Terminal device, base station device, and control method for improving performance of handover by using artificial intelligence (AI) / machine learning (ML)

Abstract

This terminal device: measures the radio quality of a signal transmitted in a cell provided by a currently connected base station device; and transmits, to the base station device, a prescribed notification based on values acquired by inputting the radio quality acquired by the terminal device in an inference phase to a trained model acquired by machine learning that is performed by using, as an input, the radio quality acquired in a training phase and using, as training data, prescribed values associated with the occurrence of a radio link failure (RLF) when handover has not occurred within a prescribed length period after the timing at which the radio quality was acquired.

Description

Terminal device, base station device, and control method for improving handover performance using artificial intelligence (AI) / machine learning (ML)

[0001] The present invention relates to an advanced handover technique in a cellular communication system.

[0002] In wireless communication systems compliant with the cellular communication standard of the Third Generation Partnership Project (3GPP (registered trademark)), a handover process is defined for switching the base station to which a mobile terminal device is connected, so that the mobile terminal device can continue to communicate with high quality. The handover process is performed, for example, when the wireless quality of a signal transmitted from a base station to which the mobile terminal device is currently connected falls below a predetermined level for a certain period of time, by transmitting a measurement result to the base station, and the base station then instructs the mobile terminal device to perform a handover.

[0003] In one example, if the wireless quality in the terminal device further deteriorates between the time when the wireless quality falls below a predetermined level and the time when the handover command is transmitted from the base station device, the terminal device may be unable to receive the handover command from the base station device. In this case, if the handover process is not performed in the terminal device and the wireless quality between the terminal device and the base station device does not improve, an unintended disconnection of the wireless link may occur. Furthermore, after the handover process is performed in the terminal device and the base station device to which the terminal device is connected is changed, depending on the base station device selected as the handover destination, the wireless quality between the base station device and the terminal device may deteriorate within a short period of time, which may necessitate a further handover or result in a disconnection of the wireless link. Furthermore, unnecessary handovers may result in a decrease in communication quality, such as average throughput, compared to when no handover is performed. In other words, communication efficiency may decrease if the handover process is not performed in a timely and appropriate manner.

[0004] The present invention provides a technique for enhancing handover processing and improving communication efficiency.

[0005] A terminal device according to one aspect of the present invention comprises: a measurement means for measuring the wireless quality of a signal transmitted in a cell provided by a base station device to which the terminal device is connected; and a transmission means for transmitting a predetermined notification to the base station device based on a value obtained by inputting the wireless quality obtained by the terminal device in the inference phase into a trained model obtained by machine learning using, as training data, a predetermined value associated with the occurrence of a radio link failure (RLF) when a handover is not performed within a predetermined length of time after a predetermined time period from the time when the wireless quality was obtained.

[0006] A base station device according to one aspect of the present invention has: a receiving means for receiving a predetermined notification based on a value acquired by inputting the wireless quality of a signal transmitted from the base station device in a learning phase from a terminal device connected to a cell provided by the base station device, and inputting the wireless quality acquired by the terminal device in an inference phase into a trained model acquired by machine learning using, as training data, a predetermined value associated with the occurrence of a radio link failure (RLF) if a handover is not performed within a predetermined length of time after a predetermined time period from the time the wireless quality was acquired; and a control means for performing control to hand over the terminal device from the cell to another cell based on the receipt of the predetermined notification.

[0007] According to the present invention, the handover process can be enhanced to improve communication efficiency.

[0008] Other features and advantages of the present invention will become apparent from the following description taken in conjunction with the accompanying drawings, in which the same or similar elements are designated by the same reference numerals.

[0009] The accompanying drawings are incorporated in and constitute a part of the specification, illustrate embodiments of the present invention, and together with the description are used to explain the principles of the present invention. Figure 1 is a diagram showing an example of the configuration of a wireless communication system. Figure 2 is a diagram showing an example of the hardware configuration of a base station device and a terminal device. Figure 3 is a diagram showing an example of the functional configuration of a terminal device. Figure 4 is a diagram showing an example of the functional configuration of a base station device. Figure 5 is a diagram showing an example of the flow of processing executed in the wireless communication system.

[0010] Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention as claimed, and not all combinations of features described in the embodiments are necessarily essential to the invention. Two or more of the features described in the embodiments may be combined in any desired manner. Furthermore, the same reference numerals are used to designate identical or similar components, and redundant descriptions will be omitted.

[0011] (System Configuration) FIG. 1 shows an example configuration of a communication system according to this embodiment. This communication system is, for example, a wireless communication system conforming to the Long Term Evolution (LTE) or fifth generation (5G) cellular communication standards of the Third Generation Partnership Project (3GPP (registered trademark)), or their successor standards. This wireless communication system includes a base station device 101 and a terminal device 111. The terminal device 111 establishes a connection with the base station device 101 and performs wireless communication in a cell 121 provided by the base station device 101. The terminal device 111 is assumed to be, for example, portable by a user and to be mobile. If the terminal device 111 moves beyond the range of the cell 121, it will no longer be able to maintain communication in that cell 121. Therefore, for example, the terminal device 111 performs handover to another cell (cell 122 or cell 123) upon the occurrence of a predetermined event, such as a deterioration in communication quality in the currently connected cell 121. Note that the cells 122 and 123 may be provided by a base station device other than the base station device 101, or may be provided by the base station device 101. In other words, the base station device 101 may provide multiple cells.

