Training device, program, system, and training method

Abstract

Provided is a training device comprising: a training execution unit that generates, by machine learning, an autoencoder, which includes an encoder and a decoder and in which an input to the encoder is a bit string to be transmitted from one communication device to another communication device via a wireless environment, a latent space representation generated by the encoder is a signal to be included in a radio wave propagating in the wireless environment in order for the one communication device to transmit the bit string to the other communication device, an input to the decoder is the latent space representation to which noise added to the signal before the signal reaches the other communication device is added, and an output from the decoded is the bit string input to the encoder; and a provision unit that provides the encoder to the one communication device and provides the decoder to the other communication device.

Description

Learning Device, Program, System, and Learning Method

[0001] The present invention relates to a learning device, a program, a system, and a learning method.

[0002] In Patent Document 1, when an information processing device causes a cloud server to process all or part of the processing of an application including an AI (Artificial Intelligence) / ML (Machine Learning) model via wireless communication, it is described that the processing of the AI / ML model is processed by the NWDAF (Network Data Analytics Function) of a 5G system. [Prior Art Documents] [Patent Documents] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2024-029281

[0003] The essence of the RAN (Radio Access Network) is the realization of the physical layer (+ data link layer) that delivers bit sequences to the destination as needed. Until now, performance optimization has been attempted by performing rule-based processing for each block that humans have considered. Standardization protocols such as 3GPP (registered trademark) fall under this. Such performance optimization for each block falls into local optimization, and the frequency utilization efficiency may reach its peak.

[0004] According to an embodiment of the present invention, a learning device having a configuration that contributes to solving such problems is provided. The learning device is an auto encoder including an encoder and a decoder, wherein the input to the encoder is a bit sequence transmitted from one communication device to another communication device via a wireless environment, and the latent space representation generated by the encoder is used as a signal included in the radio wave propagating through the wireless environment for the one communication device to transmit the bit sequence to the other communication device. The input to the decoder is the latent space representation with noise added to the signal until the signal reaches the other communication device, and the output from the decoder is the bit sequence input to the encoder. A learning execution unit that generates an auto encoder by machine learning may be provided. The learning device may include a providing unit that provides the encoder to the one communication device and provides the decoder to the other communication device.

[0005] In the learning device, the learning execution unit may generate an autoencoder by machine learning, wherein the input to the encoder is a bit sequence transmitted from one communication device to the other communication device via a wireless environment, the latent space representation generated by the encoder is a plurality of complex number signals generated from the bit sequence by OFDM (Orthogonal Frequency-Division Multiplexing) for the one communication device to transmit the bit sequence to the other communication device and transmitted by a plurality of antennas provided by the one communication device, the input to the decoder is the latent space representation with noise added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the other communication device, and the output from the decoder is the bit sequence input to the encoder.

[0006] Any of the learning devices described above may include a learning data acquisition unit that acquires a dataset including a bit sequence transmitted from a wireless communication terminal to a wireless base station via a wireless environment, and noise added to the signal transmitted by the wireless communication terminal, which includes the bit sequence in radio waves for transmission to the wireless base station, until it reaches the wireless base station. The learning execution unit may generate a first autoencoder including a first encoder and a first decoder by machine learning using a plurality of the datasets, and the providing unit may provide the first encoder to the wireless communication terminal and the first decoder to the wireless base station. The learning data acquisition unit may generate a plurality of types of noise added to the signal transmitted by the wireless communication terminal, which includes the bit sequence in radio waves for transmission to the wireless base station, until it reaches the wireless base station, using a channel model of the wireless environment between the wireless communication terminal and the wireless base station.

[0007] In any of the learning devices described above, the learning execution unit may generate a second autoencoder including a second encoder and a second decoder by machine learning using a plurality of datasets that include bit sequences transmitted from a wireless base station to a wireless communication terminal via a wireless environment, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the bit sequences to the wireless communication terminal before it reaches the wireless communication terminal. The providing unit may provide the second encoder to the wireless base station and the second decoder to the wireless communication terminal. The learning execution unit may generate a second autoencoder including a second encoder that takes a plurality of bit sequences and a plurality of terminal IDs as inputs, and a second decoder that takes a plurality of bit sequences and a plurality of terminal IDs as outputs, by machine learning using a plurality of datasets that include a plurality of bit sequences transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, the terminal IDs of each of the plurality of wireless communication terminals, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals before it reaches the plurality of wireless communication terminals.

[0008] In any of the learning devices described above, the learning execution unit may generate a second autoencoder, which includes a second encoder that takes a plurality of bit sequences and a plurality of channel information as inputs, and a second decoder that outputs the plurality of bit sequences, by machine learning using a plurality of datasets that include a plurality of bit sequences transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, channel information for each of the plurality of wireless communication terminals, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals, until the signal reaches the plurality of wireless communication terminals.

[0009] In any of the learning devices described above, the learning execution unit may generate the second autoencoder, including the second encoder and the second decoder, by machine learning using a plurality of datasets that include a plurality of bit sequences to be transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, each of which has the priority of each of the plurality of wireless communication terminals applied, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals, until it reaches the plurality of wireless communication terminals.

[0010] Any of the learning devices described above may include a RAN control unit for controlling RAN and an AI processing unit for performing AI processing, and the AI ​​processing unit may include the learning execution unit.

[0011] According to one embodiment of the present invention, a program is provided for causing a computer to function as the learning device.

[0012] According to one embodiment of the present invention, a system comprising the learning device and the wireless base station is provided.

[0013] According to one embodiment of the present invention, a learning method is provided that is performed by a computer, and the learning method may include a learning execution step of generating an autoencoder by machine learning, wherein the input to the encoder is a bit sequence to be transmitted from one communication device to another communication device via a wireless environment, the latent space representation generated by the encoder is a signal to be included in radio waves propagating the wireless environment for the one communication device to transmit the bit sequence to the other communication device, the input to the decoder is the latent space representation with noise added to the signal by the time the signal reaches the other communication device, and the output from the decoder is the bit sequence input to the encoder. The learning method may also include a providing step of providing the encoder to the one communication device and providing the decoder to the other communication device.

[0014] The types of AI processing performed by the AI ​​processing unit include AI processing related to RAN control (sometimes referred to as RAN control AI processing) and AI processing not related to RAN control (sometimes referred to as non-RAN control AI processing).

[0015] An example of AI-based RAN control processing is RIC (RAN Intelligent Controller). RIC is a technology that uses AI to optimize RAN wireless resources and automate RAN operations. RIC includes Non-RT RIC (Non-Real Time RIC) and Near-RT RIC (Near-Real Time RIC). Non-RT RIC is sometimes called Centralized RIC. Non-RT RIC is located within the SMO (Service Management and Orchestration) that manages and orchestrates the RAN. Non-RT RIC generates and notifies policies related to RAN control and transmits information to Near-RT RIC. For example, a Non-RT RIC generates a learning model for RAN control by performing machine learning using data collected from the RAN, and sends it to a Near-RT RIC. A Near-RT RIC is sometimes called a Distributed RIC. Compared to a Non-RT RIC, a Near-RT RIC is located closer to the RAN nodes (RU (Radio Unit), DU (Distributed Unit), CU (Central Unit)) and performs control of the RAN nodes and resources. Compared to a Non-RT RIC, a Near-RT RIC performs processing with higher real-time capabilities. For example, a Near-RT RIC performs inference processing related to RAN control using the learning model obtained from a Non-RT RIC. RAN control AI processing is not limited to RICs.

[0016] Non-RAN controlled AI processing may be a so-called MEC AI application. Non-RAN controlled AI processing includes learning and inference processing of any AI that is not related to RAN control.

[0017] It should be noted that the above summary of the invention does not enumerate all the necessary features of the present invention. Furthermore, subcombinations of these features may also constitute an invention.

[0018] A schematic example of system 10 is shown. A schematic example of the logical configuration 80 of the base station function in a 5G mobile communication network is shown. A schematic example of the autoencoder 400 generated by the learning device 100 is shown. A schematic example of the autoencoder 500 generated by the learning device 100 is shown. A schematic example of the autoencoder 600 generated by the learning device 100 is shown. A schematic example of the autoencoder 600 generated by the learning device 100 is shown. A schematic example of the autoencoder 600 generated by the learning device 100 is shown. A schematic example of the autoencoder 600 generated by the learning device 100 is shown. A schematic example of the functional configuration of the learning device 100 is shown. A schematic example of the functional configuration of the wireless base station 200 is shown. A schematic example of the functional configuration of the wireless communication terminal 300 is shown. A schematic example of the environment of the learning device 100 is shown. A schematic example of the functional configuration of the learning device 100 is shown. A schematic example of the hardware configuration of a computer 1200 that functions as a learning device 100, a wireless base station 200, or a wireless communication terminal 300 is shown.

