communicate with one or more radio units (RUs) 240 via corresponding fronthaul links. RUs 240 may communicate with a corresponding UE 104 via one or more radio frequency (RF) access links. In some specific implementations, UE 104 may be served simultaneously by multiple RUs 240.

[0061] Each unit in a cell (e.g., CU 210, DU 230, RU 240, and near-RT RIC 225, non-RT RIC 215, and SMO frame 205) may include or be coupled to one or more interfaces configured to receive or transmit signals, data, or information (collectively, signals) via a wired or wireless transmission medium. Each of the cells, or an associated processor or controller that provides instructions to the cell's communication interface, may be configured to communicate with one or more other cells via the transmission medium. For example, these cells may include a wired interface configured to receive signals or transmit signals to one or more other cells via a wired transmission medium. Additionally or alternatively, a cell may include a wireless interface that may include a receiver, transmitter, or transceiver (such as a radio frequency (RF) transceiver) configured to receive signals on a wireless transmission medium or transmit signals to one or more other cells, or both.

[0062] In some aspects, CU 210 can host one or more higher-level control functions. Such control functions may include Radio Resource Control (RRC), Packet Data Convergence Protocol (PDCP), Serving Data Adaptation Protocol (SDAP), etc. Each control function can be implemented using an interface configured to signal to other control functions hosted by CU 210. CU 210 can be configured to handle user plane functions (e.g., Central Unit-User Plane (CU-UP)), control plane functions (e.g., Central Unit-Control Plane (CU-CP)), or combinations thereof. In some implementations, CU 210 can be logically divided into one or more CU-UP units and one or more CU-CP units. When implemented in an O-RAN configuration, CU-UP units can communicate bidirectionally with CU-CP units via an interface such as an E1 interface. CU 210 can be implemented to communicate with DU 230 for network control and signaling purposes, as needed.

[0063] DU 230 may correspond to a logical unit that includes one or more base station functions for controlling the operation of one or more RU 240s. In some aspects, DU 230 may at least partially host one or more of the Radio Link Control (RLC) layer, Medium Access Control (MAC) layer, and one or more high physical (PHY) layers (such as modules for forward error correction (FEC) encoding and decoding, scrambling, modulation and demodulation, etc.) according to functional splits (such as those defined by the 3rd Generation Partnership Project (3GPP). In some aspects, DU 230 may further host one or more low PHY layers. Each layer (or module) may be implemented using an interface configured to communicate signals with other layers (and modules) hosted by DU 230 or with control functions hosted by CU 210.

[0064] Lower-layer functionality can be implemented by one or more RU 240s. In some deployments, the RU240 controlled by the DU 230 may correspond to a logical node that at least partially hosts RF processing functions or low-PHY layer functions (such as performing Fast Fourier Transform (FFT), Inverse FFT (iFFT), digital beamforming, Physical Random Access Channel (PRACH) extraction and filtering, etc.) based on functional decomposition such as lower-layer functional decomposition, or both. In such architectures, the RU 240 may be implemented to handle over-the-air (OTA) communications with one or more UE 104s. In some specific implementations, the real-time and non-real-time aspects of control plane and user plane communications with the RU 240 may be controlled by the corresponding DU 230. In some scenarios, this configuration allows the DU 230 and CU 210 to be implemented in cloud-based RAN architectures such as vRAN architectures.

[0065] SMO framework 205 can be configured to support RAN deployment and provisioning of both non-virtualized and virtualized network elements. For non-virtualized network elements, SMO framework 205 can be configured to support the deployment of dedicated physical resources for RAN coverage requirements, which can be managed via operation and maintenance interfaces such as the O1 interface. For virtualized network elements, SMO framework 205 can be configured to interact with cloud computing platforms such as Open Cloud (O-Cloud) 290 to perform network element lifecycle management (such as instantiating virtualized network elements) via cloud computing platform interfaces such as the O2 interface. Such virtualized network elements may include, but are not limited to, CU 210, DU 230, RU 240, and near-RT RIC 225. In some specific implementations, SMO framework 205 may communicate with hardware aspects of the 4G RAN, such as Open eNB (O-eNB) 211, via the O1 interface. Additionally, in some implementations, the SMO framework 205 may communicate directly with one or more DU 230s and / or one or more RU 240s via the O1 interface. The SMO framework 205 may also include a non-RT RIC 215 configured to support the functionality of the SMO framework 205.

[0066] The non-RT RIC 215 can be configured to include logical functions that enable non-real-time control and optimization of RAN elements and resources, including artificial intelligence / machine learning (AI / ML) workflows for model training and updates, or policy-based guidance for applications / features in the near-RT RIC 225. The non-RT RIC 215 can be coupled to or communicate with the near-RT RIC 225, such as via an A1 interface. The near-RT RIC 225 can be configured to include logical functions that enable near real-time control and optimization of RAN elements and resources via an interface, such as via an E2 interface, through data collection and actions, connecting one or more CU 210s, one or more DU 230s, or both, and O-eNBs to the near-RT RIC 225.

[0067] In some implementations, to generate AI / ML models to be deployed in the near-RT RIC 225, the non-RT RIC 215 may receive parameters or external enrichment information from an external server. This information can be utilized by the near-RT RIC 225 and may be received from non-network data sources or network functions at the SMO framework 205 or the non-RT RIC 215. In some examples, the non-RT RIC 215 or the near-RT RIC 225 may be configured to tune RAN behavior or performance. For example, the non-RT RIC 215 may monitor long-term trends and patterns in performance and employ AI / ML models to perform corrective actions via the SMO framework 205 (such as reconfiguration via O1) or by creating RAN management policies (such as A1 policies).

[0068] Figure 3 Various aspects of examples BS 102 and UE 104 are described.

[0069] Generally, BS 102 includes various processors (e.g., 318, 320, 330, 338, and 340), antennas 334a to 334t (collectively referred to as 334), transceivers 332a to 332t (collectively referred to as 332) including modulators and demodulators, and other aspects that enable the wireless transmission of data (e.g., data source 312) and the wireless reception of data (e.g., data sink 314). For example, BS 102 can transmit and receive data between BS 102 and UE 104. BS 102 includes a controller / processor 340 that can be configured to implement the various wireless communication-related functions described herein.

[0070] Generally, UE 104 includes various processors (e.g., 358, 364, 366, 370, and 380), antennas 352a to 352r (collectively referred to as 352), transceivers 354a to 354r (collectively referred to as 354) including modulators and demodulators, and other aspects that enable the wireless transmission of data (e.g., retrieval from data source 362) and the wireless reception of data (e.g., provision to data sink 360). UE 104 includes a controller / processor 380 that can be configured to implement the various wireless communication-related functions described herein.

[0071] Regarding example downlink transmission, BS 102 includes a transmission processor 320 that can receive data from data source 312 and control information from controller / processor 340. This control information may be for a Physical Broadcast Channel (PBCH), Physical Control Format Indicator Channel (PCFICH), Physical Hybrid Automatic Repeat Request (HARQ) Indicator Channel (PHICH), Physical Downlink Control Channel (PDCCH), Group Common PDCCH (GC PDCCH), and / or others. In some examples, this data may be for a Physical Downlink Shared Channel (PDSCH).

[0072] The transmitter processor 320 can process data and control information (e.g., encoding and symbol mapping) to obtain data symbols and control symbols, respectively. The transmitter processor 320 can also generate reference symbols (such as those for the primary synchronization signal (PSS), secondary synchronization signal (SSS), PBCH demodulation reference signal (DMRS), and channel state information reference signal (CSI-RS)).

[0073] The transmit (TX) multiple-input multiple-output (MIMO) processor 330 can perform spatial processing (e.g., pre-decoding) on ​​data symbols, control symbols, and / or reference symbols where applicable, and can provide the output symbol stream to the modulators (MODs) in transceivers 332a to 332t. Each modulator in transceivers 332a to 332t can process the corresponding output symbol stream to obtain an output sample stream. Each modulator can further process (e.g., convert to analog, amplify, filter, and up-convert) the output sample stream to obtain a downlink signal. The downlink signal from the modulators in transceivers 332a to 332t can be transmitted via antennas 334a to 334t, respectively.

[0074] To receive downlink transmissions, UE 104 includes antennas 352a to 352r that receive downlink signals from BS 102 and provide the received signals to demodulators (DEMODs) in transceivers 354a to 354r, respectively. Each demodulator in transceivers 354a to 354r can adjust (e.g., filter, amplify, down-convert, and digitize) the corresponding received signal to obtain an input sample. Each demodulator can further process the input sample to obtain the received symbols.

[0075] The RX MIMO detector 356 acquires received symbols from all demodulators in transceivers 354a to 354r, performs MIMO detection on the received symbols where applicable, and provides the detected symbols. The receive processor 358 processes the detected symbols (e.g., demodulation, deinterleaving, and decoding), provides the decoded data of UE 104 to data sink 360, and provides the decoded control information to controller / processor 380.

[0076] Regarding the example uplink transmission, UE 104 further includes a transmission processor 364 that receives and processes data from data source 362 (e.g., for PUSCH) and control information from controller / processor 380 (e.g., for Physical Uplink Control Channel (PUCCH)). Transmission processor 364 can also generate reference symbols for reference signals (e.g., for Sounding Reference Signal (SRS)). Symbols from transmission processor 364 may be pre-decoded by TX MIMO processor 366, where applicable, further processed by modulators in transceivers 354a to 354r (e.g., for SC-FDM), and transmitted to BS 102.

[0077] At BS 102, uplink signals from UE 104 can be received by antennas 334a to 334t, processed by demodulators in transceivers 332a to 332t, detected where applicable by RX MIMO detector 336, and further processed by receiver processor 338 to obtain decoded data and control information transmitted by UE 104. Receiver processor 338 can provide the decoded data to data sink 314 and the decoded control information to controller / processor 340.

[0078] Memory 342 and memory 382 can store data and program code for BS 102 and UE 104, respectively.

[0079] Scheduler 344 can schedule UE to send data on the downlink and / or uplink.

[0080] In various respects, BS 102 can be described as transmitting and receiving various types of data associated with the methods described herein. In these contexts, “transmitting” can refer to various mechanisms that output data, such as from data source 312, scheduler 344, memory 342, transmit processor 320, controller / processor 340, TX MIMO processor 330, transceivers 332a to 332t, antennas 334a to 334t, and / or other aspects described herein. Similarly, “receiving” can refer to various mechanisms that acquire data, such as from antennas 334a to 334t, transceivers 332a to 332t, RX MIMO detector 336, controller / processor 340, receive processor 338, scheduler 344, memory 342, and / or other aspects described herein.

[0081] In various respects, UE 104 can also be described as transmitting and receiving various types of data associated with the methods described herein. In these contexts, “transmitting” can refer to various mechanisms that output data, such as from data source 362, memory 382, ​​transmit processor 364, controller / processor 380, TX MIMO processor 366, transceivers 354a to 354t, antennas 352a to 352t, and / or other aspects described herein. Similarly, “receiving” can refer to various mechanisms that acquire data, such as from antennas 352a to 352t, transceivers 354a to 354t, RX MIMO detector 356, controller / processor 380, receive processor 358, memory 382, ​​and / or other aspects described herein.

[0082] In some respects, the processor can be configured to perform various operations (such as those associated with the methods described herein) and to send (output) data to or receive data from another interface configured to send or receive data, respectively.

[0083] In various aspects, artificial intelligence (AI) processors 318 and 370 may perform AI processing for BS 102 and / or UE 104, respectively. AI processor 318 may include AI accelerator hardware or circuitry, such as one or more neural processing units (NPUs), one or more neural network processors, one or more tensor processors, one or more deep learning processors, etc. AI processor 370 may similarly include AI accelerator hardware or circuitry. As an example, AI processor 370 may perform AI-based beam management, AI-based channel state feedback (CSF), AI-based antenna tuning, and / or AI-based positioning (e.g., Global Navigation Satellite System (GNSS) positioning). In some cases, AI processor 318 may use hardware-accelerated AI inference and / or AI training to process feedback (e.g., CSF) from UE 104. AI processor 318 may, for example, use hardware-accelerated AI inference associated with the CSF to decode compressed CSF from UE 104. In some cases, AI processor 318 may perform certain RAN-based functions, including, for example, network planning, network performance management, energy-efficient network operation, etc.

[0084] Figure 4A , Figure 4B , Figure 4C and Figure 4D Describes the use of wireless communication networks (such as Figure 1 All aspects of the data structure of the wireless communication network 100.

[0085] Specifically, Figure 4A Figure 400 is an example of the first subframe within a 5G (e.g., 5G NR) frame structure. Figure 4B Figure 430 illustrates an example of a DL channel within a 5G subframe. Figure 4C Figure 450 illustrates an example of the second subframe within a 5G frame structure, and Figure 4D Figure 480 illustrates an example of a UL channel within a 5G subframe.

[0086] Wireless communication systems can utilize Orthogonal Frequency Division Multiplexing (OFDM) with a cyclic prefix (CP) on both the uplink and downlink. Such systems can also support half-duplex operation using Time Division Duplex (TDD). OFDM and Single-Carrier Frequency Division Multiplexing (SC-FDM) will (e.g., as...) Figure 4B and Figure 4D The system bandwidth (as depicted in the text) is divided into multiple orthogonal subcarriers. Each subcarrier can be modulated with data. Modulation symbols can be transmitted in the frequency domain using OFDM and / or in the time domain using SC-FDM.

