Information processing system, training device, training method, wireless base station, information processing method, and program

Abstract

Provided is a training device comprising: a training data storage unit that stores training data including a plurality of reference signals that are used for estimating an uplink channel state and that have been received by a wireless base station from a user terminal in a reference reception cycle; and a model generation unit that generates, through machine learning by using the training data, a first AI model that infers, from a plurality of reference signals having a long reception cycle longer than the reference reception cycle, the following reference signal in the reference reception cycle of the last reference signal among said plurality of reference signals, and a second AI model that infers, from the plurality of reference signals corresponding to the long reception cycle and the following reference signal in the reference reception cycle of the last reference signal among said plurality of reference signals, the second following reference signal in the reference reception cycle of said following reference signal.

Description

Information processing system, learning device, learning method, wireless base station, information processing method, and program

[0001] This invention relates to an information processing system, a learning device, a learning method, a wireless base station, an information processing method, and a program.

[0002] Patent Document 1 states, "UE 100 transmits the full SRS to gNB 200 according to the SRS transmission settings (step S201). The receiving unit 220 of gNB 200 receives the full SRS. In learning mode, the CSI generation unit 231 generates (or estimates) CSI based on the full SRS. The data collection unit A1 collects the full SRS and CSI. The model learning unit A2 creates a trained model using the full SRS and CSI as training data." It also states, "gNB 200 identifies the SRS transmission pattern (puncture pattern) to be input as inference data to the trained model and sets the identified SRS transmission pattern in UE 100. gNB 200 may also transmit the SRS transmission settings including the identified SRS transmission pattern to gNB 200." Patent Document 2 states that "Estimation Model M is a model that, when input data INP#n is input, which is generated from time-series data SRS#n of received power RP#n of wireless communication between the wireless base station 1 and the user terminal 2#n, outputs an estimated result of the factors that degrade the communication quality between the wireless base station 1 and the user terminal 2#n. For this reason, the model building device 4 uses time-series data SRS#n of received power RP#n of wireless communication between the wireless base station 1 and the user terminal 2#n, and the wireless base station 1 and user terminal 2#n under the circumstances in which the time-series data SRS#n was observed and The description states that machine learning is performed to construct an estimation model M using a training dataset LDS which includes training data LD#n containing the correct label y#n that indicates the actual degradation factors of communication quality during the period. Specifically, the model construction device 4 generates input data INP#n based on the time series data SRS#n included in the training data LD#n, inputs the generated input data INP#n into the estimation model M, and updates the parameters of the estimation model M based on a loss function relating to the error between the correct label y#n included in the training data LD#n and the output of the estimation model M. [Prior Art Documents] [Patent Documents] [Patent Document 1] International Publication No. 2024 / 166876 [Patent Document 2] International Publication No. 2022 / 195755

[0003] In Time Division Duplex (TDD), the same frequency band is used for both the downlink (DL) and the uplink (UL), so channel information calculated from the uplink reference signal can be used for downlink precoding. Downlink precoding is used, for example, for beamforming from the base station to the user terminal. In wireless communication systems such as LTE (Long Term Evolution) and NR (New Radio), SRS (Sounding Reference Signal) is used as the uplink reference signal. In TDD, the same frequency is switched between the downlink and uplink in time division, so the base station allocates uplink resources to the user terminal so that the user terminal transmits SRS at a predetermined timing. There is a maximum number of user terminals that can transmit SRS at the same time. If more user terminals than this maximum are connected to the wireless base station, the SRS transmission interval between user terminals will increase. For example, if the SRS transmission interval is 5 ms when fewer user terminals than the maximum number are connected to the wireless base station, the SRS transmission interval will increase to 40 ms or more when more user terminals than the maximum number are connected to the wireless base station. When the SRS transmission interval increases, beamforming that can handle channel fluctuations due to user terminal movement becomes difficult. The information processing system according to this embodiment provides a technology that contributes to solving these problems.

[0004] According to one embodiment of the present invention, an information processing system is provided. The information processing system may comprise a learning apparatus and a radio base station. The learning apparatus may comprise a storage unit that stores learning data including a plurality of reference signals, which are received by the radio base station from a user terminal at a reference reception period and used for estimating an uplink channel state. Using the learning data, the learning apparatus obtains, from a plurality of reference signals having a long reception period longer than the reference reception period, a first AI model that infers a next reference signal in the reference reception period after the last reference signal among the plurality of reference signals; and the learning apparatus may comprise a model generation unit that generates by machine learning a second AI model which infers a next reference signal in the reference reception period after the aforementioned next reference signal, from a plurality of reference signals corresponding to the long reception period and the next reference signal in the reference reception period after the last reference signal among the plurality of reference signals. The radio base station selectively uses the first AI model and the second AI model generated by the model generation unit, and may comprise an inference unit that infers a next reference signal in the reference reception period after the last reference signal among the plurality of reference signals, from a plurality of reference signals including one or more reference signals received from a user terminal that has established a communication connection. The radio base station may comprise a state estimation unit that estimates a channel state with the user terminal using the reference signal inferred by the inference unit.

[0005] In the information processing system, the reference signal may be SRS (Sounding Reference Signal).

[0006] In the information processing system, the inference unit selectively uses the first AI model and the second AI model generated by the model generation unit, and may infer a next reference signal in the reference reception period after the last reference signal among the plurality of reference signals, from the plurality of reference signals including one or more reference signals received from a user terminal that has established a communication connection and one or more reference signals inferred in the past.

[0007] In the information processing system, the model generation unit may further generate a third AI model that infers the next reference signal for the subsequent reference signals in the reference reception period of a plurality of reference signals from a single reference signal corresponding to the long reception period and a plurality of subsequent reference signals in the reference reception period of the single reference signal. The inference unit may selectively use the first AI model, the second AI model, and the third AI model generated by the model generation unit to infer the next reference signal for the plurality of reference signals in the reference reception period of the plurality of reference signals from a plurality of reference signals, including one or more reference signals received from a user terminal with which the communication connection has been established.

[0008] In the information processing system, the model generation unit may further generate a fourth AI model using the learning data to infer the next reference signal in the reference reception cycle of the last reference signal among a plurality of reference signals with consecutive reference reception cycles. The inference unit stores the reference signals inferred using the first AI model and the second AI model selectively. If inference using the fourth AI model is possible using one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, the fourth AI model may infer the next reference signal in the reference reception cycle of the last reference signal among a plurality of reference signals with consecutive reference reception cycles, which are composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored.

[0009] In the information processing system, the model generation unit may further generate a plurality of fifth AI models using the learning data to infer a target reference signal, which is the next reference signal in the reference reception cycle of the last reference signal, which is the last reference signal among the first number of reference signals, from a predetermined first number of reference signals in the reference reception cycle. The inference unit stores the reference signals inferred using the first AI model and the second AI model selectively. If it is possible to perform inference using any of the plurality of fifth AI models based on one or more reference signals received from the user terminal with which the communication connection has been established and the stored one or more inferred reference signals, the inference unit may use the executable fifth AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from one or a plurality of duplicated reference signals received from the user terminal. The plurality of fifth AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0010] According to one embodiment of the present invention, an information processing system is provided. The information processing system may include a learning device and a wireless base station. The learning device may include a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by the wireless base station from a user terminal in a reference reception cycle. The learning device may include a model generation unit that uses the learning data to generate an AI model by machine learning that infers the next reference signal in the reference reception cycle of the last of the plurality of reference signals from a plurality of consecutive reference signals in the reference reception cycle. The wireless base station may include a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals. The wireless base station may include an inference unit that, if it is possible to perform inference using the AI ​​model based on one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, uses the AI ​​model to infer the next reference signal in the reference reception cycle of the last reference signal among a plurality of reference signals that have a continuous reference reception cycle, which are composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored.

[0011] According to one embodiment of the present invention, an information processing system is provided. The information processing system may include a learning device and a wireless base station. The learning device may include a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by the wireless base station from a user terminal in a reference reception cycle. The learning device may include a model generation unit that uses the learning data to generate a plurality of AI models by machine learning that infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, which is the last reference signal among the first number of reference signals. The wireless base station may include a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals. The wireless base station may include an inference unit that, if it is possible to perform inference using one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, uses the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored. The plurality of AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0012] According to one embodiment of the present invention, a learning device is provided. The learning device may include a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal at a reference reception cycle. The learning device may include a model generation unit that uses the learning data to generate a first AI model that infers the next reference signal at the reference reception cycle of the last of a plurality of reference signals with a long reception cycle longer than the reference reception cycle, and a second AI model that infers the next reference signal at the reference reception cycle of the next reference signal from a plurality of reference signals corresponding to the long reception cycle and the next reference signal at the reference reception cycle of the last of the plurality of reference signals.

[0013] According to one embodiment of the present invention, a learning device is provided. The learning device may include a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal in a reference reception cycle. The learning device may also include a model generation unit that uses the learning data to generate an AI model by machine learning that infers the next reference signal in the reference reception cycle of the last of the plurality of reference signals from a plurality of consecutive reference signals in the reference reception cycle.

[0014] According to one embodiment of the present invention, a learning device is provided. The learning device may include a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal in a reference reception cycle. The learning device may include a model generation unit that uses the learning data to generate a plurality of AI models by machine learning, which infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, to the last reference signal, which is the final reference signal, among the first number of reference signals. The plurality of AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0015] According to one embodiment of the present invention, a learning method performed by a computer is provided. The learning method may include a model generation step of generating, by machine learning, a first AI model that uses learning data including a plurality of reference signals used to estimate the channel state of an uplink received by a wireless base station from a user terminal at a reference reception period, to infer the next reference signal of the last of the plurality of reference signals at the reference reception period from a plurality of long reception period reference signals with a period longer than the reference reception period, and a second AI model that infers the next reference signal of the next reference signal at the reference reception period from a plurality of reference signals corresponding to the long reception period and the next reference signal of the last of the plurality of reference signals at the reference reception period.

[0016] According to one embodiment of the present invention, a learning method performed by a computer is provided. The learning method may include a model generation step of generating an AI model by machine learning, which uses learning data including a plurality of reference signals used to estimate the channel state of an uplink, received by a wireless base station from a user terminal in a reference reception cycle, to infer the next reference signal in the reference reception cycle of the last of the plurality of reference signals from a plurality of consecutive reference signals in the reference reception cycle.

[0017] According to one embodiment of the present invention, a learning method performed by a computer is provided. The learning method may include a model generation step of generating a plurality of AI models by machine learning, which use learning data containing a plurality of reference signals used to estimate the channel state of an uplink received by a wireless base station from a user terminal in a reference reception cycle, to infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, where the final reference signal is the last reference signal among the first number of reference signals. The plurality of AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0018] According to one embodiment of the present invention, a program is provided for causing a computer to execute the learning method.

[0019] According to one embodiment of the present invention, a wireless base station is provided. The wireless base station may include a model acquisition unit that acquires a first AI model which infers the next reference signal in the reference reception period of the last reference signal among a plurality of long reception period reference signals, which have a longer reception period than the reference reception period, from a plurality of long reception period reference signals that have a longer period than the reference reception period, generated using training data which includes a plurality of reference signals used to estimate the channel state of an uplink that the wireless base station receives from a user terminal at a reference reception period; and a second AI model which infers the next reference signal in the reference reception period of the next reference signal, from a plurality of reference signals corresponding to the long reception period and the next reference signal in the reference reception period of the last reference signal among the plurality of reference signals. The wireless base station may include an inference unit which selectively uses the first AI model and the second AI model to infer the next reference signal in the reference reception period of the last reference signal among a plurality of reference signals, which include one or more reference signals received from a user terminal with which a communication connection has been established. The wireless base station may include a state estimation unit which estimates the channel state between the wireless base station and the user terminal using the reference signals inferred by the inference unit.