[0019] The present invention will be described below through embodiments, but these embodiments are not intended to limit the scope of the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0020] Figure 1 schematically shows an example of system 10. System 10 includes a learning device 100. System 10 may include a plurality of wireless base stations 200. System 10 may also include a plurality of wireless communication terminals 300.

[0021] The learning device 100 communicates with multiple wireless base stations 200 via the network 50. The network 50 includes a mobile communication network. The mobile communication network may conform to any of the following mobile communication systems: LTE (Long Term Evolution), 5G (5th Generation), 3G (3rd Generation), and 6G (6th Generation) or later.

[0022] The learning device 100 may be located within the core network of the mobile communication network. The learning device 100 may also be located outside the core network of the mobile communication network. The network 50 may include the Internet, and the learning device 100 may be located on the Internet.

[0023] In the mobile communication network, a RAN is provided that plays a role in connecting the wireless communication terminal 300 and the core network. The learning device 100 according to this embodiment performs learning that contributes to the optimization of the RAN.

[0024] Figure 2 schematically shows an example of the logical configuration 80 of a base station function in a 5G mobile communication network. As mentioned above, the essence of RAN is the realization of a physical layer (and data link layer) that sends bit sequences to the destination as needed, and conventionally, performance optimization has been achieved by human-designed rule-based processing for each block, as illustrated in Figure 2. Such block-by-block performance optimization can lead to local optimization, and frequency utilization efficiency may plateau.

[0025] The system 10 according to this embodiment basically enables overall optimization of the RAN by treating the RAN as an autoencoder having the same set of bit sequences as input and output. Specifically, in system 10, the RAN is implemented by an encoder-decoder model having a bandpass filter and carrier frequency conversion as a normalization layer.

[0026] An autoencoder is a neural network designed to output the same data as the input data, and it has a very high affinity with RANs, which transmit the same bit sequence from the source to the destination. In RANs, the bit sequence is transmitted from the source in radio waves propagating through the wireless environment, but noise is added before it reaches the destination. In existing RANs, rule-based processing for each block, devised by humans, is used to ensure that the original bit sequence can be reproduced at the destination even if noise is added in the wireless environment. In the system 10 according to this embodiment, the input to the encoder is a bit sequence, and noise added to the latent space representation, which is the output from the encoder, is removed by the decoder. An autoencoder is then generated so that the decoder can output the bit sequence input to the encoder, and this autoencoder is used in the RAN.

[0027] Figure 3 schematically shows an example of an autoencoder 400 generated by the learning device 100. The autoencoder 400 shown in Figure 3 targets a bit sequence 302 transmitted from a wireless communication terminal 300 to a wireless base station 200.

[0028] The autoencoder 400 includes an encoder 410 and a decoder 420. The learning device 100 generates an autoencoder 400 by machine learning, with the input to the encoder 410 being a bit sequence 302 transmitted from the wireless communication terminal 300 to the wireless base station 200, the latent space representation 430 generated by the encoder 410 being a signal to be included in the radio waves propagating through the wireless environment by the wireless communication terminal 300 to transmit the bit sequence 302 to the wireless base station 200, the input to the decoder 420 being the latent space representation 430 with noise 440 added to the signal by the time it reaches the wireless base station 200, and the output from the decoder 420 being the bit sequence 302.

[0029] As a specific example, the learning device 100 generates an autoencoder 400 by machine learning, in which the input to the encoder 410 is a bit sequence 302, the latent space representation 430 generated by the encoder 410 is a plurality of complex number signals generated from the bit sequence 302 by OFDM for the wireless communication terminal 300 to transmit the bit sequence 302 to the wireless base station 200, the input to the decoder 420 is the latent space representation 430 with noise 440 added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the wireless base station 200, and the output from the decoder 420 is a bit sequence 302.

[0030] The learning device 100 may generate an autoencoder 400 by machine learning using multiple datasets that include bit sequences 302 actually transmitted from the wireless communication terminal 300 to the wireless base station 200 in the past, and noise 440 added to the signal transmitted by the wireless communication terminal 300 in radio waves to transmit the bit sequence 302 to the wireless base station 200, before it reaches the wireless base station 200.

[0031] The noise 440 included in the dataset may be noise obtained through actual measurements. The noise 440 included in the dataset may be noise obtained through simulation. For example, the learning device 100 may generate multiple types of noise 440 using a channel model of the wireless environment between the wireless communication terminal 300 and the wireless base station 200.

[0032] The learning device 100 generates an autoencoder 400 by performing learning that adjusts the encoder 410 and decoder 420 so that even when noise 440 is added to the latent space representation 430 generated by the encoder 410, the decoder 420 outputs the same bit sequence 302 as the bit sequence 302 input to the encoder 410. The autoencoder 400 may be a so-called DAE (Denoising Autoencoder).

[0033] The learning device 100 provides the encoder 410 included in the generated autoencoder 400 to the wireless communication terminal 300 and the decoder 420 to the wireless base station 200. The wireless communication terminal 300 inputs the bit sequence 302 to the encoder 410 and transmits the signal output from the encoder 410 as part of the radio waves. For example, the wireless communication terminal 300 transmits the complex number signal output from the encoder 410 as part of the radio waves. The wireless base station 200 receives the signal with noise added in the wireless environment, inputs it to the decoder 420, and obtains the bit sequence 302.

[0034] By performing training using multiple datasets, it is possible to generate an autoencoder 400 that can generate an optimal complex number signal so that even when noise is added to the complex number signal in a wireless environment, the wireless base station 200 can obtain the same bit sequence 302 as the bit sequence 302 transmitted by the wireless communication terminal 300. This makes it possible to achieve a RAN with higher frequency efficiency and time efficiency compared to existing block-by-block performance optimization.

[0035] Figure 4 schematically shows an example of an autoencoder 500 generated by the learning device 100. The autoencoder 500 shown in Figure 4 is intended for the bit sequence 202 transmitted from the wireless base station 200 to the wireless communication terminal 300.

[0036] The autoencoder 500 includes an encoder 510 and a decoder 520. The learning device 100 generates an autoencoder 500 by machine learning, with the input to the encoder 510 being a bit sequence 202 to be transmitted from the radio base station 200 to the wireless communication terminal 300, the latent space representation 530 generated by the encoder 510 being a signal to be included in the radio waves propagated by the radio base station 200 in the wireless environment to transmit the bit sequence 202 to the wireless communication terminal 300, the input to the decoder 520 being the latent space representation 530 with noise 540 added to the signal by the time the signal reaches the wireless communication terminal 300, and the output from the decoder 520 being the bit sequence 202.

[0037] As a specific example, the learning device 100 generates an autoencoder 500 by machine learning, in which the input to the encoder 510 is a bit sequence 202, the latent space representation 530 generated by the encoder 510 is a plurality of complex number signals generated from the bit sequence 202 by OFDM for the wireless base station 200 to transmit the bit sequence 202 to the wireless communication terminal 300, the input to the decoder 520 is the latent space representation 530 with noise 540 added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the wireless communication terminal 300, and the output from the decoder 520 is a bit sequence 202.

[0038] The learning device 100 may generate an autoencoder 500 by machine learning using multiple datasets that include bit sequences 202 actually transmitted from the wireless base station 200 to the wireless communication terminal 300 in the past, and noise 540 added to the signal transmitted by the wireless base station 200 in radio waves to transmit the bit sequence 202 to the wireless communication terminal 300, before it reaches the wireless communication terminal 300.

[0039] The noise 540 included in the dataset may be noise obtained through actual measurements. The noise 540 included in the dataset may be noise obtained through simulation. For example, the learning device 100 may generate multiple types of noise 540 using a channel model of the wireless environment between the wireless base station 200 and the wireless communication terminal 300.

[0040] The learning device 100 generates an autoencoder 500 by performing learning that adjusts the encoder 510 and decoder 520 so that even when noise 540 is added to the latent space representation 530 generated by the encoder 510, the decoder 520 outputs the same bit sequence 302 as the bit sequence 302 input to the encoder 510. The autoencoder 500 may be a so-called DAE.

