Radio base station apparatus, radio communication system, and radio communication method

Abstract

To provide a technique capable of appropriately executing AI processing.SOLUTION: A wireless base station device according to one aspect includes a request acquisition unit that receives an inference request made by a wireless terminal device including input data to be input to a trained model corresponding to a task and identification information regarding the task or the trained model, a model storage unit that stores the trained model, an inference execution unit that inputs the input data to the trained model stored in the model storage unit based on the identification information to acquire an inference result of the task, and a result output unit that transmits the inference result to the wireless terminal device.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to a radio base station device, a radio communication system, and a radio communication method. [Background technology]

[0002] In recent years, various AI (Artificial Intelligence) technologies have been widely used and are being used or have the potential to be used in various fields such as business, academia, art, etc. (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7504507 Summary of the Invention [Problem to be solved by the invention]

[0004] There is room for consideration when it comes to performing AI processing (especially inference). [Means for solving the problem]

[0005] A wireless base station device according to one embodiment includes a request acquisition unit that receives an inference request made by a wireless terminal device, the inference request including input data to be input into a trained model corresponding to a task and identification information regarding the task or the trained model; a model memory unit that stores the trained model; an inference execution unit that inputs the input data into the trained model stored in the model memory unit based on the identification information to obtain an inference result for the task; and a result output unit that transmits the inference result to the wireless terminal device. [Effects of the Invention]

[0006] According to non-limiting aspects of the present disclosure, AI processing can be appropriately performed. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a block diagram illustrating a configuration example of an edge AI processing system according to an embodiment. [Figure 2] 1 is a block diagram illustrating an example of a hardware configuration of an apparatus according to an embodiment. [Figure 3] FIG. 2 is a block diagram illustrating an example of a functional configuration of a wireless base station device according to an embodiment. [Figure 4] FIG. 2 is a block diagram illustrating an example of a functional configuration of a wireless terminal device according to an embodiment. [Figure 5] FIG. 2 is a block diagram illustrating an example of a functional configuration of a server device according to an embodiment. [Figure 6] FIG. 1 is a sequence diagram illustrating an example of the operation of an edge AI processing system according to an embodiment. [Figure 7] 1 is a schematic diagram illustrating an example of a wireless mesh network according to an embodiment; DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, embodiments will be described with reference to the drawings as appropriate. The same elements throughout this specification will be designated by the same reference numerals unless otherwise specified. The following description, along with the accompanying drawings, is intended to explain exemplary embodiments and is not intended to represent the only embodiments. For example, when an order of operations is indicated in an embodiment, the order of operations may be changed as appropriate within the scope of the overall operation.

[0009] When multiple embodiments and / or variants are illustrated, some configurations, functions and / or operations in one embodiment and / or variant may be included in other embodiments and / or variants to the extent that no contradiction occurs, or may be replaced with corresponding configurations, functions and / or operations in other embodiments and / or variants.

[0010] In addition, in the embodiments, more detailed explanations than necessary may be omitted. For example, detailed explanations of publicly known or well-known technical matters may be omitted to avoid unnecessary redundancy in the explanation and / or to avoid ambiguity in technical matters or concepts, and to facilitate understanding by those skilled in the art. Furthermore, duplicated explanations of substantially identical configurations, functions, and / or operations may be omitted.

[0011] The accompanying drawings and the following description are provided to aid in understanding the embodiments, and are not intended to limit the subject matter described in the claims. Furthermore, the terms used in the following description may be appropriately replaced with other terms to aid in the understanding of those skilled in the art.

[0012] [Findings that led to this disclosure] Various AI technologies, including generative artificial intelligence (AI), are being widely used over networks including the Internet, or in standalone environments.

[0013] For example, by training a machine learning model using training data and then inputting input data into the trained machine learning model (trained model), it is possible to perform inference such as predicting unknown information corresponding to the input data.

[0014] In recent years, there has been an increasing demand for AI processing, such as the recognition and generation of voice, images, documents, etc., directly on mobile terminals such as smartphones, tablets, and PCs (Personal Computers), which are examples of wireless terminal devices, without going through a cloud or data center. As a result, development is underway for processing devices (chips, etc.) specialized for AI processing that can be installed on such mobile terminals.

[0015] Currently, data centers that perform AI processing are located far from the mobile devices that require the AI ​​processing results. This results in problems such as large traffic volumes and long delay times related to AI processing. Furthermore, data centers that perform AI processing are often managed and operated by third parties, which raises concerns about information security.

[0016] To address these issues, one possible solution is to perform AI processing on mobile terminals, as described above. AI processing (especially inference performed on mobile terminals) involves large-scale matrix operations. When attempting to perform matrix operations sequentially using a conventional central processing unit (CPU), the execution of AI processing takes a long time due to insufficient CPU power, resulting in a problem of large power consumption.

[0017] Therefore, the inventors of the present application have developed a technology that offloads the AI ​​processing that causes these problems when executed on a mobile terminal to a wireless base station.

[0018] The following describes non-limiting examples of techniques for offloading AI processing to wireless base stations.

[0019] [Embodiment Mode] <Edge AI processing system configuration> FIG. 1 is a block diagram illustrating an example of the configuration of an edge AI processing system according to an embodiment.

[0020] As shown in FIG. 1, the edge AI processing system 1 includes a wireless base station device 10, a wireless terminal device 20, and a server device 30. The wireless base station device may simply be called a base station device or a base station. The wireless terminal device may simply be called a terminal device or a terminal. The server device may simply be called a server. While FIG. 1 shows an example in which there is one each of the wireless base station device 10, the wireless terminal device 20, and the server device 30, there may be two or more of each of these in the edge AI processing system 1.