[0020] According to one embodiment of the present invention, a wireless base station is provided. The wireless base station may include a model acquisition unit that acquires an AI model for inferring the next reference signal in the reference reception period from a plurality of reference signals in a plurality of reference signals in a plurality of reference signals in the reference reception period, which is generated using training data that includes a plurality of reference signals used to estimate the channel state of an uplink, received by the wireless base station from a user terminal in a reference reception period. The wireless base station may include a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals. If the wireless base station is able to perform inference using the AI ​​model with one or more reference signals received from the user terminal with which a communication connection has been established and one or more inferred reference signals in storage, the wireless base station may include an inference unit that uses the AI ​​model to infer the next reference signal in the reference reception period from a plurality of reference signals in, which are composed of one or more reference signals received from the user terminal and one or more inferred reference signals in storage.

[0021] According to one embodiment of the present invention, a wireless base station is provided. The wireless base station may include a model acquisition unit that acquires a plurality of AI models that infer a target reference signal, which is the next reference signal in the reference reception cycle of a predetermined first number of reference signals, from a first number of reference signals that are the last reference signal among the first number of reference signals, which is the final reference signal. The model acquisition unit generates a plurality of AI models that infer a target reference signal, which is the next reference signal in the reference reception cycle of a final reference signal, which is the last reference signal among the first number of reference signals. The wireless base station may include a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals. The wireless base station may include an inference unit that, if it is possible to perform inference using one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, uses the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored. The plurality of AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0022] According to one embodiment of the present invention, an information processing method performed by a computer is provided. The information processing method may include a model acquisition step of acquiring a first AI model that infers the next reference signal of the last reference signal in the reference reception period from a plurality of long reception period reference signals, which have a longer reception period than the reference reception period, from a plurality of reference signals that have a longer reception period than the reference reception period, and which are generated using training data that includes a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal in a reference reception period; and a second AI model that infers the next reference signal of the next reference signal in the reference reception period from a plurality of reference signals corresponding to the long reception period and the next reference signal of the last reference signal in the reference reception period. The information processing method may include an inference step of selectively using the first AI model and the second AI model to infer the next reference signal of the last reference signal in the reference reception period from a plurality of reference signals, which include one or more reference signals received from a user terminal with which a communication connection has been established. The information processing method may include a state estimation step in which the channel state between the user terminal and the reference signal inferred by the inference unit is estimated.

[0023] According to one embodiment of the present invention, an information processing method performed by a computer is provided. The information processing method may include a model acquisition step of acquiring an AI model that infers the next reference signal in the reference reception cycle from a plurality of reference signals in

[0024] According to one embodiment of the present invention, an information processing method performed by a computer is provided. The information processing method may include a model acquisition step of acquiring a plurality of AI models that infer a target reference signal, which is the next reference signal in the reference reception cycle of a predetermined first number of reference signals, from a first number of reference signals, which is the last reference signal among the first number of reference signals, from the first number of reference signals, which is the final reference signal. The information processing method may include a storage step of storing one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals, in a storage unit. The information processing method may include an inference step in which, if it is possible to perform inference using one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, the inference step of using the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored. The plurality of AI models may have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

[0025] According to one embodiment of the present invention, a program is provided for causing a computer to execute the information processing method.

[0026] It should be noted that the above summary of the invention does not enumerate all the necessary features of the present invention. Furthermore, subcombinations of these features may also constitute an invention.

[0027] An example of the information processing system 10 is shown in outline. This is an explanatory diagram for explaining the challenges of SRS. This is an explanatory diagram for explaining the challenges of SRS. This is an explanatory diagram for explaining the training data. This is an explanatory diagram for explaining examples of training and inference. This is an explanatory diagram for explaining examples of training and inference. This is an explanatory diagram for explaining examples of training and inference. This is an explanatory diagram for explaining examples of training and inference. This is an explanatory diagram for explaining the no-problem pattern. This is an explanatory diagram for explaining the problematic pattern. This is an explanatory diagram for explaining an example of inference by the wireless base station 200. This is an explanatory diagram for explaining an example of inference by the wireless base station 200. This is an explanatory diagram for explaining an example of training and inference for pattern Aa. This is an explanatory diagram for explaining an example of training and inference for pattern Ab. This is an explanatory diagram for explaining an example of training and inference for pattern Ac. This is an explanatory diagram for explaining an example of inference for pattern Ba. This is an explanatory diagram for explaining an example of inference for pattern Bb. This is an explanatory diagram for explaining an example of inference for pattern Bc. This is an explanatory diagram for explaining an example of training and inference for pattern Ca. This is an explanatory diagram for explaining an example of training and inference for pattern Cb. This is an explanatory diagram for explaining an example of training and inference for pattern Cc. A schematic example of the functional configuration of the learning device 100 is shown. A schematic example of the functional configuration of the wireless base station 200 is shown. A schematic example of the processing flow by the wireless base station 200 is shown. A schematic example of an environment to which the information processing system 10 is applied is shown. A schematic example of the hardware configuration of the computer 1200 that functions as the learning device 100, wireless base station 200, management infrastructure 500, or distributed infrastructure 600 is shown.

[0028] The present invention will be described below through embodiments, but these embodiments are not intended to limit the scope of the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0029] Figure 1 schematically shows an example of an information processing system 10. The information processing system 10 includes a learning device 100. The information processing system 10 includes a wireless base station 200. The information processing system 10 may include multiple wireless base stations 200.

[0030] The learning device 100 and the wireless base station 200 communicate via the network 20. The network 20 may include a mobile communication core network. The learning device 100 may be located on the mobile communication core network. The network 20 may include the core network and the internet. In this case, the learning device 100 may be located on the internet.

[0031] The wireless base station 200 provides mobile communication services to multiple user terminals 300. The user terminals 300 can be any type of terminal as long as they are capable of using the mobile communication services. Examples of user terminals 300 include, but are not limited to, smartphones, tablet terminals, PCs (Personal Computers), mobile Wi-Fi (registered trademark), wearable devices such as smartwatches and smart glasses, IoT (Internet of Things) terminals such as sensors, smart home appliances, and smart meters, in-vehicle devices, robots, and game consoles. User terminals 300 may include any terminal that falls under the category of IoE (Internet of Everything).

[0032] The wireless base station 200 according to this embodiment performs communication with the user terminal 300 using the TDD method at a minimum. In this embodiment, among the various functions of the wireless base station 200, the function of utilizing the channel state estimated from the SRS received from the user terminal 300 for the downlink will be described in detail. The SRS may be an example of a reference signal used to estimate the channel state of the uplink. Examples of utilizing the channel information estimated from the SRS for the downlink include, but are not limited to, utilization for beamforming, utilization for MCS (Modulation and Coding Scheme), and utilization for power control based on interference prediction. In this embodiment, the utilization for beamforming will be described as an example.

[0033] Figures 2 and 3 are explanatory diagrams illustrating the challenges of the SRS400.

[0034] User terminals 300 can only transmit SRS 400 from special slots (S slots). There is a limit to the number of user terminals 300 that can transmit SRS 400 from a single special slot. When a large number of user terminals 300 are connected to the wireless base station 200, the transmission cycle of SRS 400 by each user terminal 300 becomes longer, that is, the reception cycle of SRS 400 received by the wireless base station 200 from each user terminal 300 becomes longer. When the reception cycle of SRS 400 becomes longer, the frequency of channel state estimation decreases, making it difficult to perform beamforming in response to channel fluctuations caused by the movement of user terminals 300, etc.

[0035] Figure 2 illustrates the status of SRS 400 received by the wireless base station 200 from a particular user terminal 300 when the number of user terminals 300 connected to the wireless base station 200 is relatively small. In the example shown in Figure 2, the wireless base station 200 receives SRS 400 from the user terminal 300 in every special slot.

[0036] Figure 3 shows the status of SRS 400 received by the wireless base station 200 from a particular user terminal 300 when the number of user terminals 300 connected to the wireless base station 200 is relatively large. Nothing 402 indicates that the wireless base station 200 has not received SRS 400 from that user terminal 300. In the example shown in Figure 3, the reception period of SRS 400 is eight times longer compared to the example shown in Figure 2.

[0037] As shown in Figures 2 and 3, if the SRS 400 is received frequently, the beam direction can be determined in response to channel fluctuations caused by the movement of the user terminal 300, and communication with the user terminal 300 can be performed appropriately. On the other hand, if the frequency of SRS 400 reception decreases, the response to channel fluctuations caused by the movement of the user terminal 300 will be delayed, making it difficult to perform communication with the user terminal 300 appropriately.

[0038] In contrast, the information processing system 10 according to this embodiment performs SRS inference using AI. When the information processing system 10 fails to receive SRS 400, the learning device 100 generates an AI model for inferring SRS 400 at that time, and when the wireless base station 200 fails to receive SRS 400, it uses this AI model to infer SRS 400 at that time. This improves the accuracy of downlink beamforming.

[0039] Figure 4 is an explanatory diagram illustrating the learning data used by the learning device 100. The learning device 100 acquires learning data that includes multiple SRS 400s received by a wireless base station 200 from a user terminal 300 at a reference reception cycle. The learning data may include multiple SRS 400s that a particular wireless base station 200 has previously received from a user terminal 300 at a reference reception cycle. The learning data may also include multiple SRS 400s that each of multiple wireless base stations 200 has previously received from a user terminal 300 at a reference reception cycle.

[0040] Figure 4 illustrates a case where the reference reception period is the period during which SRS 400 is received in each special slot, but it is not limited to this. The reference reception period may be the period during which SRS 400 is received in every other special slot. The reference reception period may be the shortest period among multiple SRS 400s received by the wireless base station 200 from the user terminal 300 at various intervals. In this embodiment, the case where the reference reception period is the period during which SRS 400 is received in each special slot will be mainly used as an example for explanation.

[0041] The learning device 100 may use the learning data to generate an AI model by machine learning that infers SRSs corresponding to reception periods shorter than the long reception period, which are later than the multiple SRSs with long reception periods longer than the reference reception period. For example, the learning device 100 uses the learning data to generate an AI model by machine learning that infers SRSs corresponding to the reference reception period, which are later than the multiple SRSs with long reception periods, which are later than the multiple SRSs.

[0042] FIG. 5 is an explanatory diagram for explaining an example of learning and an example of inference. FIG. 5 exemplifies a case where a reference reception cycle 410 is a cycle of receiving SRS 400 in each special slot, and a long reception cycle 420 is a cycle twice as long as the reference reception cycle 410.

[0043] In the example shown in FIG. 5, the learning device 100 receives three SRS 400 corresponding to the long reception cycle 420 among the plurality of SRS 400 of the reference reception cycle 410 as inputs, and performs machine learning with the next SRS 400 after the last SRS 400 among the three SRS 400 in the reference reception cycle 410 as the correct answer, thereby generating an AI model that infers the next SRS 400 after the last SRS 400 among the three SRS 400 in the reference reception cycle 410 from the three SRS 400 of the long reception cycle 420.

[0044] The radio base station 200 uses the AI model generated by the learning device 100 when inferring the SRS 400 at the inference target timing shown in FIG. 5. The inference target timing may be the timing of a special slot for which the SRS 400 is to be inferred. The radio base station 200 inputs three SRS 400 received before the inference target timing to the AI model, and acquires the SRS 400 output from the AI model as an inference result. When the learning device 100 generates such an AI model, even if the number of connections of user terminals 300 to the radio base station 200 increases, so that the SRS 400 cannot be received in the reference reception cycle 410 and the SRS 400 can only be received every other special slot, the missing SRS 400 can be inferred. This can contribute to improving the accuracy of downlink beamforming.