[0087] Wireless communication frame structures can be frequency division duplex (FDD), where for a specific set of subcarriers, subframes within that set are dedicated to either deep (DL) or ultra-low (UL). Wireless communication frame structures can also be time division duplex (TDD), where for a specific set of subcarriers, subframes within that set are dedicated to both DL and UL.

[0088] exist Figure 4A and Figure 4C In this example, the wireless communication frame structure is TDD, where D stands for DL, U for UL, and X is flexibly used between DL and UL. The UE can configure the time slot format via the received Slot Format Indicator (SFI) (dynamically configured via DL Control Information (DCI) or semi-statically / statically configured via Radio Resource Control (RRC) signaling). In the depicted example, a 10ms frame is divided into 10 equal-sized 1ms subframes. Each subframe may include one or more time slots. In some examples, each time slot may include 12 or 14 symbols, depending on the Cyclic Prefix (CP) type (e.g., 12 symbols per time slot for extended CP, or 14 symbols per time slot for regular CP). Subframes may also include micro-slots, which typically have fewer symbols than the entire time slot. Other wireless communication technologies may have different frame structures and / or different channels.

[0089] In some respects, the number of time slots within a subframe (e.g., the time slot duration within a subframe) is based on a parameter set that defines the frequency-domain subcarrier spacing and symbol duration, as further described herein. In some respects, given a parameter set μ, each subframe has 2 μ The number of time slots is 1. Therefore, parameter sets (µ) 0 through 6 allow for 1, 2, 4, 8, 16, 32, and 64 time slots per subframe, respectively. In some cases, extended CP (e.g., 12 symbols per time slot) can be used with specific parameter sets; for example, parameter set 2 allows for 4 time slots per subframe. Subcarrier spacing and symbol length / duration are functions of the parameter set. The subcarrier spacing can be equal to... kHz, where μ is the parameter set from 0 to 6. As an example, the parameter set... Corresponding to a subcarrier spacing of 15 kHz, and the parameter set This corresponds to a subcarrier spacing of 960 kHz. Symbol length / duration is negatively correlated with subcarrier spacing. Figure 4A , Figure 4B , Figure 4C and Figure 4D It provides a slot format with 14 symbols per slot (e.g., regular CP) and a parameter set with 4 slots per subframe. Example. In this case, the slot duration is 0.25 ms, the subcarrier spacing is 60 kHz, and the symbol duration is approximately 16.67 μs.

[0090] like Figure 4A , Figure 4B , Figure 4C and Figure 4D As depicted, the resource grid can be used to represent the frame structure. Each time slot includes a resource block (RB) (also known as a physical RB (PRB)) extending for, for example, 12 consecutive subcarriers. The resource grid is divided into multiple resource elements (REs). The number of bits carried by each RE depends on the modulation scheme, including, for example, quadrature phase shift keying (QPSK) or quadrature amplitude modulation (QAM).

[0091] like Figure 4A As illustrated, some REs in the RE carry information for the UE (e.g., Figure 1 and Figure 3 The reference (pilot) signal (RS) for the UE (104) may include a demodulation RS (DMRS) and / or a channel state information reference signal (CSI-RS) for channel estimation at the UE. The RS may also include a beam measurement RS (BRS), a beam refinement RS (BRRS), and / or a phase tracking RS (PT-RS).

[0092] Figure 4B Examples of various DL channels within a subframe of a frame are illustrated. The Physical Downlink Control Channel (PDCCH) carries the DCI within one or more Control Channel Elements (CCEs), each CCE comprising, for example, nine RE groups (REGs), each REG comprising, for example, four consecutive REs in an OFDM symbol.

[0093] The Primary Synchronization Signal (PSS) can be located within symbol 2 of a specific subframe of the frame. The PSS is generated by the UE (e.g., Figure 1 and Figure 3 104) is used to determine subframe / symbol timing and physical layer identifier.

[0094] The secondary synchronization signal (SSS) can be located in symbol 4 of a specific subframe of the frame. The SSS is used by the UE to determine the physical layer cell identifier group number and radio frame timing.

[0095] Based on the Physical Layer Identifier and Physical Layer Cell Identifier Group Number, the UE can determine the Physical Cell Identifier (PCI). Based on the PCI, the UE can determine the location of the aforementioned DMRS. The Physical Broadcast Channel (PBCH), carrying the Master Information Block (MIB), can logically group with the PSS and SSS to form a Synchronization Signal (SS) / PBCH Block (SSB). The MIB provides the System Frame Number (SFN) and the number of Restricted Frames (RBs) in the system bandwidth. The Physical Downlink Shared Channel (PDSCH) carries user data, broadcast system information (such as System Information Blocks (SIBs)) not transmitted via the PBCH, and / or paging messages.

[0096] like Figure 4C As illustrated, some REs in the REs carry DMRS for channel estimation at the base station (indicated as R for a particular configuration, but other DMRS configurations are possible). The UE can transmit DMRS for PUCCH and DMRS for PUSCH. PUSCH DMRS can be transmitted, for example, in the first or second symbol before the PUSCH. PUCCH DMRS can be transmitted in different configurations depending on whether a short or long PUCCH is being transmitted and depending on the specific PUCCH format used. UE104 can transmit a Sounding Reference Signal (SRS). SRS can be transmitted, for example, in the last symbol of a subframe. SRS can have a comb structure, and the UE can transmit SRS on one of the comb teeth. SRS can be used by the base station for channel quality estimation to enable frequency-dependent scheduling of the UL.

[0097] Figure 4D Examples of various UL channels within a subframe of a frame are illustrated. The PUCCH can be located as indicated in one configuration. The PUCCH carries uplink control information (UCI), such as scheduling requests, channel quality indicators (CQI), pre-decoding matrix indicators (PMI), rank indicators (RI), and HARQ ACK / NACK feedback. The PUSCH carries data and may additionally be used to carry buffer status reports (BSR), power clearance reports (PHR), and / or UCI.

[0098] Example Synchronization Signal Scan

[0099] In some wireless communication systems (e.g., 5G NR systems and / or future wireless communication systems), a UE can scan frequencies to find synchronization signals (e.g., SSBs) broadcast by network entities, thereby establishing a communication link with the network entities. In some cases, the UE can perform a full frequency scan (FFS) to search for synchronization signals. As an example, a device enabled for global deployment can perform scans on thousands of candidate frequencies to search for synchronization signals. When there is no explicit signaling for synchronization signal location (e.g., via system information), a synchronization grid can indicate the frequency location of the synchronization signal that the UE can use for system acquisition. The frequency location of the SSB can be defined as an SSB with a corresponding Global Synchronization Channel Number (GSCN). REF The global synchronization channel number defines the center frequency of the SSB, as further described herein. As an example synchronization grid, Table 1 below provides the SSBs defining certain frequency ranges. REF And the parameters of GSCN.

[0100] Table 1: GSCN parameters used for global frequency grid

[0101] To perform FFS, the UE may assume that the synchronization signal is transmitted at a specific periodicity (e.g., 20 ms) centered on certain GSCNs provided by the synchronization grid. For example, FFS may involve the UE sequentially monitoring SSBs on possible GSCNs, as provided in Table 1. Due to the large number of possible GSCNs (e.g., thousands of GSCNs) and the varying periodicity that can be implemented for SSB transmission (e.g., ranging from 5 ms to 160 ms), this scanning operation may take a relatively long time.

[0102] Figure 5 An example SSB 500 in the time-frequency domain is illustrated. In this example, the SSB 500 occupies 20 resource blocks 502 in the frequency domain and 4 symbols 504a to 504d (collectively referred to as symbol 504) in the time domain. The SSB 500 may have synchronization grids (such as those provided in Table 1 above) with GSCN and SS. REF The corresponding center frequency is 506. SSB 500 may include a primary synchronization signal (PSS) 508, a secondary synchronization signal (SSS) 510, and a physical broadcast channel (PBCH) 512. PSS 508 occupies a first portion (e.g., 127 subcarriers) of resource block 502 in the first symbol 504a; SSS 510 occupies a first portion (e.g., 127 subcarriers) of resource block 502 in the third symbol 504c; PBCH 512 occupies resource blocks 502 in the second and fourth symbols 504b and 504d; and PBCH 512 occupies a second portion (e.g., 8 resource blocks) of resource block 502 in the fourth symbol 504c. Therefore, space-time-frequency resources 514 may exist arranged in the first and third symbols 504a and 504c. It should be noted that the SSB500 is merely an exemplary structure for synchronization signaling, and other structures (e.g., different time-domain and / or frequency-domain arrangements for PSS, SSS, and / or PBCH) may be used as a supplement to or alternative to the structure depicted for the SSB 500. In some cases, synchronization signaling may be delivered via a discovery reference signal having one or more synchronization signals (such as PSS, SSS, and / or Level 3 SS (TSS)). In some cases, some synchronization signaling may not have a PBCH.

[0103] The UE can use PSS 508 and SSS 510 for time and frequency synchronization for wireless communication with network entities. As discussed herein, PBCH 512 may carry certain system information (e.g., MIB) that enables the UE to communicate with network entities. Note that in some cases, the term "SSB" may refer to the Synchronization Signal (SS) / Physical Broadcast Channel (PBCH) block.

[0104] Example of artificial intelligence for wireless communication

[0105] Some aspects described in this paper can be implemented, at least in part, using some form of artificial intelligence (AI), such as the process of using a machine learning (ML) model to infer or predict output data based on input data. Example ML models may include mathematical representations of one or more relationships between various objects to provide outputs representing one or more predictions or inferences. Once an ML model has been trained, it can be deployed to process data that is wholly or partially similar to or related to the training data, and to provide outputs representing one or more predictions or inferences based on the input data.

[0106] ML is typically characterized by a learning type that generates a specific type of learning model that performs a particular type of task. For example, different types of machine learning include supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

[0107] Supervised learning algorithms typically model the relationships and dependencies between input features (e.g., feature vectors) and one or more target outputs. Supervised learning uses labeled training data, which consists of data including one or more inputs and the desired output. Supervised learning can be used to train models to perform tasks such as classification (where the goal is to predict discrete values) or regression (where the goal is to predict continuous values). Some example supervised learning algorithms include nearest neighbor, Naive Bayes, decision trees, linear regression, support vector machines (SVM), and artificial neural networks (ANN).

[0108] Unsupervised learning algorithms process unlabeled input data and train models that take the input and transform it into output to solve real-world problems. Examples of unsupervised learning tasks are clustering (where the model's output might be cluster labels), dimensionality reduction (where the model's output is an output feature vector with fewer features than the input feature vector), and outlier detection (where the model's output is a value indicating how the input differs from typical examples in the dataset). An example unsupervised learning algorithm is k-means.

[0109] Semi-supervised learning algorithms process datasets containing both labeled and unlabeled examples, where the number of unlabeled examples is typically much greater than the number of labeled examples. However, the goal of semi-supervised learning is to achieve the goals of supervised learning. Typically, a semi-supervised model involves a model trained to generate pseudo-labels for unlabeled data, which are then combined with labeled data to train a second classifier that leverages a larger volume of overall training data to improve task performance.

[0110] Reinforcement learning algorithms use observations gathered by an agent from its interactions with the environment to take actions that maximize reward or minimize risk. Reinforcement learning is a continuous and iterative process in which the agent learns from its experience with the environment until it has explored, for example, the entire range of possible states. An example type of reinforcement learning algorithm is adversarial networks. Reinforcement learning can be particularly beneficial when used to improve or attempt to optimize the behavior of models deployed in dynamically changing environments, such as wireless communication networks.

[0111] ML models can be deployed in one or more devices (e.g., network entities such as base stations and / or user equipment) to support various wired and / or wireless communication aspects of a communication system. For example, ML models can be trained to identify patterns and relationships in data corresponding to networks, devices, air interfaces, etc. ML models can improve operations associated with one or more aspects, such as transceiver circuitry control, frequency synchronization, timing synchronization, channel state estimation, channel equalization, channel state feedback, modulation, demodulation, device location, transceiver tuning, beamforming, signal decoding / decoding, network routing, load balancing, and energy saving (to name just a few). AI-enhanced transceiver circuitry control may include, for example, filter tuning, transmit power control, gain control (including automatic gain control), phase control, power management, etc.

[0112] The aspects described herein can be used to describe technical solutions for performing certain tasks and various technical problems by applying specific types of ML models, such as ANNs. However, it should be understood that other types of AI models can be used as supplements or alternatives to ANNs. ML models can be examples of AI models, and any suitable AI model can be used as a supplement or alternative to any ML model described herein. Therefore, unless explicitly stated otherwise, the topic of ML models is not necessarily intended to be limited to ANN solutions or machine learning. Furthermore, it should be understood that, unless specifically stated otherwise, terms such as “AI model,” “ML model,” “AI / ML model,” or “trained ML model” are intended to be used interchangeably.

[0113] Figure 6 An example AI architecture 600 is illustrated and can be used for AI-enhanced wireless communication. As illustrated, architecture 600 includes multiple logical entities such as a model training host 602, a model inference host 604, a data source 606, and an agent 608. The AI ​​architecture can be used in any of the various use cases for wireless communication, such as those listed above.