[0021] Generally, the wireless base station device 10 is subject to fewer restrictions on housing size and power consumption, which is an advantage that the wireless terminal device 20 does not have. Furthermore, the wireless base station device 10 often has more powerful hardware resources (memory, processor (e.g., CPU), etc.) than the wireless terminal device 20. While the wireless base station device 10 controls communications with the wireless terminal device 20, it does not always use these hardware resources for that control. For this reason, it may be effective for the wireless base station device 10 to perform AI processing. Therefore, in this embodiment, in the edge AI processing system 1, the wireless terminal device 20 addresses at least one of the above-mentioned problems by requesting (submitting) AI processing to the wireless base station device 10, which can be said to be located at the edge of the wireless communication system or wireless communication network.

[0022] The wireless base station device 10 is an example of a wireless device capable of wireless communication. The wireless base station device 10 may be a wireless base station used in a wireless communication standard such as a fourth generation mobile communication system (4G) or a fifth generation mobile communication system (5G), or may be an access point (e.g., a Wi-Fi (registered trademark) access point) or a router used in a wireless LAN-related standard.

[0023] The wireless base station device 10 forms an area where wireless communication is possible. The "area where wireless communication is possible" may be referred to as a "wireless communication area," "wireless area," "communication area," "service area," "coverage area," "cover area," or the like. The wireless communication area formed by the wireless base station device 10 that complies with or is based on a wireless LAN-related standard may be considered to correspond to a "cell," which is a term used in cellular communication. For example, the wireless communication area formed by the wireless base station device 10 may be considered to correspond to a "femtocell," which is classified as a "small cell."

[0024] The wireless base station device 10 communicates wirelessly with the wireless terminal device 20 that exists in an area formed by the wireless base station device 10 where wireless communication is possible.

[0025] The wireless base station device 10 communicates with the server device 30 via a backbone network (not shown), which is an example of a large-scale communication network such as the Internet, via a wired (and wireless, if necessary) connection. The "backbone network" may also be called a "core network," a "global network," or the like.

[0026] The wireless base station device 10 receives an inference request (inference request) transmitted by the wireless terminal device 20. The inference request may include input data to be input into a trained model corresponding to the task, and identification information related to the task or the trained model. The wireless base station device 10 transmits to the server device 30 a model request requesting a trained model corresponding to the task for executing the inference requested (requested) by the inference request (inference request). The wireless base station device 10 receives the trained model requested by the model request transmitted by the server device 30. The wireless base station device 10 inputs the received input data into the received trained model to obtain an inference result for the task. The wireless base station device 10 transmits the obtained inference result to the wireless terminal device 20.

[0027] The inference request and the inference result may be communicated between the wireless base station device 10 and the wireless terminal device 20 via an NPI (Neural Processing Interface), which is an example of a predefined application programming interface (API). A predefined API may mean that the parameters communicated through the API have a predefined format. The NPI may have input / output parameters such as an "AI processing class that specifies the type of AI processing to be performed" corresponding to identification information related to the task or the trained model, "data (set) to be subjected to AI processing" corresponding to input data to be input to the trained model corresponding to the task, and "result data (set) resulting from AI processing" corresponding to the inference result of the task. The NPI may also have an "option parameter that specifies detailed options for the specified AI processing class."

[0028] The wireless terminal device 20 is a terminal owned or managed by a user and directly operated by the user as an end of a wireless network, and may be a mobile terminal such as a mobile phone, smartphone, tablet, or notebook PC, or a terminal installed in a predetermined location such as a stationary PC, game console, or people flow monitoring camera. For example, when the wireless terminal device 20 is located in the service area of ​​the wireless base station device 10, it connects to the wireless base station device 10 via a wireless access line.

[0029] The wireless access line may be a public wireless line such as a communication network operated by a so-called carrier company or the Internet, or a local wireless line used for communication within a home, office, or store. For example, any of Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Orthogonal Frequency Division Multiple Access (OFDMA), and Single Carrier Frequency Division Multiple Access (SC-FDMA) may be applied. OFDMA may be implemented using wireless technologies such as IEEE802.11, IEEE802.16, Long Term Evolution (LTE), LTE-Advanced, and New Radio (NR). A local wireless line is typically owned or managed by an entity that owns or manages the home, office, or store within the coverage area of ​​the line. A public wireless line is typically configured by an external carrier company or a third-party PC that provides Internet functionality, rather than by such an entity.

[0030] In the wireless access line, a MIMO (Multiple Input Multiple Output) technique using an antenna array having multiple antenna elements may be applied to the downlink (from the wireless base station device 10 to the wireless terminal device 20) and / or the uplink (from the wireless terminal device 20 to the wireless base station device 10). For example, beamforming using an antenna array may be performed in the downlink and / or the uplink.

[0031] The wireless terminal device 20 transmits an inference request to the wireless base station device 10. The inference request may include input data (parameters possessed by the NPI) to be input to a trained model corresponding to a task, based on information designated by a user of the wireless terminal device 20 and input to the wireless terminal device 20, and identification information (parameters possessed by the NPI) related to the task or the trained model. The parameters possessed by the NPI may be set by the wireless terminal device 20 in response to an input (designation) by the user of the wireless terminal device 20. The inference request may also be transmitted to the wireless base station device 10 in response to an input by the user (for example, pressing an "execute inference button"). The wireless terminal device 20 receives the inference result of the task transmitted by the wireless base station device 10. In this way, the wireless terminal device 20 transmits the inference request to the wireless base station device 10 by calling the NPI for the wireless base station device 10 after setting the above parameters, and acquires the inference result via result data (set) that is an NPI parameter. The wireless terminal device 20 may display the received inference result of the task.

[0032] The server device 30 is a device that stores multiple trained models and may be installed in a cloud or a data center. Each of the multiple trained models may be a machine learning model trained using training data by a training device, which may or may not be the server device 30.