[0045] Note that the number of SRS 400 used as inputs to the AI model is not limited to three, and may be two, or may be four or more.

[0046] Figures 6 and 7 are explanatory diagrams for explaining an example of learning and an example of inference. In Figures 6 and 7, an example is illustrated in which a reference reception period 410 is a period for receiving SRS 400 in each special slot, and a long reception period 422 is a period four times the length of the reference reception period 410. INF 404 represents the SRS 400 inferred by the learning apparatus 100.

[0047] As shown in Learning Example A in FIG. 6, the learning apparatus 100 receives three SRSs 400 corresponding to the long reception period 422 among a plurality of SRSs 400 in the reference reception period 410 as inputs, and performs machine learning with the next SRS 400 in the reference reception period 410 after the last SRS 400 among the three SRSs 400 as a correct label. Accordingly, as shown in Inference Example A in FIG. 7, an AI model (which may be referred to as AI model α) that infers the next SRS 400 in the reference reception period 410 after the last SRS 400 among the three SRSs 400 from the three SRSs 400 in the long reception period 422 is generated.

[0048] As shown in Learning Example B in FIG. 6, the learning apparatus 100 receives three SRSs 400 corresponding to the long reception period 422 among a plurality of SRSs 400 in the reference reception period 410 as inputs, and performs machine learning with the second next SRS 400 in the reference reception period 410 after the last SRS 400 among the three SRSs 400 as a correct label. Accordingly, as shown in Inference Example B in FIG. 7, an AI model (which may be referred to as AI model β) that infers the second next SRS 400 in the reference reception period 410 after the last SRS 400 among the three SRSs 400 from the three SRSs 400 in the long reception period 422 is generated.

[0049] As shown in learning example C in Figure 6, the learning device 100 takes three SRS 400s corresponding to the long reception period 422 from among a plurality of SRS 400s of the reference reception period 410 as input, and performs machine learning with the next-next-next SRS 400 of the last of the three SRS 400s in the reference reception period 410 as the correct answer. As shown in inference example C in Figure 7, the device generates an AI model (sometimes referred to as AI model γ) that infers the next-next-next SRS 400 of the last of the three SRS 400s of the long reception period 422 from the three SRS 400s of the reference reception period 410.

[0050] In the case of inference example A in Figure 7, when the wireless base station 200 infers the SRS 400 at the timing to be inferred, it selects AI model α, inputs the three SRS 400s received before the timing to be inferred into AI model α, and obtains the SRS 400 output from AI model α as the inference result. Next, in the case of inference example B in Figure 7, when the wireless base station 200 infers the SRS 400 at the timing to be inferred, it selects AI model β, inputs the three SRS 400s received before the timing to be inferred into AI model β, and obtains the SRS 400 output from AI model β as the inference result. Next, in the case of inference example C in Figure 7, when the wireless base station 200 infers the SRS 400 at the timing to be inferred, it selects AI model γ, inputs the three SRS 400s received before the timing to be inferred into AI model γ, and obtains the SRS 400 output from AI model γ as the inference result.

[0051] The learning device 100 generates these AI models, and the wireless base station 200 uses these AI models to infer SRS 400 during long reception cycles 422 when SRS 400 cannot be received from the user terminal 300. In other words, during long reception cycles 422, it is possible to infer SRS 400 that can be received during the reference reception cycle 410 (in the example shown in Figure 7, SRS 400 corresponding to three Nothing 402 between two consecutive SRS 400). This contributes to improving the accuracy of channel fluctuation estimation.

[0052] The wireless base station 200 may continuously use the same AI model. In particular, the wireless base station 200 may continuously use the same AI model when the reception interval of the SRS 400 is relatively short.

[0053] Figure 8 is an explanatory diagram illustrating the case where the same AI model is continuously used to infer SRS400 when the reception interval of SRS400 is relatively short. Here, an example is shown where the reference reception period 410 is the period in which SRS400 is received in each special slot, and the long reception period 420 is the period in which SRS400 is received in every other special slot (i.e., the long reception period 420 is twice the reference reception period 410), and the next SRS400 in the reference reception period 410 for the last of the three SRS400s in the long reception period 420 is inferred.

[0054] Here, the wireless base station 200 continuously uses the AI ​​model shown in the learning example in Figure 5. At timing A, in order to infer the SRS 400 of the timing to be inferred, the wireless base station 200 inputs the SRS 400 received from the three preceding user terminals 300 into the AI ​​model and obtains the SRS 400 output from the AI ​​model as the inference result. At timing B, which follows timing A, the wireless base station 200 inputs the SRS 400 received from the three preceding user terminals 300 into the AI ​​model in order to infer the SRS 400 of the timing to be inferred, and obtains the SRS 400 output from the AI ​​model as the inference result.

[0055] In this case, the three SRS400s input to the AI ​​model for inferring SRS400 at timing A will be different from the three SRS400s input to the AI ​​model for inferring SRS400 at timing B, and the inference results may also be different, making it possible to adapt to channel fluctuations.

[0056] On the other hand, if the reception interval for SRS400 becomes relatively long, and the radio base station 200 continues to use the same AI model, the SRS400 during the reception interval will be inferred using the limited number of SRS400 received previously. This raises concerns that the inference results will be the same, making it impossible to adapt to channel fluctuations. This point will be explained with a specific example using Figure 9.

[0057] Figure 9 is an explanatory diagram illustrating the case where the same AI model is continuously used to infer SRS400 when the reception interval of SRS400 is relatively long. Here, the reference reception period 410 is the period in which SRS400 is received in each special slot, and the long reception period 422 is the period in which SRS400 is received every three special slots (i.e., the long reception period 422 is four times the reference reception period 410), and this example illustrates the case in which the subsequent SRS400 in the reference reception period 410 is inferred from the three SRS400 in the long reception period 420.

[0058] Here, the wireless base station 200 continuously uses the AI ​​model α shown in learning example A in Figure 6. At timing A, in order to infer the SRS 400 of the timing to be inferred, the wireless base station 200 inputs the SRS 400 received from the three preceding user terminals 300 into the AI ​​model α, and obtains the SRS 400 output from the AI ​​model α as the inference result. At timing B, which follows timing A, the wireless base station 200 inputs the SRS 400 received from the three preceding user terminals 300 into the AI ​​model α, in order to infer the SRS 400 of the timing to be inferred, and obtains the SRS 400 output from the AI ​​model α as the inference result.

[0059] In this case, the three SRS400 inputs to AI model α for inferring SRS400 at timing A and the three SRS400 inputs to AI model α for inferring SRS400 at timing B will be the same, and the inference results will also be the same. Therefore, it may be difficult to adapt to fine channel fluctuations.

[0060] In this embodiment, the wireless base station 200 may use the SRS 400 inferred using an AI model from the SRS 400 received from a user terminal 300 with which a communication connection has been established, as input to the AI ​​model for subsequent SRS 400 inference.

[0061] Figure 10 is an explanatory diagram illustrating an example of inference by the wireless base station 200. Here, the reference reception period 410 is the period for receiving the SRS 400 in each special slot, and the long reception period 422 is four times the period of the reference reception period 410.

[0062] At timing A, in order to infer the SRS 400 of the timing to be inferred, the wireless base station 200 uses the three SRS 400s with a long reception period 422 that it received from the user terminal 300 earlier to infer the SRS 400 of the timing to be inferred (the next SRS 400 in the reference reception period 410 of the last of the three SRS 400s).

[0063] At timing B, following timing A, the radio base station 200 uses one SRS 400 (INF 404) inferred at timing A and two SRS 400s of the long reception period 422 received earlier from the user terminal 300 to infer the SRS 400 of the timing to be inferred (the next SRS 400 of the INF 404 in the reference reception period 410).

[0064] At timing C following timing B, the wireless base station 200 uses the SRS 400 (INF 404) inferred at timing B, the SRS 400 (INF 404) inferred at timing A, and one SRS 400 previously received from the user terminal 300 to infer the next SRS 400 for the timing to be inferred (the next SRS 400 in the reference reception cycle 410 of the SRS 400 (INF 404) inferred at timing B).

[0065] By performing inference as shown in Figure 10, the wireless base station 200 can prevent the inference results from becoming the same when inferring multiple SRS 400s in a reference reception cycle between two consecutive SRS 400s in a long reception cycle, and can contribute to improving the accuracy of the SRS 400 inference results at each timing.

[0066] Figure 11 is an explanatory diagram illustrating an example of inference by the wireless base station 200. Here, the wireless base station 200 performs the inference shown in Figure 10, and describes an example of new SRS 400 inference by the wireless base station 200 in a state where the SRS 400 in the reference reception period has been inferred between two consecutive SRS 400 in the long reception period 422.

[0067] Pattern A is a pattern in which the next SRS400 in the reference reception cycle of the last SRS400 among a predetermined number of SRS400s is inferred from a predetermined number of SRS400s. The SRS400 used as input may be an SRS400 received from the user terminal 300, or an SRS400 (INF404) that has been inferred in the past. Figure 11 shows an example where the predetermined number is three, but it is not limited to three; it may be two, four or more, etc. In Pattern A shown in Figure 11, the radio base station 200 uses three SRS400s with a long reception cycle 422 that were previously received from the user terminal 300 to infer the SRS400 of the timing to be inferred (the next SRS400 in the reference reception cycle 410 of the last SRS400 among the three SRS400s).

[0068] Pattern B is a pattern in which inference is performed using a predetermined number of SRS400s immediately preceding the timing to be inferred. In this embodiment, the timing to be inferred may be described as Nothing402 of the inferred timing. The SRS400 used as input may be an SRS400 received from the user terminal 300, or an SRS400 (INF404) that has been inferred in the past. Figure 11 shows an example where the predetermined number is three, but it is not limited to three; it may be two, four or more, or any number. In Pattern B shown in Figure 11, the wireless base station 200 infers the SRS400 of the timing to be inferred using one SRS400 received from the user terminal 300 immediately preceding it, an SRS400 (INF404) that was inferred before that, and an SRS400 (INF404) that was inferred before that.

[0069] Pattern C is a pattern in which inference is performed using a predetermined number of SRS 400s received from a predetermined number of user terminals 300 immediately before the timing of the inference target, and a predetermined number of inferred SRS 400s (INF 404s). Figure 11 shows an example where two SRS 400s are received from the user terminals 300 and three SRS 400s (INF 404s) are inferred, but it is not limited to these and other numbers may be used. In pattern C shown in Figure 11, the wireless base station 200 uses two received SRS400s and three inferred SRS400s to infer the SRS400 at the timing to be inferred. These SRS400s consist of the inferred SRS400 (INF404) immediately preceding the current SRS400, the SRS400 received from the previous user terminal 300, the inferred SRS400 (INF404) from the previous inferred SRS400 (INF404) from the previous inferred SRS400 (INF404) from the user terminal 300 before that.

[0070] The wireless base station 200 may infer the subsequent SRS400 using pattern A, pattern B, or pattern C.

[0071] Figures 12, 13, and 14 are explanatory diagrams illustrating an example of learning and inference for Pattern A. As described above, Pattern A is a pattern that infers the next SRS400 in the reference reception cycle of the last SRS400 among a predetermined number of SRS400s. The SRS400 used as input in Pattern A may be an SRS400 received from the user terminal 300 or an inferred SRS400 (INF404). Figures 12, 13, and 14 illustrate the case where the reference reception cycle 410 is the cycle for receiving an SRS400 in each special slot, and the long reception cycle 422 is four times the period of the reference reception cycle 410. In this case, the wireless base station 200 may use three types of AI models.