[0114] The model inference host 604 in architecture 600 is configured to run an ML model based on inference data 612 provided by data source 606. The model inference host 604 may produce output 614 (e.g., prediction or inference, such as discrete or continuous values) based on the inference data 612 and then provide it as input to agent 608.

[0115] Agent 608 can be a component or entity of a wireless communication system, including, for example, a radio access network (RAN), a wireless local area network, a device-to-device (D2D) communication system, etc. As an example, Agent 608 can be user equipment (e.g., Figure 1 UE 104 in the base station (e.g., Figure 1 The BS 102 or any of its decomposed network entities, including centralized units (CUs), distributed units (DUs) and / or radio units (RUs), access points, radio stations, RAN intelligent controllers (RICs) in cloud-based RANs, and so on. Additionally, the type of agent 608 may also depend on the type of task performed by the model inference host 604, the type of inference data 612 provided to the model inference host 604, and / or the type of output 614 generated by the model inference host 604.

[0116] For example, if the output 614 from the model inference host 604 is associated with beam management, then the agent 608 may be or may include a UE, DU, or RU. As another example, if the output 614 from the model inference host 604 is associated with transmit and / or receive scheduling, then the agent 608 may be a CU or DU.

[0117] After agent 608 receives output 614 from model inference host 604, agent 608 may determine whether to take action based on the output. For example, if agent 608 is a DU or RU, and the output from model inference host 604 is associated with beam management, agent 608 may determine whether to change or modify the transmit and / or receive beams based on output 614. If agent 608 determines to take action based on output 614, agent 608 may instruct the action to at least one action subject 610. For example, if agent 608 determines to change or modify the transmit and / or receive beams used for communication between agent 608 and action subject 610 (e.g., UE), agent 608 may transmit a beam switching instruction to action subject 610 (e.g., UE). As another example, agent 608 may be a UE, and output 614 from model inference host 604 may be one or more predicted channel characteristics of one or more beams. For example, model inference host 604 may predict the channel characteristics of a beam set based on measurements of another beam set. Based on predicted channel characteristics, an agent 608, such as a UE, can transmit a request to an action subject 610, such as a BS, to switch to a different beam for communication. In some cases, the agent 608 and the action subject 610 are the same entity.

[0118] Data source 606 can be configured to collect data that can be used as training data 616 for training an ML model or as inference data 612 for feeding ML model inference operations. Specifically, data source 606 can collect data from any of various entities that may include action subject 610 (e.g., UE and / or BS) and provide the collected data to model training host 602 for ML model training. For example, after action subject 610 (e.g., UE) receives beam configuration from agent 608, action subject 610 can provide data source 606 with performance feedback associated with the beam configuration, where the performance feedback can be used by model training host 602 to monitor and / or evaluate ML model performance, such as whether the output 614 provided to agent 608 is accurate. In some examples, if the output 614 provided to agent 608 is inaccurate (or the accuracy is below an accuracy threshold), model training host 602 can determine, for example, to modify or retrain the ML model used by model inference host 604 via ML model deployment / update.

[0119] In some respects, the model training host 602 may be deployed at the same or different entity as the entity where the model inference host 604 is deployed, or deployed with the same or different entity as the entity where the model inference host 604 is deployed. For example, to offload model training processing that may affect the performance of the model inference host 604, the model training host 602 may be deployed at a model server, as further described herein. Furthermore, in some cases, training and / or inference may be distributed among devices in a decentralized or federated manner.

[0120] In some other respects, ML models are deployed at or on the UE for SSB pre-scanning. More specifically, the model infers the host (such as...) Figure 6 The model infers that the host (604) can be deployed at or on the UE to indicate the probability that the GSCN is the center frequency of the SSB.

[0121] Figure 7 An example AI architecture is illustrated for a first wireless device 702 communicating with a second wireless device 704. The first wireless device 702 can be as described in this article... Figure 1 and Figure 3 The UE 104 described herein. Similarly, the second wireless device can be as described herein. Figure 1 and Figure 3 The BS 102 described. Note that the AI ​​architecture of the first wireless device 702 can be applied to the second wireless device 704.

[0122] The first wireless device 702 may be or may include a chip, system-on-a-chip (SoC), system-in-package (SiP), chipset, package, or device, which includes one or more processors, processing blocks, or processing elements (collectively, “processor 710”) and one or more memory blocks or elements (collectively, “memory 720”).

[0123] As an example, in transmission mode, processor 710 can transform information (e.g., packets or data blocks) into modulated symbols. As digital baseband signals (e.g., digital in-phase (I) and / or quadrature (Q) baseband signals representing corresponding symbols), processor 710 can output the modulated symbols to transceiver 740. Processor 710 may be coupled to transceiver 740 for transmitting and / or receiving signals via one or more antennas 746. In this example, transceiver 740 includes radio frequency (RF) circuitry 742, which may be coupled to antenna 746 via interface 744. As an example, interface 744 may include a switch, duplexer, double-ended converter, multiplexer, etc. RF circuitry 742 may, for example, use a digital-to-analog converter to convert digital signals into analog baseband signals. RF circuitry 742 may include any circuitry of various types, including, for example, baseband filters, mixers, frequency synthesizers, power amplifiers, and / or low-noise amplifiers. In some cases, RF circuitry 742 may up-convert baseband signals to one or more carrier frequencies for transmission. Antenna 746 can transmit RF signals, which can be received at the second wireless device 704.

[0124] In receive mode, RF signals received via antenna 746 (e.g., from a second wireless device 704) can be amplified and converted to baseband frequencies (e.g., down-conversion). The received baseband signals can be filtered and converted to digital I or Q signals for digital signal processing. Processor 710 can receive the digital I or Q signals and further process them, for example, by demodulating them.

[0125] One or more ML models 730 may be stored in memory 720 and accessible by processor 710. In some cases, different ML models 730 with different characteristics may be stored in memory 720, and a particular ML model 730 may be selected based on its characteristics and / or application and the characteristics and / or conditions of the first wireless device 702 (e.g., power state, mobility state, battery reserve, temperature, etc.). For example, ML models 730 may have different inference data and output pairings (e.g., different types of inference data produce different types of outputs), and predictions (e.g., Figure 6 The different accuracy levels (e.g., 80%, 90%, or 95% accuracy) associated with the output (614), the different time delays associated with generating the prediction (e.g., processing time less than 10ms, 100ms, or 1 second), the different ML model sizes (e.g., file size), the different coefficients or weights, etc.

[0126] Processor 710 can use ML model 730 based on input data (e.g., Figure 6 The inferred data 612) is used to generate output data (e.g., Figure 6 (of 614), for example, as in this article relative to Figure 6 The inference host 604 is described. The ML model 730 can be used to perform any AI augmentation task in a variety of AI augmentation tasks, such as those listed above.

[0127] As this article is about Figures 9 to 12 Further described, the ML model 730 can obtain input including a spectral energy image. The ML model 730 can provide an output indicating whether a GSCN corresponding to the center frequency of the SSB is detected in the spectral energy image. It should be noted that other input and / or output data may be used as a supplement to or alternative to the examples described herein.

[0128] In some respects, model server 750 may perform any ML model lifecycle management (LCM) task for first wireless device 702 and / or second wireless device 704. Model server 750 may operate as model training host 602 and use training data to update ML model 730. In some cases, model server 750 may operate as data source 606 to collect and host training data, inference data, and / or performance feedback associated with ML model 730. In some respects, model server 750 may host various types and / or versions of ML model 730 for download by first wireless device 702 and / or second wireless device 704.

[0129] In some cases, model server 750 can monitor and evaluate the performance of ML model 730 to trigger one or more LCM tasks. For example, model server 750 can determine whether to activate or deactivate the use of a specific ML model at first wireless device 702 and / or second wireless device 704, and model server 750 can provide such instructions to the respective first wireless device 702 and / or second wireless device 704. In some cases, model server 750 can determine whether to switch to a different ML model 750 used at first wireless device 702 and / or second wireless device 704, and model server 750 can provide such instructions to the respective first wireless device 702 and / or second wireless device 704. In yet other examples, model server 750 can also act as a central server for decentralized machine learning tasks such as federated learning.

[0130] Example Artificial Intelligence Model

[0131] Figure 8 This is an exemplary block diagram of an example Artificial Neural Network (ANN) 800.

[0132] ANN 800 may receive input data 806, which may include one or more bits of data 802, preprocessed data (optionally) output from preprocessor 804, or some combination thereof. Here, data 802 may include training data, validation data, application-related data, etc., for example, depending on the development and / or deployment phase of ANN 800. In some other embodiments, preprocessor 804 may be included within ANN 800. Preprocessor 804 may, for example, process all or part of data 802, which may result in some data in data 802 being changed, replaced, deleted, etc. In some embodiments, preprocessor 804 may add additional data to data 802.

[0133] ANN 800 includes at least one first layer 808 of artificial neurons 810 to process input data 806 and provide the resulting first layer output data to at least a portion of at least one second layer 814 via edge 812. The second layer 814 processes the data received via edge 812 and provides the second layer output data to at least a portion of at least one third layer 818 via edge 816. The third layer 818 processes the data received via edge 816 and provides the third layer output data to at least a portion of a final layer 822 comprising one or more neurons via edge 820 to provide output data 824. All or part of the output data 824 may be further processed in some way by (optionally) a post-processor 826. Thus, in some examples, ANN 800 may provide output data 828 based on output data 824, post-processed data output from post-processor 826, or some combination thereof. In some other embodiments, post-processor 826 may be included within ANN 800. Postprocessor 826 may, for example, process all or part of the output data 824, which may result in output data 828 being at least partially different from output data 824, for example, due to data being altered, replaced, deleted, etc. In some specific implementations, postprocessor 826 may be configured to add additional data to output data 824. In this example, the second layer 814 and the third layer 818 represent intermediate or hidden layers that may be arranged in a hierarchical or other similar structure. Although not explicitly shown, one or more additional intermediate layers may exist between the second layer 814 and the third layer 818.

[0134] The structure and training of the artificial neurons 810 in each layer can be customized according to the specific requirements of the application. Within a given layer of an ANN, some or all of the neurons can be configured to process the information provided to that layer and output corresponding transformed information from that layer. For example, the transformed information from a layer can represent a weighted sum of input information associated with a nonlinear activation function or another activation function used to “activate” the artificial neurons in the next layer, or otherwise based on that nonlinear activation function or the other activation function used to “activate” the artificial neurons in the next layer. Artificial neurons in such layers can be activated by or in response to weights and biases that can be adjusted during the training process. The weights of various artificial neurons can act as parameters controlling the connection strength between layers or between artificial neurons, while the biases can act as parameters controlling the connection direction between layers or between artificial neurons. Activation functions can select or determine whether an artificial neuron sends its output to the next layer in response to the data received by the artificial neuron. Different activation functions can be used to model different types of nonlinear relationships. By introducing nonlinearity into the ML model, activation functions allow the ML model to “learn” the input data (e.g., Figure 6 Complex patterns and relationships in the inferred data (612). Some non-exhaustive example activation functions include linear functions, binary step functions, sigmoid, hyperbolic tangent (tanh), rectified linear unit (ReLU) or its variants, exponential linear unit (ELU), Swish, Softmax, etc.

[0135] Design tools (such as computer applications, programs, etc.) can be used to select the appropriate structure and number of layers for the ANN 800, as well as the number of artificial neurons in each layer, and to select activation functions, loss functions, training procedures, etc. Once the initial model is designed, it can be trained using training data. Training data may include one or more datasets within which the ANN 800 can detect, determine, identify, or discover patterns. Training data can represent various types of information, including written, visual, audio, environmental context, operational attributes, etc. During training, the parameters of the artificial neurons 810 can be changed, such as minimizing or otherwise reducing the loss function or cost function. The training process can be repeated multiple times to fine-tune the ANN 800 with each iteration.

[0136] Various ANN model architectures are available for consideration. For example, in a feedforward ANN architecture, each artificial neuron 810 in a layer receives information from the previous layer and similarly generates information for the next layer. In a convolutional ANN architecture, some layers can be organized as filters that extract features from data (e.g., training data and / or input data). In a recursive ANN architecture, some layers may have connections that allow data to be processed across time, such as for processing information with temporal structure (e.g., time series data prediction).

[0137] In the autoencoder ANN architecture, compact representations of data can be processed, and models can be trained to make predictions or potentially reconstruct the original data from a reduced set of features. The autoencoder ANN architecture can be used for tasks related to dimensionality reduction and data compression.

[0138] Generative adversarial network (GAN) architectures can include generator ANNs and discriminator ANNs trained to compete against each other. GANs are ANN architectures that can be used for tasks related to generating synthetic data or improving the performance of other models.

[0139] Transformer ANN structures utilize an attention mechanism that enables the model to process input sequences in a parallel and efficient manner. The attention mechanism allows the model to focus on different parts of the input sequence at different times. This attention mechanism can be implemented using a series of layers called attention layers to compute, compute, determine, or select a weighted sum of input features based on the similarity between different elements of the input sequence. A transformer ANN structure may include a series of feedforward ANN layers that "learn" the non-linear relationship between the input and output sequences. The output of the transformer ANN structure can be obtained by applying a linear transformation to the output of the final attention layer. Transformer ANN structures are particularly useful for tasks involving sequence modeling or other similar processing.