[0033] Regarding training of a machine learning model, for example, if the machine learning model to be trained is realized as a neural network, training data is input to the input layer of the machine learning model to be trained, and an output result is obtained from the output layer of the machine learning model to be trained. The obtained output result is compared with the corresponding label, and the parameters of the machine learning model to be trained are updated according to the comparison result using a backpropagation algorithm or the like. When a predetermined termination condition (e.g., all prepared training data has been processed, or the error has converged to a predetermined threshold or less) is satisfied, the training of the machine learning model to be trained ends. When the training ends, the machine learning model to be trained having the updated parameters is obtained as a trained model. Note that the generation of the trained model described above is merely an example, and it goes without saying that any other well-known training method may be used to generate a trained model.

[0034] Server device 30 receives the model request transmitted by wireless base station device 10. Server device 30 acquires the trained model requested by the model request from among multiple trained models present (stored) in server device 30, and transmits the acquired trained model to wireless base station device 10.

[0035] Hereinafter, the term "apparatus" may be interchangeably read as "circuit," "device," "unit," "module," etc. The term "interface" may be interchangeably read as "adapter," "board," "card," "module," "chip," etc.

[0036] <Configuration of various devices> 2 is a block diagram showing an example of a hardware configuration of an apparatus according to an embodiment. The apparatus 1000 may include a wireless base station apparatus 10, a wireless terminal apparatus 20, and a server apparatus 30.

[0037] 2, the device 1000 may include a storage device 1001, a processing device 1002, a user interface (UI) device 1003, a communication device 1004, and a bus 1005. The storage device 1001, the processing device 1002, the UI device 1003, and the communication device 1004 may be connected to each other via the bus 1005.

[0038] The programs or instructions for realizing the functions and processes described above and below of the wireless base station device 10, the wireless terminal device 20, and the server device 30 may be downloaded from some external device (e.g., a server) via a network, etc. Furthermore, such programs or instructions may be provided from a removable storage medium such as a CD-ROM (Compact Disc Read Only Memory), flash memory, etc.

[0039] Storage device 1001 may be implemented by RAM (Random Access Memory), flash memory, a hard disk drive, or the like, and stores installed programs or instructions as well as files, data, and the like used to execute the programs or instructions. Storage device 1001 may also include a non-transitory storage medium.

[0040] The processing device 1002 may be realized, for example, by a general-purpose processor or controller (circuit), or by a dedicated processor or controller (circuit). When realized by a general-purpose processor, the processing device 1002 may be realized by one or more central processing units (CPUs), graphics processing units (GPUs), processing circuitry, etc., each of which may be composed of one or more processor cores. In this case, the processing device 1002 performs the functions and processes of the wireless base station device 10, the wireless terminal device 20, and the server device 30 described above and below in accordance with programs or instructions stored in the storage device 1001, data such as parameters used to execute the programs or instructions, etc. Such a program, when executed by a computer or the processing device 1002, may cause the computer or the processing device 1002 to function as the device 1000 according to the present disclosure. Furthermore, such a program, when executed by the computer or the processing device 1002, may cause the computer or the processing device 1002 to execute a method related to inference execution according to the present disclosure.

[0041] The device 1000 may be configured to include hardware such as a microprocessor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array). For example, the processing device 1002 may be implemented to include at least one of these pieces of hardware. For example, an NPU (Neural Processing Unit) specialized for AI processing (inference), which will be described below, may be configured to include at least one of these pieces of hardware when realized as hardware.

[0042] The UI device 1003 may be composed of input devices such as a keyboard, a mouse, a camera, a microphone, etc., output devices such as a display, a speaker, a headset, a printer, etc., and input / output devices such as a haptic device, a touch panel, etc., and realizes an interface between the user of the device 1000 and the device 1000. For example, the user of the device 1000 operates the device 1000 by manipulating a graphical user interface (GUI) displayed on a display or a touch panel using a keyboard, a mouse, a stylus, a finger, etc.

[0043] Communication device 1004 may be realized by various communication circuits that perform communication processing with external devices and communication networks such as the Internet, a LAN, a wireless communication network, etc., and that correspond to the communications that device 1000 is capable of communicating with. For example, examples of such communication circuits include a communication circuit for cellular communications, a communication circuit for wireless LAN communications (e.g., for Wi-Fi communications), a communication circuit for wired communications (e.g., for Ethernet communications, optical communications), etc.

[0044] When the communication device 1004 is, for example, a device for transmitting and receiving radio signals over an access line with a wireless terminal device 20 or a wireless base station device 10, the communication device 1004 may include, for example, one or more antennas, a baseband (BB) signal processing circuit, a MAC processing circuit, an upconverter, a downconverter, an amplifier, etc. The BB signal processing circuit may include, for example, an encoding circuit and a modulation circuit for encoding and modulating a transmission signal, and a demodulation circuit and a decoding circuit for demodulating and decoding a received signal.

[0045] The above-described hardware configuration of the device 1000 is merely an example, and the device 1000 according to the present disclosure may be realized by any other appropriate hardware configuration.

[0046] It should be noted that the amount of hardware may be increased or decreased as appropriate in the example hardware configuration illustrated in Fig. 2. For example, any hardware block may be added, deleted, divided, or integrated in any combination as appropriate.

[0047] Next, an example of the functional configuration of the wireless base station device 10 will be described.

[0048] FIG. 3 is a block diagram illustrating an example of a functional configuration of a radio base station apparatus according to an embodiment.

[0049] For example, as shown in FIG. 3 , the radio base station device 10 includes a request acquisition unit 101, a model request unit 102, a model acquisition unit 103, an inference execution unit 104, a result output unit 105, and a model storage unit 151. The request acquisition unit 101, the model request unit 102, the model acquisition unit 103, the inference execution unit 104, and the result output unit 105 may be realized as hardware, software, or a combination of hardware and software by a processing device 1002 of the radio base station device 10. For example, the inference execution unit 104 may be realized as hardware (e.g., including a large-scale matrix calculation circuit) by the above-mentioned NPU or GPU, or as software, or a combination of hardware and software. The model storage unit 151 may be realized by the storage device 1001 of the radio base station device 10.