[0041] The learning device 100 provides the encoder 510 included in the generated autoencoder 500 to the wireless base station 200 and the decoder 520 to the wireless communication terminal 300. The wireless base station 200 inputs the bit sequence 202 to the encoder 510 and transmits the signal output from the encoder 510 as part of the radio waves. For example, the wireless base station 200 transmits the complex number signal output from the encoder 510 as part of the radio waves. The wireless communication terminal 300 receives the signal with noise added in the wireless environment, inputs it to the decoder 520, and obtains the bit sequence 202.

[0042] Figure 5 schematically shows an example of an autoencoder 600 generated by the learning device 100. Here, we will mainly explain the differences from the autoencoder 500. The autoencoder 600 shown in Figure 5 is intended for multiple bit sequences 202 to be transmitted from the wireless base station 200 to multiple wireless communication terminals 300. In Figure 5, for the sake of simplicity, wireless communication terminals 320 and 340 are shown as examples, but the number of wireless communication terminals 300 targeted by the autoencoder 600 is not limited to two.

[0043] The autoencoder 600 includes an encoder 610 and a plurality of decoders 620. In Figure 5, decoders 622 and 624 are shown as examples of the plurality of decoders 620. The learning device 100 uses machine learning to generate an autoencoder 500 in which the input to encoder 610 is bit sequences 202 and 204 to be transmitted to wireless communication terminals 320 and 340, respectively, the latent space representation 630 generated by encoder 610 is a signal to be included in radio waves propagating in the wireless environment by the wireless base station 200 to transmit bit sequences 202 and 204 to wireless communication terminals 320 and 340, the input to decoders 622 and 624 is the latent space representation 630 with noise 640 added to the signal by the time it reaches wireless communication terminals 320 and 340, the output from decoder 622 is bit sequence 202, and the output from decoder 624 is bit sequence 204.

[0044] As a specific example, the learning device 100 uses the input to the encoder 610 as the bit sequence 202, and uses the latent space representation 630 generated by the encoder 610 as a plurality of complex signals generated from the bit sequences 202 and 204 by OFDM for the radio base station 200 to transmit the bit sequences 202 and 204 to the radio communication terminals 320 and 340, and transmitted by a plurality of antennas included in the radio base station 200. The input to the decoders 622 and 624 is the latent space representation 630 added with the noise 640 added to the plurality of complex signals until the plurality of complex signals transmitted by the plurality of antennas reach the radio communication terminals 320 and 340. The output from the decoder 622 is used as the bit sequence 202, and the output from the decoder 624 is used as the bit sequence 204, and the autoencoder 600 is generated by machine learning.

[0045] The learning device 100 may generate the autoencoder 600 by machine learning using a plurality of data sets including a plurality of bit sequences actually transmitted from the radio base station 200 to a plurality of radio communication terminals 300 in the past and the noise 640 added to the signals transmitted by the radio base station 200 included in the radio waves until the signals reach the plurality of radio communication terminals 300.

[0046] The noise 640 included in the data set may be noise obtained by actual measurement. The noise 640 included in the data set may be noise obtained by simulation. For example, the learning device 100 may generate a plurality of types of noise 640 using a channel model of the radio environment between the radio base station 200 and the plurality of radio communication terminals 300.

[0047] The learning device 100 generates the autoencoder 600 by performing learning to adjust the encoder 610, the decoder 622, and the decoder 624 so that, even when noise 640 is added to the latent space representation 630 generated by the encoder 610, the same bit sequence 202 as the bit sequence 202 input to the encoder 610 is output from the decoder 622 and the same bit sequence 204 as the bit sequence 204 input to the encoder 610 is output from the decoder 624.

[0048] The learning device 100 provides the encoder 610 included in the generated autoencoder 600 to the radio base station 200, provides the decoder 622 to the wireless communication terminal 320, and provides the decoder 624 to the wireless communication terminal 340. The radio base station 200 inputs the bit sequence 202 and the bit sequence 204 to the encoder 610 and transmits the signal output from the encoder 610 included in radio waves. For example, the radio base station 200 transmits the complex number signal output from the encoder 610 included in radio waves. The wireless communication terminal 320 receives the signal with noise added in the wireless environment, inputs it to the decoder 622, and acquires the bit sequence 202. The wireless communication terminal 340 receives the signal with noise added in the wireless environment, inputs it to the decoder 624, and acquires the bit sequence 204.

[0049] The learning device 100 may apply the technical idea of RAG (Retrieval Augmented Generation) in an LLM (Large Language Model). The learning device 100, for example, performs machine learning assuming that, in addition to the plurality of bit sequences transmitted to the plurality of wireless communication terminals 300, information corresponding to the plurality of wireless communication terminals 300 is acquired from the core side of the mobile communication network and input to the encoder 610.

[0050] An example of information corresponding to the wireless communication terminal 300 is a terminal ID that can identify the wireless communication terminal 300. Examples of terminal IDs include IMSI (International Mobile Subscriber Identity) and IMEI (International Mobile Equipment Identity), but are not limited to these; any information that can identify the wireless communication terminal 300 is acceptable.

[0051] An example of information corresponding to the wireless communication terminal 300 is the channel information of the wireless communication terminal 300. The channel information includes, for example, the CSI (Channel State Information) of the wireless communication terminal 300. The channel information may also include information about the physical channel of the wireless communication terminal 300 or information about the logical channel of the wireless communication terminal 300.

[0052] An example of information corresponding to the wireless communication terminal 300 is the priority of the wireless communication terminal 300. In existing mobile communication networks, priority is used to provide a better communication environment (bandwidth, latency, signal quality, etc.) to wireless communication terminals 300 with higher priority.

[0053] Figure 6 schematically shows an example of an autoencoder 600 generated by the learning device 100. Here, we will mainly explain the differences from the autoencoder 600 shown in Figure 5. In the example shown in Figure 6, the learning device 100 inputs the terminal ID of the wireless communication terminal 300 to the encoder 610, in addition to the bit sequence to be transmitted to the wireless communication terminal 300.

[0054] The autoencoder 600 shown in Figure 6 includes an encoder 610, a decoder 622, and a decoder 624. The learning device 100 takes the input to the encoder 610 as a bit sequence 202 and the terminal ID 212 of the wireless communication terminal 320 to be transmitted to the wireless communication terminal 320, and a bit sequence 204 and the terminal ID 214 of the wireless communication terminal 340 to be transmitted to the wireless communication terminal 340. The latent space representation 630 generated by the encoder 610 is used by the wireless base station 200 to propagate the wireless environment in order to transmit the bit sequence 202 and terminal ID 212 and the bit sequence 204 and terminal ID 214 to the wireless communication terminals 320 and 340. An autoencoder 600 is generated by machine learning, in which the signals to be included in the wave are set as a latent space representation 630 with noise 640 added to the signal by the time it reaches the wireless communication terminals 320 and 340, the inputs to decoders 622 and 624 are set as bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214, and the output from decoder 624 is bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214.

[0055] As a specific example, the learning device 100 takes bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214 as inputs to the encoder 610, and the latent space representation 630 generated by the encoder 610 is used by the wireless base station 200 to transmit bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214 to wireless communication terminals 320 and 340, and is generated from bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214 by OFDM and transmitted by multiple antennas provided by the wireless base station 200. Autoencoder 600 is generated by machine learning, which takes a complex number signal as the input to decoders 622 and 624 as a latent space representation 630 in which noise 640 is added to the complex number signals transmitted by the multiple antennas as they reach wireless communication terminals 320 and 340, and outputs from decoder 622 as bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214, and outputs from decoder 624 as bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214.

[0056] The learning device 100 may generate an autoencoder 600 by machine learning using multiple datasets that include multiple bit sequences and multiple terminal IDs that have actually been transmitted from the wireless base station 200 to multiple wireless communication terminals 300 in the past, and noise 640 that is added to the signal transmitted by the wireless base station 200 in radio waves to transmit the bit sequences and terminal IDs to the multiple wireless communication terminals 300, before it reaches the multiple wireless communication terminals 300.

[0057] The learning device 100 generates an autoencoder 600 by performing learning that adjusts the encoder 610, decoder 622, and decoder 624 so that even when noise 640 is added to the latent space representation 630 generated by the encoder 610, the decoder 622 and decoder 624 output the same bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214 as the bit sequence 202 and terminal ID 212 and bit sequence 204 and terminal ID 214 input to the encoder 610.