[0140] Another example type of ANN structure is a model with one or more invertible layers. This type of model can be inverted or "unfolded" to reveal the input data used to generate the output of the layers.

[0141] Other examples of ANN model architectures include fully connected neural networks (FCNN) and long short-term memory (LSTM) networks.

[0142] ANN 800 or other ML models can be implemented in various types of processing circuits, as well as their memory and applicable instructions, for example, as described in this paper. Figure 6 and Figure 7 As described. For example, the model can be implemented using general-purpose hardware circuitry such as one or more central processing units (CPUs) and one or more graphics processing units (GPUs). One or more ML accelerators, such as tensor processing units (TPUs), embedded neural processing units (eNPUs) or other dedicated processors, and / or field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), etc., can also be used to develop ANN models. Various programming tools can be used to develop ANN models.

[0143] Various aspects of artificial intelligence model training

[0144] There are deployable ML models (such as...) Figure 8 The various model training techniques and processes used at some point before or after the ANN 800.

[0145] As part of the model development process, information may be collected or otherwise created in a suitable training data format for training the ML model accordingly. For example, training data may be collected or otherwise created relating to data on received / transmitted signal strength, interference, and resource usage, as well as any other relevant data that can be used to train the model to solve one or more problems or challenges in a communication system. In some instances, all or part of the training data may originate from one or more user equipment (UEs), one or more network entities, or one or more other devices in a wireless communication system. In some cases, all or part of the training data may be aggregated from multiple sources, such as one or more UEs, one or more network entities, the Internet, etc. For example, wireless network architectures (such as self-organizing networks (SONs) or mobile-driven test (MDT) networks) may be adapted to support data collection for ML model applications. In another example, training data may be generated or collected online, offline, or both online and offline by UEs, network entities, or other devices, and all or part of such training data may be transmitted or shared (real-time or near real-time), such as through store-and-forward functions. Offline training may refer to, for example, creating and using a static training dataset in a batch manner, while online training may refer to collecting and using training data in real-time or near real-time. For example, online or offline training can be used to train and / or fine-tune the ML model at a network device (e.g., a UE). For offline training, data collection and training can occur offline at the network side (e.g., at a base station or other network entity) or at the UE side. For online training, training of the ML model at the UE side can be performed locally at the UE or by a server device (e.g., a server hosted by the UE vendor) based on data provided from the UE to the server device, in a real-time or near real-time manner.

[0146] In some instances, all or part of the training data may be shared within the wireless communication system, or even shared (or obtained from outside) the wireless communication system.

[0147] Once the ML model has been trained on the training data, its performance can be evaluated. In some scenarios, evaluation / validation tests can be conducted using a validation dataset, which may include data not present in the training data, to compare the model's performance against a baseline or other benchmark information. If the model's performance is deemed unsatisfactory, fine-tuning the model may be beneficial, for example, by changing its architecture, retraining it on the data, or using different optimization techniques. Once the model's performance is deemed satisfactory, it can be deployed accordingly. In some instances, the model may be updated in some way, such as by changing or replacing all or part of the model, or by undergoing further training, to name just a few examples.

[0148] As for use in ANNs (such as Figure 8 As part of the training process of an ANN (Artificial Neural Network 800), parameters affecting the operation of artificial neurons and layers can be tuned. For example, backpropagation can be used to train an ANN by iteratively adjusting the weights and / or biases of certain artificial neurons associated with the error between the model's predicted output and the expected output, which may be known or otherwise considered acceptable. Backpropagation may include forward propagation, loss function, backward propagation, and parameter updates that can be performed during training iterations. This process can be repeated a certain number of times for each training dataset until the weights of the artificial neurons / layers are properly tuned.

[0149] Backpropagation, associated with the loss function, measures how well a model can predict the desired output for a given input. Optimization algorithms can be used during training to adjust weights and / or biases to reduce or minimize the loss function, which improves model performance. Various optimization algorithms exist that can be used with backpropagation or other training techniques. Some initial examples include gradient descent-based and stochastic gradient descent-based optimization algorithms. Stochastic gradient descent (or ascent) can be used to adjust weights / biases to minimize or otherwise reduce the loss function. Mini-batch gradient descent, a variant of gradient descent, involves updating weights / biases using mini-batch training data instead of the entire dataset. Momentum techniques can accelerate the optimization process by adding momentum terms to update or otherwise influence certain weights / biases.

[0150] Adaptive learning rate techniques adjust the learning rate of an optimization algorithm that is associated with one or more characteristics of the training data. Batch normalization techniques can be used to normalize the input to a model in order to stabilize the training process and potentially improve the model's performance.

[0151] The "drop-out" technique can be used to randomly discard some of the artificial neurons from the model during the training process, for example, to reduce overfitting and potentially improve the model's generalization.

[0152] The “early stopping” technique can be used to stop an ongoing training process early, such as when the performance of a model using a validation dataset begins to degrade.

[0153] Another example technique includes data augmentation, which generates additional training data by applying transformations to all or part of the training information.

[0154] Transfer learning techniques can be used, which involve using a pre-trained model as a starting point for training a new model. This can be useful when training data is limited or when there are multiple tasks that are related to each other.

[0155] Multi-task learning techniques can be used, which involve training a model to perform multiple tasks simultaneously to potentially improve the model's performance on one or more of these tasks. In some instances, hyperparameters can be input and applied during the training process.

[0156] Another example technique that can be useful for ML models is some form of "pruning." Pruning techniques, which can be performed during the training process or after the model has been trained, involve removing unnecessary (e.g., because they have no effect on the output), less necessary (e.g., because their effect on the output is negligible), or potentially redundant features from the model. In some instances, pruning techniques can reduce the complexity of the model or improve its efficiency without compromising its expected performance.

[0157] Pruning techniques can be particularly useful in the context of wireless communication, where available resources, such as power and bandwidth, may be limited. Some example pruning techniques include weight pruning, neuron pruning, layer pruning, structural pruning, and dynamic pruning. Pruning techniques can, for example, reduce the amount of data corresponding to a model that may need to be transmitted or stored.

[0158] Weight pruning techniques may involve removing some weights from a model. Neuron pruning techniques may involve removing some neurons from a model. Layer pruning techniques may involve removing some layers from a model. Structural pruning techniques may involve removing some connections between neurons in a model. Dynamic pruning techniques may involve adapting a pruning strategy for a model to one or more characteristics associated with the data or environment. For example, in some wireless communication devices, dynamic pruning techniques may be more aggressive in pruning models used in low-power or low-bandwidth environments and less aggressive in pruning models used in high-power or high-bandwidth environments. In some aspects, pruning techniques may also be applied to training data, such as to remove outliers. In some specific implementations, preprocessing techniques on all or part of the training dataset can improve model performance or facilitate faster model convergence. For example, training data may be preprocessed to alter or remove unnecessary, irrelevant, incorrect, or otherwise identifiable data. Such preprocessing of training data can, for example, lead to a reduction in potential overfitting or otherwise improve the performance of the trained model.

[0159] One or more of the example training techniques presented above can be used as part of the training process. As mentioned above, some example training processes that can be used to train ML models include supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning techniques.

[0160] Decentralized, distributed, or shared learning, such as federated learning, enables training on data distributed across multiple devices or organizations without the need for centralized data or training. Federated learning can be particularly useful in data-sensitive or privacy-constrained situations, or when centralized data is impractical, inefficient, or expensive. For example, in the context of wireless communications, federated learning can be used to improve performance by allowing ML models to be trained on data collected from a wide range of devices and environments. For instance, ML models can be trained on data collected from a large number of wireless devices in a network, such as distributed wireless communication nodes, smartphones, or Internet of Things (IoT) devices, to improve network performance and efficiency. With federated learning, a user equipment (UE) or other device can receive a full or partial copy of the model and perform local training on such a copy of the model using locally available training data. Such devices can provide updated information about the locally trained model (e.g., trainable parameter gradients) to one or more other devices, such as network entities or servers, where updates from other similar devices (such as other UEs) can be aggregated and used to provide updates to shared models, etc. The federated learning process can be iteratively repeated until all or part of the model achieves a satisfactory level of performance. Federated learning enables devices to protect the privacy and security of local data while supporting collaboration on training and updating shared models, in whole or in part.

[0161] In some implementations, one or more devices or services may support processes related to the use, maintenance, activation, or reporting of ML models. In some instances, all or part of a dataset or model may be shared across multiple devices, for example, to provide or otherwise enhance or improve processing. In some examples, signaling mechanisms may be used at various nodes in a wireless network to signal capabilities for performing specific functions related to the ML model, support for a particular ML model, capabilities for collecting, creating, and transmitting training data, or other ML-related capabilities. ML models in wireless communication systems may, for example, support decisions related to wireless resource allocation or selection, wireless channel condition estimation, interference mitigation, beam management, positioning accuracy, energy saving, or modulation or decoding schemes. In some implementations, model deployment may be conducted jointly or individually at various network levels, such as central units (CUs), distributed units (DUs), radio units (RUs), etc.

[0162] Various aspects related to AI-based synchronous signal scanning

[0163] This disclosure provides techniques for AI-based synchronous signal scanning and techniques for training AI models for such scanning.

[0164] Figure 9 Examples are given for use by UEs (such as...) Figure 1 and Figure 3 Example operation 900 of UE 104 performing SSB pre-scan. In this example, the UE (e.g., UE 104) may perform SSB pre-scan via one or more antennas (e.g., antennas Rx0 and Rx1). N The frequency bandwidth 902 is monitored over time windows 904a to 904n (collectively referred to as time windows 904) (e.g., period 1 to period N). The time windows 904 can be arranged continuously in time to form a continuous monitoring time window.

[0165] In some respects, the UE may generate one or more spectral energy images for each time window in time window 904. As an example, the UE may obtain samples of the signal received via the antenna (Rx0), as monitored in frequency bandwidth 902 within the first time window 904a. The UE may convert these samples into a first spectral energy image 906a, for example, as described herein. Figure 10Further described. The spectral energy image may be or may include a time-frequency grid representing the spectral energy (e.g., a spectrogram or matrix indicating the value of the spectral energy over time) and / or an indication of the spectral energy over time. The duration of the frequency bandwidth 902 and the time window 904a may be selected to suit at least one SSB (e.g., SSB 500) in the first spectral energy image 906a. In some cases, the UE may combine multiple spectral energy images 906a to 906n corresponding to samples obtained via multiple antennas (e.g., antennas Rx0 and Rx1) in a given time window (e.g., the first time window 904a). The multiple antennas (e.g., antennas Rx0 and Rx1) may be arranged at different locations across the UE to achieve spatial diversity for pre-scanning. In some cases, the multiple antennas may be tuned to different frequency bands to achieve frequency diversity for pre-scanning, for example, as described herein. Figure 10 Further description.

[0166] For the nth time window 904n, the UE may generate one or more spectral energy images 908a to 908n corresponding to samples obtained via antennas (e.g., antennas Rx0 and Rx1), such as those described herein with respect to the first time window 904a. The UE may combine the spectral energy images generated across time window 904 into a multi-period spectral energy image. For example, the UE may perform incoherent combination of the spectral energy images, which may improve the performance of the image compression techniques described herein (e.g., reduced latency, memory usage, etc.).

[0167] At position 910, the UE can perform AI-based SSB detection on the combined spectral energy image, for example, as described in this paper regarding... Figure 10 Further described. The UE can use an AI model to predict SSB candidates, which is trained to detect one or more features of SSBs in a combined spectral energy image. For example, the AI ​​model may have a kernel tuned to the aspect ratio of the SSB in a time-frequency grid. In some aspects, the kernel may have weights tuned to other suitable features for detecting SSBs. Such an AI model can achieve improved accuracy in detecting SSBs in a spectral energy image, and therefore, the improved accuracy of AI-based SSB detection can reduce the latency and / or power consumption of SSB scanning. As an example, the UE may provide input including a combined spectral energy image to one or more AI models (e.g., ML model 730). The AI ​​model may output an indication of whether an SSB is detected in the combined spectral energy image. In some aspects, the AI ​​model may, for example, be based on the features described herein. Figure 5 The described synchronization grid outputs an indication of GSCN candidate 912 corresponding to the center frequency of the SSB. The UE can monitor the SSB at the center frequency corresponding to the GSCN candidate 912, and the UE can perform cell acquisition via any detected SSB.

[0168] Figure 10 Examples are provided for performing actions as described in this article. Figure 9 The example operation 1000 of the described SSB pre-scan is as follows. In this example, the UE obtains digital samples 1002 of the signal generated from a monitored frequency bandwidth (e.g., frequency bandwidth 902) via one or more antennas (e.g., antennas Rx0 and Rx1). The digital samples can indicate the RF energy in the frequency bandwidth, as further described herein. As an example, the UE can use... Figure 7 The transceiver 740 is used to monitor the frequency bandwidth. The UE can receive various radio waves in the frequency bandwidth, such as noise, interference, ambient radio waves, wireless communication signals, and / or pilot signals (e.g., SSB 500). The UE can convert the analog signal generated using RF circuitry 742 into a digital signal. In some cases, the UE can perform digital preprocessing operations (e.g., digital filtering and / or amplification) on the digital signal to generate a digital sample 1002.