[0012] Conventionally, the terminal device 111 measures the wireless quality of signals transmitted from the currently connected cell 121 and other adjacent cells (cells 122 and 123), and determines whether an event for initiating handover processing has occurred based on whether the wireless quality satisfies a predetermined condition. Events are defined in a cellular communication standard. Examples of events include event A2, which is determined to have occurred when the wireless quality of the currently connected cell falls below a predetermined threshold; event A3, which is determined to have occurred when the wireless quality of the adjacent cell becomes higher (by a predetermined offset) than the wireless quality of the currently connected cell; event A4, which is determined to have occurred when the wireless quality of the adjacent cell exceeds a predetermined threshold; and event A5, which is determined to have occurred when the wireless quality of the currently connected cell falls below a first predetermined threshold and the wireless quality of the adjacent cell exceeds a second predetermined threshold. When the terminal device 111 determines that any of the events set by the base station device 101 has occurred, it transmits a predetermined report to the base station device 101. Then, in response to receiving the predetermined report, the base station device 101 executes processing to hand over the terminal device 111 from the currently connected cell 121 to another cell (for example, cell 122 or cell 123). Note that the handover of the terminal device 111 can be performed, for example, by transmitting setting information for communication in a neighboring cell from the base station device 101 to the terminal device 111 before the occurrence of an event, and then transmitting a command (Handover command) from the base station device 101 to the terminal device 111 instructing the handover. Also, after the occurrence of an event, the handover processing can be started by transmitting setting information for communication in a neighboring cell of the handover from the base station device 101 to the terminal device 111 using an RRC Reconfiguration message. Note that RRC is an abbreviation for Radio Resource Control.

[0013] Conventionally, handover processing is initiated when the wireless quality measured at that time in the terminal device 111 satisfies one of the above-mentioned events. In this procedure, for example, the wireless quality of the connected cell (serving cell) deteriorates in event A2 or event A3, and a predetermined report is transmitted to the base station device 101 after the event occurs. Then, a handover instruction is transmitted from the base station device 101. Therefore, a certain period of time is required from the deterioration of wireless quality to the execution of handover. Then, during the period from when the terminal device 111 detects the occurrence of the event to when the base station device 101 transmits the handover instruction, the wireless quality may further deteriorate, and the handover instruction may not be received by the terminal device 111.

[0014] The terminal device 111 is supposed to start timer T310 when it is deemed unable to establish downlink synchronization with the base station device 101. More specifically, the terminal device 111 starts timer T310 when it receives N310 consecutive out-of-sync indications from the physical layer in the RRC layer. The terminal device 111 receives information on the number of N310 and the expiration time of T310 in advance as configuration information from the connected base station device 101. When timer T310 expires after being started, the terminal device 111 determines that a radio link failure (RLF) has occurred. Timer T310 is stopped when a handover command is received. However, as described above, if wireless quality deteriorates after an event occurs and the terminal device 111 is unable to receive the command, it cannot stop timer T310. As a result, the terminal device 111 is unable to execute handover processing, resulting in the occurrence of a handover failure (HOF). Furthermore, the terminal device 111 cannot start a connection re-establishment process (RRC Connection Re-establishment process) unless an RLF occurs after waiting until timer T310 expires. For this reason, in order to improve the efficiency of communication in a wireless communication system, it is important to reduce the probability of occurrence of HOF or RLF.

[0015] In recent years, the application of artificial intelligence (AI) / machine learning (ML) to cellular communication systems has been discussed. Using AI / ML, for example, it is possible to predict whether an RLF will occur in the future based on the wireless quality of the currently connected cell. For example, a trained model can be acquired by machine learning using the wireless quality of the cell 121 acquired in the learning phase as input and a predetermined value associated with the occurrence of an RLF if a handover does not occur within a predetermined period of time after the wireless quality is acquired as training data. Once such a trained model is acquired, the terminal device 111 can input the results of measuring the wireless quality of a signal transmitted from the cell 121 in the inference phase to determine whether an RLF is predicted to occur within a predetermined period of time after the wireless quality is acquired. Furthermore, the terminal device 111 provides the base station device 101 with the output of the trained model, enabling the base station device 101 to make this determination. Here, the predetermined period can be set arbitrarily, for example, to a period of several seconds. The predetermined length of the period is also set arbitrarily, for example, to a period of several tens to several hundreds of milliseconds. Note that information about these settings can be notified in advance from the base station device 101 to the terminal device 111.

[0016] In this embodiment, the base station device 101 or the terminal device 111 makes such a determination, and processing such as handover of the terminal device 111 is performed based on an inference regarding the occurrence of a future RLF. That is, the terminal device 111 transmits a predetermined report to the base station device 101 including information about an RLF that is not currently occurring but is expected to occur in the future. Based on the predetermined report, the base station device 101 can instruct the terminal device 111 to start handover processing regardless of the current situation. This makes it possible to change the cell to which the terminal device 111 is connected before an RLF actually occurs (before timer T310 expires), thereby preventing the terminal device 111 from waiting for timer T310 to expire. Furthermore, by performing handover before timer T310 is started in the terminal device 111, it is possible to prevent the handover command from not reaching the terminal device 111, and to reliably complete the handover. In addition, for example, even if the wireless quality in the terminal device 111 temporarily deteriorates, it is predicted that the connection can be maintained without the timer T310 expiring, and unnecessary handover can be prevented.

[0017] Note that multiple trained models for estimating future RLF occurrences may be prepared, and the base station device 101 may notify the terminal device 111 in advance (for example, when connecting with the terminal device 111) of information indicating which of the trained models to use. Examples of multiple trained models will be described later. The base station device 101 may determine whether to permit the execution of RLF occurrence estimation by inference using a trained model, and notify the terminal device 111 of the determination. The terminal device 111 may also transmit capability information (UE Capability) including information indicating whether or not it is possible to execute RLF occurrence estimation by inference using a trained model to the base station device 101. The base station device 101 may permit such RLF occurrence estimation processing only for terminal devices 111 that are capable of executing RLF occurrence estimation by inference using a trained model. Note that the following description focuses on the cell 121 to which the terminal device 111 is connected, but the same applies to other cells. Although a different trained model can be generated for each cell, a trained model common to multiple cells may also be generated.