[0058] The learning device 100 provides the encoder 610 included in the generated autoencoder 600 to the wireless base station 200, the decoder 622 to the wireless communication terminal 320, and the decoder 624 to the wireless communication terminal 340. The wireless base station 200 obtains the terminal ID 212 of the wireless communication terminal 320 and the terminal ID 214 of the wireless communication terminal 340 from the core side, inputs the bit sequence 202 and terminal ID 212 and the bit sequence 204 and terminal ID 214 to the encoder 610, and transmits the signal output from the encoder 610 as part of the radio waves. For example, the wireless base station 200 transmits the complex number signal output from the encoder 610 as part of the radio waves. The wireless communication terminal 320 receives a signal with noise added in the wireless environment, inputs it to the decoder 622, and obtains the bit sequence 202 and terminal ID 212 and the bit sequence 204 and terminal ID 214. Wireless communication terminal 320 acquires bit sequence 202, which corresponds to its own terminal ID 212, as a bit sequence addressed to itself. Wireless communication terminal 340 receives a signal with noise added in the wireless environment and inputs it to decoder 624 to acquire bit sequence 202 and terminal ID 212, and bit sequence 204 and terminal ID 214. Wireless communication terminal 340 acquires bit sequence 204, which corresponds to its own terminal ID 214, as a bit sequence addressed to itself.

[0059] Figure 7 schematically shows an example of an autoencoder 600 generated by the learning device 100. Here, we will mainly explain the differences from the autoencoder 600 shown in Figure 5. In the example shown in Figure 7, learning is realized in which a bit sequence is transmitted using channel information as a hyperparameter.

[0060] The autoencoder 600 shown in Figure 7 includes an encoder 610, a decoder 622, and a decoder 624. The learning device 100 takes the input to the encoder 610 as a bit sequence 202 and channel information 222 of the wireless communication terminal 320 to be transmitted to the wireless communication terminal 320, and a bit sequence 204 and channel information 224 of the wireless communication terminal 340 to be transmitted to the wireless communication terminal 340. The latent space representation 630 generated by the encoder 610 is used as a signal to be included in the radio waves propagating in the wireless environment by the wireless base station 200 to transmit the bit sequence 202 and channel information 222 and the bit sequence 204 and channel information 224 to the wireless communication terminals 320 and 340. The autoencoder 600 uses a latent space representation 630, which includes noise 640 added to the signal by the time it reaches the wireless communication terminals 320 and 340, as input to decoders 622 and 624, and the output from decoder 622 as a bit sequence 202 and the output from decoder 624 as a bit sequence 204, and the autoencoder 600 is generated by machine learning such that the bit sequence 202 is transmitted with a signal corresponding to channel information 222 and the bit sequence 204 is transmitted with a signal corresponding to channel information 224.

[0061] As a specific example, the learning device 100 takes bit sequence 202 and channel information 222 and bit sequence 204 and channel information 224 as inputs to the encoder 610, and the latent space representation 630 generated by the encoder 610 is a plurality of complex number signals generated from bit sequence 202 and bit sequence 204 by OFDM and transmitted by a plurality of antennas provided by the wireless base station 200 for the wireless base station 200 to transmit bit sequence 202 and channel information 222 and bit sequence 204 and channel information 224 to wireless communication terminals 320 and 340, and the decoder 622 and decoder The autoencoder 600 takes as input to 624 a latent space representation 630 in which noise 640 is added to the multiple complex number signals transmitted by the multiple antennas by the time they reach the wireless communication terminals 320 and 340, the output from decoder 622 is a bit sequence 202, and the output from decoder 624 is a bit sequence 204, and the autoencoder 600 is generated by machine learning such that bit sequence 202 is transmitted as a complex number signal corresponding to channel information 222, and bit sequence 204 is transmitted as a complex number signal corresponding to channel information 224.

[0062] The learning device 100 provides the encoder 610 included in the generated autoencoder 600 to the wireless base station 200, the decoder 622 to the wireless communication terminal 320, and the decoder 624 to the wireless communication terminal 340. The wireless base station 200 obtains the channel information 222 of the wireless communication terminal 320 and the channel information 224 of the wireless communication terminal 340 from the core side, inputs the bit sequence 202 and channel information 222 and the bit sequence 204 and channel information 224 to the encoder 610, and transmits the signal output from the encoder 610 as part of the radio waves. For example, the wireless base station 200 transmits the complex number signal output from the encoder 610 as part of the radio waves. The wireless communication terminal 320 receives a signal with noise added in the wireless environment, inputs it to the decoder 622, and obtains the bit sequence 202. The wireless communication terminal 340 receives a signal with noise added in the wireless environment, inputs it to the decoder 624, and obtains the bit sequence 204.

[0063] The learning device 100 may acquire communication feedback. For example, the learning device 100 acquires feedback on communication quality, such as throughput, from multiple wireless communication terminals 300, and determines whether the communication quality has deteriorated by inputting channel information. The learning device 100 trains the autoencoder 600 by performing at least one of the following: giving a penalty when the communication quality deteriorates by inputting channel information, and giving a reward when the communication quality improves by inputting channel information. By continuously performing such training, it is possible to realize an autoencoder and decoder that can perform encoding and decoding in a way that improves communication quality when channel information is input.

[0064] Figure 8 schematically shows an example of an autoencoder 600 generated by the learning device 100. Here, we will mainly explain the differences from the autoencoder 600 shown in Figure 5.

[0065] The autoencoder 600 shown in Figure 8 includes an encoder 610, a decoder 622, and a decoder 624. The learning device 100 takes the input to encoder 610 as a bit sequence 202 reflecting the priority 232 of wireless communication terminal 320 and a bit sequence 204 reflecting the priority 234 of wireless communication terminal 340, and uses the latent space representation 630 generated by encoder 610 as a signal to be included in radio waves propagating in the wireless environment for wireless base station 200 to transmit bit sequences 202 and 204 to wireless communication terminals 320 and 340, and takes the input to decoders 622 and 624 as the latent space representation 630 with noise 640 added to the signal by the time it reaches wireless communication terminals 320 and 340, and uses bit sequence 202 as the output from decoder 622 and bit sequence 204 as the output from decoder 624 to generate autoencoder 600 by machine learning.

[0066] As a specific example, the learning device 100 uses machine learning to generate an autoencoder 600 in which the input to the encoder 610 is a bit sequence 202 reflecting priority 232 and a bit sequence 204 reflecting priority 234, the latent space representation 630 generated by the encoder 610 is a plurality of complex number signals generated from bit sequences 202 and 204 by OFDM for the wireless base station 200 to transmit bit sequences 202 and 204 to wireless communication terminals 320 and 340, and transmitted by a plurality of antennas provided by the wireless base station 200, the input to the decoders 622 and 624 is the latent space representation 630 with noise 640 added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the wireless communication terminals 320 and 340, the output from decoder 622 is bit sequence 202 and the output from decoder 624 is bit sequence 204.

[0067] Reflecting priority in a bit sequence can be done, for example, by multiplying the bit sequence by a value corresponding to its priority. For instance, if the priority is 2, each bit in the bit sequence is multiplied by 2, and if the priority is 3, each bit in the bit sequence is multiplied by 3. This allows for larger values ​​in the bit sequence for higher priority bits, enabling learning that prioritizes higher-priority bit sequences.

[0068] The learning device 100 acquires communication feedback. For example, the learning device 100 acquires feedback on communication quality, such as throughput, from multiple wireless communication terminals 300 and determines changes in communication quality by reflecting priority. The learning device 100 trains the autoencoder 600 by performing at least one of the following: rewarding the high-priority wireless communication terminal 300 when its communication quality is high and the low-priority wireless communication terminal 300 when its communication quality is low, and penalizing the high-priority wireless communication terminal 300 when its communication quality is low and the low-priority wireless communication terminal 300 when its communication quality is high. The learning device 100 also trains the autoencoder 600 by rewarding the higher-priority wireless communication terminal 300 when its communication quality is higher, and penalizing the lower-priority wireless communication terminal 300 when its communication quality is not higher. By continuously performing such training, it is possible to realize an autoencoder and decoder that can perform encoding and decoding such that the communication quality of the higher-priority wireless communication terminal 300 is higher than that of the lower-priority wireless communication terminal 300.

[0069] Figure 9 schematically shows an example of the functional configuration of the learning device 100. The learning device 100 comprises a storage unit 102, a learning data acquisition unit 104, a learning execution unit 106, and a providing unit 108.

[0070] The training data acquisition unit 104 acquires training data. The training data may include multiple datasets. The training data acquisition unit 104 stores the acquired training data in the storage unit 102.