[0169] At 1004, the UE can, for example, press the sampling ratio (e.g., 1 / N t ,in N t =8) Buffered digital sample 1002. The UE can downsample digital sample 1002 by removing a portion of the sample per OFDM symbol (e.g., merging one-eighth of the samples corresponding to the OFDM symbol). At 1006, the UE can perform a Fast Fourier Transform (FFT) on the downsampled sample, for example, at a resolution of eight subcarrier spacings (SCS). The FFT generates a frequency domain representation of the sample. In some cases, at 1008, the UE can perform droop compensation on the frequency domain representation of the sample, for example, to flatten the frequency response.

[0170] At 1010, the UE may, for example, use an absolute value function of the frequency domain representation of the sample to determine the spectral energy information for each symbol. The UE may arrange the spectral energy information for each symbol across a time series to form a spectral energy image 1050 (e.g., a spectrum or matrix indicating the values ​​of spectral energy over time). The spectral energy image may be a representation of the spectral energy observed in a frequency bandwidth (e.g., frequency bandwidth 902) over a time window (e.g., time window 904a). In some cases, the spectral energy image may be or may include a matrix of spectral energy values ​​arranged in a time-frequency grid. It should be noted that the spectral energy image 1050 is depicted at a higher sampling resolution (time and frequency) than the sampling resolution discussed above. In addition, the spectral energy image 1050 depicts an example of the spectral energy over time associated with an SSB (e.g., SSB 500), and therefore, an example SSB is depicted in the spectral energy image 1050. The UE can perform operations at 1004 to 1010 for each of the digital samples 1002 obtained from all or some of the receiving antennas (e.g., antennas Rx0 and Rx1), which enables frequency and / or spatial diversity for SSB pre-scanning.

[0171] At 1012, the UE can, for example, use an averaging function (or mean or median) to combine the spectral energy images derived from multiple receiving antennas. For example, the UE can determine the average energy for each frequency-time location in the set of spectral energy images. For instance, for each time-frequency location, the UE can select energy values ​​in the spectral energy images corresponding to a specific frequency and time (e.g., the first symbol with a frequency of 3 GHz in the time window), and the UE can determine the average energy value of such energy values. It should be noted that each frequency-time location in the spectral energy image can effectively be a pixel of the spectral energy image, and therefore, the averaging function can be conceived as determining the average value at each pixel across the set of spectral energy images. The combined spectral energy image formed at 1012 can be referred to as a multi-antenna spectral energy image.

[0172] At 1014, the UE can combine the multi-antenna spectral energy image of the current time window (e.g., the nth time window 904n) with the spectral energy image 1052 generated for a previous time window (e.g., a time window that occurred before the nth time window 904n), which can be stored in memory 1016 (e.g., ...). Figure 7The UE can store the spectral energy image in memory 720 and access it via that memory. For example, the UE can perform an averaging function (or mean or median) for each frequency-time location between the spectral energy images (1050 and 1052) of the current time window and the previous time window. If no previous spectral energy image exists, the UE can store the multi-antenna spectral energy image of the current time window in memory 1016. In some cases, spectral energy image 1052 can be a multi-period spectral energy image representing the spectral energy across multiple time windows, for example, the multiple time windows are compressed into the duration of a single time window.

[0173] UE can target Nc The UE can perform operations from 1004 to 1014 within a time window (e.g., time window 904). In other words, the UE can target... Nc The multi-antenna spectral energy images generated in each time window are combined to form a multi-period spectral energy image, as mentioned above. Figure 9 As described above. In each iteration following the initial iteration, the UE can generate a multi-period spectral energy image representing the combined spectral energy associated with the current spectral energy image 1050 and the previous spectral energy image 1052, and store the multi-period spectral energy image in memory 1016, as discussed above regarding 1014. As an example, the total number of combined time windows can be 12 for two receiving antennas or 6 for four receiving antennas, where each time window has a duration of 20 ms. In some cases, the multi-period spectral energy image can represent spectral energy compressed to represent the spectral energy within 120 ms to 240 ms of a 20 ms time window. This spectral energy compression described above enables efficient processing of SSB pre-scanning, for example, using AI-based processing, as further described herein.

[0174] At any image combining operation described herein (e.g., at 1012 or 1014), the UE may perform incoherent combining, which ignores or disregards the phase associated with the combined images. Such incoherent combining allows the UE to perform image combining with improved performance (e.g., reduced processing latency, reduced processing usage, reduced memory usage, etc.).

[0175] At 1018, the UE can demultiplex (e.g., crop) the multi-period spectral energy image into sub-images, where each sub-image may span a portion of the bandwidth and / or time window of the source image. As an example, due to multi-antenna monitoring supporting frequency diversity (e.g., monitoring system bandwidth), the multi-period spectral image may have spectral energy information spanning bandwidths, such as one or more frequency ranges (e.g., including a range of GSCNs from 2 to 7498). Demultiplexing can segment the multi-period spectral energy image into sub-images, where each sub-image represents a sub-bandwidth of the source image (e.g., 80 MHz to 100 MHz). In some cases, demultiplexing can segment the multi-period spectral energy image into sub-images, where each sub-image represents a portion of the time window of the source image (e.g., one time slot). The time-frequency dimension of the sub-images can be selected to fit at least one SSB, for example, as illustrated in spectral energy image 1050. In some aspects, the frequency bandwidth of the sub-images may include multiple GSCNs. In some cases, the sub-images may overlap each other in the time and / or frequency dimensions. In some cases, the sub-images may not overlap each other in the time and / or frequency dimensions.

[0176] As a representative example of SSB detection using spectral energy images, at position 1020, the UE can normalize the spectral energy image obtained from the demultiplexing at position 1018. For example, the spectral energy image can be normalized based on its mean and standard deviation, as described in this paper regarding... Figure 11 Further described. At 1022, the UE can provide a normalized spectral energy image to the AI ​​model 1022 (e.g., ML model 730), which is trained to detect SSBs (e.g., SSB 500) and / or SSB timings, as described herein. Figure 11 and Figure 12Further described. In some aspects, the AI ​​model 1022 may be trained to detect one or more features of SSBs, such as the aspect ratio of the SSB in a time-frequency grid. For example, the AI ​​model 1022 may have a kernel tuned to the aspect ratio of the SSB. Such training and / or configuration of the AI ​​model 1022 can achieve improved accuracy in detecting SSBs in a spectral energy image, and thus reduce the latency and / or power consumption of SSB scanning. In some aspects, the AI ​​model 1022 may be or may include a convolutional neural network (CNN). As a more specific example, the AI ​​model 1022 may be or may include a two-dimensional (2D) CNN. The UE may obtain an output from the AI ​​model 1022, wherein the output may include the probability of detecting an SSB at a specific GSCN and the corresponding SSB timing (e.g., symbol position). At 1024, the UE may evaluate the output of the AI ​​model 1022 based on one or more thresholds 1026. For example, if the probability is greater than or equal to the threshold 1026 (e.g., ≥50%, 75%, or 95%), the UE can determine that a GSCN candidate corresponding to the SSB has been detected in the spectral energy image.

[0177] The UE can repeat the operations at points 1020 to 1024 for all or some of the spectral energy images obtained from the demultiplexing at point 1018. For example, the UE can process the spectral energy images for the GSCN in the bandwidth of all time slots and multi-cycle spectral energy images within a 20ms time window. In some aspects, the UE can perform the operations at points 1020 to 1024 for multiple spectral energy images via parallel processing to efficiently pre-scan the bandwidth for the SSB. The UE can process multiple spectral energy images concurrently via AI model 1022. Through AI-based pre-scanning, the UE can obtain GSCN candidates 1028 corresponding to the center frequency of the SSB and / or the SSB timing in which the SSB can be received.

[0178] For SSB Example convolutional neural network for detection

[0179] Figure 11 An example CNN 1100 trained to detect SSBs and / or their SSB timing is shown. A CNN 1100 can be as follows: Figure 10 Examples of the described AI model 1022 and / or as about Figure 8 An example of the described ANN 800. The CNN 1100 may include a feedforward neural network and / or a recurrent neural network. The CNN 1100 may receive input 1102, which may include a spectral energy image. In some cases, the CNN 1100 may include a preprocessor 1104, for example, as described herein. Figure 8 As described. In this example, preprocessor 1104 can process input 1102 according to the following expression ( x Normalize: (1) in The input is the mean of 1102, and Enter the standard deviation of 1102. Note that preprocessor 1104 can... Figure 10 Normalization is performed at position 1020. Therefore, in some cases, normalization can be integrated with AI models.

[0180] CNN 1100 processes the spectral energy image of input 1102 through a pipeline of layers. CNN 1100 may include multiple convolutional layers 1106a to 1106d, pooling layer sets 1108a to 1108d, and a fully connected layer 1110. As shown, at least one pooling layer is arranged between two convolutional layers. The first convolutional layer 1106a may receive input from the preprocessor 1104 and provide output to the first pooling layer 1108a. The first pooling layer 1108a processes the output of the first convolutional layer 1106a and provides output to the second convolutional layer 1106c. The second convolutional layer 1108b processes the output of the first pooling layer 1108a and provides output to the second pooling layer 1108b, and so on for subsequent convolutional and pooling layers arranged in CNN 1100.

[0181] The first convolutional layer 1106a may include one or more filters, for example, one to eight filters. Each of these filters may output a feature map associated with the input. As an example, eight filters output eight feature maps for a spectral energy image. In some respects, each of these filters (or some of these filters) may be configured or trained to detect specific features of the SSB, for example, via specific weights or coefficients of the respective filter. Features of the SSB (for its configuration and / or training of filters for detection) may include, for example, the PSS, SSS, TSS, empty regions (e.g., space-time-frequency resource 514), PBCH, its time-frequency arrangement / dimension, sequence signature of any SS, etc. The time-frequency arrangement and / or dimension may refer to the arrangement and / or dimension of the PSS, SSS, TSS and / or PBCH in the SSB or any other suitable synchronization signal specification, for example, as described herein. Figure 5 As described.

[0182] In some respects, each (or some) of these filters may have a kernel size tuned to the aspect ratio of the SSB in the time-frequency domain. The kernel may have a dimension that matches (or corresponds to) the time-frequency dimension of the SSB in the spectral energy image of input 1102. The aspect ratio (e.g., size or dimension) of the kernel may match the aspect ratio of the SSB in the spectral energy image. For example, the kernel size may be set to effectively form a bounding box around the SSB in the spectral energy image of input 1102. As an example, the kernel size may be four by five (e.g., 4×5). Note that the kernel size may depend on the time-frequency resolution of the spectral energy image of input 1102.

[0183] In some respects, the first convolutional layer 1106a may apply padding (e.g., the same padding) to the filter so that the output can have the same dimension as the input. The first convolutional layer 1106a may apply activation layers (e.g., activation functions) to prepare the output for the next convolutional layer. For example, the activation layer may be or may include a ReLU function.

[0184] The first pooling layer 1108a may perform a pooling operation (e.g., max pooling) on ​​the input data received from the first convolutional layer 1106a. The first pooling layer 1108a may downsample the input data received from the first convolutional layer 1106a. In some aspects, the first pooling layer 1108a may be or may include a max pooling layer. As an example, the first pooling layer 1108a may have a 2×2 size. It should be noted that, as a supplement to or alternative to max pooling, the first pooling layer 1108a may perform other types of pooling, such as average pooling, median pooling, etc.

[0185] In some respects, the first convolutional layer 1106a can be an example of the second convolutional layer 1106b, the third convolutional layer 1106c, and the fourth convolutional layer 1106d. In some cases, the second convolutional layer 1106b and the third convolutional layer 1106c may have the same convolutional filtering architecture as the first convolutional layer 1106a. For example, each of the second convolutional layer 1106b and the third convolutional layer 1106c may have multiple filters (e.g., eight filters) that have an aspect ratio tuned to the SSB and have a padding kernel size applied. In some cases, the fourth convolutional layer 1106d may be unpadded to reduce the dimensionality of the feature map extracted from the input.

[0186] In some respects, the first pooling layer 1108a can be an example of the second pooling layer 1108b, the third pooling layer 1108c, and the fourth pooling layer 1108d. In some cases, the second pooling layer 1108b and the third pooling layer 1108c can have the same architecture as the first pooling layer 1108a. For example, each of the second pooling layer 1108b and the third pooling layer 1108c can perform max pooling with a size of 2×2. In some cases, the fourth pooling layer 1108d can have a size (e.g., 1×16) of the input downsampled from the fourth convolutional layer 1106d into a feature array (e.g., 8×1).

[0187] Fully connected layer 1110 can apply weights to features extracted from previous layers to transform the features into output 1112 associated with SSB detection. For example, output 1112 may include the probability of detecting an SSB in the spectral energy image of input 1102, the corresponding center frequency GSCN of the SSB (e.g., a GSCN candidate), and / or the corresponding candidate SSB timing. In some aspects, fully connected layer 1110 may use a sigmoid activation function. SSB timing may indicate the predicted time when an SSB is expected to occur. As an example, SSB timing may be indicated based on the symbol index in a time slot and / or half-frame.

