Wireless communication device and received data estimation method

Abstract

In a wireless communication device 10, a storage unit 15 stores a plurality of AI models including a (1, 1)-th AI model, a (1, 2)-th AI model, ..., and an (M, N)-th AI model. The plurality of AI models are generated by machine learning using, as teacher data, the SNRs and the UE movement speeds for a plurality of combinations each consisting of one of a plurality of SNR ranges and one of a plurality of speed ranges, and are respectively specialized for the plurality of combinations. In the wireless communication device 10, a processor 14: selects, from among the plurality of AI models, any one AI model corresponding to any one SNR range to which the SNR of a received signal belongs from among the plurality of SNR ranges and any one speed range to which the current UE movement speed belongs from among the plurality of speed ranges; and estimates, by using the selected AI model, received data on the basis of the SNR of the received signal and the current UE movement speed.

Description

Wireless Communication Device and Received Data Estimation Method

[0001] The present disclosure relates to a wireless communication device and a received data estimation method.

[0002] In mobile communication, received data may be estimated using an AI (Artificial Intelligence) model.

[0003] Japanese Patent Application Publication No. 2008-507216

[0004] The wireless communication device of the present disclosure has a processor. The processor selects any one AI model corresponding to any one SNR range to which the SNR of the received signal belongs among a plurality of SNR ranges and any one speed range to which the current UE movement speed belongs among a plurality of speed ranges from among a plurality of AI models. Then, the processor estimates the received data using the selected AI model. Each of the plurality of AI models is an AI model generated by machine learning with SNR and UE movement speed as teacher data for each of a plurality of combinations of each of the plurality of SNR ranges and each of the plurality of speed ranges, and is an AI model specialized for each of the plurality of combinations.

[0005] Figure showing a configuration example of the wireless communication device of the present disclosure Figure showing an example of the AI model selection table of the present disclosure

[0006] In mobile communication using RIC (RAN Intelligent Controller), which is one of the 5G wireless network technologies, received data may be estimated using an AI (Artificial Intelligence) model. By using an AI model for estimating received data, an improvement in the estimation performance of received data can be expected.

[0007] In mobile communication, since the communication environment changes variously as the UE (User Equipment) moves, if received data is estimated using a single AI model, the estimation performance of received data may not be improved even when using an AI model.

[0008] Hereinafter, embodiments of the technology of the present disclosure will be described based on the drawings. The same components are denoted by the same reference numerals in the following embodiments.

[0009] The technology disclosed herein can improve communication quality and streamline operations through the use of AI, and will serve as an innovative technological foundation in the telecommunications business, thereby contributing to the achievement of Sustainable Development Goal (SDG) 9, "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation."

[0010] <Configuration of Wireless Communication Device> Figure 1 shows an example configuration of the wireless communication device of this disclosure. In Figure 1, the wireless communication device 10 includes an antenna 11, a wireless receiving unit 12, an A / D conversion unit 13, a separation unit 141, an SNR (Signal-to-Noise Ratio) estimation unit 142, a received data estimation unit 143, and a storage unit 15. The separation unit 141, the SNR estimation unit 142, and the received data estimation unit 143 are implemented by, for example, a processor 14. The storage unit 15 is implemented by, for example, memory or storage. Examples of the processor 14 include a CPU (Central Processing Unit) and a DSP (Digital Signal Processor). The wireless communication device 10 is, for example, a UE or a base station.

[0011] The memory unit 15 stores M × N AI models, including the (1,1) AI model, the (1,2) AI model, ..., the (M,N) AI model.

[0012] In the wireless communication device 10, the wireless receiving unit 12 converts the high-frequency signal (RF signal), which is the received signal received via the antenna 11, into a baseband signal and outputs the converted received signal to the A / D conversion unit 13. If the wireless communication device 10 is a UE (Uninterruptible Equipment), the received signal in the wireless communication device 10 is a downlink signal. On the other hand, if the wireless communication device 10 is a base station, the received signal in the wireless communication device 10 is an uplink signal.

[0013] The A / D conversion unit 13 converts the analog received signal output from the wireless receiving unit 12 into a digital received signal by sampling it at a predetermined sampling frequency, and outputs the converted received signal to the separation unit 141.

[0014] The separation unit 141 separates the digital received signal output from the A / D conversion unit 13 into a DMRS (Demodulation Reference Signal) and a data signal, outputs the DMRS to the SNR estimation unit 142, and outputs the data signal to the received data estimation unit 143.

[0015] The SNR estimation unit 142 estimates the SNR of the received signal using the DMRS output from the separation unit 141, and outputs the estimated SNR (hereinafter sometimes referred to as "estimated SNR") to the received data estimation unit 143.