[0072] Figure 12 is an explanatory diagram illustrating an example of learning and inference for pattern Aa. In the example shown in Figure 12, the learning device 100 takes three SRS 400s corresponding to the long reception period 422 from among a plurality of SRS 400s of the reference reception period 410 included in the learning data as input, and performs machine learning with the next SRS 400 of the last of the three SRS 400s in the reference reception period 410 as the correct answer, thereby generating an AI model Aa that infers the next SRS 400 of the last of the three SRS 400s in the reference reception period 410 from the three SRS 400s of the long reception period 422. The wireless base station 200 uses the AI ​​model Aa generated by the learning device 100 to infer the next SRS 400 in the reference reception period 410 of the last of the three SRS 400s with a long reception period 422 received from the user terminal 300.

[0073] Figure 13 is an explanatory diagram illustrating an example of learning and inference of pattern Ab. In the example shown in Figure 13, the learning device 100 takes as input the next SRS 400 in the reference reception period 410 of two SRS 400s corresponding to the long reception period 422 from among the multiple SRS 400s of the reference reception period 410 included in the learning data, and the last SRS 400 of those two SRS 400s as the correct answer. By performing machine learning with the next SRS 400 in the reference reception period 410 of the last SRS 400 of the three input SRS 400s as the correct answer, the learning device 100 generates an AI model Ab that infers the next SRS 400 in the reference reception period 410 of INF 404 from the inferred SRS 400 (INF 404) of the two SRS 400s of the long reception period 422 and the last SRS 400 of those two SRS 400s. The wireless base station 200 uses the AI ​​model Ab generated by the learning device 100 to infer the next SRS 400 of INF 404 at the reference reception period 410 from two SRS 400s with long reception periods 422 received from the user terminal 300, and the next inferred SRS 400 (INF 404) at the reference reception period 410 of the last of the two SRS 400s.

[0074] Figure 14 is an explanatory diagram illustrating an example of learning and inference of pattern Ac. In the example shown in Figure 14, the learning device 100 takes three consecutive SRS 400s from among a plurality of SRS 400s of a reference reception period 410 included in the learning data as input, and performs machine learning with the next SRS 400 of the last of the three SRS 400s in the reference reception period 410 as the correct answer, thereby generating an AI model Ac that infers the next SRS 400 of the last of the three SRS 400s in the reference reception period 410 from three consecutive SRS 400s of the reference reception period 410. The wireless base station 200 uses the AI ​​model Ac generated by the learning device 100 to infer the next SRS 400 for the reference reception cycle 410 of the INF 404 from the SRS 400 received from the user terminal 300, the next inferred SRS 400 (INF 404) for the reference reception cycle 410 of the SRS 400, and the next inferred SRS 400 (INF 404) for the INF 404.

[0075] Here, an example of learning and inference has been described in the case where the long reception period 422 is four times the period of the reference reception period 410. However, the learning device 100 may generate multiple AI models of pattern A for each long reception period, and the radio base station 200 may use the AI ​​model of pattern A for each long reception period. By performing inference as shown in Figures 12, 13, and 14, the radio base station 200 can prevent the inference results from becoming the same when inferring multiple SRS 400s in the reference reception period between two consecutive SRS 400s in the long reception period, and can contribute to improving the accuracy of the SRS 400 inference results at each timing.

[0076] Figures 15, 16, and 17 are explanatory diagrams illustrating an example of learning and inference for Pattern B. As described above, Pattern B is a pattern that performs inference using a predetermined number of SRS400 immediately preceding the timing of the inference target. The SRS400 used as input in Pattern B may be an SRS400 received from the user terminal 300 or an inferred SRS400 (INF404). Figures 15, 16, and 17 illustrate the case where the reference reception period 410 is the period for receiving an SRS400 in each special slot, and the long reception period 422 is four times the period of the reference reception period 410.

[0077] In pattern B, the learning device 100 takes a plurality of consecutive SRS 400s from among the plurality of SRS 400s of a reference reception period 410 included in the learning data as input, and performs machine learning with the next SRS 400 of the last SRS 400 among the plurality of SRS 400s in the reference reception period 410 as the correct answer, thereby generating an AI model B that infers the next SRS 400 of the last SRS 400 among the plurality of SRS 400s of a reference reception period 410 from a plurality of consecutive SRS 400s of a reference reception period 410. The learning device 100 may also use AI model Ac as AI model B. Here, we will explain using the case where there are three SRS 400 inputs to AI model B as an example, but there may be two SRS 400 inputs to AI model B, or four or more.

[0078] Figure 15 is an explanatory diagram illustrating an example of inference for pattern Ba. The AI ​​model used by the wireless base station 200 is AI model B. In pattern Ba, the wireless base station 200 uses AI model B generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402 from the SRS 400 received from the user terminal 300 immediately preceding the Nothing 402 to be inferred, the SRS 400 (INF 404) inferred before that, and the SRS 400 (INF 404) inferred before that.

[0079] Figure 16 is an explanatory diagram illustrating an example of inference for pattern Bb. The AI ​​model used by the wireless base station 200 is AI model B. In pattern Bb, the wireless base station 200 uses AI model B generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402 from the inferred SRS 400 (INF 404) immediately preceding the inferred Nothing 402, the SRS 400 received from the user terminal 300 before that, and the inferred SRS 400 (INF 404) before that.

[0080] Figure 17 is an explanatory diagram illustrating an example of inference for pattern Bc. The AI ​​model used by the wireless base station 200 is AI model B. In pattern Bc, the wireless base station 200 uses AI model B generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402, using the inferred SRS 400 (INF 404) immediately preceding the inferred SRS 400 (INF 404) before that, and the SRS 400 received from the user terminal 300 before that.

[0081] By performing inference as shown in Figures 15, 16, and 17, the wireless base station 200 can prevent the inference results from becoming the same when inferring multiple SRS 400s in a reference reception cycle between two consecutive SRS 400s in a long reception cycle, and can contribute to improving the accuracy of the SRS 400 inference results at each timing.

[0082] Figures 18, 19, and 20 are explanatory diagrams illustrating an example of learning and inference for pattern C. As described above, pattern C is a pattern that uses a predetermined number of SRS 400s received from user terminals 300 immediately before the timing of the inference target, and a predetermined number of inferred SRS 400s (INF 404). Figures 18, 19, and 20 illustrate the case where two SRS 400s are received from user terminals 300 and three SRS 400s (INF 404) are inferred. Figures 18, 19, and 20 illustrate the case where the reference reception period 410 is the period for receiving SRS 400 in each special slot, and the long reception period 422 is four times the period of the reference reception period 410.

[0083] Figure 18 is an explanatory diagram illustrating an example of learning and inference of pattern Ca. In the example shown in Figure 18, the learning device 100 takes five consecutive SRS 400s from among a plurality of SRS 400s of a reference reception period 410 included in the learning data as input, and performs machine learning with the next SRS 400 of the last SRS 400 of the five SRS 400s in the reference reception period 410 as the correct answer, thereby generating an AI model Ca that infers the next SRS 400 of the last SRS 400 of the five consecutive SRS 400s of the reference reception period 410 from the five consecutive SRS 400s of the reference reception period 410. The wireless base station 200 uses the AI ​​model Ca generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402 from five consecutive SRS 400s, each consisting of at least one of the SRS 400 received from the user terminal 300 and the inferred SRS 400 (INF 404) immediately preceding the Nothing 402 to be inferred. In the example shown in Figure 18, the wireless base station 200 inputs the SRS 400 received from the user terminal 300 immediately before the Nothing 402 to be inferred, the previously inferred SRS 400 (INF 404), the previously inferred SRS 400 (INF 404), the previously inferred SRS 400 (INF 404), and the SRS 400 received from the user terminal 300 before that into the AI ​​model Ca, and obtains the SRS 400 output from the AI ​​model Ca as the inference result.

[0084] Figure 19 is an explanatory diagram illustrating an example of learning and inference of pattern Cb. In the example shown in Figure 19, the learning device 100 takes four consecutive SRS 400s from among a plurality of SRS 400s in a reference reception period 410, and the SRS 400 two positions prior to the first SRS 400 of the four SRS 400s in the reference reception period 410 as inputs, and performs machine learning with the next SRS 400 of the last SRS 400 of the five input SRS 400s in the reference reception period 410 as the correct answer. In this way, the learning device 100 generates an AI model Cb that infers the next SRS 400 of the last SRS 400 of the five SRS 400s in the reference reception period 410 from the four consecutive SRS 400s in a reference reception period 410 and the SRS 400 two positions prior to the first SRS 400 of the four SRS 400s in the reference reception period 410. The wireless base station 200 uses the AI ​​model Cb generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402, which is the target of inference, from four consecutive SRS 400s, each consisting of at least one of the SRS 400 received from the user terminal 300 immediately preceding the Nothing 402 to be inferred, and the SRS 400 or inferred SRS 400 (INF 404) received from the user terminal 300 two reference reception cycles 410 prior to the first SRS 400 of the four SRS 400s. In the example shown in Figure 19, the wireless base station 200 inputs the inferred SRS 400 (INF 404) immediately preceding the inferred Nothing 402, the SRS 400 received from the user terminal 300 before that, the inferred SRS 400 (INF 404) before that, the inferred SRS 400 (INF 404) before that, and the SRS 400 received from the user terminal 300 two steps prior in the reference reception cycle 410 of the INF 404 into the AI ​​model Cb, and obtains the SRS 400 output from the AI ​​model Cb as the inference result.

[0085] Figure 20 is an explanatory diagram illustrating an example of learning and inference of pattern Cc. In the example shown in Figure 20, the learning device 100 takes four consecutive SRS 400s from among a plurality of SRS 400s in a reference reception period 410, and the SRS 400 three positions prior to the first of the four SRS 400s in the reference reception period 410 as input, and performs machine learning with the next SRS 400 in the reference reception period 410 of the last of the five input SRS 400s as the correct answer. In doing so, it generates an AI model Cb that infers the next SRS 400 in the reference reception period 410 of the last of the five SRS 400s from the four consecutive SRS 400s in a reference reception period 410, and the SRS 400 three positions prior to the first of the four SRS 400s in the reference reception period 410. The wireless base station 200 uses the AI ​​model Cb generated by the learning device 100 to infer the SRS 400 corresponding to Nothing 402, which is the target of inference, from four consecutive SRS 400s, each consisting of at least one of the SRS 400 received from the user terminal 300 and the inferred SRS 400 (INF 404) immediately preceding Nothing 402, and the SRS 400 or inferred SRS 400 (INF 404) received from the user terminal 300 three reference reception cycles 410 prior to the first SRS 400 of the four SRS 400s. In the example shown in Figure 20, the wireless base station 200 inputs the inferred SRS 400 (INF 404) immediately preceding the inferred Nothing 402, the inferred SRS 400 (INF 404) before that, the SRS 400 received from the user terminal 300 before that, the inferred SRS 400 (INF 404) before that, and the SRS 400 received from the user terminal 300 three steps prior in the reference reception cycle 410 of the INF 404 into the AI ​​model Cc, and obtains the SRS 400 output from the AI ​​model Cc as the inference result.

[0086] By performing inference as shown in Figures 18, 19, and 20, the wireless base station 200 can prevent the inference results from becoming the same when inferring multiple SRS 400s in a reference reception cycle between two consecutive SRS 400s in a long reception cycle, and can contribute to improving the accuracy of the SRS 400 inference results at each timing.