[0071] The learning data acquisition unit 104 acquires multiple datasets for generating an autoencoder 400, which includes, for example, an encoder 410 provided to a wireless communication terminal 300 and a decoder 420 provided to a wireless base station 200. The dataset may include a bit sequence transmitted from the wireless communication terminal 300 to the wireless base station 200 via the wireless environment, and noise added to the signal transmitted by the wireless communication terminal 300 in radio waves to transmit the bit sequence to the wireless base station 200. The dataset may also include a bit sequence transmitted from the wireless communication terminal 300 to the wireless base station 200 via the wireless environment, and noise added to the multiple complex number signals generated from the bit sequence by OFDM and transmitted by multiple antennas of the wireless communication terminal 300 to transmit the bit sequence to the wireless base station 200. The learning data acquisition unit 104 may, for example, include in the dataset bit sequences actually transmitted in the past by any wireless communication terminal 300 to the wireless base station 200 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset bit sequences generated by simulation. The learning data acquisition unit 104 may, for example, include in the dataset noise measured from actual noise added to signals transmitted in the past by any wireless communication terminal 300 to the wireless base station 200 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset noise generated by simulation. The learning data acquisition unit 104 may, for example, use a channel model of the wireless environment between the wireless communication terminal 300 and the wireless base station 200 to generate and include in the dataset multiple types of noise added to signals transmitted by the wireless communication terminal 300 in radio waves to the wireless base station 200, and include them in the dataset.

[0072] The learning data acquisition unit 104 acquires multiple datasets for generating an autoencoder 500, which includes, for example, an encoder 510 provided to a wireless base station 200 and a decoder 520 provided to a wireless communication terminal 300. The dataset may include a bit sequence to be transmitted from the wireless base station 200 to the wireless communication terminal 300 via the wireless environment, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the bit sequence to the wireless communication terminal 300. The dataset may also include a bit sequence to be transmitted from the wireless base station 200 to the wireless communication terminal 300 via the wireless environment, and noise added to the multiple complex number signals generated from the bit sequence by OFDM and transmitted by multiple antennas of the wireless base station 200 to transmit the bit sequence to the wireless communication terminal 300. The learning data acquisition unit 104 may, for example, include in the dataset bit sequences that were actually transmitted in the past by any wireless base station 200 to the wireless communication terminal 300 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset bit sequences generated by simulation. The learning data acquisition unit 104 may, for example, include in the dataset noise measured from the noise that was actually added to the signal transmitted in the past by any wireless base station 200 to the wireless communication terminal 300 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset noise generated by simulation. The learning data acquisition unit 104 may, for example, use a channel model of the wireless environment between the wireless base station 200 and the wireless communication terminal 300 to generate multiple types of noise that are added to the signal transmitted by the wireless base station 200 in radio waves to transmit bit sequences to the wireless communication terminal 300 before it reaches the wireless communication terminal 300, and include these in the dataset.

[0073] The learning data acquisition unit 104 acquires multiple datasets for generating an autoencoder 600, which includes, for example, an encoder 610 provided to a wireless base station 200 and a plurality of decoders 620 provided to a plurality of wireless communication terminals 300. The dataset may include a plurality of bit sequences to be transmitted from the wireless base station 200 to the plurality of wireless communication terminals 300 via the wireless environment, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals 300 before it reaches the plurality of wireless communication terminals 300. The dataset may also include a plurality of bit sequences to be transmitted from the wireless base station 200 to the plurality of wireless communication terminals 300 via the wireless environment, and noise added to the plurality of complex number signals generated from the plurality of bit sequences by OFDM and transmitted by a plurality of antennas provided by the wireless base station 200 before it reaches the plurality of wireless communication terminals 300. The learning data acquisition unit 104 may, for example, include in the dataset multiple bit sequences that were actually transmitted in the past by any wireless base station 200 to multiple wireless communication terminals 300 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset multiple bit sequences generated by simulation. The learning data acquisition unit 104 may, for example, include in the dataset noise measured from signals that were actually added to signals transmitted in the past by any wireless base station 200 to multiple wireless communication terminals 300 via the wireless environment. The learning data acquisition unit 104 may also include in the dataset noise generated by simulation. The learning data acquisition unit 104 may, for example, use a channel model of the wireless environment between the wireless base station 200 and the multiple wireless communication terminals 300 to generate multiple types of noise that are added to signals transmitted by the wireless base station 200 in radio waves to transmit bit sequences to the multiple wireless communication terminals 300 before they reach the multiple wireless communication terminals 300, and include these in the dataset.

[0074] The dataset may include multiple bit sequences transmitted from the radio base station 200 to multiple wireless communication terminals 300 via the wireless environment, the terminal ID of each of the multiple wireless communication terminals 300, and noise added to the signal transmitted by the radio base station 200, which includes the multiple bit sequences in radio waves, before it reaches the multiple wireless communication terminals 300.

[0075] The dataset may include multiple bit sequences transmitted from the radio base station 200 to multiple wireless communication terminals 300 via the wireless environment, channel information for each of the multiple wireless communication terminals 300, and noise added to the signal transmitted by the radio base station 200, which includes the multiple bit sequences in radio waves, before it reaches the multiple wireless communication terminals 300.

[0076] The dataset may include multiple bit sequences to which the priority of each of the multiple wireless communication terminals is applied to each of the multiple bit sequences transmitted from the wireless base station 200 to the multiple wireless communication terminals 300 via the wireless environment, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the multiple bit sequences to the multiple wireless communication terminals 300 before it reaches the multiple wireless communication terminals 300.

[0077] The learning execution unit 106 generates an autoencoder including an encoder and a decoder by machine learning. The learning execution unit 106 may generate an autoencoder in which the input to the encoder is a bit sequence transmitted from one communication device to another communication device via a wireless environment, the latent space representation generated by the encoder is a signal to be included in radio waves propagating in the wireless environment for one communication device to transmit the bit sequence to another communication device, the input to the decoder is a latent space representation with noise added to the signal by the time it reaches the other communication device, and the output from the decoder is the bit sequence input to the encoder. The learning execution unit 106 may generate an autoencoder by machine learning, wherein the input to the encoder is a bit sequence transmitted from one communication device to another communication device via a wireless environment, the latent space representation generated by the encoder is a plurality of complex number signals generated from the bit sequence by OFDM for the first communication device to transmit the bit sequence to the other communication device and transmitted by a plurality of antennas provided by the first communication device, the input to the decoder is a latent space representation with noise added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the other communication device, and the output from the decoder is the bit sequence input to the encoder. The providing unit 108 provides the autoencoder generated by the learning execution unit 106. The providing unit 108 may provide the encoder to one communication device and the decoder to another communication device. One communication device may be a wireless communication terminal 300 and the other communication device may be a wireless base station 200. One communication device may be a wireless base station 200 and the other communication device may be a wireless communication terminal 300.

[0078] The learning execution unit 106 may generate an autoencoder 400, including an encoder 410 and a decoder 420, by machine learning using multiple datasets stored in the storage unit 102. The supply unit 108 may provide the encoder 410 to the wireless communication terminal 300 and provide the decoder 420 to the wireless communication terminal 300. The encoder 410 may be an example of a first encoder, and the decoder 420 may be an example of a first decoder.

[0079] The learning execution unit 106 may generate an autoencoder 500, including an encoder 510 and a decoder 520, by machine learning using multiple datasets stored in the storage unit 102. The supply unit 108 may provide the encoder 510 to the wireless base station 200 and the decoder 520 to the wireless communication terminal 300. The encoder 510 may be an example of a second encoder, and the decoder 520 may be an example of a second decoder.

[0080] The learning execution unit 106 may generate an autoencoder 600, which includes an encoder 610 and multiple decoders 620, by machine learning using multiple datasets stored in the storage unit 102. The supply unit 108 may provide the encoder 610 to the wireless base station 200 and the multiple decoders 620 to multiple wireless communication terminals 300.

[0081] The learning execution unit 106 may generate an autoencoder 600, which includes an encoder 610 that takes multiple bit sequences and multiple terminal IDs as inputs, and multiple decoders 620 that output multiple bit sequences and multiple terminal IDs, by machine learning using multiple datasets that include bit sequences transmitted from the wireless base station 200 to multiple wireless communication terminals 300 via the wireless environment, the terminal IDs of each of the multiple wireless communication terminals 300, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the multiple bit sequences to the multiple wireless communication terminals 300, until the signal reaches the multiple wireless communication terminals 300.