[0018] Furthermore, the "radio quality" in this embodiment may be, for example, reference signal received power (RSRP), but other indices such as reference signal received quality (RSRQ), signal-to-noise ratio (SNR), and signal-to-noise and interference ratio (SINR) may also be used. However, a common indices are used in the learning phase and inference phase of machine learning. That is, when RSRP is used as input in the learning phase, RSRP is also used as input in the inference phase. Note that when the output of the model is radio quality, the same indices may be used for the input and output of the model, but the input and output may be expressed by different indices, such as when RSRP is used as input and RSRQ is used as output. In one example, RSRP of layer 1 may be used as input and RSRP of layer 3 may be used as output. In addition, the reference signal used for measuring RSRP etc. may be, for example, a synchronization signal (SS) / physical broadcast channel (PBCH) block (SSB) or a channel state information-reference signal (CSI-RS), but other reference signals may also be used.

[0019] Below, several examples of trained models are described. Various trained models can be obtained by setting various predetermined values ​​used as training data in the learning phase. Therefore, below, examples of predetermined values ​​used as training data, examples of expected outputs in the trained models, and examples of how the terminal device 111 estimates that an RLF will occur in the future are described.

[0020] The predetermined value used as training data may be, for example, a value indicating whether an RLF occurred within a predetermined length of time after the wireless quality of the cell 121 was acquired during the learning phase. That is, each terminal device that previously communicated through the cell 121 measures the wireless quality of the cell 121, and whether an RLF occurred when a handover was not performed within a certain period of time after the wireless quality was acquired is collected for machine learning. This information may be collected in an environment where a conventional handover process is performed (i.e., an environment where a handover process based on an estimation of future RLF occurrence, as in the present embodiment, is not performed). That is, an environment where an RLF occurs in an environment where a conventional handover process is performed is learned, and the trained model obtained by this learning makes it possible to determine what wireless quality measurement values ​​are likely to cause an RLF. Furthermore, the base station device 101 may set stricter criteria for transmitting a handover command than usual during the learning phase. That is, the base station device 101 may be configured to issue fewer handover instructions than usual during the learning phase, so that an environment where degradation of wireless quality is likely to occur and where RLF is more likely to occur is learned.

[0021] In the inference phase, the wireless quality measured in the cell 121 is input to the trained model obtained by the machine learning, and information indicating, for example, whether or not an RLF will occur within a certain period of time after the measurement value is obtained, or the probability of an RLF occurring within that certain period of time, is output. Then, for example, if it is inferred in the inference phase that an RLF will occur, the terminal device 111 may transmit a predetermined report to the base station device 101. Furthermore, for example, if the probability of an RLF occurrence is obtained in the inference phase, the terminal device 111 may transmit a predetermined report including information about the probability to the base station device 101. Note that the terminal device 111 may transmit a predetermined report including information about the probability to the base station device 101 based on the probability exceeding a predetermined value. That is, the predetermined report may be transmitted only when the probability of an RLF occurring within a certain period of time in the future is high above a predetermined level, and may not be transmitted when the probability is sufficiently low. This prevents unnecessary reports from being transmitted. Furthermore, when the probability of RLF occurrence exceeds a predetermined value, the terminal device 111 may determine that RLF is estimated to occur within a certain period of time in the future, and may transmit the determination result to the base station device 101. Note that the information on the above-mentioned predetermined value related to the probability and the information on the predetermined length of the period may be notified in advance from the base station device 101 to the terminal device 111 as configuration information. In other words, configuration information to be used for determining future RLF based on AI / ML may be prepared.

[0022] The terminal device 111 may be configured to, for example, acquire whether or not RLF will occur or the probability of RLF occurrence for multiple timings within a predetermined period of time targeted for RLF estimation. In this case, in the learning phase, information indicating whether RLF occurred at multiple different timings within a predetermined period of time after the timing at which the input wireless quality was acquired is used as training data. Then, in the inference phase, the trained model is configured to output, for the input wireless quality measurement value, an estimated value of whether or not RLF will occur or the probability of RLF occurrence at multiple different timings within a predetermined period of time after the timing at which the measurement value was acquired. Then, the terminal device 111 may transmit to the base station device 101 a predetermined report including information on timings at which it was determined that RLF will occur among the multiple timings within the predetermined period of time, or the probability of RLF occurrence at each of the multiple timings. Furthermore, the terminal device 111 may transmit a predetermined report including information indicating the timing at which the probability of RLF occurrence exceeds a predetermined value among multiple timings within the predetermined period (i.e., the timing corresponding to the probability of RLF occurrence exceeding the predetermined value) to the base station device 101. Furthermore, the terminal device 111 may transmit the predetermined report to the base station device 101 when a state in which RLF is estimated to occur is maintained within the predetermined period or when a state in which the probability of RLF occurrence exceeds a predetermined value is maintained.

[0023] Furthermore, the predetermined value used as the training data may be, for example, a second wireless quality of one or more signals transmitted in the cell 121 within a predetermined period after a timing at which a first wireless quality of a signal transmitted in the cell 121 is acquired in the learning phase. That is, the first wireless quality acquired in the cell 121 is used as input in the learning phase, and (one or more) second wireless qualities acquired in the same cell 121 within a predetermined period after the timing at which the first wireless quality is acquired may be used as training data. Then, in the inference phase, the wireless quality measured in the cell 121 is input to the trained model obtained by the machine learning, thereby outputting, for example, an estimated value of the wireless quality in the cell 121 a predetermined period after the timing at which the wireless quality is measured. The terminal device 111 measures the wireless quality of a signal transmitted in the cell 121 (from the base station device 101) and inputs the measurement result into the trained model, thereby obtaining an estimated value of the wireless quality in the cell 121 a predetermined period after the timing at which the measurement value is acquired. The terminal device 111 can then use the estimated value of wireless quality to determine whether RLF will occur. For example, if the estimated value of wireless quality falls below a predetermined level that is a criterion for the likelihood of RLF occurring, the terminal device 111 can estimate that RLF will occur within a certain period of time after the timing of measuring the wireless quality. Note that information such as a threshold value (the above-mentioned predetermined level) for determining whether RLF has occurred may be notified from the base station device 101 to the terminal device 111, for example, by separately preparing an information element for transmitting setting information for AI / ML.