[0087] Figure 21 schematically shows an example of the functional configuration of the learning device 100. Here, the functional parts related to the functions of the learning device 100 according to this embodiment will be described. The learning device 100 includes a storage unit 102, a data acquisition unit 104, a model generation unit 106, and a model provision unit 108. The learning device 100 may also include other functional parts.

[0088] The data acquisition unit 104 acquires various types of data. The data acquisition unit 104 stores the acquired data in the storage unit 102.

[0089] The data acquisition unit 104 acquires, for example, a reference signal used to estimate the channel state of the uplink, which is received by the wireless base station 200 from the user terminal 300 at a reference reception cycle. In the example shown in Figure 21, the case in which SRS 400 is used as an example of a reference signal will be mainly described. The data acquisition unit 104 acquires, for example, an SRS 400 that a single wireless base station 200 has received from one or more user terminals 300 at a reference reception cycle. The data acquisition unit 104 acquires, for example, an SRS 400 that a wireless base station 200 has received from one or more user terminals 300 at a reference reception cycle from each of multiple wireless base stations 200. The data acquisition unit 104 may also acquire the SRS 400 that each of the multiple wireless base stations 200 has received from one or more user terminals 300 at a reference reception cycle from a core network device that manages the multiple wireless base stations 200. The data acquisition unit 104 stores the learning data, which includes multiple SRS 400s, in the storage unit 102.

[0090] The model generation unit 106 generates an AI model using machine learning with the training data stored in the memory unit 102.

[0091] The model generation unit 106 may generate an AI model of pattern A. That is, the model generation unit 106 generates an AI model that infers the next SRS 400 in the reference reception cycle of the last SRS 400 among a predetermined number of SRS 400s. The predetermined number may be two or more.

[0092] For example, the model generation unit 106 generates an AI model (sometimes referred to as the first AI model) that infers the next SRS in the reference reception period of the last SRS among a plurality of SRSs with long reception periods longer than the reference reception period. For example, the model generation unit 106 generates an AI model (sometimes referred to as the second AI model) that infers the next SRS in the reference reception period of the next SRS from a plurality of SRSs corresponding to long reception periods and the next SRS in the reference reception period of the last SRS among the plurality of SRSs. For example, the model generation unit 106 generates an AI model (sometimes referred to as the third AI model) that infers the next SRS in the reference reception period of the last SRS among the plurality of SRSs from one SRS corresponding to a long reception period and a plurality of SRSs following the first SRS in the reference reception period of that first SRS.

[0093] The model generation unit 106 may generate an AI model of pattern A that infers one SRS using a number of SRS corresponding to the difference between the reference reception period and the long reception period.

[0094] For example, if the long reception period is four times the reference reception period, the model generation unit 106 generates multiple AI models that infer one SRS using three SRSs. For example, the model generation unit 106 generates an AI model that infers the next SRS in the reference reception period for the last of the three SRSs from the three SRSs of the long reception period; an AI model that infers the next SRS in the reference reception period for the next SRS from two SRSs corresponding to the long reception period and the next SRS in the reference reception period for the last of the two SRSs; and an AI model that infers the next SRS in the reference reception period for the last of the two subsequent SRSs from one SRS corresponding to the long reception period and the two subsequent SRSs of that one SRS.

[0095] For example, if the long reception period is five times the reference reception period, the model generation unit 106 generates multiple AI models that infer one SRS using four SRSs. For example, the model generation unit 106 generates an AI model that infers the next SRS in the reference reception period of the last of the four SRSs with long reception periods from four SRSs with long reception periods, an AI model that infers the next SRS in the reference reception period of the next SRS from three SRSs corresponding to long reception periods and the next SRS in the reference reception period of the last of the three SRSs, an AI model that infers the next SRS in the reference reception period of the last of the two subsequent SRSs from two SRSs corresponding to long reception periods and the two subsequent SRSs in the reference reception period of those two SRSs, and an AI model that infers the next SRS in the reference reception period of the last of the three subsequent SRSs from one SRS corresponding to a long reception period and the three subsequent SRSs in the reference reception period of that one SRS.

[0096] For example, if the long reception period is six times the reference reception period, the model generation unit 106 generates multiple AI models that infer one SRS using five SRSs. For example, the model generation unit 106 generates an AI model that infers the next SRS in the reference reception period of the last of the five SRSs from the five SRSs of the long reception period, an AI model that infers the next SRS in the reference reception period of the next SRS from the four SRSs corresponding to the long reception period and the next SRS in the reference reception period of the last of the four SRSs, and an AI model that infers the next SRS in the reference reception period of the next SRS from the three SRSs corresponding to the long reception period and the next two SRSs in the reference reception period of the next two SRSs The system generates an AI model that infers the next SRS in the reference reception cycle of an SRS, an AI model that infers the next SRS in the reference reception cycle of the last of the three subsequent SRSs from two SRSs corresponding to long reception cycles and the three subsequent SRSs in the reference reception cycles of those two SRSs, and an AI model that infers the next SRS in the reference reception cycle of the last of the four subsequent SRSs from one SRS corresponding to a long reception cycle and the four subsequent SRSs in the reference reception cycle of that one SRS.

[0097] The model generation unit 106 may also generate multiple AI models when the long reception period is seven times or more the reference reception period.

[0098] The model generation unit 106 may generate an AI model of pattern B. That is, the model generation unit 106 takes a plurality of consecutive SRSs of a reference reception period as input and performs machine learning with the next SRS of the last SRS 400 among the plurality of SRSs in the reference reception period as the correct answer, thereby generating an AI model (sometimes referred to as the fourth AI model) that infers the next SRS of the last SRS among the plurality of SRSs in the reference reception period from a plurality of consecutive SRSs of a reference reception period.

[0099] The model generation unit 106 may generate multiple AI models of pattern C. That is, the model generation unit 106 may generate multiple AI models (sometimes referred to as multiple fifth AI models) that infer the target SRS, which is the next SRS in the reference reception cycle of the last SRS, which is the last SRS among the first number of SRSs, from a predetermined first number of SRSs in the reference reception cycle. The time arrangement of the SRSs other than the last SRS among the first number of SRSs may differ from one another in these multiple fifth AI models. For example, if the first number is 5, the model generation unit 106 may generate AI models for various time arrangements of SRS, such as an AI model that takes five consecutive SRS signals immediately preceding the timing to be inferred as input in the reference reception cycle, an AI model that takes four consecutive SRS signals immediately preceding the timing to be inferred and SRS signals prior to those four that are not consecutive with those four, an AI model that takes three consecutive SRS signals immediately preceding the timing to be inferred and two SRS signals prior to those three that are not consecutive with those three, an AI model that takes two consecutive SRS signals immediately preceding the timing to be inferred and three SRS signals prior to those two that are not consecutive with those two, and so on.

[0100] The model provision unit 108 provides the AI ​​model generated by the model generation unit 106. The model provision unit 108 may provide the AI ​​model generated by the model generation unit 106 to the wireless base station 200.

[0101] Figure 22 schematically shows an example of the functional configuration of the wireless base station 200. Here, the functional units related to the functions of the wireless base station 200 according to this embodiment will be described. The wireless base station 200 includes a storage unit 202, a communication control unit 204, a state estimation unit 206, a model acquisition unit 208, and an inference unit 210. The wireless base station 200 may also include other functional units.

[0102] The communication control unit 204 controls various communications. The communication control unit 204 may communicate with devices on the core network side. The communication control unit 204 may communicate with user terminals 300. The communication control unit 204 may establish communication connections with multiple user terminals 300 and provide mobile communication services to multiple user terminals 300.

[0103] The communication control unit 204 determines the transmission interval of the reference signal according to the number of user terminals 300 connected. In the example shown in Figure 22, the case in which SRS 400 is used as an example of the reference signal will be mainly described. The communication control unit 204 allocates uplink resources to the multiple user terminals 300 so that each of the multiple user terminals 300 transmits SRS 400 according to the determined transmission interval of SRS 400. For each of the multiple user terminals 300, the communication control unit 204 stores and retains the SRS 400 received from the user terminal 300 in the storage unit 202. The SRS 400 received from the user terminal 300 may be referred to as received SRS 430.

[0104] The state estimation unit 206 estimates the channel state for each of the user terminals 300. The state estimation unit 206 estimates the channel state of the uplink of the user terminal 300 using the SRS 400 stored in the storage unit 202. In this embodiment, since the communication control unit 204 performs communication with the user terminal 300 using the TDD method, the channel state of the uplink can be utilized for the downlink.

[0105] The communication control unit 204 generates a precode for each of the multiple user terminals 300 using the channel state estimated by the state estimation unit 206. The communication control unit 204 may generate a precode for beamforming of the downlink. Specifically, the communication control unit 204 generates a precode matrix suitable for the channel state. The communication control unit 204 may use the generated precode to communicate with the user terminals 300. The communication control unit 204 may generate a precode for each of the multiple user terminals 300 using any existing method other than SRS 400 until it receives SRS 400. Examples of any existing method other than SRS 400 include, but are not limited to, methods using CSI-RS (Channel State Information Reference Signal), methods using pilot signals, and methods using statistical information of received data.

[0106] The model acquisition unit 208 acquires an AI model. The model acquisition unit 208 stores the acquired AI model in the storage unit 202. The model acquisition unit 208 may acquire an AI model that has been generated by the model generation unit 106 and provided by the model provision unit 108.

[0107] The inference unit 210 infers one SRS400 from multiple SRS400s. The SRS400 inferred by the inference unit 210 may be referred to as the inferred SRS440. INF404 and inferred SRS440 may have the same meaning. The inference unit 210 stores and retains the inferred SRS440 in the storage unit 202.

[0108] The inference unit 210 does not perform inference if it can acquire SRS 400 at a reference reception cycle from a user terminal 300 with which a communication connection has been established. Instead, it may perform inference if the reception cycle of SRS 400 from the user terminal 300 with which a communication connection has been established becomes a long reception cycle that is longer than the reference reception cycle, and if the conditions are met. The conditions being met may be that the number of received SRS 430s held and the number of inferred SRS 440s held meet the number required to be input into the AI ​​model. For example, if the inference unit 210 uses an AI model that is generated to infer one SRS 400 from three SRS 400s with long reception cycles, it determines that the conditions are met if the number of SRS 400s with long reception cycles received from the user terminal 300 is three or more.

[0109] If the SRS 400 is acquired from the user terminal 300 at a reference reception cycle, the state estimation unit 206 may estimate the channel state using the received SRS 430 received from the user terminal 300, and the communication control unit 204 may generate a precode using the channel state estimated by the state estimation unit 206, and use the generated precode to perform communication with the user terminal 300.

[0110] If the reception period of the SRS 400 from the user terminal 300 becomes a long reception period which is longer than the reference reception period, and there are not yet enough received SRS 430s to use the AI ​​model of pattern A, the state estimation unit 206 may estimate the channel state using the most recent received SRS 430 from the user terminal 300, and the communication control unit 204 may generate a precode using the channel state estimated by the state estimation unit 206, and use the generated precode to perform communication with the user terminal 300.

[0111] After a sufficient number of received SRS 430s have been acquired to use the AI ​​model of pattern A, the inference unit 210 performs inference on the SRS 400 using the AI ​​model acquired by the model acquisition unit 208. For example, the inference unit 210 performs inference on the SRS 400 using the AI ​​model of pattern A. Here, as shown in Figure 10, we will explain using the case where the long reception period is four times the reference reception period and the inference unit 210 uses AI models Aa, Ab, and Ac as an example.