[0050] The request acquisition unit 101 acquires an inference request made by the wireless terminal device 20, the inference request including input data to be input to a trained model corresponding to a task and identification information related to the task or the trained model, by receiving the inference request from the wireless terminal device 20 (via the communication device 1004 of the wireless base station device 10). The request acquisition unit 101 outputs the acquired inference request to the model request unit 102 and the inference execution unit 104.

[0051] The model request unit 102 identifies a trained model corresponding to a task based on identification information related to the task or the trained model included in the inference request output by the request acquisition unit 101. Correspondence information between identification information related to the task and (identification information related to) the trained model may be stored in the storage device 1001 of the wireless base station device 10, and the model request unit 102 may identify a trained model corresponding to the task by referring to the correspondence information. The model request unit 102 transmits a model request including identification information related to the identified trained model to the server device 30 (via the communication device 1004 of the wireless base station device 10). Alternatively, the model request unit 102 may transmit a model request including the identification information related to the task output by the request acquisition unit 101 to the server device 30 (via the communication device 1004 of the wireless base station device 10). The model request unit 102 outputs the identification information related to the trained model or task to be transmitted to the server device 30 to the model acquisition unit 103.

[0052] The model acquisition unit 103 acquires a trained model by receiving from the server device 30 (via the communication device 1004 of the wireless base station device 10) a trained model corresponding to the identification information related to the trained model or task transmitted to the server device 30 by the model request unit 102. The model acquisition unit 103 associates the acquired trained model with the identification information of the trained model or the corresponding task and stores the trained model in the model storage unit 151. The model acquisition unit 103 outputs information indicating that the acquired trained model has been stored in the model storage unit 151 to the inference execution unit 104.

[0053] Upon receiving the information output by the model acquisition unit 103, the inference execution unit 104 identifies a trained model corresponding to the task based on the identification information relating to the task or trained model included in the inference request output by the request acquisition unit 101, and inputs the input data included in the inference request output by the request acquisition unit 101 to the identified trained model stored in the model storage unit 151. As a result, the inference execution unit 104 acquires an inference result for the task. The inference execution unit 104 outputs the acquired inference result for the task to the result output unit 105.

[0054] The result output unit 105 transmits the inference result of the task output by the inference execution unit 104 to the wireless terminal device 20 (via the communication device 1004 of the wireless base station device 10).

[0055] The model storage unit 151 stores the trained model transmitted by the server device 30 in association with identification information relating to the trained model or the corresponding task.

[0056] Next, an example of the functional configuration of the wireless terminal device 20 will be described.

[0057] FIG. 4 is a block diagram illustrating an example of a functional configuration of a wireless terminal device according to an embodiment.

[0058] 4, the wireless terminal device 20 includes an inference request unit 201, a result acquisition unit 202, a display control unit 203, and a display unit 251. The inference request unit 201, the result acquisition unit 202, and the display control unit 203 may be realized as hardware, software, or a combination of hardware and software by a processing device 1002 of the wireless terminal device 20. The display unit 251 may be realized by a UI device 1003 (e.g., a display device) of the wireless terminal device 20.

[0059] The inference request unit 201 transmits an inference request including input data to be input to a trained model corresponding to a task and identification information related to the task or the trained model to the wireless base station device 10 (via the communication device 1004 of the wireless terminal device 20). As a result, the wireless terminal device 20 (the inference request unit 201) requests (requests) inference from the wireless base station device 10.

[0060] The result acquisition unit 202 acquires the inference result of the task requested by the inference request transmitted to the wireless base station device 10 by receiving the inference result from the wireless base station device 10 (via the communication device 1004 of the wireless terminal device 20). The result acquisition unit 202 outputs the acquired inference result of the task to the display control unit 203.

[0061] The display control unit 203 controls the display unit 251 to display the inference result of the task output by the result acquisition unit 202 on the display unit 251 .

[0062] The display unit 251 displays the inference results of the task under the control of the display control unit 203.

[0063] Next, an example of the functional configuration of the server device 30 will be described.

[0064] FIG. 5 is a block diagram illustrating an example of a functional configuration of a server device according to an embodiment.

[0065] 5 , the server device 30 includes a request acquisition unit 301, a model output unit 302, and a model storage unit 351. The request acquisition unit 301 and the model output unit 302 may be realized as hardware, software, or a combination of hardware and software by a processing device 1002 of the server device 30. The model storage unit 351 may be realized by the storage device 1001 of the server device 30.

[0066] The request acquisition unit 301 acquires a model request including identification information related to a trained model or a task by receiving the model request from the wireless base station device 10 (via the communication device 1004 of the server device 30). The request acquisition unit 301 outputs the acquired model request to the model output unit 302.

[0067] The model output unit 302 identifies a trained model corresponding to a task based on identification information related to the trained model or task included in the model request output by the request acquisition unit 301. Correspondence information between tasks and trained models may be stored in the storage device 1001 of the server device 30, and the model output unit 302 may identify a trained model corresponding to the task (identification information of the task) by referring to the correspondence information. The model output unit 302 acquires the identified trained model from the model storage unit 351 that stores a plurality of trained models, and transmits the acquired trained model to the radio base station device 10 (via the communication device 1004 of the server device 30).

[0068] The model storage unit 351 stores a trained model in association with identification information related to the trained model or the corresponding task. The trained model (and the identification information) may be stored in the model storage unit 351 in response to a user of the server device 30 specifying (inputting) the trained model (and the identification information).

[0069] <Operation of edge AI processing system> Next, an example of the operation (flow) of the edge AI processing system 1 will be described.

[0070] FIG. 6 is a sequence diagram illustrating an example of the operation of the edge AI processing system according to one embodiment.

[0071] In step S101, the wireless terminal device 20 generates an inference request including input data to be input to a trained model corresponding to a task and identification information related to the task or the trained model. The wireless terminal device 20 may also include, in the inference request, option parameters that specify detailed options for the AI ​​processing class (inference).