[0024] The terminal device 111 may be configured to acquire estimated values ​​of wireless quality for multiple timings within a predetermined period of time. In this case, in the learning phase, wireless quality values ​​acquired at multiple different timings within a predetermined period of time after the timing at which the input wireless quality was acquired are used as training data. Then, in the inference phase, the trained model is configured to output, for an input wireless quality measurement value, estimated values ​​of wireless quality at multiple different timings after the predetermined period of time after the timing at which the measurement value was acquired. Then, if the estimated value of wireless quality continues to fall below a predetermined level, which is a criterion for the likelihood of RLF occurring, over a predetermined period of time, the terminal device 111 may estimate that RLF will occur within a certain period of time after the timing at which the wireless quality was measured. Note that the length of the period for determining the occurrence of RLF may be the same as the length of the period for which the trained model outputs wireless quality. That is, the terminal device 111 may estimate that RLF will occur if all of the estimated values ​​corresponding to multiple timings corresponding to a predetermined period of time output from the trained model are below the predetermined level. Furthermore, the terminal device 111 may estimate that an RLF will occur when all of the estimated values ​​corresponding to a predetermined number of consecutive timings among the multiple timings corresponding to a predetermined period of time output from the trained model are below a predetermined level. That is, the length of the period for determining the occurrence of an RLF may be shorter than the length of the period for which the wireless quality is output by the trained model. Note that information about the predetermined period of time may be notified in advance from the base station device 101 to the terminal device 111, for example.

[0025] When the terminal device 111 determines that an RLF will occur, it may transmit a predetermined report including information indicating the occurrence of the RLF to the base station device 101. The terminal device 111 may also identify the timing at which an RLF is expected to occur based on the wireless quality, and transmit a predetermined report including information indicating the timing to the base station device 101. The terminal device 111 may also transmit a predetermined report including information indicating the wireless quality, such as the correspondence between the wireless quality and the timing, to the base station device 101.

[0026] In the above example, the terminal device 111 predicts whether or not an RLF will occur in the connected cell within a certain period of time in the future. However, the terminal device 111 may also predict the occurrence of an RLF when handed over to a neighboring cell that is a candidate handover destination. In order to estimate the RLF in the neighboring cell, machine learning must be performed. For example, in the learning phase, the wireless quality measured in the cell 121 is used as input, and machine learning is performed using as training data whether or not an RLF occurred in a terminal device that handed over to another cell within a predetermined period after the wireless quality was obtained, thereby acquiring a trained model. As a result, in the inference phase, when a measurement value of the wireless quality in the cell 121 is input to the trained model, whether or not an RLF will occur when handed over to each of the candidate handover destination cells, such as the cell 122 or the cell 123, within a predetermined period of time after the measurement value was obtained can be output.

[0027] Furthermore, for example, in the learning phase, a trained model can be acquired by performing machine learning using, as input, the wireless quality measured in cell 121 and the wireless quality of the cell obtained by a terminal device that has handed over to another cell within a predetermined period of time after the timing at which the wireless quality was obtained, as training data. As a result, in the inference phase, when a measurement value of the wireless quality in cell 121 is input to the trained model, estimated values ​​of the wireless quality in each of candidate cells for handover, such as cell 122 and cell 123, when handover occurs within a predetermined period of time after the timing at which the measurement value was obtained can be output.

[0028] Furthermore, in the learning phase, instead of or in addition to the wireless quality measured in cell 121, the wireless quality in each of candidate handover destination cells (e.g., cell 122 and cell 123) may be input, and a trained model may be acquired by performing machine learning using as training data whether an RLF occurred in a terminal device that handed over to the candidate cell after a predetermined period from the timing at which the wireless quality was obtained. As a result, in the inference phase, when the measurement values ​​of the wireless quality in the candidate handover destination cells are input to the trained model, whether or not an RLF will occur if handover is performed to each of the candidate cells within a predetermined period of time after the timing at which the measurement values ​​were obtained, or the probability of RLF occurrence may be output. Furthermore, for example, in the learning phase, instead of or in addition to the wireless quality measured in cell 121, the wireless quality in each of candidate handover destination cells (e.g., cell 122 or cell 123) may be used as input, and a trained model may be acquired by performing machine learning using, as training data, the wireless quality of the candidate cell obtained by a terminal device that has handed over to the candidate cell within a predetermined period of time after the wireless quality was obtained. As a result, in the inference phase, when the measured values ​​of the wireless quality in the candidate handover destination cell are input to the trained model, estimated values ​​of the wireless quality in each of the candidate handover destination cells, such as cell 122 or cell 123, when a handover is performed to those cells within a predetermined period of time after the measured values ​​are obtained may be output.

[0029] The terminal device 111 may transmit to the base station device 101 a predetermined report about a future RLF regarding a candidate handover destination cell, similar to the predetermined report about a future RLF regarding the cell 121. The terminal device 111 may transmit to the base station device 101 a predetermined report including information similar to the information about the cell 121 described above, such as, for example, that an RLF is estimated to occur when handing over to the candidate cell, the probability of the RLF occurring, information indicating the timing at which the probability of RLF occurrence exceeds a predetermined value, values ​​indicating the probability of RLF occurrence at multiple timings, the timing at which RLF is expected to occur, and the radio quality of the candidate cell. Note that the trained model for the candidate cell may use information about radio quality at multiple timings as input, or may use information about radio quality at multiple timings as output. That is, the occurrence of RLF when handing over to the candidate handover destination cell may be estimated in a format similar to the estimation of the occurrence of RLF regarding the cell to which the terminal device 111 is connected. In addition, a separate trained model may be prepared for each candidate cell as a handover destination, or one trained model may be prepared for multiple candidate cells.