[0082] The learning execution unit 106 may generate an autoencoder 600, which includes an encoder 610 that takes multiple bit sequences and multiple channel information as inputs, and multiple decoders 620 that output multiple bit sequences, by machine learning using multiple datasets that include multiple bit sequences transmitted from the wireless base station 200 to multiple wireless communication terminals 300 via the wireless environment, channel information for each of the multiple wireless communication terminals 300, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the multiple bit sequences to the multiple wireless communication terminals 300, until the signal reaches the multiple wireless communication terminals 300.

[0083] The learning execution unit 106 may generate an autoencoder 600, including an encoder 610 and a decoder 620, by machine learning using multiple datasets that include multiple bit sequences to be transmitted from the wireless base station 200 to multiple wireless communication terminals 300 via the wireless environment, each of which has the priority of each of the multiple wireless communication terminals 300 applied, and noise added to the signal transmitted by the wireless base station 200 in radio waves to transmit the multiple bit sequences to the multiple wireless communication terminals 300, until it reaches the multiple wireless communication terminals 300.

[0084] Figure 10 schematically shows an example of the functional configuration of the wireless base station 200. Here, only the functional configuration for executing the processing according to this embodiment will be described. The wireless base station 200 includes a storage unit 242, a communication unit 244, a management unit 246, and an acquisition unit 248.

[0085] The communication unit 244 performs communication with the wireless communication terminal 300. The communication unit 244 communicates wirelessly with the wireless communication terminal 300 via the wireless environment. The communication unit 244 may communicate wirelessly with the wireless communication terminal 300 according to an existing communication protocol. The communication unit 244 may communicate wirelessly with the wireless communication terminal 300 according to a standard protocol such as 3GPP. The communication unit 244 may communicate with the core network of the mobile communication network included in network 50.

[0086] The management unit 246 manages communications by the communication unit 244. The management unit 246 manages data received from the wireless communication terminal 300. The management unit 246 manages data to be transmitted to the wireless communication terminal 300. The management unit 246 manages data received from the core network. The management unit 246 manages data to be transmitted to the core network. This data is stored in the storage unit 242.

[0087] The acquisition unit 248 acquires the decoder 420 generated by the learning device 100. The acquisition unit 248 may receive the decoder 420 from the learning device 100. The acquisition unit 248 stores the acquired decoder 420 in the storage unit 242. The management unit 246 may use the decoder 420 in communication with the wireless communication terminal 300.

[0088] The acquisition unit 248 acquires the encoder 510 generated by the learning device 100. The acquisition unit 248 may receive the encoder 510 from the learning device 100. The acquisition unit 248 stores the acquired encoder 510 in the storage unit 242. The management unit 246 may use the encoder 510 in communication with the wireless communication terminal 300.

[0089] The acquisition unit 248 acquires the encoder 610 generated by the learning device 100. The acquisition unit 248 may receive the encoder 610 from the learning device 100. The acquisition unit 248 stores the acquired encoder 610 in the storage unit 242. The management unit 246 may use the encoder 610 in communication with the wireless communication terminal 300.

[0090] Figure 11 schematically shows an example of the functional configuration of the wireless communication terminal 300. Here, only the functional configuration for executing the processing according to this embodiment will be described. The wireless communication terminal 300 includes a storage unit 342, a communication unit 344, a management unit 346, and an acquisition unit 348.

[0091] The communication unit 344 performs communication with the radio base station 200. The communication unit 344 communicates wirelessly with the radio base station 200 via the wireless environment. The communication unit 344 may communicate wirelessly with the radio base station 200 according to an existing communication protocol. The communication unit 344 may communicate wirelessly with the radio base station 200 according to a standard protocol such as 3GPP.

[0092] The management unit 346 manages communications by the communication unit 344. The management unit 346 manages data received from the radio base station 200. The management unit 346 manages data to be transmitted to the radio base station 200. This data is stored in the storage unit 342.

[0093] The acquisition unit 348 acquires the encoder 410 generated by the learning device 100. The acquisition unit 348 may receive the encoder 410 from the learning device 100. The acquisition unit 348 stores the acquired encoder 410 in the storage unit 342. The management unit 346 may use the encoder 410 in communication with the wireless base station 200.

[0094] The acquisition unit 348 acquires the decoder 520 generated by the learning device 100. The acquisition unit 348 may receive the decoder 520 from the learning device 100. The acquisition unit 348 stores the acquired decoder 520 in the storage unit 342. The management unit 346 may use the decoder 520 in communication with the wireless base station 200.

[0095] The acquisition unit 348 acquires the decoder 620 generated by the learning device 100. The acquisition unit 348 may receive the decoder 620 from the learning device 100. The acquisition unit 348 stores the acquired decoder 620 in the storage unit 342. The management unit 346 may use the decoder 620 in communication with the wireless base station 200.

[0096] For example, when a wireless communication terminal 300 and a wireless base station 200 communicate using an autoencoder 400, the management unit 346 inputs the bit sequence 302 to be transmitted to the wireless base station 200 into the encoder 410 and obtains the latent space representation 430 output from the encoder 410. The latent space representation 430 may be a signal included in radio waves propagating in the wireless environment. For example, the latent space representation 430 may be a plurality of complex number signals transmitted by a plurality of antennas provided by the wireless communication terminal 300. The management unit 346 causes the communication unit 344 to transmit a plurality of complex number signals using a plurality of antennas. The management unit 246 inputs the latent space representation 430, which has noise added in the wireless environment and has been received by the communication unit 244, into the decoder 420 and obtains the bit sequence 302 output from the decoder 420.

[0097] For example, when a wireless base station 200 and a wireless communication terminal 300 communicate using an autoencoder 500, the management unit 246 inputs the bit sequence 202 to be transmitted to the wireless communication terminal 300 into the encoder 510 and obtains the latent space representation 530 output from the encoder 510. The latent space representation 530 may be a signal included in radio waves propagating in the wireless environment. For example, the latent space representation 530 may be a plurality of complex number signals transmitted by a plurality of antennas provided by the wireless base station 200. The management unit 246 causes the communication unit 244 to transmit the plurality of complex number signals using the plurality of antennas. The management unit 346 inputs the latent space representation 530, which has noise added in the wireless environment and has been received by the communication unit 344, into the decoder 520 and obtains the bit sequence 202 output from the decoder 520.

[0098] For example, when a wireless base station 200 and multiple wireless communication terminals 300 communicate using an autoencoder 600, the management unit 246 inputs multiple bit sequences to be transmitted to the multiple wireless communication terminals 300 into the encoder 610 and obtains the latent space representation 630 output from the encoder 610. The latent space representation 630 may be a signal included in radio waves propagating in the wireless environment. For example, the latent space representation 630 may be multiple complex number signals transmitted by multiple antennas provided by the wireless base station 200. The management unit 246 causes the communication unit 244 to transmit multiple complex number signals using multiple antennas. The management unit 346 of each of the multiple wireless communication terminals 300 inputs the latent space representation 630, which has noise added in the wireless environment and has been received by the communication unit 344, into the decoder 620 and obtains the bit sequence output from the decoder 620.

[0099] For example, when a wireless base station 200 and multiple wireless communication terminals 300 communicate using an autoencoder 600, the management unit 246 obtains the terminal ID of each of the multiple wireless communication terminals 300, inputs the multiple bit sequences and multiple terminal IDs to be transmitted to the multiple wireless communication terminals 300 into the encoder 610, and obtains the latent space representation 630 output from the encoder 610. The latent space representation 630 may be a signal included in the radio waves propagating in the wireless environment. For example, the latent space representation 630 may be a plurality of complex number signals transmitted by the plurality of antennas provided by the wireless base station 200. The management unit 246 causes the communication unit 244 to transmit the plurality of complex number signals using the plurality of antennas. Each management unit 346 of the multiple wireless communication terminals 300 inputs the latent space representation 630, which has noise added in the wireless environment and has been received by the communication unit 344, to the decoder 620. The decoder 620 outputs multiple bit sequences and multiple terminal IDs, and the management unit 346 obtains the bit sequence corresponding to the terminal ID of the wireless communication terminal 300 (sometimes referred to as "the self") on which it is installed.

[0100] For example, when a wireless base station 200 and multiple wireless communication terminals 300 communicate using an autoencoder 600, the management unit 246 searches for and obtains the channel information of each of the multiple wireless communication terminals 300, inputs the multiple bit sequences and multiple channel information to be transmitted to the multiple wireless communication terminals 300 into the encoder 610, and obtains the latent space representation 630 output from the encoder 610. The latent space representation 630 may be a signal included in the radio waves propagating in the wireless environment. For example, the latent space representation 630 may be multiple complex number signals transmitted by multiple antennas provided by the wireless base station 200. The management unit 246 causes the communication unit 244 to transmit multiple complex number signals using multiple antennas. The management unit 346 of each of the multiple wireless communication terminals 300 inputs the latent space representation 630, which has noise added in the wireless environment and has been received by the communication unit 344, into the decoder 620, and obtains the bit sequence output from the decoder 620.