[0016] The UE moving speed is input to the received data estimation unit 143. The received data estimation unit 143 estimates the received data based on the data signal output from the separation unit 141, the estimated SNR output from the SNR estimation unit 142, and the UE moving speed input to the received data estimation unit 143. If the wireless communication device 10 is a UE, the UE moving speed measured by the wireless communication device 10 is input to the received data estimation unit 143. On the other hand, if the wireless communication device 10 is a base station, the UE moving speed measured by the UE communicating with the wireless communication device 10 is notified from the UE to the wireless communication device 10, and the UE moving speed notified by the UE is input to the received data estimation unit 143.

[0017] The received data estimation unit 143 selects one AI model from among the M x N AI models stored in the memory unit 15 based on the estimated SNR and the current UE movement speed, and estimates the received data using the selected AI model (hereinafter sometimes referred to as the "selected AI model"). The received data estimation unit 143 outputs the received data estimated using the selected AI model (hereinafter sometimes referred to as the "estimated received data") as a received bit sequence. Each of the M x N AI models outputs estimated received data by performing processing equivalent to demodulation and decoding when it receives a data signal from the received data estimation unit 143.

[0018] <Operation of the Received Data Estimation Unit> The operation of the received data estimation unit 143 will be explained below as an example. In the following explanation, we will take the case where M=4, N=5, and 20 AI models, namely the (1,1) AI model, the (1,2) AI model, the (1,3) AI model, the (1,4) AI model, the (1,5) AI model, the (2,1) AI model, ..., the (4,4) AI model, and the (4,5) AI model, are stored in the storage unit 15 as an example.

[0019] Figure 2 shows an example of the AI ​​model selection table of this disclosure. The AI ​​model selection table is set, for example, in the received data estimation unit 143. In the AI ​​model selection table, 20 AI models are associated with multiple ranges of different estimated SNRs (hereinafter sometimes referred to as "SNR ranges") and multiple ranges of different UE movement speeds (hereinafter sometimes referred to as "MS") (hereinafter sometimes referred to as "MS ranges"). In the AI ​​model selection table shown in Figure 2, "a_(m,n)" indicates the "(m,n) AI model".

[0020] In the AI ​​model selection table, the SNR range is divided into four ranges, for example, m=1, 2, 3, and 4. "m=1" corresponds to the range "-20dB ≤ SNR < -10dB" (sometimes referred to as the "m1 range" below), "m=2" corresponds to the range "-10dB ≤ SNR < 0dB" (sometimes referred to as the "m2 range" below), "m=3" corresponds to the range "0dB ≤ SNR < 10dB" (sometimes referred to as the "m3 range" below), and "m=4" corresponds to the range "10dB ≤ SNR < 20dB" (sometimes referred to as the "m4 range" below).

[0021] Furthermore, in the AI ​​model selection table, the MS range is divided into five ranges, for example, n=1, 2, 3, 4, and 5. "n=1" corresponds to the range "0 km / h ≤ MS < 20 km / h" (sometimes referred to as the "n1 range" below), "n=2" corresponds to the range "20 km / h ≤ MS < 40 km / h" (sometimes referred to as the "n2 range" below), "n=3" corresponds to the range "40 km / h ≤ MS < 60 km / h" (sometimes referred to as the "n3 range" below), "n=4" corresponds to the range "60 km / h ≤ MS < 80 km / h" (sometimes referred to as the "n4 range" below), and "n=5" corresponds to the range "80 km / h ≤ MS < 100 km / h" (sometimes referred to as the "n5 range" below).

[0022] On the other hand, the memory unit 15 stores 20 AI models, including the (1,1) AI model, the (1,2) AI model, the (1,3) AI model, the (1,4) AI model, the (1,5) AI model, the (2,1) AI model, ..., the (4,4) AI model, and the (4,5) AI model, which were generated by machine learning using SNR and UE movement speed as training data.

[0023] Specifically, the memory unit 15 stores the following: the (1,1) AI model generated by machine learning under the conditions of m1 range and n1 range; the (1,2) AI model generated by machine learning under the conditions of m1 range and n2 range; the (1,3) AI model generated by machine learning under the conditions of m1 range and n3 range; the (1,4) AI model generated by machine learning under the conditions of m1 range and n4 range; and the (1,5) AI model generated by machine learning under the conditions of m1 range and n5 range. In other words, the (1,1) AI model is an AI model specialized in the ranges m1 and n1, the (1,2) AI model is an AI model specialized in the ranges m1 and n2, the (1,3) AI model is an AI model specialized in the ranges m1 and n3, the (1,4) AI model is an AI model specialized in the ranges m1 and n4, and the (1,5) AI model is an AI model specialized in the ranges m1 and n5.