[0030] The terminal device 111 may generate a trained model by performing machine learning, or may acquire a separately generated trained model from the base station device 101. When machine learning is performed in the terminal device 111, a trained model cannot be acquired until training data is acquired a sufficient number of times, but a trained model suitable for the terminal device 111 can be acquired. On the other hand, when a trained model is provided from the base station device 101, a large amount of training data can be obtained from communications by a large number of terminal devices previously connected to the cell 121 (or the cell 122 or the cell 123), allowing learning to be completed quickly. Furthermore, the base station device 101 may notify the terminal device 111 of information to be input into the trained model and information indicating the output of the trained model at that time. In accordance with the notification, the terminal device 111 may collect, for example, information to be input into the trained model provided from the base station device 101 (such as the wireless quality of the currently connected cell 121 or another cell) and input it into the trained model. Then, in accordance with the notification, the terminal device 111 may identify what information is output from the trained model and, accordingly, determine, for example, whether or not an RLF is estimated to occur. Furthermore, the terminal device 111 may perform machine learning in accordance with the notification.

[0031] When the base station device 101 receives the predetermined report, it can initiate handover processing for the terminal device 111 based on the predetermined report. For example, when the base station device 101 receives a predetermined report indicating that the terminal device 111 has determined that an RLF is estimated to occur in the future or indicating that the probability of an RLF occurrence is equal to or greater than a predetermined level, it can cause the terminal device 111 to execute handover even at a timing when handover processing would not be initiated in conventional processing. For example, the base station device 101 can transmit a handover command to the terminal device 111 before timer T310 is started in the terminal device 111. That is, the base station device 101 initiates handover processing when, before an RLF actually occurs, it is estimated by inference using a trained model that an RLF will occur within a predetermined period of time. This allows handover to be performed earlier than when conventional handover processing is performed, thereby reducing the probability of handover failure and improving communication efficiency.

[0032] Furthermore, for example, even if the wireless quality temporarily deteriorates and the timer T310 is started, the terminal device 111 may not transmit a predetermined report if the probability of RLF occurrence is low. Furthermore, the terminal device 111 may also transmit a predetermined report, for example, even if it is assumed that RLF will not occur. In one example, the terminal device 111 may periodically transmit a predetermined report and, in the predetermined report, notify the base station device 101 of information such as whether or not an RLF is expected to occur and the probability of RLF occurrence. This allows the base station device 101 to not initiate handover processing even if the wireless quality of the terminal device 111 temporarily deteriorates if the probability of RLF occurrence in the terminal device 111 is low. In other words, it is possible to prevent unnecessary handover processing from being performed in cases where the wireless quality only deteriorates temporarily due to, for example, the presence of an obstacle between the terminal device 111 and the base station device 101. Furthermore, by predicting the probability of RLF occurrence when the terminal device 111 hands over to another cell, the base station device 101 can select, for example, another cell in which the probability of RLF occurrence is below a predetermined value as the handover destination. In other words, in conventional processing, a cell with high wireless quality is usually selected as the handover destination during handover processing. However, if an RLF occurs immediately after the terminal device 111 connects to that cell, another handover processing will be required, and therefore such a cell can be prevented from being selected as the handover destination. For example, when the base station device 101 receives a predetermined report about another cell from the terminal device 111, the base station device 101 can prevent the selection of the other cell as the handover destination. Furthermore, the terminal device 111 can transmit a predetermined report to the base station device 101 including information indicating the probability of RLF occurrence for each of the candidate handover destination cells, even if the probability is low. In this case, the base station device 101 may select cells in which the probability of RLF occurrence is lower than a predetermined value as candidate cells for the handover destination, and select a cell with good wireless quality from among these candidate cells as the handover destination.For example, in response to receiving a predetermined report indicating that the terminal device 111 has entered a state in which RLF is likely to occur in relation to the cell to which it is connected, the base station device 101 can select a candidate cell for handover based on the RLF probability in that cell.The base station device 101 can then instruct the terminal device 111 to measure and report the wireless quality of the selected candidate cell for handover, and determine to handover the terminal device 111 to a cell with good measurement results.In this way, a cell with good wireless quality and low RLF risk after handover can be selected for handover of the terminal device 111.

[0033] Note that performance evaluation of the trained model may be performed periodically, for example. For example, in some cases where the base station device 101 receives the above-mentioned predetermined report, the base station device 101 may perform handover processing using a conventional procedure rather than performing handover processing based on the estimation. The base station device 101 may then evaluate the trained model as unsuitable for the real environment and decide to perform re-learning (additional learning) if, for example, the error between the estimated timing of RLF occurrence and the actual timing of RLF occurrence is equal to or greater than a predetermined value. Furthermore, the terminal device 111 may evaluate the trained model as unsuitable for the real environment if, for example, the difference between the estimated value and the actual measured value of wireless quality at a specific timing is equal to or greater than a predetermined value, and notify the base station device 101 of the evaluation result. Furthermore, the terminal device 111 may evaluate the trained model as unsuitable for the real environment if, for example, the difference between the estimated value and the actual measured value of wireless quality for the maximum value or a predetermined number of highest values ​​of wireless quality for multiple cells is equal to or greater than a predetermined value, and notify the base station device 101 of the evaluation result. In response to the notification, the base station device 101 may decide to perform re-learning (additional learning), which allows the trained model to be maintained in an appropriate state and enables handover processing based on prediction to be performed at an appropriate time.