[0072] In step S102, the wireless terminal device 20 transmits the generated inference request to the wireless base station device 10, and the wireless base station device 10 acquires (receives) the inference request from the wireless terminal device 20.

[0073] In step S103, the radio base station device 10 generates a model request including identification information related to the task or the trained model, based on the identification information related to the task or the trained model included in the acquired inference request.

[0074] In step S104, the wireless base station device 10 transmits the generated model request to the server device 30, and the server device 30 acquires (receives) the model request from the wireless base station device 10.

[0075] In step S105, the server device 30 acquires a trained model corresponding to the task based on the acquired model request and transmits it to the radio base station device 10, and the radio base station device 10 acquires (receives) the trained model corresponding to the task from the server device 30 and stores it.

[0076] In step S106, the radio base station device 10 executes inference. Specifically, the radio base station device 10 inputs the input data included in the inference request acquired in step S102 into the trained model acquired in step S105 and stored in the radio base station device 10, and acquires an inference result for the task.

[0077] In step S107, the wireless base station device 10 transmits the inference result of the task to the wireless terminal device 20, and the wireless terminal device 20 acquires (receives) the inference result of the task from the wireless base station device 10.

[0078] As described above, according to the embodiment of the present disclosure, the wireless terminal device 20 does not perform the AI ​​processing, but the wireless base station device 10 located at the edge of the wireless communication network performs the AI ​​processing upon receiving a request from the wireless terminal device 20.

[0079] This allows wireless terminal device 20 to reduce the amount of calculation required for AI processing, thereby reducing power consumption and extending battery life of wireless terminal device 20. Furthermore, since the latest AI processing model is obtained from server device 30 in response to a request from wireless terminal device 20, AI inference is performed using the latest trained model that is constantly evolving, and results equivalent to those achieved by server device 30 can be obtained.

[0080] The wireless base station device 10 is operated by a communications platform company with which the company has a contract, enabling edge processing in a closed communications environment from the wireless terminal device 20 before it goes out into the open internet environment where an unspecified number of PCs and the like may be present to the specific wireless base station device 10. This makes it possible to largely eliminate concerns about information security. In addition, because many wireless terminal devices 20 are connected to the wireless base station device 10 of the communications platform company, there is a high demand for AI processing, and the utilization frequency of the CPU resources of the wireless base station device 10 can be maintained high and leveled out.

[0081] In addition, the delay time associated with AI processing can be reduced compared to when AI processing is performed in the cloud or a data center.

[0082] Furthermore, since the wireless base station device 10 often has more powerful hardware resources than the wireless terminal device 20, it is possible to increase the possibility of shortening the time required to execute AI processing.

[0083] In addition to the above-mentioned information, the request acquisition unit 101 may additionally acquire external AI usage permission information from the wireless terminal device 20. This information indicates whether the wireless terminal device 20 is permitted or prohibited from using an external AI, typically an AI implemented in the server device 30. If the use of external AI is permitted, and if the wireless base station device 10 does not have sufficient spare CPU resources, the server device 30 executes AI inference as in the conventional manner, and transmits the execution results to the wireless terminal device 20. If the use of external AI is prohibited by the external AI usage permission information, the wireless base station device 10 executes AI inference and transmits the execution results to the wireless terminal device 20, regardless of the spare CPU resources of the wireless base station device 10.

[0084] In this embodiment, wireless terminal device 20 simply requests AI processing from an external device and receives the results, and does not need to be aware of whether the AI ​​processing was performed by server device 30 or wireless base station device 10. However, if the external AI usage permission information prohibits the use of external AI, flag information indicating that the AI ​​processing was performed by wireless base station device 10 may be added to the results of the AI ​​processing and transmitted to wireless terminal device 20.

[0085] <Modification> A modification of the above embodiment will now be described.

[0086] (Variation 1) If the radio base station device 10 constantly stores the trained model used for inference in the model storage unit 151, the storage capacity of the model storage unit 151 will gradually become insufficient. Therefore, the model acquisition unit 103 of the radio base station device 10 may store the trained model acquired (received) from the server device 30, along with the date and time when the trained model was acquired from the server device 30, in the model storage unit 151, and may delete the trained model from the model storage unit 151 when a predetermined time (for example, one month (which may be fixed or may be set by the user)) has elapsed from the date and time.

[0087] In this case, the model request unit 102 may determine whether a trained model corresponding to the task is stored in the model storage unit 151, based on identification information related to the task or the trained model included in the inference request output by the request acquisition unit 101. If the trained model is not stored in the model storage unit 151, the model request unit 102 may transmit the identification information to the server device 30, and the model acquisition unit 103 may acquire the trained model from the server device 30 and store it in the model storage unit 151. Conversely, if the trained model is stored in the model storage unit 151, the inference execution unit 104 may input the input data included in the inference request to the trained model stored in the model storage unit 151 to acquire an inference result for the task, without the model request unit 102 transmitting the identification information to the server device 30. This eliminates the need to inquire about the trained model from the server device 30 and acquire the trained model, thereby increasing the response speed of AI processing.

[0088] Additionally or alternatively, when a trained model is stored in the model storage unit 151, the model request unit 102 may inquire of the server device 30 about whether the trained model has been updated. When the trained model has been updated, the model acquisition unit 103 may acquire the updated trained model from the server device 30 and store it in the model storage unit 151, and the inference execution unit 104 may input input data into the updated trained model stored in the model storage unit 151 to acquire an inference result for the task. This makes it possible to execute AI processing using the latest trained model.

[0089] (Variation 2) As a wireless communication network, it is possible to utilize a local wireless circuit network called a wireless mesh network, a wireless multi-hop network, a wireless relay network, etc., in which wireless communication is performed by arranging, for example, a plurality of repeaters equivalent to a plurality of wireless base station devices (for example, Wi-Fi access points) in a hierarchical manner. Furthermore, in this modified example, edge processing is possible in an extremely limited local area of ​​a Wi-Fi environment managed by the communication platform company, rather than the wireless base station device 10 operated by the communication platform company shown in the embodiment, so there is no need to send any information outside the company's range, and further it is possible to dispel concerns about information security.