[0024] Furthermore, the memory unit 15 stores the following: the (2,1) AI model generated by machine learning under the conditions of m2 range and n1 range; the (2,2) AI model generated by machine learning under the conditions of m2 range and n2 range; the (2,3) AI model generated by machine learning under the conditions of m2 range and n3 range; the (2,4) AI model generated by machine learning under the conditions of m2 range and n4 range; and the (2,5) AI model generated by machine learning under the conditions of m2 range and n5 range. In other words, the (2,1) AI model is an AI model specialized in the ranges m2 and n1, the (2,2) AI model is an AI model specialized in the ranges m2 and n2, the (2,3) AI model is an AI model specialized in the ranges m2 and n3, the (2,4) AI model is an AI model specialized in the ranges m2 and n4, and the (2,5) AI model is an AI model specialized in the ranges m2 and n5.

[0025] Furthermore, the memory unit 15 stores the following: the (3,1) AI model generated by machine learning under the conditions of m3 range and n1 range; the (3,2) AI model generated by machine learning under the conditions of m3 range and n2 range; the (3,3) AI model generated by machine learning under the conditions of m3 range and n3 range; the (3,4) AI model generated by machine learning under the conditions of m3 range and n4 range; and the (3,5) AI model generated by machine learning under the conditions of m3 range and n5 range. In other words, the (3,1) AI model is an AI model specialized in the ranges m3 and n1, the (3,2) AI model is an AI model specialized in the ranges m3 and n2, the (3,3) AI model is an AI model specialized in the ranges m3 and n3, the (3,4) AI model is an AI model specialized in the ranges m3 and n4, and the (3,5) AI model is an AI model specialized in the ranges m3 and n5.

[0026] Furthermore, the memory unit 15 stores the following AI models: the (4,1) AI model generated by machine learning under the conditions of m4 range and n1 range; the (4,2) AI model generated by machine learning under the conditions of m4 range and n2 range; the (4,3) AI model generated by machine learning under the conditions of m4 range and n3 range; the (4,4) AI model generated by machine learning under the conditions of m4 range and n4 range; and the (4,5) AI model generated by machine learning under the conditions of m4 range and n5 range. In other words, the (4,1) AI model is an AI model specialized in the m4 range and the n1 range, the (4,2) AI model is an AI model specialized in the m4 range and the n2 range, the (4,3) AI model is an AI model specialized in the m4 range and the n3 range, the (4,4) AI model is an AI model specialized in the m4 range and the n4 range, and the (4,5) AI model is an AI model specialized in the m4 range and the n5 range.

[0027] As described above, each of the multiple (m,n) AI models, m=1 to 4 and n=1 to 5, is generated by machine learning using SNR and UE movement speed as training data for each of the 20 combinations of the four SNR ranges from m1 to m4 and the five MS ranges from n1 to n5.

[0028] The received data estimation unit 143 determines whether the estimated SNR falls within the m1, m2, m3, or m4 range (hereinafter sometimes referred to as "SNR range determination") and whether the current UE movement speed falls within the n1, n2, n3, n4, or n5 range (hereinafter sometimes referred to as "MS range determination"). Then, using the AI ​​model selection table shown in Figure 2, the received data estimation unit 143 selects one AI model from the 20 AI models stored in the storage unit 15 to be used for estimating the received data, based on the determination results of both the SNR range determination and the MS range determination. For example, the received data estimation unit 143 performs the SNR range determination and MS range determination every Δt. Δt is, for example, a predetermined time of 1 second to 10 seconds.

[0029] For example, if the received data estimation unit 143 determines that the estimated SNR falls within the m1 range and that the current UE movement speed falls within the n1 range, it selects the (1,1) AI model based on the AI ​​model selection table and uses the (1,1) AI model to estimate the received data based on the estimated SNR and the current UE movement speed.

[0030] For example, if the received data estimation unit 143 determines that the estimated SNR falls within the m1 range and that the current UE movement speed falls within the n2 range, it selects the (1st, 2nd) AI model based on the AI ​​model selection table and uses the (1st, 2nd) AI model to estimate the received data based on the estimated SNR and the current UE movement speed.

[0031] For example, if the received data estimation unit 143 determines that the estimated SNR falls within the m2 range and that the current UE movement speed falls within the n1 range, it selects the (2,1) AI model based on the AI ​​model selection table and uses the (2,1) AI model to estimate the received data based on the estimated SNR and the current UE movement speed.