[0034] (Device Configuration) FIG. 2 shows an example of the hardware configuration of the base station device 101 and the terminal device 111 according to this embodiment. In one example, the base station device 101 and the terminal device 111 are configured to include a processor 201, a ROM 202, a RAM 203, a storage device 204, and a communication circuit 205. The processor 201 is a computer configured to include one or more processing circuits, such as a general-purpose CPU (Central Processing Unit) or an ASIC (Application Specific Integrated Circuit), and performs the overall processing of the device and each of the above-mentioned processes by reading and executing programs stored in the ROM 202 or the storage device 204. The ROM 202 is a read-only memory that stores information such as programs and various parameters related to the processing performed by the base station device 101 and the terminal device 111. The RAM 203 functions as a workspace when the processor 201 executes a program and is also a random access memory that stores temporary information. The storage device 204 is configured, for example, by a removable external storage device. The communication circuit 205 is configured, for example, by a circuit for wireless communication of 5G or a successor standard. While FIG. 2 illustrates one communication circuit 205, the base station device 101 and the terminal device 111 may have multiple communication circuits. For example, the base station device 101 and the terminal device 111 may have wireless communication circuits for 5G and a successor standard, respectively, and a common antenna for these circuits. The base station device 101 and the terminal device 111 may also have separate antennas suitable for each standard. The base station device 101 may also have a wired communication circuit used when communicating with other base station devices or nodes in the core network. The terminal device 111 may also have a communication circuit compliant with a wireless communication standard other than a cellular communication standard, such as a wireless local area network (LAN) or Bluetooth (registered trademark). The base station device 101 and the terminal device 111 may have separate communication circuits 205 for each of the multiple available frequency bands, or may have a common communication circuit 205 for at least some of the frequency bands.

[0035] FIG. 3 shows an example of the functional configuration of the terminal device 111. The terminal device 111 includes, for example, a measurement unit 301, an RLF estimation unit 302, a reporting unit 303, and a setting information receiving unit 304. The terminal device 111 may also include, as an option, a learning unit 305. Note that FIG. 3 only shows functions particularly related to this embodiment, and does not illustrate various other functions that the terminal device 111 may have. For example, the terminal device 111 naturally has other functions that terminal devices compliant with LTE, 5G, or subsequent standards generally have. The functional blocks in FIG. 3 are shown schematically, and the respective functional blocks may be integrated or further subdivided. The functions in FIG. 3 may be realized, for example, by the processor 201 executing a program stored in the ROM 202 or the storage device 204, or by a processor within the communication circuit 205 executing predetermined software. A storage medium storing a program for implementing the functions may also be provided. Since the details of the processes executed by each functional unit have been described above, only the general functions of the terminal device 111 will be outlined here.

[0036] The measurement unit 301 observes signals (e.g., reference signals) transmitted in the destination cell 121 and other candidate handover destination cells (e.g., cell 122 or cell 123) and measures their wireless quality. The RLF estimation unit 302 inputs the wireless quality of the cell 121 measured by the measurement unit 301 into a trained model and performs inference processing, and determines whether or not RLF will occur in the cell 121 if handover is not performed within a predetermined length of time after the timing at which the wireless quality is acquired, and the probability of RLF occurrence, etc. Furthermore, the RLF estimation unit 302 inputs the wireless qualities of the cell 121 and other cells measured by the measurement unit 301 into a trained model and performs inference processing, and determines whether or not RLF will occur in the handover destination cell (cell 122 or cell 123) if handover is performed within a predetermined length of time after the timing at which the wireless quality is acquired, and the probability of RLF occurrence, etc. The reporting unit 303 transmits a predetermined report including the estimation result of the RLF estimation unit 302 to the base station device 101. The setting information receiving unit 304 receives setting information for determining the occurrence of RLF based on future predictions using a trained model from the connected base station device 101. The setting information receiving unit 304 may receive, for example, the trained model itself to be used in the terminal device 111. Furthermore, when multiple trained models are available in the terminal device 111, the setting information receiving unit 304 may receive information from the base station device 101 specifying which trained model to use. Furthermore, the setting information receiving unit 304 may receive information from the base station device 101 specifying the input and output of the trained model. Furthermore, the setting information receiving unit 304 may receive various other information as described above. When a trained model should be generated in the terminal device 111, the learning unit 305 generates the trained model through machine learning. Furthermore, the learning unit 305 may re-learn the trained model, for example, in response to an instruction from the base station device 101. For example, if handover processing is not started after a predetermined report is transmitted, the learning unit 305 may identify the difference between the RLF occurrence timing and the estimated value of the wireless quality by the trained model and the conventional actual measured value of the RLF occurrence timing and the wireless quality, and notify the base station device 101 of the difference.Furthermore, when the difference between the estimated value and the actual measured value exceeds a predetermined value, the learning unit 305 may determine that relearning is necessary and may notify the base station device 101 of information indicating that relearning should be performed. In other words, the need for relearning may be determined by the base station device 101 or may be determined by the terminal device 111. Note that the learning unit 305 may perform relearning without obtaining permission from the base station device 101.

[0037] FIG. 4 shows an example of the functional configuration of the base station device 101. The base station device 101 includes a setting information notification unit 401, a report receiving unit 402, and a handover control unit 403. The base station device 101 may also include a learning unit 404 as an option. Note that FIG. 4 only illustrates functions particularly related to this embodiment, and does not illustrate various other functions that the base station device 101 may have. For example, the base station device 101 naturally has other functions that are generally possessed by base station devices compliant with LTE, 5G, or subsequent standards. The functional blocks in FIG. 4 are shown schematically, and the respective functional blocks may be integrated or further subdivided. The functions in FIG. 4 may be realized, for example, by the processor 201 executing a program stored in the ROM 202 or the storage device 204, or by a processor within the communication circuit 205 executing predetermined software. A storage medium storing a program for implementing the functions may also be provided. Since the details of the processes executed by each functional unit have been described above, only the general functions of the base station device 101 will be outlined here.