[0188] Training machine learning models for various aspects of SSB detection

[0189] Figure 12 An example operation 1200 for training an AI model to detect SSBs is illustrated. Operation 1200 can be generated by a model training host (e.g., Figure 6 The model training host (602) is used to perform the training. In some cases, the model training host may be or may include a UE (e.g., UE 104) and / or a network entity (e.g., BS 102). In some aspects, the model training host may be or may include a base station (e.g., BS 102), its decomposed entities (e.g., CU 210, DU 230 and / or RU 240), a core network entity (e.g., 5GC 190), and / or a cloud-based RAN network entity (e.g., ...). Figure 2 (Near-RT RIC 225, non-RTTRIC 215 and / or SMO frame 205).

[0190] The model training host obtains training data 1202, which includes training input data 1204 and corresponding labels 1206 for training input data 1204. Training input data 1204 may include spectral energy images, for example, as described in this paper... Figure 10As described, spectral energy images can be simulated (e.g., computer-generated) and / or collected from SSB scans. In some aspects, spectral energy images may include the distribution of spectral energy images with or without SSBs. In some cases, spectral energy images may include a portion of the SSBs, where a portion of the SSBs is outside the spectral energy image.

[0191] In some respects, the spectral energy image can include spectral energy images obtained from SSB scans under various channel conditions. For example, the spectral energy image can include spectral energy information measured at various frequency ranges (e.g., FR1 and FR2), signal quality (e.g., low to high signal-to-noise ratio (SNR)), signal strength (e.g., low to high reference signal received power (RSRP)), signal propagation effects (e.g., scattering, fading, Doppler effect, etc.), UE mobility status (e.g., low, medium, and high mobility), interference level, noise level, transmission range (e.g., proximity to the cell), line-of-sight conditions, non-line-of-sight conditions, etc. AI models can be trained to detect SSBs in a wide range of channel conditions (e.g., FR1 and FR2) and / or specific channel conditions (e.g., FR2).

[0192] Each tag in tag 1206 may be associated with at least one spectral energy image. As an example, each tag in tag 1206 may include an indication of whether any SSB is present in the corresponding spectral energy image. In some cases, for a tag indicating that an SSB is present in the corresponding spectral energy image, the tag may include an indication of the timing of the corresponding SSB (e.g., the time interval in which the SSB occurred).

[0193] The model training host provides training input data 1204 to the AI ​​model 1208. In some aspects, the AI ​​model 1208 may include... Figure 11 CNN 1100, and AI model 1208 can be the subject of this paper. Figures 6 to 10 An example of the AI ​​model described. AI model 1208 provides output 1210, which may include, as described in this paper, relative to... Figures 9 to 11 The described SSB detection information.

[0194] The model training host provides the output 1210 of the AI ​​model 1208 to a performance evaluator 1212, which evaluates the quality and / or accuracy of the output 1210. The performance evaluator 1212 can determine whether the output 1210 matches the corresponding label of the training input data 1204. For example, the performance evaluator 1212 can determine whether the prediction of detecting an SSB in the spectral energy image is correct based on the label associated with the spectral energy image. The performance evaluator 1212 can adjust the AI ​​model 1208 (e.g., any weights in the weights of the convolutional layers) to reduce the loss associated with the AI ​​model 1208. The model training host can continue to provide the training input data 1204 to the AI ​​model 1208 and adjust the AI ​​model 1208 until the loss of the AI ​​model 1208 meets a threshold and / or reaches a minimum loss. In some aspects, the loss may include sum of squared errors (SSE) loss, intersection-to-union (IoU) loss, cross-entropy loss, or a combination thereof.

[0195] In some respects, the model training host can train multiple AI models. These AI models can be trained with different performance characteristics and / or for different channel conditions. For example, AI models can be trained to detect SSBs with different accuracy levels (e.g., 70%, 80%, or 99% accuracy), different latency (e.g., processing time for predicting SSBs), and / or different throughput (e.g., the ability to predict SSBs from one or more spectral energy images). In some cases, AI models can be trained to detect SSBs under different channel conditions, as described above. Therefore, the UE can select an AI model capable of detecting SSBs based on certain specifications and / or conditions (such as the current channel condition, a specific power consumption level, a specific latency, and / or a specific accuracy). In some respects, such AI models enable the UE to perform SSB scanning with improved accuracy in detecting GSCN candidates, and thus, improved accuracy in detecting GSCN candidates can lead to reduced cell acquisition time and / or reduced power consumption.

[0196] Figure 13 An example of AI-based SSB scanning technology (e.g., Figure 10The figure presents example scan times for scanning operations 1000 and the FFS technique at various signal qualities. As shown, the first set of scan times 1302 is associated with the AI-based SSB scanning technique, and the second set of scan times 1304 is associated with the FFS technique. Compared to the FFS technique, the AI-based SSB scanning technique provides shorter SSB scan times across a wider range of signal qualities. In some cases, the AI-based SSB scanning technique can reduce scan time by more than 60%. Furthermore, the AI-based SSB scanning technique provides improved accuracy regarding the detection of SSBs in the frequency bandwidth using spectral energy correlation techniques. This improved accuracy can be attributed to the training of AI models (e.g., CNNs) and / or robust AI models configured to detect and extract various features of SSBs, such as those described in this paper. Figure 12 As described herein, the AI-based SSB scanning technique described in this paper enables reduced latency and / or power consumption for synchronization and cell acquisition.

[0197] Although this article describes the detection of SSB via AI models Figures 9 to 13 The examples depicted are for the purpose of facilitating understanding, but aspects of this disclosure can also be applied to other synchronization signaling schemes, such as those with one or more synchronization signals and, in some cases, without a discovery reference signal (DRS) for the PBCH.

[0198] Example Operations of AI-Based SSB Scanning and Model Training

[0199] Figure 14 A process flow 1400 is described for communication between network entity 1402 and user equipment (UE) 1404 in the system. In some aspects, network entity 1402 may be related to... Figure 1 and Figure 3 The BS 102 depicted and described or related to Figure 2 Examples of its decomposed entities depicted and described. Similarly, UE 1404 could be about Figure 1 and Figure 3 Examples of UE 104 depicted and described herein. However, in other respects, UE 1404 may be another type of wireless communication device, and network entity 1402 may be another type of network entity or network node, such as those network entities or network nodes described herein.

[0200] At point 1406, UE 1404 and / or network entity 1402 can train AI models, for example, as described in this article regarding Figure 12As described. In some cases, UE 1404 and / or network entity 1402 can use training data collected from UE 1404, which performs SSB scanning with or without AI, to perform model training. In some cases, UE 1404 and / or network entity 1402 can use data from data sources (such as...) Figure 6 The network entity 1402 obtains training data from the data source 606. In some respects, the network entity 1402 may transmit the trained AI model and / or information used to reproduce the AI ​​model to the UE 1404. It should be noted that the UE 1404 and / or the network entity 1402 may perform online model training and / or batch model training as described above. Therefore, the UE 1404 and / or the network entity 1402 may perform AI model training at various times.

[0201] At 1408, UE 1404 can monitor SSBs in one or more frequency bandwidths, for example, as described in this article. Figure 9 and Figure 10 As described herein. Network entity 1402 may transmit at certain frequencies and times, for example, as described herein. Figure 5 The described synchronization grid broadcasts SSBs. UE 1404 can generate a spectral energy image indicating the spectral energy observed in the frequency bandwidth. In some cases, UE 1404 can receive SSBs while monitoring radio waves in the frequency bandwidth.

[0202] At 1410, UE 1404 can perform AI-based SSB pre-scan operations, for example, as described in this article regarding Figures 9 to 11 As described. For example, UE 1404 may obtain GSCN candidates from an AI model (e.g., AI model 1022). GSCN candidates may correspond to the center frequencies of possible SSBs.

[0203] At 1412, UE 1404 can receive SSBs from network entity 1402. UE 1404 can monitor SSBs at GSCN candidates obtained at 1410, and some of these GSCN candidates may be the actual center frequency of the SSB. UE 1404 can use the received SSBs to perform time and / or frequency synchronization.

[0204] At 1414, UE 1404 can perform cell acquisition and establish a communication link with network entity 1402. For example, the received SSB may carry certain system information that enables UE 1404 to establish a communication link with network entity 1402.

[0205] Example Operation

[0206] Figure 15A method 1500 for training an AI model by a device (such as UE 104) is illustrated. In some cases, the device may be or may include a network entity, such as [the entity mentioned in the original text]. Figure 1 and Figure 3 The BS 102 depicted and described, or as per the description Figure 2 The decomposed entity depicted and described. In some cases, the apparatus may be or may include a model training host 602.

[0207] Method 1500 begins at box 1505, where training data associated with the AI ​​model is obtained, for example, as discussed in this paper. Figure 12 As described. In some respects, the training data includes multiple spectral energy images and multiple labels, wherein at least one of the multiple labels includes at least one indication of whether the corresponding spectral energy image contains at least a portion of an SSB.

[0208] Then, method 1500 proceeds to box 1510, where training data is used to train an AI model to provide one or more probabilities of detecting one or more SSBs within a frequency bandwidth spanning a set of time windows, which includes multiple GSCNs, such as those described in this paper. Figure 12 As described.

[0209] Then, method 1500 may proceed to block 1515, in which the trained AI model is transmitted to a wireless communication device. For example, as a network entity, the device may transmit the trained AI model (or information for reproducing the AI ​​model) to a UE (e.g., UE 104).

[0210] In some aspects, each of these tags is associated with at least one of a plurality of spectral energy images. In some aspects, the at least one spectral energy image includes a time-frequency grid representing the spectral energy (e.g., a spectrum or matrix indicating the value of the spectral energy over time). In some aspects, each of these tags includes an indication of whether the corresponding spectral energy image includes any SSB. In some aspects, for each tag indicating that the corresponding spectral energy image includes at least one SSB, the corresponding tag also includes an indication of the time interval in which the at least one SSB occurred. In some aspects, the plurality of spectral energy images includes a set of spectral energy images associated with different frequency bandwidths across the system bandwidth.

[0211] In some aspects, a plurality of spectral energy images includes one or more of the following: a first set of spectral energy images, the first set of spectral energy images including at least one SSB in one or more first frequency bandwidths; a second set of spectral energy images, the second set of spectral energy images not having any SSB in one or more second frequency bandwidths; or a third set of spectral energy images, the third set of spectral energy images including a first portion of an SSB in one or more third frequency bandwidths, wherein a second portion of the SSB is disposed outside the one or more third frequency bandwidths. In some aspects, each spectral energy image in the first set of spectral energy images includes at least one SSB disposed in different locations in the time-frequency grid of the respective spectral energy image.

[0212] In some aspects, box 1505 includes generating training data at least in part based on the system bandwidth for one or more SSB scans. In some aspects, the device can generate training data from SSB scans with and / or without AI.

[0213] In some respects, box 1510 includes reducing the loss associated with the AI ​​model, at least in part based on the training data. In some respects, this loss includes: sum of squared errors (SSE) loss; intersection-to-union (IoU) loss; cross-entropy loss; or a combination thereof.

[0214] In some aspects, method 1500 also includes monitoring at least a first frequency bandwidth across a set of time windows. In some aspects, method 1500 also includes using a trained AI model to identify at least one SSB centered on the GSCN in the first frequency bandwidth. In some aspects, method 1500 also includes communicating with network entities at least in part based on at least one SSB.

[0215] In some aspects, identifying at least one SSB includes: compressing a plurality of samples across a time-frequency grid into a spectral energy image based on a first frequency bandwidth and at least one time window, the plurality of samples representing a first frequency bandwidth across a set of time windows; providing an AI model with input including the spectral energy image; and obtaining from the AI ​​model an output including an indication that at least one SSB is in the first frequency bandwidth.

[0216] In some respects, AI models include CNNs with kernel sizes tuned to an SSB aspect ratio, for example, as discussed in this paper. Figure 11 As described. In some aspects, a CNN includes: a first convolutional layer, which is cascaded and coupled to a first max-pooling layer, which is padded and has a first plurality of filters; a second convolutional layer, which is coupled to a second max-pooling layer, which is unpadded and has a second plurality of filters; and a fully connected layer having sigmoid activation.

[0217] In some aspects, method 1500 also includes storing the trained AI model in one or more memories. For example, the device may store the trained AI model in... Figure 7 In the memory 720.

[0218] In some respects, method 1500 or any aspect thereof may be made by means of a device (such as...) Figure 17 The communication device 1700 is used to perform the method, which includes various components capable of operating, configured, or adapted to perform the method 1500. The communication device 1700 is described in further detail below.

[0219] It should be noted that Figure 15 This is merely one example of a method, and other methods that include fewer, additional, or alternative operations may also be consistent with this disclosure.

[0220] Figure 16 It shows a device (such as) Figure 1 and Figure 3 Method 1600 for wireless communication of UE 104.

[0221] Method 1600 begins at box 1605, wherein at least a first frequency bandwidth is monitored across a time window set, for example, as described in this paper regarding Figure 9 and Figure 10 As described.

[0222] Then, method 1600 proceeds to block 1610, where an AI model is used to identify at least one SSB centered on the GSCN in the first frequency bandwidth. This AI model includes the core size tuned to the SSB aspect ratio, for example, as described in this paper regarding... Figure 10 and Figure 11 As described.