[0101] For example, when a wireless base station 200 and multiple wireless communication terminals 300 communicate using an autoencoder 600, the management unit 246 searches for and obtains the priority of each of the multiple wireless communication terminals 300, generates multiple bit sequences that reflect the respective priorities for each of the multiple bit sequences to be transmitted to the multiple wireless communication terminals 300, inputs the generated bit sequences to the encoder 610, and obtains the latent space representation 630 output from the encoder 610. The latent space representation 630 may be a signal included in radio waves propagating in the wireless environment. For example, the latent space representation 630 may be multiple complex number signals transmitted by multiple antennas provided by the wireless base station 200. The management unit 246 causes the communication unit 244 to transmit multiple complex number signals using multiple antennas. The management unit 346 of each of the multiple wireless communication terminals 300 inputs the latent space representation 630, which has noise added in the wireless environment and has been received by the communication unit 344, to the decoder 620, and obtains the bit sequence output from the decoder 620.

[0102] The learning execution unit 106 of the learning device 100 may acquire feedback from the communication between the wireless base station 200 using the autoencoder 600 and the multiple wireless communication terminals 300. The learning execution unit 106 may acquire this feedback from the core network of the mobile communication network.

[0103] The learning execution unit 106 obtains feedback on the communication quality, such as throughput, of multiple wireless communication terminals 300 when channel information is used, and determines whether the communication quality has deteriorated by inputting channel information. The learning execution unit 106 may train the autoencoder 600 by performing at least one of the following: giving a penalty when the communication quality deteriorates by inputting channel information, and giving a reward when the communication quality improves by inputting channel information.

[0104] The learning execution unit 106 may obtain feedback on the communication quality, such as throughput, of multiple wireless communication terminals 300 when priorities are used, and determine the change in communication quality due to the reflection of priorities. The learning execution unit 106 trains the autoencoder 600 by performing at least one of the following: rewarding the high-priority wireless communication terminal 300 when its communication quality is high and the low-priority wireless communication terminal 300 when its communication quality is low, and penalizing the high-priority wireless communication terminal 300 when its communication quality is high. Alternatively, the learning execution unit 106 trains the autoencoder 600 by rewarding the higher-priority wireless communication terminal 300 when its communication quality is higher, and penalizing the lower-priority wireless communication terminal 300 when its communication quality is lower.

[0105] Figure 12 schematically shows an example of the environment of the learning device 100. The learning device 100 may be located on a distributed infrastructure 800 in a system 20 which comprises a management infrastructure 700, a plurality of distributed infrastructures 800, and a plurality of wireless base stations 200. In system 20, the management infrastructure 700 and the plurality of distributed infrastructures 800 may cooperate to control the RAN 250 and perform AI processing. The RAN 250 provides mobile communication services to the wireless communication terminal 300.

[0106] RAN250 may be a virtualized vRAN (Virtual RAN). RAN250 may also be a physical RAN.

[0107] The AI ​​processing performed by the management infrastructure 700 and the multiple distributed infrastructures 800 may include RAN control AI processing. The AI ​​processing performed by the management infrastructure 700 and the multiple distributed infrastructures 800 may also include non-RAN control AI processing.

[0108] The distributed infrastructure 800 may be data centers located in various locations. The distributed infrastructure 800 may be composed of multiple devices. The distributed infrastructure 800 may be implemented on a virtualization infrastructure consisting of multiple devices. The distributed infrastructure 800 may be implemented by a single device. The distributed infrastructure 800 may function as a BBU (BaseBand Unit), and the wireless base station 200 may function as an RRU (Remote Radio Unit). The distributed infrastructure 800 may implement a CU. The distributed infrastructure 800 may implement a DU. The distributed infrastructure 800 may implement a UPF (User Plane Function).

[0109] The management infrastructure 700 may be a data center that manages multiple distributed infrastructures 800. The management infrastructure 700 may be composed of multiple devices. The management infrastructure 700 may be implemented on a virtualization infrastructure consisting of multiple devices. The management infrastructure 700 may be implemented by a single device. In other words, the management infrastructure 700 may be a management device.

[0110] The management infrastructure 700 may be called the Core Brain, and the distributed infrastructure 800 may be called the Regional Brain. Note that Figure 12 illustrates a case where a single-layer distributed infrastructure 800 is located below the management infrastructure 700, but this is not the only case. The distributed infrastructure 800 may have multiple layers. For example, if two layers of distributed infrastructure 800 are located below the management infrastructure 700, the management infrastructure 700 may be called the Core Brain, the lower-layer distributed infrastructure 800 may be called the Regional Brain, and the further lower-layer distributed infrastructure 800 may be called the Sub-Regional Brain.

[0111] The distributed infrastructure 800 may have one or more CPUs (Central Processing Units). The distributed infrastructure 800 may have one or more GPUs (Graphics Processing Units). The distributed infrastructure 800 may have multiple superchips, each connected to a CPU and a GPU by an interconnect. This interconnect may be memory consistent and capable of achieving high bandwidth and low latency. Thus, the distributed infrastructure 800 may have both CPU resources and GPU resources as computing resources.

[0112] The learning device 100 may be located on the distributed infrastructure 800 and perform some or all of the functions of the distributed infrastructure 800.

[0113] Figure 13 schematically shows an example of the functional configuration of the learning device 100. Here, we will mainly explain the differences from Figure 9. The learning device 100 comprises a storage unit 102, a learning data acquisition unit 104, a data provision unit 108, a RAN control unit 110, and an AI processing unit 120.

[0114] The RAN control unit controls the RAN 250. The AI ​​processing unit 120 performs AI processing. The AI ​​processing unit 120 includes a learning execution unit 106.

[0115] Figure 14 schematically shows an example of the hardware configuration of a computer 1200 that functions as a learning device 100, a wireless base station 200, or a wireless communication terminal 300. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the device according to this embodiment, or to cause the computer 1200 to execute operations associated with the device according to this embodiment or such one or more "parts", and / or to cause the computer 1200 to execute a process or a stage of such process according to this embodiment. Such a program may be executed by the CPU 1212 to cause the computer 1200 to execute specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0116] The computer 1200 according to this embodiment includes a CPU 1212, a GPU 1213, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive and a DVD-RAM drive, etc. The storage device 1224 may be a hard disk drive and a solid-state drive, etc. The computer 1200 also includes legacy input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0117] The CPU 1212 operates according to the programs stored in the ROM 1230 and RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires the image data generated by the CPU 1212 and stores it in the frame buffer provided in the RAM 1214 or within itself, so that the image data is displayed on the display device 1218.

[0118] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive 1226 reads programs or data from the DVD-ROM 1227, etc., and provides them to the storage device 1224. The IC card drive reads programs and data from the IC card and / or writes programs and data to the IC card.

[0119] The ROM 1230 stores boot programs and / or hardware-dependent programs of the computer 1200, which are executed by the computer 1200 when activated. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via USB ports, parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0120] The program is provided on a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The program is read from the computer-readable storage medium and installed on a storage device 1224, RAM 1214, or ROM 1230, which are examples of computer-readable storage media, and executed by the CPU 1212. The information processing described within these programs is read by the computer 1200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the operation or processing of information in accordance with the use of the computer 1200.

[0121] For example, when communication is performed between a computer 1200 and an external device, the CPU 1212 may execute a communication program loaded into the RAM 1214 and, based on the processing described in the communication program, instruct the communication interface 1222 to perform communication processing. Under the control of the CPU 1212, the communication interface 1222 reads transmission data stored in a transmission buffer area provided in a recording medium such as the RAM 1214, storage device 1224, DVD-ROM 1227, or IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0122] Furthermore, the CPU 1212 may read all or necessary parts of a file or database stored on an external recording medium such as a storage device 1224, a DVD drive 1226 (DVD-ROM 1227), or an IC card into the RAM 1214, and perform various types of processing on the data in the RAM 1214. The CPU 1212 may then write the processed data back to the external recording medium.