[0038] The setting information notification unit 401 notifies the connected terminal device 111 of setting information for determining the occurrence of RLF based on future predictions using the trained model. The setting information transmitted here is information received by the setting information receiving unit 304 of the terminal device 111. The report receiving unit 402 receives, for example, a predetermined report from the terminal device 111, which is transmitted in response to the terminal device 111 determining that the occurrence of RLF is estimated based on future predictions using the trained model. In response to receiving the predetermined report, the handover control unit 403 initiates handover processing to transfer the cell to which the terminal device 111 is connected to another cell before an RLF actually occurs in the terminal device 111. Furthermore, for example, in some cases where the predetermined report is received, the handover control unit 403 may initiate handover processing according to a conventional procedure and determine whether or not re-learning of the trained model is necessary. For example, depending on the type of communication being performed by the terminal device 111, the handover control unit 403 may initiate handover processing for a terminal device 111 that requires highly reliable communication upon receiving a predetermined report, and may perform conventional handover processing for a terminal device 111 that performs other communication. Furthermore, when the handover control unit 403 receives a predetermined report including information on RLF prediction regarding a candidate cell for the handover destination, the handover control unit 403 may select a handover destination, such as not selecting a cell with a high probability of RLF occurrence as a candidate cell for the handover destination of the terminal device 111. The learning unit 404 performs machine learning to generate a trained model for determining the occurrence of RLF based on future predictions. It is sufficient that either the learning unit 305 of the terminal device 111 or the learning unit 404 of the base station device 101 performs machine learning; it is not necessary for both to perform machine learning. Furthermore, machine learning may be performed in a network node different from both the base station device 101 and the terminal device 111. In addition, when conventional handover processing is performed, the base station device 101 may send an instruction to the terminal device 111 to report to the base station device 101 the difference between the estimated value of the RLF occurrence timing and wireless quality using the learned model and the actual measured value of the RLF occurrence timing and wireless quality.Then, if the difference exceeds a predetermined value, either the learning unit 305 of the terminal device 111 or the learning unit 404 of the base station device 101 performs re-learning of the trained model. Also, instead of the difference value between the estimated value and the actual measured value described above, information indicating that re-learning is necessary may be notified from the terminal device 111 to the base station device 101. Note that when machine learning is performed in a network node different from both the base station device 101 and the terminal device 111, the base station device 101 may notify the network node that re-learning should be performed.

[0039] (Processing Flow) An example of a processing flow according to this embodiment is shown in Fig. 5. Note that, since the details of the processing executed in the wireless communication system are as described above, only an overview of the processing flow will be given here, and the details will not be repeated.

[0040] First, the base station device 101 transmits configuration information for estimating future RLF occurrences using the trained model (AI / ML) to the terminal device 111 (S501). This configuration information includes whether to activate estimation of future RLF occurrences using AI / ML, configuration values ​​for RLF determination, and the like. This configuration information is notified from the base station device 101 to the terminal device 111, for example, using an RRC (Radio Resource Control) message including ReportConfig. The configuration information may be notified using other existing information elements (such as MeasConfig), or a new information element for notifying the configuration information may be defined. The configuration information may also be notified as information on the MAC (Media Access Control) layer or the physical layer. The terminal device 111 retains the configuration information and, when instructed to activate estimation of future RLF occurrences using AI / ML, starts the process of estimating future RLF occurrences. In the process of estimating a future RLF occurrence, the terminal device 111 measures the wireless quality of the cell 121 based on signals such as a reference signal transmitted from the base station device 101 (S502). Furthermore, the terminal device 111 measures the wireless quality of neighboring cells (e.g., cells 122 and 123) from signals transmitted in the neighboring cells as needed. The terminal device 111 may measure only the wireless quality of the neighboring cells. The terminal device 111 then inputs the measurement results into a trained model to perform inference (S503), and determines whether the inference predicts that an RLF will occur within a predetermined period of time from the timing at which the measurement value was acquired (S504). If the terminal device 111 determines that an RLF will occur within the predetermined period of time, it transmits a predetermined report to the base station device 101 (S505). In response to receiving this predetermined report, the base station device 101 starts handover processing in the terminal device 111 earlier than conventional handover processing and at a timing such as before timer T310 is started, and transmits a handover instruction (RRC Reconfiguration message or Handover command) to the terminal device 111 (S506).This allows the base station device 101 to quickly hand over the terminal device 111 that is expected to require handover in the near future to another cell, thereby preventing future handover failures in the terminal device 111.

[0041] The above-mentioned "cell" may be read as "beam." That is, the terminal device 111 may measure the wireless quality of at least one of a beam used for communication and a beam not used. In this case, machine learning is performed based on the wireless quality of each beam, and, for example, when it is predicted that the wireless quality will deteriorate in all of multiple cells of a specific cell, control may be performed to execute handover.

[0042] As described above, in this embodiment, by determining whether or not an RLF will occur in the future based on the current wireless quality measured in the terminal device 111, it becomes possible to start handover processing before an RLF actually occurs (for example, before timer T310 is started). This improves the stability of communication in the terminal device 111 and improves communication efficiency in the entire system. This makes it possible to contribute to Goal 9 of the Sustainable Development Goals (SDGs) led by the United Nations, which is to "build resilient infrastructure, promote sustainable industrialization and foster innovation."

[0043] The invention is not limited to the above-described embodiment, and various modifications and variations are possible within the scope of the gist of the invention.

[0044] This application claims priority based on Japanese Patent Application No. 2024-131179, filed August 7, 2024, the entire contents of which are incorporated herein by reference.

Claims

1. A terminal device comprising: a measurement means for measuring wireless quality of a signal transmitted in a cell provided by a base station device to which the terminal device is currently connected; and a transmission means for transmitting a predetermined notification to the base station device based on a value obtained by inputting the wireless quality obtained in a learning phase into a trained model obtained by machine learning using as training data a predetermined value associated with the occurrence of a radio link failure (RLF) when a handover is not performed within a predetermined period of time after the timing at which the wireless quality was obtained.