[0223] Then, method 1600 proceeds to block 1615, wherein communication with the network entity is based at least in part on at least one SSB, for example, as described herein. Figure 14 As described.

[0224] In some respects, the AI ​​model is trained to detect one or more SSB features. In some respects, one or more SSB features include one or more of the following: Primary Synchronization Signal (PSS), Secondary Synchronization Signal (SSS), Third-Level Synchronization Signal (TSS), Physical Broadcast Channel (PBCH), or one or more space-time-frequency resources in the SSB (e.g., space-time-frequency resource 514). In some respects, one or more SSB features include the time-frequency arrangement of the PSS, SSS, TSS, PBCH, and / or space-time-frequency resources in the SSB or synchronization signal specification, as described herein. Figure 5As described herein. In some respects, one or more SSB features include the time-frequency dimension of the PSS, SSS, TSS, PBCH, and / or space-time-frequency resources in the SSB or synchronization signal specification, as described herein. Figure 5 As described.

[0225] In some respects, box 1610 includes: using an AI model to identify at least one GSCN candidate in a first frequency bandwidth; and searching for at least one SSB based at least in part on at least one GSCN candidate.

[0226] In some respects, method 1600 further includes: monitoring multiple frequency bandwidths across a set of time windows, the multiple frequency bandwidths including a first frequency bandwidth; using an AI model to identify at least one GSCN candidate corresponding to at least one potential SSB in the multiple frequency bandwidths; and searching for one or more SSBs, the one or more SSBs including at least one SSB, based at least in part on the at least one GSCN candidate.

[0227] In some aspects, box 1610 includes: compressing a plurality of samples across a time-frequency grid into a spectral energy image based on a first frequency bandwidth and at least one time window, the plurality of samples representing a first frequency bandwidth across a set of time windows; providing an input including the spectral energy image to an AI model; and obtaining an output from the AI ​​model including an indication that at least one SSB is within the first frequency bandwidth. In some aspects, compressing the plurality of samples into a spectral energy image includes: converting the plurality of samples into a plurality of spectral energy images, wherein each of the plurality of spectral energy images corresponds to a different time window in the set of time windows; and compressing the plurality of spectral energy images into a spectral energy image. In some aspects, compressing the plurality of spectral energy images includes incoherently combining the plurality of spectral energy images into a spectral energy image, for example, as described herein. Figure 10 As described.

[0228] In some respects, AI models include CNNs, for example, as discussed in this article. Figure 11 As described. In some aspects, a CNN includes multiple convolutional layers and a fully connected output layer with sigmoid activation. In some aspects, the multiple convolutional layers include: a first convolutional layer coupled to a first max-pooling layer, the first convolutional layer being padded and having a first plurality of filters; and a second convolutional layer coupled to a second max-pooling layer, the second convolutional layer being unpadded and having a second plurality of filters. In some aspects, the first plurality of filters includes at least two, four, six, or eight filters.

[0229] In some respects, method 1600 or any aspect thereof may be made by means of a device (such as...) Figure 17The communication device 1700 is used to perform the method 1600, which includes various components capable of operating, configured, or adapted to perform the method. The communication device 1700 is described in further detail below.

[0230] It should be noted that Figure 16 This is merely one example of a method, and other methods that include fewer, additional, or alternative operations may also be consistent with this disclosure.

[0231] Example communication device

[0232] Figure 17 Various aspects of the example communication device 1700 are described. In some aspects, the communication device 1700 is user equipment, such as those mentioned above. Figure 1 and Figure 3 The UE 104 is described. In some respects, the communication device 1700 is a network entity, such as regarding... Figure 1 and Figure 3 The BS 102 depicted and described, or as per the description Figure 2 The decomposed entities that are depicted and described.

[0233] Communication device 1700 includes a processing system 1702 coupled to transceiver 1738 (e.g., transmitter and / or receiver). Transceiver 1738 is configured to transmit and receive signals for communication device 1700 via antenna 1740, such as various signals as described herein. Processing system 1702 may be configured to perform processing functions of communication device 1700, including processing signals received by communication device 1700 and / or to be transmitted by communication device 1700.

[0234] Processing system 1702 includes one or more processors 1704. In various aspects, the one or more processors 1704 may represent one or more of a receive processor 358, a transmit processor 364, a TX MIMO processor 366, and / or a controller / processor 380, as per [reference to...]. Figure 3 As described. One or more processors 1704 are coupled to a computer-readable medium / memory 1720 via a bus 1736. In some aspects, the computer-readable medium / memory 1720 is configured to store instructions (e.g., computer-executable code) that, when executed by one or more processors 1704, enable one or more processors 1704 to execute and cause the one or more processors to perform actions related to... Figure 15 The described method 1500 or any aspect thereof, including relative to the method, is as follows: Figure 15 Any additional operations described, and / or regarding Figure 16 The described method 1600 or any aspect thereof, including relative to the method, is as follows: Figure 16Any additional operations described. Note that references to processors performing the functions of communication device 1700 may include one or more processors, such as performing the functions of communication device 1700 in a distributed manner.

[0235] In the depicted example, computer-readable medium / memory 1720 stores code 1722 for acquisition, code 1724 for training, code 1726 for storage, code 1728 for monitoring, code 1730 for identification, code 1732 for communication, code 1734 for transmission, or any combination thereof. Processing of codes 1722 to 1734 enables communication device 1700 to perform and cause the communication device to perform actions related to… Figure 15 The described method 1500 or any aspect related to the method, and / or about Figure 16 The described method 1600 or any aspect related to that method.

[0236] One or more processors 1704 include circuitry configured to implement (e.g., execute) code stored in computer-readable medium / memory 1720, including circuitry 1706 for acquisition, circuitry 1708 for training, circuitry 1710 for storage, circuitry 1712 for monitoring, circuitry 1714 for identification, circuitry 1717 for communication, circuitry 1718 for transmission, or any combination thereof. Processing using circuitry 1706 to 1718 enables communication device 1700 to perform and allow the communication device to perform actions related to... Figure 15 The described method 1500 or any aspect related to the method, and / or about Figure 16 The described method 1600 or any aspect related to that method.

[0237] More generally, components used for conveying, sending, transmitting, or outputting for transmission may include Figure 3 The UE104 illustrated includes a transceiver 354, an antenna 352, a transmit processor 364, a TX MIMO processor 366, and / or a controller / processor 380. Figure 17 The transceiver 1738 and / or antenna 1740 of the communication equipment 1700 in the middle Figure 17 One or more processors 1704 of the communication device 1700. Components for communicating, receiving, monitoring, or acquiring may include... Figure 3 The UE 104 illustrated includes a transceiver 354, an antenna 352, a receiver processor 358, and / or a controller / processor 380. Figure 17 The transceiver 1738 and / or antenna 1740 of the communication device 1700 and / or Figure 17The communication device 1700 includes one or more processors 1704. Components for training, storage, or monitoring may include... Figure 3 The UE 104 illustrated includes a transceiver 354, an antenna 352, and / or a controller / processor 380. Figure 17 The transceiver 1738 and / or antenna 1740 of the communication equipment 1700 in the middle Figure 17 One or more processors 1704 of the communication device 1700 in the middle.

[0238] Example Terms

[0239] Specific implementation examples are described in the following numbered clauses: Clause 1: A method for wireless communication by a device, the method comprising: monitoring at least a first frequency bandwidth across a time window set; using an artificial intelligence (AI) model to identify at least one synchronization signal block (SSB) in the first frequency bandwidth centered on a globally synchronized channel number (GSCN), the AI ​​model including a kernel size tuned to the aspect ratio of the SSB; and communicating with a network entity at least in part based on the at least one SSB.

[0240] Clause 2: The method described in Clause 1, wherein the AI ​​model is trained to detect one or more SSB features.

[0241] Clause 3: The method according to Clause 2, wherein the one or more SSB features include one or more of the following: primary synchronization signal, secondary synchronization signal, physical broadcast channel, or one or more space-time-frequency resources of SSB.

[0242] Clause 4: The method according to any one of Clauses 1 to 3, wherein identifying the at least one SSB comprises: using the AI ​​model to identify at least one GSCN candidate in the first frequency bandwidth; and searching for the at least one SSB based at least in part on the at least one GSCN candidate.

[0243] Clause 5: The method according to any one of Clauses 1 to 4, the method further comprising: monitoring a plurality of frequency bandwidths across the set of time windows, the plurality of frequency bandwidths including the first frequency bandwidth; using the AI ​​model to identify at least one GSCN candidate corresponding to at least one potential SSB in the plurality of frequency bandwidths; and searching for one or more SSBs, the one or more SSBs including the at least one SSB, based at least in part on the at least one GSCN candidate.

[0244] Clause 6: The method according to any one of Clauses 1 to 5, wherein identifying the at least one SSB comprises: compressing a plurality of samples across a time-frequency grid into a spectral energy image based on the first frequency bandwidth and at least one time window, the plurality of samples representing the first frequency bandwidth across the set of said time windows; providing the AI ​​model with an input including the spectral energy image; and obtaining from the AI ​​model an output including an indication that the at least one SSB is in the first frequency bandwidth.

[0245] Clause 7: The method according to Clause 6, wherein compressing the plurality of samples into the spectral energy image comprises: converting the plurality of samples into a plurality of spectral energy images, wherein each of the plurality of spectral energy images corresponds to a different time window in the set of time windows; and compressing the plurality of spectral energy images into the spectral energy image.

[0246] Clause 8: The method according to Clause 7, wherein compressing the plurality of spectral energy images includes incoherently combining the plurality of spectral energy images into the spectral energy image.

[0247] Clause 9: The method according to any one of Clauses 1 to 8, wherein the AI ​​model comprises a convolutional neural network (CNN).

[0248] Item 10: The method according to Item 9, wherein the CNN comprises multiple convolutional layers and a fully connected output layer with sigmoid activation.

[0249] Clause 11: The method according to Clause 10, wherein the plurality of convolutional layers comprises: a first convolutional layer coupled to a first max-pooling layer, the first convolutional layer being padded and having a first plurality of filters; and a second convolutional layer coupled to a second max-pooling layer, the second convolutional layer being unpadded and having a second plurality of filters.

[0250] Clause 12: The method according to Clause 11, wherein the first plurality of filters comprises at least eight filters.

[0251] Clause 13: A method comprising: obtaining training data comprising: a plurality of spectral energy images and a plurality of labels, wherein at least one of the plurality of labels comprises at least one indication of whether the corresponding spectral energy image includes at least a portion of a synchronization signal block (SSB); and using the training data to train an artificial intelligence (AI) model to provide one or more probabilities of detecting one or more SSBs within a frequency bandwidth spanning a set of time windows, the frequency bandwidth including a plurality of global synchronization channel numbers (GSCNs).

[0252] Clause 14: The method according to Clause 13, wherein each of the tags is associated with at least one of the plurality of spectral energy images and includes an indication of whether the corresponding spectral energy image includes at least a portion of a synchronization signal block (SSB).

[0253] Clause 15: The method described in accordance with Clause 13 or 14 further includes transmitting the trained AI model to a wireless communication device.

[0254] Clause 16: The method according to any one of Clauses 13 to 15, wherein for each of the tags indicating that the corresponding spectral energy image includes at least one SSB, the corresponding tag further includes an indication of the time interval in which the at least one SSB occurs.

[0255] Clause 17: The method according to any one of Clauses 13 to 14, wherein the plurality of spectral energy images comprises a set of spectral energy images associated with different frequency bandwidths across the system bandwidth.

[0256] Clause 18: The method according to any one of Clauses 13 to 15, wherein the plurality of spectral energy images comprises one or more of the following: a first set of spectral energy images, the first set of spectral energy images comprising at least one SSB in one or more first frequency bandwidths; a second set of spectral energy images, the second set of spectral energy images not having any SSB in one or more second frequency bandwidths; or a third set of spectral energy images, the third set of spectral energy images comprising a first portion of the SSB in one or more third frequency bandwidths, wherein a second portion of the SSB is disposed outside the one or more third frequency bandwidths.

[0257] Clause 19: The method according to Clause 18, wherein each spectral energy image in the first set of spectral energy images includes the at least one SSB arranged in different locations in the time-frequency grid of the respective spectral energy image.

[0258] Clause 20: The method according to any one of Clauses 13 to 19, wherein obtaining the training data includes generating the training data based at least in part on the bandwidth of any SSB scanning system.

[0259] Clause 21: The method according to any one of Clauses 13 to 20, wherein training the AI ​​model includes at least in part based on the training data to reduce the loss associated with the AI ​​model.

[0260] Clause 22: The method described in Clause 21, wherein the loss comprises: sum of squared errors (SSE) loss; intersection-to-union (IoU) loss; cross-entropy loss; or a combination thereof.

[0261] Clause 23: The method according to any one of Clauses 13 to 22, the method further comprising: monitoring at least a first frequency bandwidth across the set of said time windows; using a trained AI model to identify at least one SSB centered on the GSCN in the first frequency bandwidth; and communicating with network entities at least in part based on said at least one SSB.

[0262] Clause 24: The method according to Clause 23, wherein identifying the at least one SSB comprises: compressing a plurality of samples across a time-frequency grid into a spectral energy image based on the first frequency bandwidth and at least one time window, the plurality of samples representing the first frequency bandwidth across the set of said time windows; providing the AI ​​model with input including the spectral energy image; and obtaining from the AI ​​model an output including an indication that the at least one SSB is in the first frequency bandwidth.