[0032] For example, if the received data estimation unit 143 determines that the estimated SNR falls within the m2 range and that the current UE movement speed falls within the n2 range, it selects the (2,2) AI model based on the AI ​​model selection table and uses the (2,2) AI model to estimate the received data based on the estimated SNR and the current UE movement speed.

[0033] Furthermore, even if the SNR range to which the estimated SNR at past time t1 (hereinafter sometimes referred to as "SNR_t1") corresponds differs from the SNR range to which the estimated SNR at the current time t2 (hereinafter sometimes referred to as "SNR_t2") corresponds after Δt has elapsed from time t1 differs, if the absolute value of the difference between SNR_t1 and SNR_t2 (hereinafter sometimes referred to as "SNR change amount") is less than ΔSNR, the received data estimation unit 143 may not change the AI ​​model used for estimating the received data and may instead adopt the selected AI model at time t1 as the selected AI model at time t2.

[0034] Furthermore, even if the MS range to which the UE movement speed at past time t1 (hereinafter sometimes referred to as "MS_t1") corresponds and the MS range to which the UE movement speed at the current time t2 (hereinafter sometimes referred to as "MS_t2") after Δt has elapsed from time t1 correspond differ, the received data estimation unit 143 may not change the AI ​​model used for estimating the received data and may instead adopt the AI ​​model selected at time t1 as the selected AI model at time t2, provided that the absolute value of the difference between MS_t1 and MS_t2 (hereinafter sometimes referred to as "MS change amount") is less than ΔMS.

[0035] However, the received data estimation unit 143 prefers to use the AI ​​model selected at time t1 as the selected AI model at time t2 only when the SNR change is less than ΔSNR and the MS change is less than ΔMS. In other words, the received data estimation unit 143 prefers to change the AI ​​model used for estimating received data between time t1 and time t2 if at least one of the following conditions is met: the SNR change is greater than or equal to ΔSNR, or the MS change is greater than or equal to ΔMS.

[0036] Therefore, for example, if ΔSNR, which is a predetermined threshold for the SNR change, is 3 dB, and ΔMS, which is a predetermined threshold for the MS change, is 5 km / h, the received data estimation unit 143 determines the selected AI model as follows.

[0037] For example, if the selected AI model at time t1 is the (1,1) AI model, and the estimated SNR at time t2 is between -7 dB and 0 dB, and the UE movement speed at time t2 is less than 25 km / h, then at time t2, the selected AI model is changed to the (2,1) AI model.

[0038] For example, if the selected AI model at time t1 is the (1,1) AI model, and the estimated SNR at time t2 is less than -7 dB, and the UE movement speed at time t2 is 25 km / h or more but less than 40 km / h, then at time t2, the selected AI model is changed to the (1,2) AI model.

[0039] Also, for example, when the selected AI model at time t1 is the (1,1) AI model, if the estimated SNR at time t2 is -7 dB or more and less than 0 dB, and the UE movement speed at time t2 is 25 km / h or more and less than 40 km / h, then at time t2, the selected AI model is changed to the (2,2) AI model.

[0040] Also, for example, when the selected AI model at time t1 is the (1,1) AI model, if the estimated SNR at time t2 is 0 dB or more and less than 10 dB, and the UE movement speed at time t2 is 40 km / h or more and less than 60 km / h, then at time t2, the selected AI model is changed to the (3,2) AI model.

[0041] Also, for example, when the selected AI model at time t1 is the (3,5) AI model, if the estimated SNR at time t2 is 13 dB or more, and the UE movement speed at time t2 is 75 km / h or more, then at time t2, the selected AI model is changed to the (4,5) AI model.

[0042] Also, for example, when the selected AI model at time t1 is the (4,4) AI model, if the estimated SNR at time t2 is 7 dB or more, and the UE movement speed at time t2 is 85 km / h or more, then at time t2, the selected AI model is changed to the (4,5) AI model.

[0043] On the other hand, for example, when the selected AI model at time t1 is the (1,1) AI model, if the estimated SNR at time t2 is less than -7 dB, and the UE movement speed at time t2 is less than 25 km / h, then at time t2, the selected AI model is not changed.

[0044] Also, for example, when the selected AI model at time t1 is the (1,2) AI model, if the estimated SNR at time t2 is less than -7 dB, and the UE movement speed at time t2 is 15 km / h or more and less than 45 km / h, then at time t2, the selected AI model is not changed.