2. The terminal device of claim 1, wherein the predetermined value includes a value indicating whether or not an RLF has occurred within the predetermined period of time after the timing at which the wireless quality was acquired in the learning phase, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability that the RLF will occur within the predetermined period of time after the timing at which the wireless quality was acquired, and the transmitting means transmits the predetermined notification including the probability to the base station device when the probability exceeds a predetermined value.

3. The terminal device of claim 1, wherein the predetermined value includes a value indicating whether or not RLF has occurred at each of a plurality of timings within the predetermined period after the timing at which the wireless quality input in the learning phase is acquired, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability of RLF occurring at the plurality of timings within the predetermined period after the timing at which the wireless quality is acquired, and the transmitting means transmits the predetermined notification to the base station device, the notification including information indicating the probability and the timing corresponding to the probability.

4. The terminal device of claim 1, wherein the predetermined value includes a value indicating whether or not RLF has occurred at each of a plurality of timings within the predetermined period after the timing at which the wireless quality input in the learning phase is acquired, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability that RLF has occurred at the plurality of timings within the predetermined period after the timing at which the wireless quality is acquired, and the transmitting means transmits the predetermined notification to the base station device, the notification including information on the timing among the plurality of timings corresponding to the probability exceeding the predetermined value.

5. The terminal device described in claim 1, wherein the predetermined value includes a value indicating the wireless quality of the cell within the predetermined period after the predetermined period from the time when the wireless quality was acquired, the trained model uses the wireless quality acquired in the inference phase as input and outputs an estimated value of the wireless quality of the cell within the predetermined period after the time when the wireless quality was acquired, and the transmitting means transmits the predetermined notification to the base station device if the estimated value falls below a predetermined value over the predetermined period.

6. The terminal device according to claim 1, wherein the measurement means further measures a second wireless quality of a signal transmitted in a cell different from the cell, and the transmission means transmits to the base station device the predetermined notification further based on a value acquired by inputting the second wireless quality acquired by the terminal device in the inference phase into a trained model acquired by machine learning using as training data the second wireless quality acquired in the learning phase and a second predetermined value associated with the occurrence of RLF when handover to the other cell occurs within the predetermined length period after the predetermined period from the time when the second wireless quality was acquired.

7. A base station device comprising: a receiving means for receiving a predetermined notification based on a value acquired by inputting the wireless quality of a signal transmitted from the base station device in a learning phase from a terminal device connected to a cell provided by the base station device, and inputting the wireless quality acquired by the terminal device in an inference phase into a trained model acquired by machine learning using as training data a predetermined value associated with the occurrence of a radio link failure (RLF) when a handover is not performed within a predetermined length of time after the timing at which the wireless quality was acquired; and a control means for performing control to hand over the terminal device from the cell to another cell based on the reception of the predetermined notification.

8. The base station device according to claim 7, wherein the predetermined value includes a value indicating whether or not an RLF has occurred within the predetermined period of time after the timing at which the wireless quality was acquired in the learning phase, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability of the RLF occurring within the predetermined period of time after the timing at which the wireless quality was acquired, and the receiving means receives the predetermined notification from the terminal device, the notification including the probability exceeding the predetermined value.

9. The base station device of claim 7, wherein the predetermined value includes a value indicating whether or not RLF has occurred at each of a plurality of timings within the predetermined period after the timing at which the wireless quality input in the learning phase is acquired, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability of RLF occurring at the plurality of timings within the predetermined period after the timing at which the wireless quality is acquired, and the receiving means receives the predetermined notification from the terminal device, the notification including information indicating the probability and the timing corresponding to the probability.

10. The base station device of claim 7, wherein the predetermined value includes a value indicating whether or not RLF has occurred at each of a plurality of timings within the predetermined period after the timing at which the wireless quality input in the learning phase is acquired, the trained model is configured to use the wireless quality acquired in the inference phase as input and output the probability that RLF has occurred at the plurality of timings within the predetermined period after the timing at which the wireless quality is acquired, and the receiving means receives from the terminal device the predetermined notification including information on a timing among the plurality of timings corresponding to the probability exceeding the predetermined value.

11. The base station device described in claim 7, wherein the predetermined value includes a value indicating the wireless quality of the cell within the predetermined period after the predetermined period from the time when the wireless quality is acquired, the trained model uses the wireless quality acquired in the inference phase as input and outputs an estimated value of the wireless quality of the cell within the predetermined period after the time when the wireless quality is acquired, and the receiving means receives the predetermined notification transmitted from the terminal device when the estimated value falls below a predetermined value over the predetermined period.

12. The base station device according to claim 7, wherein the predetermined notification is further based on a second value obtained by inputting the second wireless quality obtained by the terminal device in the inference phase into a trained model obtained by machine learning using as training data a second predetermined value associated with the occurrence of RLF when handing over to the other cell within the predetermined period of time after the timing at which the second wireless quality was obtained, the second wireless quality being obtained in a learning phase, and the control means selects the other cell to which the terminal device will be handed over based on the predetermined notification.

13. A control method executed by a terminal device, comprising: measuring wireless quality of a signal transmitted in a cell provided by a base station device to which the terminal device is connected; and transmitting a predetermined notification to the base station device based on a value obtained by inputting the wireless quality obtained by the terminal device in an inference phase into a trained model obtained by machine learning using, as training data, a predetermined value associated with the occurrence of a radio link failure (RLF) when a handover is not performed within a predetermined period of time after a predetermined period from the time when the wireless quality was obtained.

14. A control method executed by a base station device, comprising: receiving a predetermined notification based on a value acquired by inputting, from a terminal device connected to a cell provided by the base station device, the wireless quality of a signal transmitted from the base station device in a learning phase and the wireless quality acquired by the terminal device in an inference phase into a trained model acquired by machine learning using, as training data, a predetermined value associated with the occurrence of a radio link failure (RLF) when a handover is not performed within a predetermined period of time after a predetermined period from the time the wireless quality was acquired; and performing control to hand over the terminal device from the cell to another cell based on the receipt of the predetermined notification.

Images