[0263] Clause 25: The method according to any one of Clauses 13 to 24, wherein the AI ​​model comprises a convolutional neural network (CNN) having a kernel size tuned to an SSB aspect ratio.

[0264] Clause 26: The method according to Clause 25, wherein the CNN comprises: a first convolutional layer, the first convolutional layer being cascaded to a first max-pooling layer, the first convolutional layer being padded and having a first plurality of filters; a second convolutional layer, the second convolutional layer being coupled to a second max-pooling layer, the second convolutional layer being unpadded and having a second plurality of filters; and a fully connected layer, the fully connected layer having sigmoid activation.

[0265] Clause 27: One or more means comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to cause the one or more means to perform the method according to any one of Clauses 1 to 26.

[0266] Clause 28: One or more apparatuses, said apparatuses comprising components for performing the method according to any one of Clauses 1 to 26.

[0267] Clause 29: One or more non-transitory computer-readable media, the one or more non-transitory computer-readable media comprising executable instructions that, when executed by one or more processors of one or more devices, cause the one or more devices to perform the method according to any one of Clauses 1 to 26.

[0268] Clause 30: One or more computer program products embodied on one or more computer-readable storage media, said computer-readable storage media including code for performing the method according to any one of Clauses 1 to 26.

[0269] Additional Notes

[0270] The foregoing description is provided to enable any person skilled in the art to practice the various aspects described herein. The examples discussed herein do not limit the scope, applicability, or aspects set forth in the claims. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects. For example, the function and arrangement of the elements discussed may be changed without departing from the scope of this disclosure. Various processes or components may be omitted, substituted, or added as appropriate in the various examples. For example, the described methods may be performed in a different order than described, and various actions may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined in some other examples. For example, any number of aspects set forth herein may be used to implement an apparatus or practice. Additionally, the scope of this disclosure is intended to cover such apparatuses or methods practiced using other structures, functionalities, or structures and functionalities that complement or replace the various aspects of this disclosure set forth herein. It should be understood that any aspect of the disclosure herein may be embodied by one or more elements of the claims.

[0271] The various exemplary logic blocks, modules, and circuits described in this disclosure can be implemented or executed using a general-purpose processor, AI processor, digital signal processor (DSP), ASIC, field-programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic unit, discrete hardware component, or any combination thereof designed to perform the functions described herein. While the general-purpose processor may be a microprocessor, in alternative embodiments, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors working in conjunction with a DSP core, a system-on-a-chip (SoC), or any other such configuration.

[0272] As used in this article, the phrase “at least one of” in a list of items refers to any combination of those items, including a single member. For example, “at least one of a, b, or c” is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiple identical elements (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbb, bbc, cc, and ccc, or any other ordering of a, b, and c).

[0273] As used herein, the term "determine" encompasses a wide variety of actions. For example, "determine" can include calculation, operation, processing, deduction, investigation, lookup (e.g., searching in a table, database, or other data structure), assertion, etc. Additionally, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Furthermore, "determine" can include parsing, selecting, picking, building, etc.

[0274] As used herein, unless otherwise stated, “coupled to” and “coupled with” generally encompass both direct and indirect coupling (e.g., including intermediate aspects of coupling). For example, stating that a processor is coupled to memory allows for direct coupling or coupling via an intermediate aspect such as a bus.

[0275] The methods disclosed herein include one or more actions for implementing the methods. These method actions may be interchanged without departing from the scope of the claims. In other words, unless a specified order of actions is given, the order and / or use of a particular action may be modified without departing from the scope of the claims. Furthermore, the various operations of the methods described above can be performed by any suitable component capable of performing the corresponding function. This component may include various hardware and / or software components and / or modules, including but not limited to circuits, application-specific integrated circuits (ASICs), or processors.

[0276] The following claims are not intended to be limited to the aspects shown herein, but should be given the full scope consistent with the language of the claims. References to singular elements are not intended to mean “only one” (unless specifically stated as “only one”), but rather “one or more”. Unless otherwise specified, definite articles (e.g., “the” or “described”) subsequently used with an element (e.g., “processor”) are not intended to give that element a singular meaning (e.g., “only one”). For example, unless otherwise specified, references to elements (e.g., “processor”, “controller”, “memory”, “transceiver”, “antenna”, “the processor”, “the controller”, “the memory”, “the transceiver”, “the antenna”, etc.) should be understood to refer to one or more elements (e.g., “one or more processors”, “one or more controllers”, “one or more memories”, “a plurality of transceivers”, etc.). The terms “set” and “group” are intended to include one or more elements and may be used interchangeably with “one or more”. In the case of references to one or more elements performing a function (e.g., steps of a method), one element may perform all the functions, or more than one element may collectively perform those functions. When more than one element performs these functions together, each function does not need to be performed by every single element (e.g., different functions can be performed by different elements), and / or each function does not need to be performed by only one element as a whole (e.g., different elements can perform different sub-functions of a function). Similarly, when referring to one or more elements configured to cause another element (e.g., a device) to perform a function, one element may be configured to cause another element to perform all functions, or more than one element may be jointly configured to cause another element to perform these functions. Unless otherwise specifically stated, the term "some" refers to one or more. All structural and functional equivalents of the various aspects described throughout this disclosure that are currently or hereafter known to those skilled in the art are intended to be covered by the claims. Furthermore, nothing disclosed herein is intended to be offered to the public, whether or not such disclosure is explicitly recited in the claims.

Claims

1. An apparatus configured for wireless communication, the apparatus comprising: One or more memory units; and One or more processors, said one or more processors coupled to said one or more memories, said one or more processors being configured to cause the device to: The collection of monitoring data across time windows includes at least the first frequency bandwidth. An artificial intelligence (AI) model is used to identify at least one synchronization signal block (SSB) centered on a global synchronization channel number (GSCN) in the first frequency bandwidth, the AI ​​model including a kernel size tuned to the aspect ratio of the SSB; as well as Communication with network entities is based at least in part on the at least one SSB.

2. The apparatus of claim 1, wherein the AI ​​model is trained to detect one or more SSB features.

3. The apparatus of claim 2, wherein the one or more SSB features include one or more of the following: a primary synchronization signal, a secondary synchronization signal, a physical broadcast channel, or one or more space-time-frequency resources of an SSB.

4. The apparatus of claim 1, wherein, in order to identify the at least one SSB, the one or more processors are configured to cause the apparatus to: The AI ​​model is used to identify at least one GSCN candidate in the first frequency bandwidth; and The search for the at least one SSB is based at least in part on the at least one GSCN candidate.

5. The apparatus of claim 1, wherein the one or more processors are configured to cause the apparatus to: Multiple frequency bandwidths are monitored across the set of time windows, the multiple frequency bandwidths including the first frequency bandwidth; The AI ​​model is used to identify at least one GSCN candidate corresponding to at least one potential SSB among the plurality of frequency bandwidths; and One or more SSBs are searched, at least in part, based on the at least one GSCN candidate, the one or more SSBs including the at least one SSB.

6. The apparatus of claim 1, wherein, in order to identify the at least one SSB, the one or more processors are configured to cause the apparatus to: Based on the first frequency bandwidth and at least one time window, multiple samples are compressed across a time-frequency grid into a spectral energy image, wherein the multiple samples represent the first frequency bandwidth across the set of said time windows; The AI ​​model is provided with input including the spectral energy image; and The AI ​​model generates an output including an indication that the at least one SSB is within the first frequency bandwidth.

7. The apparatus of claim 6, wherein, in order to compress the plurality of samples into the spectral energy image, the one or more processors are configured to cause the apparatus to: The plurality of samples are converted into a plurality of spectral energy images, wherein each of the plurality of spectral energy images corresponds to a different time window in the set of time windows; and The plurality of spectral energy images are compressed into the spectral energy image.

8. The apparatus of claim 7, wherein, in order to compress the plurality of spectral energy images, the one or more processors are configured to cause the apparatus to incoherently combine the plurality of spectral energy images into the spectral energy image.

9. The apparatus of claim 1, wherein the AI ​​model comprises a convolutional neural network (CNN).

10. The apparatus of claim 9, wherein the CNN comprises a plurality of convolutional layers and a fully connected output layer with sigmoid activation.

11. The apparatus of claim 10, wherein the plurality of convolutional layers comprises: A first convolutional layer, coupled to a first max-pooling layer, the first convolutional layer being filled and having a first plurality of filters; and A second convolutional layer, coupled to a second max-pooling layer, is unfilled and has a second plurality of filters.

12. The apparatus of claim 11, wherein the first plurality of filters comprises at least eight filters.

13. An apparatus comprising: One or more memory units; and One or more processors, said one or more processors coupled to said one or more memories, said one or more processors being configured to cause the device to: The training data includes: multiple spectral energy images and multiple labels, wherein at least one of the multiple labels includes at least one indication of whether the corresponding spectral energy image includes at least a portion of a synchronization signal block (SSB); as well as The training data is used to train an artificial intelligence (AI) model to provide one or more probabilities of detecting one or more SSBs within a frequency bandwidth spanning a set of time windows, the frequency bandwidth including multiple Global Synchronization Channel Numbers (GSCNs).

14. The apparatus of claim 13, wherein each of the tags is associated with at least one of the plurality of spectral energy images and includes an indication of whether the corresponding spectral energy image includes at least a portion of a synchronization signal block (SSB).

15. The apparatus of claim 13, wherein the one or more processors are configured to cause the apparatus to transmit the trained AI model to a wireless communication device.

16. The apparatus of claim 13, wherein for each of the tags indicating that the corresponding spectral energy image includes at least one SSB, the corresponding tag further includes an indication of the time interval in which the at least one SSB occurs.

17. The apparatus of claim 13, wherein the plurality of spectral energy images comprises a set of spectral energy images associated with different frequency bandwidths across the system bandwidth.

18. The apparatus of claim 13, wherein the plurality of spectral energy images comprises one or more of the following: A first set of spectral energy images, the first set of spectral energy images including at least one SSB in one or more first frequency bandwidths; A second set of spectral energy images, wherein the second set of spectral energy images does not have any SSB in one or more second frequency bandwidths; or A third set of spectral energy images, the third set of spectral energy images including a first portion of the SSB in one or more third frequency bandwidths, wherein a second portion of the SSB is arranged outside the one or more third frequency bandwidths.

19. The apparatus of claim 18, wherein each spectral energy image in the first set of spectral energy images includes the at least one SSB arranged at different locations in the time-frequency grid of the respective spectral energy image.

20. The apparatus of claim 13, wherein, in order to obtain the training data, the one or more processors are configured to cause the apparatus to generate the training data at least in part based on the system bandwidth for any SSB scan.

21. The apparatus of claim 13, wherein, in order to train the AI ​​model, the one or more processors are configured to cause the apparatus to reduce the loss associated with the AI ​​model at least in part based on the training data.

22. The apparatus of claim 21, wherein the loss comprises: Sum of Squared Errors (SSE) loss; Intersection over Union (IoU) loss; Cross-entropy loss; or Their combination.

23. The apparatus of claim 13, wherein the one or more processors are configured to cause the apparatus to: Monitor at least the first frequency bandwidth across the set of said time windows; Use the trained AI model to identify at least one SSB centered at the GSCN in the first frequency bandwidth; and Communication with network entities is based at least in part on the at least one SSB.

24. The apparatus of claim 23, wherein, in order to identify the at least one SSB, the one or more processors are configured to cause the apparatus to: Based on the first frequency bandwidth and at least one time window, multiple samples are compressed across a time-frequency grid into a spectral energy image, wherein the multiple samples represent the first frequency bandwidth across the set of said time windows; The AI ​​model is provided with input including the spectral energy image; and The AI ​​model generates an output including an indication that the at least one SSB is within the first frequency bandwidth.

25. The apparatus of claim 13, wherein the AI ​​model comprises a convolutional neural network (CNN) having a kernel size tuned to an SSB aspect ratio.

26. The apparatus of claim 25, wherein the CNN comprises: A first convolutional layer, which is cascaded to a first max-pooling layer, is filled and has a first plurality of filters; A second convolutional layer, coupled to a second max-pooling layer, the second convolutional layer being unfilled and having a second plurality of filters; and A fully connected layer having S-shaped activation.

27. A method for wireless communication by a device, the method comprising: The collection of monitoring data across time windows includes at least the first frequency bandwidth. An artificial intelligence (AI) model is used to identify at least one synchronization signal block (SSB) centered on a global synchronization channel number (GSCN) in the first frequency bandwidth, the AI ​​model including a kernel size tuned to the aspect ratio of the SSB; as well as Communication with network entities is based at least in part on the at least one SSB.

28. A method, the method comprising: Obtain training data, the training data comprising: multiple spectral energy images and multiple labels, wherein at least one of the multiple labels includes at least one indication of whether the corresponding spectral energy image includes at least a portion of a synchronization signal block (SSB); and The training data is used to train an artificial intelligence (AI) model to provide one or more probabilities of detecting one or more SSBs within a frequency bandwidth spanning a set of time windows, the frequency bandwidth including multiple Global Synchronization Channel Numbers (GSCNs).

Images