[0045] For example, when the selected AI model at time t1 is the (2,1) AI model, if the estimated SNR at time t2 is -13 dB or more and less than 3 dB, and the UE moving speed at time t2 is less than 25 km / h, then at time t2, the selected AI model is not changed.

[0046] For example, when the selected AI model at time t1 is the (2,2) AI model, if the estimated SNR at time t2 is -13 dB or more and less than 3 dB, and the UE moving speed at time t2 is 15 km / h or more and less than 45 km / h, then at time t2, the selected AI model is not changed.

[0047] For example, when the selected AI model at time t1 is the (3,5) AI model, if the estimated SNR at time t2 is -3 dB or more and less than 13 dB, and the UE moving speed at time t2 is 75 km / h or more, then at time t2, the selected AI model is not changed.

[0048] For example, when the selected AI model at time t1 is the (4,4) AI model, if the estimated SNR at time t2 is 7 dB or more, and the UE moving speed at time t2 is 55 km / h or more and 85 km / h, then at time t2, the selected AI model is not changed.

[0049] For example, when the selected AI model at time t1 is the (4,5) AI model, if the estimated SNR at time t2 is 7 dB or more, and the UE moving speed at time t2 is 75 km / h or more, then at time t2, the selected AI model is not changed.

[0050] The above are descriptions of the embodiments.

[0051] As described above, the wireless communication device of this disclosure (wireless communication device 10 in the embodiment) has a processor (processor 14 in the embodiment). A storage unit (storage unit 15 in the embodiment) stores a plurality of AI models (the (1,1) AI model, the (1,2) AI model, ..., the (M,N) AI model in the embodiment). Each of the plurality of AI models is an AI model generated by machine learning using SNR and UE movement speed as training data for each of the plurality of combinations of each of the plurality of SNR ranges and each of the plurality of speed ranges, and is an AI model specialized for each of the plurality of combinations. The processor selects one AI model from the plurality of AI models that corresponds to one of the plurality of SNR ranges to which the SNR of the received signal corresponds, and one of the plurality of speed ranges to which the current UE movement speed corresponds. The processor then estimates the received data using the selected AI model.

[0052] By doing this, when estimating received data, an appropriate AI model is selected based on the SNR of the received signal and the current UE movement speed, thereby improving the performance of estimating received data.

[0053] Furthermore, the processor changes the AI ​​model used to estimate received data when the change in the SNR of the received signal is greater than or equal to a threshold, but does not change the AI ​​model used to estimate received data when the change in the SNR of the received signal is less than the threshold. Also, the processor changes the AI ​​model used to estimate received data when the change in the UE movement speed is greater than or equal to a threshold, but does not change the AI ​​model used to estimate received data when the change in the UE movement speed is less than the threshold.

[0054] This prevents frequent changes to the AI ​​model, thereby reducing overhead.

[0055] 10 Wireless communication device 14 Processor 15 Memory unit 142 SNR estimation unit 143 Received data estimation unit

Claims

1. A wireless communication device comprising: a processor that, for each of the multiple combinations of each of the multiple SNR ranges and each of the multiple speed ranges, selects one AI model from among the multiple AI models specialized for each of the multiple combinations, which are generated by machine learning using SNR and UE movement speed as training data, and which corresponds to one of the multiple SNR ranges to which the SNR of the received signal corresponds and one of the multiple speed ranges to which the current UE movement speed corresponds, and estimates the received data using the selected AI model.

2. The wireless communication device according to claim 1, wherein the processor changes the AI ​​model used to estimate the received data in the plurality of AI models when the amount of change in the SNR of the received signal is greater than or equal to a threshold, while not changing the AI ​​model used to estimate the received data in the plurality of AI models when the amount of change is less than the threshold.

3. The wireless communication device according to claim 1 or 2, wherein the processor changes the AI ​​model used for estimating the received data in the plurality of AI models when the amount of change in UE movement speed is greater than or equal to a threshold, while not changing the AI ​​model used for estimating the received data in the plurality of AI models when the amount of change is less than the threshold.

4. UE, which is a wireless communication device according to claim 1.

5. A base station which is a wireless communication device according to claim 1.

6. A method for estimating received data, wherein a processor in a wireless communication device selects one AI model from among a plurality of AI models specialized for each of the plurality of combinations of each of the plurality of SNR ranges and each of the plurality of speed ranges, which corresponds to one of the plurality of SNR ranges to which the SNR of the received signal corresponds, and one of the plurality of speed ranges to which the current UE movement speed corresponds, for each of the plurality of combinations of each of the plurality of SNR ranges and each of the plurality of speed ranges, and estimates the received data using the selected AI model.

Images