[0112] As illustrated in timing A of Figure 10, when the inference unit 210 is receiving SRS 400 with a reception period four times longer than the reference reception period, and has three received SRS 430, which is a sufficient number to use AI model Aa, it inputs the three received SRS 430 to AI model Aa in order to infer the SRS 400 corresponding to Nothing 402, the target of inference, and obtains the inferred SRS 440 output by AI model Aa. Then, the state estimation unit 206 estimates the channel state using the inferred SRS 440, and the communication control unit 204 generates precoding using the channel state and executes communication with the user terminal 300.

[0113] Next, as illustrated in timing B of Figure 10, the inference unit 210 inputs the inferred SRS 440 inferred at timing A and the two received SRS 430s from the previous long reception period into the AI ​​model Ab in order to infer the SRS 400 corresponding to Nothing 402, the next inference target, and obtains the inferred SRS 440 output from the AI ​​model Ab. Then, the state estimation unit 206 estimates the channel state using the inferred SRS 440, and the communication control unit 204 generates precoding using the channel state and executes communication with the user terminal 300.

[0114] Next, as illustrated in timing C of Figure 10, the inference unit 210 inputs the inferred SRS 440 inferred at timing B, the inferred SRS 440 inferred at timing A, and one received SRS 430 from the previous long reception period into the AI ​​model Ac in order to infer the SRS 400 corresponding to Nothing 402, the next inference target, and obtains the inferred SRS 440 output from the AI ​​model Ac. Then, the state estimation unit 206 estimates the channel state using the inferred SRS 440, and the communication control unit 204 generates precoding using the channel state and executes communication with the user terminal 300.

[0115] After inferring multiple SRSs corresponding to the reference reception period during a series of SRSs 400 with long reception periods by using the AI ​​model of pattern A, or by using other inference methods, the inference unit 210 may use pattern A, pattern B, or pattern C, as illustrated in Figure 11.

[0116] When using pattern A, the inference unit 210 may use an AI model of pattern A that infers one SRS using a number of SRS corresponding to the difference between the reference reception period and the long reception period.

[0117] For example, if the long reception period is four times the reference reception period, the inference unit 210 uses multiple AI models that infer one SRS using three SRSs. As a specific example, after receiving a received SRS 430, the inference unit 210 inputs the three preceding received SRSs 430, including the received SRS 430, into an AI model that infers the next SRS in the reference reception period of the last of the three SRSs from the three SRSs with a long reception period, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. Next, the inference unit 210 inputs the inferred SRS 440 and the two preceding received SRS 430 into an AI model that infers the next SRS in the reference reception cycle of the next SRS from two SRSs corresponding to a long reception cycle and the next SRS in the reference reception cycle of the last of the two SRSs, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. Next, the inference unit 210, in order to infer the next SRS 400 of the reference reception cycle of the inferred inference SRS 440, inputs the inferred SRS 440, the previous inference SRS 440, and the previous received SRS 430 into an AI model that infers the next SRS of the reference reception cycle of the last of the two subsequent SRSs from one SRS corresponding to a long reception cycle and the two subsequent SRSs in the reference reception cycle of that one SRS, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. By repeating the above, the inference unit 210 infers the SRS 400 corresponding to the reference reception cycle from the SRS 400 of the long reception cycle.

[0118] For example, if the long reception period is five times the reference reception period, the inference unit 210 uses multiple AI models that infer one SRS using four SRSs. As a specific example, after receiving a received SRS 430, the inference unit 210 inputs the four preceding received SRSs 430, including the received SRS 430, into an AI model that infers the next SRS in the reference reception period of the last of the four SRSs from the four SRSs with a long reception period, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. Next, the inference unit 210 inputs the inferred SRS 440 and the three preceding received SRS 430 into an AI model that infers the next SRS in the reference reception cycle of the next SRS from three SRSs corresponding to long reception cycles and the next SRS in the reference reception cycle of the last of the three SRSs, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. Next, the inference unit 210 inputs the inferred SRS 440, the previous inferred SRS 440, and the two previous received SRS 430s into an AI model that infers the next SRS in the reference reception cycle of the last of the two subsequent SRSs, based on two SRSs corresponding to a long reception cycle and the two subsequent SRSs in the reference reception cycle of those two SRSs. The inference unit then obtains the inferred SRS 440 output from the AI ​​model as the inference result. Next, the inference unit 210, in order to infer the next SRS 400 of the reference reception cycle of the inferred inference SRS 440, inputs the inferred SRS 440, the previous inference SRS 440, the previous inference SRS 440, and the previous received SRS 430 into an AI model that infers the next SRS of the reference reception cycle of the last of the three SRSs that follow, based on one SRS corresponding to a long reception cycle and the three SRSs that follow in the reference reception cycle of that one SRS, and obtains the inferred SRS 440 output from the AI ​​model as the inference result. By repeating the above, the inference unit 210 infers the SRS 400 corresponding to the reference reception cycle from the SRS 400 of the long reception cycle.

[0119] The inference unit 210 may similarly use multiple AI models to infer the SRS 400 corresponding to the reference reception period from the SRS 400 of the long reception period, even when the long reception period is six times or more the reference reception period.

[0120] The inference of the inference unit 210 will be explained using Figures 12, 13, and 14, in the case where the long reception period is four times the reference reception period. First, as shown in Figure 12, the inference unit 210 receives the received SRS 430 immediately preceding the Nothing 402 to be inferred, and then inputs the three immediately preceding received SRS 430s, including the received SRS 430, into AI model Aa, and obtains the inferred SRS 440 output from AI model Aa as the inference result. Next, as shown in Figure 13, the inference unit 210 takes the Nothing 402 following the reference reception period of the inferred SRS 440 inferred in the example shown in Figure 12 as the inference target, and inputs the inferred SRS 440 and the two immediately preceding received SRS 430s into AI model Ab, and obtains the inferred SRS 440 output from AI model Ab as the inference result. Next, as shown in Figure 14, the inference unit 210 takes the Nothing 402 following the reference reception cycle of the inferred inference SRS 440 as the inference target, inputs the said inference SRS 440, the previous inference SRS 440, and the previous received SRS 430 into the AI ​​model Ac, and obtains the inference SRS 440 output from the AI ​​model Ac as the inference result. By repeating the above, the inference unit 210 infers the SRS 400 corresponding to the reference reception cycle from the SRS 400 of the long reception cycle.

[0121] When using pattern B, the inference unit 210 repeatedly infers the SRS 400 of Nothing 402 (which may be a received SRS 430 or an inferred SRS 440) that precedes the Nothing 402 to be inferred, thereby inferring an SRS 400 corresponding to the reference reception period from an SRS 400 with a long reception period. For example, the inference unit 210 may perform the inference as illustrated in Figures 15, 16, and 17. First, as shown in Figure 15, the inference unit 210 receives the received SRS 430 immediately preceding the Nothing 402 to be inferred, then inputs the received SRS 430, the preceding inferred SRS 440, and the preceding inferred SRS 440 into AI model B, and obtains the inferred SRS 440 output from AI model B as the inference result. Next, as shown in Figure 16, the inference unit 210 takes the Nothing 402 following the reference reception cycle of the inferred inferred SRS 440 as the inference target, inputs the said inferred SRS 440, the previous received SRS 430, and the previous inferred SRS 440 into AI model B, and obtains the inferred SRS 440 output from AI model B as the inference result. Next, as shown in Figure 17, the inference unit 210 takes the Nothing 402 following the reference reception cycle of the inferred inferred SRS 440 as the inference target, inputs the said inferred SRS 440, the previous inferred SRS 440, and the previous received SRS 430 into AI model B, and obtains the inferred SRS 440 output from AI model B as the inference result. By repeating the above, the inference unit 210 infers the SRS 400 corresponding to the reference reception cycle from the SRS 400 of the long reception cycle.

[0122] When using pattern C, the inference unit 210 repeatedly infers the SRS 400 of Nothing 402 by using a predetermined number of received SRS 430s and a predetermined number of inferred SRS 440s that precede the Nothing 402 to be inferred, thereby inferring an SRS 400 corresponding to the reference reception period from an SRS 400 with a long reception period. For example, the inference unit 210 may perform the inference as illustrated in Figures 18, 19, and 20. First, as shown in Figure 18, the inference unit 210 receives the received SRS 430 immediately preceding the Nothing 402 to be inferred, then inputs the received SRS 430, the three preceding inferred SRS 440s, and the previous received SRS 430 to the AI ​​model Ca, and obtains the inferred SRS 440 output from the AI ​​model Ca as the inference result. Next, as shown in Figure 19, the inference unit 210 takes the Nothing 402 following the reference reception cycle of the inferred inference SRS 440 as the inference target, and inputs the inferred SRS 440, the previous received SRS 430, the two previous inference SRS 440s, and the received SRS 430 from the reference reception cycle before that into the AI ​​model Cb, and obtains the inference SRS 440 output by the AI ​​model Cb as the inference result. Next, as shown in Figure 20, the inference unit 210 takes the Nothing 402 following the reference reception cycle of the inferred inference SRS 440 as the inference target, and inputs the said inference SRS 440, the previous inference SRS 440, the previous received SRS 430, the previous inference SRS 440, and the inference SRS 440 from the reference reception cycle before that into the AI ​​model Cc, and obtains the inference SRS 440 output from the AI ​​model Cc as the inference result. By repeating the above, the inference unit 210 infers the SRS 400 corresponding to the reference reception cycle from the SRS 400 of the long reception cycle. Here, we are illustrating the case where there are two received SRS 430s and three inference SRS 440s input to the AI ​​model of pattern C, but the number of received SRS 430s and the number of inference SRS 440s input to the AI ​​model of pattern C are not limited to these.

[0123] Figure 23 schematically shows an example of the processing flow by the wireless base station 200. Figure 23 explains an example of the initial processing flow using the AI ​​model of pattern A. Here, we will explain using the example where the reference reception period is the period for receiving the SRS 400 in each special slot, and the long reception period is four times the period of the reference reception period.

[0124] In step 102 (sometimes abbreviated as S), the inference unit 210 determines whether or not there is a held SRS 400. If it determines that there is no SRS, it proceeds to S104; if it determines that there is, it proceeds to S106. In S104, the system waits until an SRS is received, and during this time, the communication control unit 204 generates a precode using a method other than SRS.

[0125] In S106, the inference unit 210 determines whether there are enough received SRS 430s to use the AI ​​model Aa. If it determines that there are not enough, the process proceeds to S108; if it determines that there are enough, the process proceeds to S110. In S108, the state estimation unit 206 performs channel estimation using the most recent received SRS 430, and the communication control unit 204 generates a precode using the estimated channel state.

[0126] In S110, the inference unit 210 uses the AI ​​model Aa to infer the next SRS 400. In S112, the inference unit 210 stores the inferred SRS 440, which is the inference result, in the storage unit 202.

[0127] In S114, the inference unit 210 uses the AI ​​model Ab to infer the next SRS 400. In S116, the inference unit 210 stores the inferred SRS 440, which is the inference result, in the storage unit 202.

[0128] In S118, the inference unit 210 uses the AI ​​model Ac to infer the next SRS 400. In S120, the inference unit 210 stores the inferred SRS 440, which is the inference result, in the storage unit 202.

[0129] In S122, the communication control unit 204 stores and holds the received SRS 430 in the storage unit 202.

[0130] The initial processing is completed up to S122. From S124 onward, the inference unit 210 performs inference of the SRS400 using one of pattern A, pattern B, or pattern C.