[0090] Fig. 7 is a schematic diagram showing an example of a wireless mesh network according to an embodiment. The wireless mesh network shown in Fig. 7 includes wireless base station devices 100 to 107, which may have the configurations shown in Figs. 2 and 3, respectively. In the example shown in Fig. 7, wireless base station devices 101 and 102 are connected to wireless base station device 100, wireless base station devices 103 and 105 are connected to wireless base station device 101, wireless base station device 104 is connected to wireless base station device 103, wireless base station device 107 is connected to wireless base station device 104, and wireless base station device 106 is connected to wireless base station device 105, forming a hierarchical structure (hierarchical network). Also, in the example shown in Fig. 7, wireless terminal device 20 is connected to wireless base station device 103, and wireless base station device 100 is connected to server device 30. The wireless base station device 100 connected to the backbone network may be referred to as a parent device or node, a master device or node, a core device or node, etc., and each of the wireless base station devices 101 to 107 other than the core device may be referred to as a child device or node, a slave device or node, etc. The configuration of the wireless mesh network shown in Fig. 7 is merely an example, and any number of wireless base station devices may be present, for example.

[0091] In such a wireless mesh network, the wireless base station device 100, which is a core device, acquires (receives) information about the connection partners of each wireless base station device from the wireless base station devices 101 to 107 other than the core device, thereby calculating and storing topology information of the wireless mesh network (information about the number of hops (layers) of each wireless base station device, the connection partners, etc.). Then, the wireless base station device 100 may transmit (the topology information is relayed) the topology information to the wireless base station devices 101 to 107 (via the wireless base station device (wireless base station device 101 in the example shown in FIG. 7) to which it is connected). As a result, the wireless base station devices 101 to 107 other than the core device may also store the topology information of the wireless mesh network.

[0092] Furthermore, each of (the processing devices (e.g., CPUs) of) the radio base station devices 100 to 107 constantly monitors (measures) the traffic volume of (the processing device of) the radio base station device. Here, since the load on the processing devices of the radio base station devices used in the wireless mesh network is mainly due to data transfer, the traffic volume can be considered as the load on the processing devices or the radio base station device. Each of the radio base station devices 101 to 107 other than the core device periodically reports information about the traffic volume (load) of the radio base station device to the radio base station device 100, which is the core device (the information about the traffic volume (load) is relayed). The radio base station device 100 may transmit the reported information about the traffic volume (load) of the radio base station devices 101 to 107 (and the traffic volume (load) of the radio base station device 100) to the radio base station devices 101 to 107 (the information about the traffic volume (load) is relayed). As a result, the radio base station devices 100 to 107 may store information relating to the traffic volume (load) of the radio base station devices 100 to 107.

[0093] Here, when wireless terminal device 20 transmits an inference request as described in the above embodiment, wireless terminal device 20 transmits the inference request to wireless base station device 103, and the inference request is relayed (transferred) successively via wireless base station device 103, wireless base station device 101, and wireless base station device 100, and wireless base station device 100, which receives the inference request, transmits a model request to server device 30. Note that, although an example in which various operations are performed under the control of wireless base station device 100 will be described below, various operations may be performed under the control of a wireless base station device to which wireless terminal device 20 is connected (wireless base station device 103 in the example shown in FIG. 7) instead of wireless base station device 100.

[0094] First example As a first example, only radio base station device 100 may acquire a trained model, and radio base station device 101 to radio base station device 107 other than the core device may not acquire a trained model (i.e., may not be able to perform AI processing). For example, radio base station device 100 may perform inference in the same manner as radio base station device 10 described in the above embodiment, and transmit the inference result to radio terminal device 20. The inference result may then be relayed (transferred) in the reverse order of the inference request, i.e., via radio base station device 101 and then radio base station device 103, and then relayed (transferred) from radio base station device 103 to radio terminal device 20.

[0095] Second example As a second example, the radio base station device 100 may transmit the trained model acquired from the server device 30 to the radio base station devices 101 to 107 (the trained model is relayed). Then, the radio base station devices 101 to 107 other than the core device may also acquire the trained model and store it in their respective model storage units 151. In this case, under the control of the radio base station device 100, one radio base station device may execute AI processing (inference), or multiple radio base station devices may execute (distributed) AI processing. The one radio base station device may be the radio base station device 100, or may be another radio base station device other than the radio base station device 100. The multiple radio base station devices may include the radio base station device 100, or may not include the radio base station device 100. The radio base station device 100 may determine (select) one radio base station device or multiple radio base station devices to execute the AI ​​processing based on a predetermined determination (selection) rule. As one example, the predetermined decision rule may be to determine N radio base station devices in ascending order of traffic volume, or to determine radio base station devices with traffic volume equal to or less than M. That is, radio base station device 100 may determine one radio base station device or multiple radio base station devices to perform AI processing based on information related to the traffic volume of radio base station device 100 to radio base station device 107. As another example, the predetermined decision rule may be to determine N radio base station devices in ascending order of the number of hops from the radio base station device to which radio terminal device 20 is connected (radio base station device 103 in the example shown in FIG. 7), or to determine a radio base station device with the number of hops equal to or less than M. That is, radio base station device 100 may determine one radio base station device or multiple radio base station devices to perform AI processing based on information related to the locations of radio base station device 100 to radio base station device 107 (e.g., topology information, information related to the number of hops). As yet another example, radio base station device 100 may determine one radio base station device or multiple radio base station devices to perform AI processing based on information related to the traffic volume and information related to the locations of radio base station device 100 to radio base station device 107.For example, the radio base station device 100 may convert information related to traffic volume into a first score using a predetermined formula, convert information related to location into a second score using a predetermined formula, and determine N radio base station devices in ascending order (or ascending order) of the sum of the first score and the second score, or may determine radio base station devices for which the sum is equal to or less than M (or equal to or greater than M). Furthermore, the radio base station device 100 may determine one or more radio base station devices to perform AI processing so as to leave some capacity available to perform processing other than AI processing (e.g., data transfer). As described above, a radio base station device with capacity available to perform AI processing performs AI processing, thereby enabling appropriate distribution of processing load. Furthermore, when various operations are performed under the control of a radio base station device to which the radio terminal device 20 is connected (radio base station device 103 in the example shown in FIG. 7), a radio base station device closer to the radio terminal device 20 performs AI processing, thereby preventing an increase in delay time.