[0131] The information processing system 10 according to this embodiment may also be applied to a so-called AI-RAN (Radio Access Network). There are three types of AI-RAN: "AI for RAN," "AI on RAN," and "AI and RAN." "AI for RAN" is a technology that utilizes AI and machine learning techniques to improve the frequency utilization efficiency and performance of existing RANs, and to realize automation of base station operations and power saving. With "AI for RAN," it is expected that AI will optimize processing at all layers, such as channel estimation and scheduling processing performed by each cell of the RAN, optimize cooperation between RAN cells, and improve the equipment utilization rate of base stations. "AI on RAN" is a technology that utilizes the computing infrastructure of a base station to provide highly immediate services to users and devices around the base station with low latency. "AI on RAN" enables the deployment of AI and machine learning technology applications at the network edge via RAN, thereby facilitating the creation of new industries and solutions that leverage low latency and confidentiality. "AI and RAN" is a technology for performing RAN processing and AI and machine learning technology processing, which are not directly related to RAN, on the same computing infrastructure. By integrating AI and RAN processing with "AI and RAN," improvements in infrastructure utilization efficiency are expected. The information processing system 10 may be applied in particular to "AI for RAN."

[0132] Figure 24 schematically shows an example of an environment to which the information processing system 10 is applied. Figure 24 illustrates the case where the information processing system 10 is applied to AI-RAN.

[0133] In the example shown in Figure 24, the information processing system 10 comprises a management infrastructure 500 and a plurality of distributed infrastructures 600. In the example shown in Figure 24, the information processing system 10 may perform RAN control to control the RAN 310 and AI processing.

[0134] The distributed infrastructure 600 may be data centers located in various locations. Multiple devices may be located on the distributed infrastructure 600. The distributed infrastructure 600 may be implemented on a virtualization infrastructure consisting of multiple devices. The distributed infrastructure 600 may be implemented by a single device. In other words, the distributed infrastructure 600 may be a distributed device.

[0135] The management infrastructure 500 may be a data center that manages multiple distributed infrastructures 600. Multiple devices may be located on the management infrastructure 500. The management infrastructure 500 may be implemented on a virtualization infrastructure consisting of multiple devices. The management infrastructure 500 may be implemented by a single device. In other words, the management infrastructure 500 may be a management device.

[0136] The management infrastructure 500 may be called the Core Brain, and the distributed infrastructure 600 may be called the Regional Brain. Note that Figure 24 illustrates a case where a single-layer distributed infrastructure 600 is located below the management infrastructure 500, but this is not the only case. The distributed infrastructure 600 may have multiple layers. For example, if two layers of distributed infrastructure 600 are located below the management infrastructure 500, the management infrastructure 500 may be called the Core Brain, the lower-layer distributed infrastructure 600 may be called the Regional Brain, and the further lower-layer distributed infrastructure 600 may be called the Sub-Regional Brain.

[0137] The management infrastructure 500 or the devices located on the management infrastructure 500 may include one or more CPUs (Central Processing Units) and one or more GPUs (Graphics Processing Units). The management infrastructure 500 or the devices located on the management infrastructure 500 may include a superchip in which the CPUs and GPUs are connected by an interconnect. The interconnect may be memory-consistent and capable of achieving high bandwidth and low latency.

[0138] The distributed infrastructure 600 or the devices deployed on the distributed infrastructure 600 may comprise one or more CPUs and one or more GPUs. The distributed infrastructure 600 or the devices deployed on the distributed infrastructure 600 may comprise a superchip in which the CPUs and GPUs are connected by an interconnect. The interconnect may be memory consistent and capable of achieving high bandwidth and low latency.

[0139] The RAN control performed by the information processing system 10 may be a vRAN (Virtual RAN). Multiple distributed infrastructures 600 may constitute a vRAN. The distributed infrastructures 600 may function as, for example, a DU (Distributed Unit) and a CU (Central Unit). The distributed infrastructures 600 may function as a DU and the management infrastructure 500 may function as a CU.

[0140] The AI ​​processing performed by the information processing system 10 may include RAN control AI processing, which is AI processing related to RAN control. The AI ​​processing performed by the information processing system 10 may also include non-RAN control AI processing, which is AI processing not related to RAN control.

[0141] An example of AI-based RAN control processing is the RIC (RAN Intelligent Controller). The RIC is a technology that uses AI to optimize RAN wireless resources and automate RAN operations. The RIC includes Non-RT (Real Time) RIC and Near-RT RIC. The Non-RT RIC is sometimes called a Centralized RIC. The Non-RT RIC is located within the SMO (Service Management and Orchestration), which manages and orchestrates the RAN. The Non-RT RIC generates and notifies policies related to RAN control and transmits information to the Near-RT RIC. For example, a Non-RT RIC generates a trained model for RAN control by performing machine learning using data collected from the RAN, and sends it to a Near-RT RIC. A Near-RT RIC is sometimes called a Distributed RIC. Compared to a Non-RT RIC, a Near-RT RIC is located closer to the RAN nodes (RU (Radio Unit), DU (Distributed Unit), CU (Central Unit)) and performs control of the RAN nodes and resources. Compared to a Non-RT RIC, a Near-RT RIC performs processing with higher real-time capabilities. For example, a Near-RT RIC performs inference processing related to RAN control using the trained model obtained from a Non-RT RIC. RAN control AI processing is not limited to RICs.

[0142] Non-RAN-controlled AI processing may correspond to so-called MEC (Multi-access Edge Computing) applications. Examples of non-RAN-controlled AI processing include, but are not limited to, monitoring AI execution processing that determines the situation within the imaging range of an input image, and response AI execution processing that outputs a response to an inquiry made by a user.

[0143] In the example shown in Figure 24, the learning device 100 may be placed in each of the multiple distributed infrastructures 600 to generate AI models for the multiple wireless base stations 200 under the distributed infrastructure 600. The learning device 100 may be placed in any of the multiple distributed infrastructures 600 to generate AI models for the multiple wireless base stations 200 under the multiple distributed infrastructures 600. The learning device 100 may be placed in the management infrastructure 500 to generate AI models for the wireless base stations 200 under the multiple distributed infrastructures 600.

[0144] Figure 25 schematically shows an example of the hardware configuration of a computer 1200 that functions as a learning device 100, a wireless base station 200, a management infrastructure 500, or a distributed infrastructure 600. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the device according to this embodiment, or to cause the computer 1200 to execute operations associated with the device according to this embodiment or such one or more "parts", and / or to cause the computer 1200 to execute a process or a stage of such process according to this embodiment. Such a program may be executed by the CPU 1212 to cause the computer 1200 to execute specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0145] The computer 1200 according to this embodiment includes a CPU 1212, a GPU 1213, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive and a DVD-RAM drive, etc. The storage device 1224 may be a hard disk drive and a solid-state drive, etc. The computer 1200 also includes legacy input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0146] The CPU 1212 operates according to the programs stored in the ROM 1230 and RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires the image data generated by the CPU 1212 and stores it in the frame buffer provided in the RAM 1214 or within itself, so that the image data is displayed on the display device 1218.

[0147] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive 1226 reads programs or data from the DVD-ROM 1227, etc., and provides them to the storage device 1224. The IC card drive reads programs and data from the IC card and / or writes programs and data to the IC card.

[0148] The ROM 1230 stores boot programs and / or hardware-dependent programs of the computer 1200, which are executed by the computer 1200 when activated. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via USB ports, parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0149] The program is provided on a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The program is read from the computer-readable storage medium and installed on a storage device 1224, RAM 1214, or ROM 1230, which are examples of computer-readable storage media, and executed by the CPU 1212. The information processing described within these programs is read by the computer 1200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the operation or processing of information in accordance with the use of the computer 1200.

[0150] For example, when communication is performed between a computer 1200 and an external device, the CPU 1212 may execute a communication program loaded into the RAM 1214 and, based on the processing described in the communication program, instruct the communication interface 1222 to perform communication processing. Under the control of the CPU 1212, the communication interface 1222 reads transmission data stored in a transmission buffer area provided in a recording medium such as the RAM 1214, storage device 1224, DVD-ROM 1227, or IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0151] Furthermore, the CPU 1212 may read all or necessary parts of a file or database stored on an external recording medium such as a storage device 1224, a DVD drive 1226 (DVD-ROM 1227), or an IC card into the RAM 1214, and perform various types of processing on the data in the RAM 1214. The CPU 1212 may then write the processed data back to the external recording medium.

[0152] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 1212 may perform various types of processing on the data read from the RAM 1214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to the RAM 1214. The CPU 1212 may also retrieve information in files, databases, etc., within the recording medium. For example, if a plurality of entries are stored in the recording medium, each having an attribute value of a first attribute associated with an attribute value of a second attribute, the CPU 1212 may search among the plurality of entries for an entry that matches the specified condition for the attribute value of the first attribute, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0153] The program or software module described above may be stored on or near the computer 1200 in a computer-readable storage medium. Alternatively, a recording medium such as a hard disk or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to the computer 1200 via the network.

[0154] In this embodiment, blocks in the flowchart and block diagram may represent a stage in a process in which an operation is performed or a "part" of a device that has the role of performing an operation. A particular stage and "part" may be implemented by a dedicated circuit, a programmable circuit supplied with computer-readable instructions stored on a computer-readable storage medium, and / or a processor supplied with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuit may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuit may include reconfigurable hardware circuits, such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), which include logical AND, logical OR, exclusive OR, negated AND, negated OR, and other logical operations, flip-flops, registers, and memory elements.

[0155] A computer-readable storage medium may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, a computer-readable storage medium having instructions stored therein will comprise a product that includes instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable storage media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital multipurpose disk (DVD), Blu-ray® disk, memory stick, integrated circuit card, etc.

[0156] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and conventional procedural programming languages ​​such as the C programming language or similar programming languages.

[0157] Computer-readable instructions may be provided locally or via a wide area network (WAN) such as a local area network (LAN) or the internet to a processor or programmable circuit of a general-purpose computer, special-purpose computer, or other programmable data processing device, so that the processor or programmable circuit of the programmable data processing device, such as a computer, can execute the instructions to generate means for performing operations specified in a flowchart or block diagram. Here, the computer may be a PC (personal computer), tablet computer, smartphone, workstation, server computer, general-purpose computer, or special-purpose computer, and may also be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system and is a computer in a broad sense. In a distributed computing system, multiple computers execute a program collectively by each computer executing a part of the program and passing data during program execution between computers as needed.

[0158] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, and microcontrollers. A computer may have one or more processors. In a multiprocessor system with multiple processors, each processor executes a portion of the program, and the processors collectively execute the program by passing program execution data between them as needed. For example, in the execution of multitasks, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which part of a program each processor executes changes dynamically. Which part of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0159] By using the invention according to this embodiment, it is possible to prevent a decrease in the accuracy of channel state estimation even when the number of user terminals 300 connected to the wireless base station 200 increases, thereby contributing to preventing a decrease in the communication quality of the user terminals 300, and thus contributing to the achievement of at least one of the Sustainable Development Goals (SDGs) Goal 9 "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation" and Goal 11 "Make cities and human settlements inclusive, safe, resilient and sustainable."

[0160] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0161] It should be noted that the execution order of operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not explicitly stated as "before" or "prior to," and that these can be performed in any order unless the output of a previous operation is used in a later operation. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," and "next," for convenience, this does not mean that it is mandatory to perform the operations in that order.

[0162] 10 Information processing system, 20 Network, 100 Learning device, 102 Memory unit, 104 Data acquisition unit, 106 Model generation unit, 108 Model provision unit, 200 Wireless base station, 202 Memory unit, 204 Communication control unit, 206 State estimation unit, 208 Model acquisition unit, 210 Inference unit, 300 User terminal, 310 RAN, 400 SRS, 402 Nothing, 404 INF, 410 Reference reception cycle, 420 Long reception cycle, 422 Long reception cycle, 430 Receiving SRS, 440 Inference SRS, 500 Management infrastructure, 600 Distributed infrastructure, 1200 Computer, 1210 Host controller, 1212 CPU, 1213 GPU, 1214 RAM, 1216 Graphics controller, 1218 Display device, 1220 Input / output controller, 1222 Communication interface, 1224 Storage device, 1226 DVD drive, 1227 DVD-ROM, 1230 ROM, 1240 Input / output chip

Claims

1. An information processing system comprising a learning device and a wireless base station, wherein the learning device has a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by the wireless base station from a user terminal at a reference reception cycle, and a model generation unit that uses the learning data to generate a first AI model that uses machine learning to infer the next reference signal of the last reference signal among a plurality of reference signals with a long reception cycle longer than the reference reception cycle, and a second AI model that uses a plurality of reference signals corresponding to the long reception cycle and the next reference signal of the last reference signal among the plurality of reference signals at the reference reception cycle, and generates the next reference signal of the next reference signal at the reference reception cycle, respectively, using a plurality of reference signals and the next reference signal of the last reference signal among the plurality of reference signals at the reference reception cycle, wherein the wireless base station has an inference unit that selectively uses the first AI model and the second AI model generated by the model generation unit to infer the next reference signal of the last reference signal among a plurality of reference signals, which include one or more reference signals received from a user terminal with which a communication connection has been established, An information processing system comprising: a state estimation unit that estimates the channel state between the user terminal and the reference signal inferred by the inference unit; and a state estimation unit that estimates the channel state between the user terminal and the reference signal inferred by the inference unit.

2. The information processing system according to claim 1, wherein the reference signal is an SRS (Sounding Reference Signal).

3. The information processing system according to claim 1 or 2, wherein the inference unit selectively uses the first AI model and the second AI model generated by the model generation unit to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from the plurality of reference signals, which include one or more reference signals received from a user terminal with which a communication connection has been established and one or more reference signals that have been inferred in the past.

4. The information processing system according to any one of claims 1 to 3, wherein the model generation unit further generates a third AI model that infers the next reference signal for the subsequent reference signals in the reference reception period of a plurality of reference signals from a single reference signal corresponding to the long reception period and a plurality of subsequent reference signals in the reference reception period of the single reference signal, and the inference unit selectively uses the first AI model, the second AI model, and the third AI model generated by the model generation unit to infer the next reference signal for the plurality of reference signals in the reference reception period of the plurality of reference signals from a plurality of reference signals, including one or more reference signals received from a user terminal with which a communication connection has been established.

5. The information processing system according to claim 4, wherein the model generation unit further generates a fourth AI model using the learning data to infer the next reference signal in the reference reception cycle of the last reference signal among a plurality of reference signals with a continuous reference reception cycle, and the inference unit stores the reference signals inferred using the first AI model and the second AI model selectively, and when it is possible to perform inference using the fourth AI model with respect to one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, the fourth AI model is used to infer the next reference signal in the reference reception cycle of the last reference signal among a plurality of reference signals with a continuous reference reception cycle, which are composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored.

6. The model generation unit further generates a plurality of fifth AI models using the learning data to infer a target reference signal, which is the next reference signal in the reference reception cycle of the last reference signal, which is the last reference signal among the first number of reference signals, from a predetermined first number of reference signals in the reference reception cycle; the inference unit stores the reference signals inferred by selectively using the first AI model and the second AI model; and if it is possible to perform inference using any of the plurality of fifth AI models with respect to one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, the inference unit uses the executable fifth AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, which is composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored; and the plurality of fifth AI models have different time arrangements of the reference signals other than the last reference signal among the first number of reference signals.

7. An information processing system comprising a learning device and a wireless base station, wherein the learning device includes a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by the wireless base station from a user terminal in a reference reception cycle, and a model generation unit that uses the learning data to generate an AI model by machine learning that infers the next reference signal in the reference reception cycle of the last reference signal among a plurality of consecutive reference signals in the reference reception cycle, and the wireless base station includes a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals, An information processing system comprising: an inference unit that, when inference using the AI ​​model is possible based on one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, uses the AI ​​model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals with a continuous reference reception cycle composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored.

8. An information processing system comprising a learning device and a wireless base station, wherein the learning device includes a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by the wireless base station from a user terminal in a reference reception cycle, and a model generation unit that uses the learning data to generate a plurality of AI models by machine learning that infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, which is the last reference signal among the first number of reference signals, and the wireless base station includes a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals, An information processing system comprising: an inference unit that, when inference can be performed using any of the plurality of AI models based on one or more reference signals received from the user terminal with which the communication connection has been established and one or more inferred reference signals stored, uses the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored; wherein the plurality of AI models differ from each other in the time arrangement of the reference signals other than the last reference signal among the first number of reference signals.

9. A learning device comprising: a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, received by a wireless base station from a user terminal at a reference reception cycle; a first AI model that uses the learning data to infer the next reference signal of the last of the plurality of reference signals at the reference reception cycle from a plurality of long reception cycles with a longer period than the reference reception cycle; and a second AI model that uses machine learning to infer the next reference signal of the next reference signal at the reference reception cycle from a plurality of reference signals corresponding to the long reception cycle and the next reference signal of the last of the plurality of reference signals at the reference reception cycle.

10. A learning device comprising: a storage unit that stores learning data including a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal at a reference reception cycle; and a model generation unit that uses the learning data to generate an AI model by machine learning that infers the next reference signal in the reference reception cycle of the last of the plurality of reference signals from a plurality of consecutive reference signals in the reference reception cycle.

11. A learning device comprising: a storage unit that stores learning data including multiple reference signals used to estimate the channel state of an uplink, received by a wireless base station from a user terminal at a reference reception cycle; and a model generation unit that uses the learning data to generate a plurality of AI models by machine learning that infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, wherein the plurality of AI models differ from each other in the time arrangement of the reference signals other than the last reference signal among the first number of reference signals.

12. A learning method performed by a computer, comprising a model generation step of generating by machine learning a first AI model that uses learning data including a plurality of reference signals used to estimate the channel state of an uplink received by a wireless base station from a user terminal at a reference reception period, to infer the next reference signal of the last of the plurality of reference signals at the reference reception period from a plurality of long reception period reference signals with a period longer than the reference reception period, and a second AI model that infers the next reference signal of the next reference signal at the reference reception period from a plurality of reference signals corresponding to the long reception period and the next reference signal of the last of the plurality of reference signals at the reference reception period.

13. A learning method performed by a computer, comprising a model generation step of machine learning, which generates an AI model that infers the next reference signal in the reference reception period of the last of the multiple reference signals from a plurality of consecutive reference signals in the reference reception period, using learning data that includes a plurality of reference signals used by a wireless base station to estimate the channel state of an uplink received from a user terminal in a reference reception period.

14. A learning method performed by a computer, comprising a model generation step of generating a plurality of AI models by machine learning, which use learning data containing a plurality of reference signals used to estimate the channel state of an uplink received by a wireless base station from a user terminal in a reference reception cycle, to infer a target reference signal, which is the next reference signal in the reference reception cycle, from a predetermined first number of reference signals in the reference reception cycle, wherein the plurality of AI models differ from each other in the time arrangement of the reference signals other than the last reference signal among the first number of reference signals.

15. A program for causing a computer to perform the learning method described in any one of claims 12 to 14.

16. A wireless base station comprising: a first AI model that infers the next reference signal of the last reference signal in the reference reception period from a plurality of long reception period reference signals, which have a longer reception period than the reference reception period, from a plurality of long reception period reference signals, which are generated using training data that includes a plurality of reference signals used to estimate the channel state of the uplink, which are received by the wireless base station from a user terminal at a reference reception period; a second AI model that infers the next reference signal of the next reference signal in the reference reception period from a plurality of reference signals corresponding to the long reception period and the next reference signal of the last reference signal in the reference reception period from the plurality of reference signals, which are the next reference signal of the next reference signal in the reference reception period; an inference unit that selectively uses the first AI model and the second AI model to infer the next reference signal of the last reference signal in the reference reception period from a plurality of reference signals, which include one or more reference signals received from a user terminal with which a communication connection has been established; and a state estimation unit that estimates the channel state between the wireless base station and the user terminal using the reference signals inferred by the inference unit.

17. A wireless base station comprising: a model acquisition unit that acquires an AI model that infers the next reference signal in the reference reception period from a plurality of reference signals in 18. A model acquisition unit that acquires a plurality of AI models that infer a target reference signal, which is the next reference signal in the reference reception cycle of the last reference signal among the first number of reference signals, from a predetermined first number of reference signals in the reference reception cycle, from a plurality of reference signals, which is the last reference signal among the first number of reference signals, from a plurality of reference signals, which is the last reference signal among the first number of reference signals; a storage unit that stores one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals; and an inference unit that, if inference using any of the plurality of AI models is possible using the one or more reference signals received from the user terminal with which a communication connection has been established and the one or more inferred reference signals stored, uses the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal with which a communication connection has been established and one or more inferred reference signals stored. The plurality of AI models are wireless base stations in which the time arrangement of the reference signals, excluding the last reference signal, of the first number of reference signals differs from each other.

19. An information processing method performed by a computer, comprising: a model acquisition step of acquiring a second AI model that infers the next reference signal in the reference reception period of the last reference signal among a plurality of long reception period reference signals, from a plurality of long reception period reference signals with a period longer than the reference reception period, generated using training data which includes a plurality of reference signals used to estimate the channel state of an uplink received by a wireless base station from a user terminal at a reference reception period; a model acquisition step of acquiring a second AI model that infers the next reference signal in the reference reception period of the next reference signal from a plurality of reference signals corresponding to the long reception period and the next reference signal in the reference reception period of the last reference signal among the plurality of reference signals; an inference step of selectively using the first AI model and the second AI model to infer the next reference signal in the reference reception period of the last reference signal among a plurality of reference signals, from a plurality of reference signals which include one or more reference signals received from a user terminal with which a communication connection has been established; and a state estimation step of estimating the channel state between the user terminal and the reference signals inferred in the inference step.

20. An information processing method performed by a computer, comprising: a model acquisition step of acquiring an AI model that infers the next reference signal in the reference reception period from a plurality of reference signals in 21. A computer-based information processing method comprising: a model acquisition step of acquiring a plurality of AI models that infer a target reference signal, which is the next reference signal in the reference reception cycle of the last reference signal among the first number of reference signals, from a predetermined first number of reference signals in the reference reception cycle, using learning data that includes a plurality of reference signals used to estimate the channel state of an uplink, which are received by a wireless base station from a user terminal in a reference reception cycle; a storage step of storing one or more reference signals received from a user terminal with which a communication connection has been established, and one or more reference signals inferred using the one or more reference signals, in a storage unit; and, if inference using any of the plurality of AI models is possible using the one or more reference signals received from the user terminal with which a communication connection has been established and the one or more inferred reference signals stored, an inference step of using the executable AI model to infer the next reference signal in the reference reception cycle of the last reference signal among the plurality of reference signals, from a plurality of reference signals composed of one or more reference signals received from the user terminal and one or more inferred reference signals stored. An information processing method wherein the plurality of AI models have different time arrangements of the reference signals, excluding the last reference signal, among the first number of reference signals.

22. A program for causing a computer to perform the information processing method described in any one of claims 19 to 21.

Images