[0096] To summarize the second example, the radio base station device may operate as follows: In the following, acquisition of a trained model is omitted, and it is assumed that the trained model is stored in the model storage unit 151 of each of the radio base station devices 100 to 107.

[0097] The wireless base station device 100 may acquire (receive) an inference request transmitted by the wireless terminal device 20, which includes input data to be input into a trained model corresponding to a task and identification information related to the task or the trained model.

[0098] Next, radio base station device 100 may determine one or more radio base station devices that will execute task inference from radio base station devices 100 to 107. Here, the determination of one or more radio base station devices that will execute task inference may be based on information about the traffic volume (load) of radio base station devices 100 to 107 and / or information about their locations, as described above.

[0099] Next, the radio base station device 100 may cause the determined one or more radio base station devices to input the input data included in the inference request to the trained models stored in each of the determined one or more radio base station devices based on identification information regarding the task or trained model included in the inference request, thereby obtaining an inference result for the task.

[0100] Next, the radio base station device 100 may acquire (receive) the inference results of the task from the determined one or more radio base station devices.

[0101] Finally, the wireless base station device 100 may transmit the obtained inference result of the task to the wireless terminal device 20.

[0102] In the second example, in order to maintain the mesh topology optimally, each wireless base station device 10 constantly keeps track of the traffic volume (load) of the other wireless base station devices 10, and therefore, the moment a request for AI processing is received, it is possible to know which wireless base station device 10 has the least amount of communication traffic (load), which is advantageous because it is possible to select the optimal wireless base station device 10 that has as few hops as possible from the wireless terminal device 20 that issued the AI ​​processing request and has ample CPU resources.

[0103] Furthermore, unlike public wireless lines, a Wi-Fi environment, which is a local wireless line, can complete AI processing in a local environment without leaking any of the customer's personal information outside the store, for applications such as facial recognition and people flow analysis of customers visiting a store, eliminating the need to worry about unintended data leaks on public lines or in the server device 30.

[0104] (Variation 3) A radio base station device may be composed of multiple units. Specifically, for example, radio base station device 10 may be composed of an RU (Radio Unit), a DU (Distributed Unit), and a CU (Central Unit). In radio base station device 10 having such a configuration, the NPU may be located in the DU or the CU.

[0105] (Other variations) The names of the above-described device 1000 (wireless base station device 10, wireless terminal device 20, and server device 30) and the functional units (request acquisition unit 101, inference request unit 201, request acquisition unit 301, etc.) constituting the device 1000 are merely examples and may be changed as appropriate. Furthermore, two or more of the functional units constituting the device 1000 may be combined and integrated, or one functional unit may be divided into multiple functional units. For example, two or more of the request acquisition unit 101, model request unit 102, model acquisition unit 103, inference execution unit 104, and result output unit 105 may be integrated into a functional unit such as a processing unit or a control unit, and two or more of the inference request unit 201, result acquisition unit 202, and display control unit 203 may be integrated into a functional unit such as a processing unit or a control unit. Furthermore, the transmission function and / or reception function of each functional unit may be collectively referred to as a notification unit, a notified unit, a communication unit, a transmission / reception unit, a transmission unit, a reception unit, etc.

[0106] The order of steps in the sequence diagrams shown above is merely an example and may be changed as appropriate, or some steps may be performed in parallel.

[0107] <Summary of the embodiment> A wireless base station device according to one embodiment includes a request acquisition unit that receives an inference request made by a wireless terminal device, the inference request including input data to be input into a trained model corresponding to a task and identification information regarding the task or the trained model; a model memory unit that stores the trained model; an inference execution unit that inputs the input data into the trained model stored in the model memory unit based on the identification information to obtain an inference result for the task; and a result output unit that transmits the inference result to the wireless terminal device.

[0108] In one example, the request acquisition unit receives the inference request through a predefined application programming interface (API), and the result output unit transmits the inference result to the wireless terminal device through the predefined API.

[0109] In one example, the wireless base station device further includes a model request unit that sends a model request including the identification information to a server device that stores multiple trained models, and a model acquisition unit that receives the trained model corresponding to the identification information from the server device and stores it in the model storage unit.

[0110] In one example, the model request unit determines whether the trained model is stored in the model storage unit, and if the trained model is not stored in the model storage unit, the model request unit transmits the identification information to the server device, and the model acquisition unit receives the trained model from the server device and stores it in the model storage unit.

[0111] In one example, when the trained model is stored in the model storage unit, the model request unit does not send the identification information to the server device, and the inference execution unit inputs the input data into the trained model stored in the model storage unit to obtain the inference result of the task.

[0112] In one example, when the trained model is stored in the model storage unit, the model request unit inquires of the server device whether the trained model has been updated, and when the trained model has been updated, the model acquisition unit receives the updated trained model from the server device and stores it in the model storage unit, and the inference execution unit inputs the input data into the updated trained model stored in the model storage unit to obtain the inference result of the task.

[0113] A wireless communication system according to one embodiment is a wireless communication system having a plurality of wireless base station devices, wherein one of the plurality of wireless base station devices receives an inference request made by a wireless terminal device, the inference request including input data to be input into a trained model corresponding to a task and identification information regarding the task or the trained model, determines one or more wireless base station devices from the plurality of wireless base station devices to perform inference of the task, causes the one or more wireless base station devices to input the input data into the trained model stored in each of the one or more wireless base station devices based on the identification information to obtain an inference result for the task, receives the inference result for the task from the one or more wireless base station devices, and transmits the inference result to the wireless terminal device.

[0114] In one example, the one radio base station device receives information about traffic volumes of the plurality of radio base station devices, and determines the one or more radio base station devices based on the information about the traffic volumes.

[0115] In one example, the one wireless base station device receives information about the locations of the plurality of wireless base station devices, and determines the one or more wireless base station devices based on the information about the locations.

[0116] A wireless communication system according to one embodiment is a wireless communication system having a wireless terminal device and a wireless base station device, wherein the wireless terminal device transmits an inference request to the wireless base station device, the inference request including input data to be input into a trained model corresponding to a task and identification information regarding the task or the trained model, and the wireless base station device receives the inference request, and based on the identification information, inputs the input data into the trained model stored in the wireless base station device to obtain an inference result for the task, and transmits the inference result to the wireless terminal device.

[0117] In one embodiment, a wireless communication method includes a wireless base station device receiving an inference request made by a wireless terminal device, the inference request including input data to be input into a trained model corresponding to a task and identification information regarding the task or the trained model, and based on the identification information, inputting the input data into the trained model stored in the wireless base station device to obtain an inference result for the task, and transmitting the inference result to the wireless terminal device.

[0118] As described above, according to one aspect of the present disclosure, AI processing can be executed appropriately.

[0119] Although the embodiments have been described above with reference to the drawings, the present disclosure is not limited to such examples. It is clear that a person skilled in the art can conceive of various modifications or alterations within the scope of the claims. It is understood that such modifications or alterations also fall within the technical scope of the present disclosure. Furthermore, the components or features in the embodiments (including modifications or alterations) may be combined in any manner without departing from the spirit of the present disclosure. [Explanation of symbols]

[0120] 1. Edge AI processing system 1000 devices 1001 Storage device 1002 Processing equipment 1003 UI device 1004 Communication equipment 1005 Bus 10 Wireless base station equipment 101 Request acquisition part 102 Model request section 103 Model Acquisition Department 104 Inference execution unit 105 Result output section 151 Model Memory Unit 20 Wireless terminal equipment 201 Inference request part 202 Result acquisition part 203 Display control unit 251 Display section 30 Server device 301 Request acquisition part 302 Model Output Section 351 Model Memory Unit

Claims

1. a request acquisition unit that receives an inference request made by a wireless terminal device, the inference request including input data to be input into a trained model corresponding to a task and identification information related to the task or the trained model; a model storage unit that stores the trained model; an inference execution unit that inputs the input data into the trained model stored in the model storage unit based on the identification information and obtains an inference result of the task; a result output unit that transmits the inference result to the wireless terminal device; A wireless base station device comprising:

2. The request acquisition unit receives the inference request through a predefined application programming interface (API); The result output unit transmits the inference result to the wireless terminal device through the predefined API. The radio base station device according to claim 1 .

3. a model request unit that transmits a model request including the identification information to a server device that stores a plurality of trained models; a model acquisition unit that receives the trained model corresponding to the identification information from the server device and stores the trained model in the model storage unit; The radio base station apparatus according to claim 1 , further comprising:

4. the model request unit determines whether the trained model is stored in the model storage unit; If the trained model is not stored in the model storage unit, the model request unit transmits the identification information to the server device, and the model acquisition unit receives the trained model from the server device and stores it in the model storage unit. The radio base station device according to claim 3.

5. When the trained model is stored in the model storage unit, the model request unit does not transmit the identification information to the server device, and the inference execution unit inputs the input data into the trained model stored in the model storage unit to obtain an inference result for the task. The radio base station device according to claim 4.

6. If the trained model is stored in the model storage unit, the model request unit inquires of the server device about whether the trained model has been updated; When the trained model has been updated, the model acquisition unit receives the updated trained model from the server device and stores it in the model storage unit, and the inference execution unit inputs the input data into the updated trained model stored in the model storage unit to obtain an inference result for the task. The radio base station device according to claim 4.

7. A wireless communication system including a plurality of wireless base station devices, One of the plurality of radio base station devices, receiving an inference request made by a wireless terminal device, the inference request including input data to input into a trained model corresponding to a task and identification information related to the task or the trained model; determining one or more wireless base station devices from the plurality of wireless base station devices to execute inference of the task; causing the one or more wireless base station devices to input the input data into the trained model stored in each of the one or more wireless base station devices based on the identification information, and to obtain an inference result for the task; receiving an inference result of the task from the one or more wireless base station devices; transmitting the inference result to the wireless terminal device; Wireless communication system.

8. the one wireless base station device, receiving information about traffic volumes of the plurality of wireless base station devices; determining the one or more wireless base station devices based on information about the traffic volume; 8. The wireless communication system according to claim 7.

9. the one wireless base station device, receiving information about the locations of the plurality of wireless base station devices; determining the one or more wireless base station devices based on the information about the locations; 8. The wireless communication system according to claim 7.

10. A wireless communication system having a wireless terminal device and a wireless base station device, The wireless terminal device transmitting an inference request to the wireless base station device, the inference request including input data to be input into a trained model corresponding to a task and identification information related to the task or the trained model; the radio base station device, receiving the inference request; based on the identification information, inputting the input data into the trained model stored in the wireless base station device to obtain an inference result for the task; transmitting the inference result to the wireless terminal device; Wireless communication system.

11. The wireless base station device receiving an inference request made by a wireless terminal device, the inference request including input data to input into a trained model corresponding to a task and identification information related to the task or the trained model; based on the identification information, inputting the input data into the trained model stored in the wireless base station device to obtain an inference result for the task; transmitting the inference result to the wireless terminal device; Wireless communication method.

Images