[0123] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 1212 may perform various types of processing on the data read from the RAM 1214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to the RAM 1214. The CPU 1212 may also retrieve information in files, databases, etc., within the recording medium. For example, if a plurality of entries are stored in the recording medium, each having an attribute value of a first attribute associated with an attribute value of a second attribute, the CPU 1212 may search among the plurality of entries for an entry that matches the specified condition for the attribute value of the first attribute, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0124] The program or software module described above may be stored on or near the computer 1200 in a computer-readable storage medium. Alternatively, a recording medium such as a hard disk or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to the computer 1200 via the network.

[0125] In this embodiment, blocks in the flowchart and block diagram may represent a stage in a process in which an operation is performed or a "part" of a device that has the role of performing an operation. A particular stage and "part" may be implemented by a dedicated circuit, a programmable circuit supplied with computer-readable instructions stored on a computer-readable storage medium, and / or a processor supplied with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuit may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuit may include reconfigurable hardware circuits, such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), which include logical AND, logical OR, exclusive OR, negated AND, negated OR, and other logical operations, flip-flops, registers, and memory elements.

[0126] A computer-readable storage medium may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, a computer-readable storage medium having instructions stored therein will comprise a product that includes instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable storage media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media may include floppy disks (registered trademark), diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital versatile disk (DVD), Blu-ray (registered trademark) disk, memory stick, integrated circuit card, etc.

[0127] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and conventional procedural programming languages ​​such as the C programming language or similar programming languages.

[0128] Computer-readable instructions may be provided locally or via a wide area network (WAN) such as a local area network (LAN) or the internet to a processor or programmable circuit of a general-purpose computer, a special-purpose computer, or another programmable data processing device, so that the processor or programmable circuit of the programmable data processing device, such as a computer, may execute the computer-readable instructions to generate means for performing operations specified in a flowchart or block diagram. Here, the computer may be a PC (personal computer), a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, or a special-purpose computer, and may also be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system and is a computer in a broad sense. In a distributed computing system, multiple computers execute a program collectively by each computer executing a part of the program and passing data during program execution between computers as needed.

[0129] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, and microcontrollers. A computer may have one or more processors. In a multiprocessor system with multiple processors, each processor executes a portion of the program, and the processors collectively execute the program by passing program execution data between them as needed. For example, in the execution of multitasks, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which part of a program each processor executes changes dynamically. Which part of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0130] By using the invention according to this embodiment, it is possible to contribute to the optimization of RAN and to the achievement of Sustainable Development Goal (SDG) 9, "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation."

[0131] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0132] It should be noted that the execution order of operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not explicitly stated as "before" or "prior to," and that these can be performed in any order unless the output of a previous operation is used in a later operation. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," and "next," for convenience, this does not mean that it is mandatory to perform the operations in that order.

[0133] 10 System, 20 System, 50 Network, 80 Logical Configuration, 100 Learning Device, 102 Storage Unit, 104 Learning Data Acquisition Unit, 106 Learning Execution Unit, 108 Provision Unit, 110 RAN Control Unit, 120 AI Processing Unit, 200 Wireless Base Station, 202 Bit String, 204 Bit String, 212 Terminal ID, 214 Terminal ID, 222 Channel Information, 224 Channel Information, 232 Priority, 234 Priority, 242 Storage Unit, 244 Communication Unit, 246 Management Unit, 248 Acquisition Unit, 250 RAN, 300 Wireless Communication Terminal, 302 Bit String, 320 Wireless Communication Terminal, 340 Wireless Communication Terminal, 342 Storage Unit, 344 Communication Unit, 346 Management Unit, 348 Acquisition Unit, 400 Autoencoder, 410 Encoder, 420 Decoder, 430 Latent space representation, 440 Noise, 500 Autoencoder, 510 Encoder, 520 Decoder, 530 Latent space representation, 540 Noise, 600 Autoencoder, 610 Encoder, 620 Decoder, 622 Decoder, 624 Decoder, 630 Latent space representation, 640 Noise, 700 Management board, 800 Distributed board, 1200 Computer, 1210 Host controller, 1212 CPU, 1213 GPU, 1214 RAM, 1216 Graphics controller, 1218 Display device, 1220 Input / Output controller, 1222 Communication interface, 1224 Storage device, 1226 DVD drive, 1227 DVD-ROM, 1230 ROM, 1240 Input / Output chip

Claims

1. A learning device comprising: a learning execution unit that generates an autoencoder by machine learning, wherein the input to the encoder is a bit sequence transmitted from one communication device to another communication device via a wireless environment, the latent space representation generated by the encoder is a signal included in radio waves propagated by the one communication device in the wireless environment for transmitting the bit sequence to the other communication device, the input to the decoder is the latent space representation with noise added to the signal by the time the signal reaches the other communication device, and the output from the decoder is the bit sequence input to the encoder; and a providing unit that provides the encoder to the one communication device and provides the decoder to the other communication device.

2. The learning device according to claim 1, wherein the learning execution unit generates an autoencoder by machine learning, wherein the input to the encoder is a bit sequence transmitted from one communication device to the other communication device via a wireless environment, the latent space representation generated by the encoder is a plurality of complex number signals generated from the bit sequence by OFDM for the one communication device to transmit the bit sequence to the other communication device and transmitted by a plurality of antennas provided by the one communication device, the input to the decoder is the latent space representation with noise added to the plurality of complex number signals transmitted by the plurality of antennas before reaching the other communication device, and the output from the decoder is the bit sequence input to the encoder.

3. A learning device according to claim 1 or 2, comprising: a learning data acquisition unit that acquires a dataset including a bit sequence transmitted from a wireless communication terminal to a wireless base station via a wireless environment, and noise added to the signal transmitted by the wireless communication terminal in radio waves to transmit the bit sequence to the wireless base station, the learning execution unit that generates a first autoencoder including a first encoder and a first decoder by machine learning using a plurality of the datasets, and the providing unit that provides the first encoder to the wireless communication terminal and the first decoder to the wireless base station.

4. The learning device according to claim 3, wherein the learning data acquisition unit generates a plurality of types of noise that are added to a signal transmitted by the wireless communication terminal in radio waves to transmit the bit sequence to the wireless base station, using a channel model of the radio environment between the wireless communication terminal and the wireless base station.

5. The learning device according to claim 3, wherein the learning execution unit generates a second autoencoder including a second encoder and a second decoder by machine learning using a plurality of datasets including a bit sequence transmitted from a wireless base station to a wireless communication terminal via a wireless environment and noise added to the signal transmitted by the wireless base station in radio waves to transmit the bit sequence to the wireless communication terminal, and the providing unit provides the second encoder to the wireless base station and the second decoder to the wireless communication terminal.

6. The learning device according to claim 5, wherein the learning execution unit generates a second autoencoder including a second encoder that takes a plurality of bit sequences and a plurality of terminal IDs as inputs, and a second decoder that outputs a plurality of bit sequences and a plurality of terminal IDs as outputs, by machine learning using a plurality of datasets including a plurality of bit sequences transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, the terminal ID of each of the plurality of wireless communication terminals, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals.

7. The learning device according to claim 5, wherein the learning execution unit generates a second autoencoder including a second encoder that takes a plurality of bit sequences and a plurality of channel information as inputs, and a second decoder that outputs the plurality of bit sequences as outputs, by machine learning using a plurality of datasets including a plurality of bit sequences transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, channel information of each of the plurality of wireless communication terminals, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals.

8. The learning device according to claim 5, wherein the learning execution unit generates the second autoencoder, including the second encoder and the second decoder, by machine learning using a plurality of datasets, each of which includes a plurality of bit sequences to be transmitted from a wireless base station to a plurality of wireless communication terminals via a wireless environment, each of which has the priority of each of the plurality of wireless communication terminals applied, and noise added to the signal transmitted by the wireless base station in radio waves to transmit the plurality of bit sequences to the plurality of wireless communication terminals, until the signal reaches the plurality of wireless communication terminals.

9. A learning device according to any one of claims 1 to 8, comprising a RAN control unit for controlling RAN and an AI processing unit for performing AI processing, wherein the AI ​​processing unit has the learning execution unit.

10. A program for causing a computer to function as a learning device according to any one of claims 1 to 8.

11. A system comprising the learning device according to any one of claims 3 to 8 and the wireless base station.

12. A learning method performed by a computer, comprising: a learning execution step of generating an autoencoder by machine learning, the autoencoder including an encoder and a decoder, wherein the input to the encoder is a bit sequence transmitted from one communication device to another communication device via a wireless environment, the latent space representation generated by the encoder is a signal to be included in radio waves propagating in the wireless environment for the one communication device to transmit the bit sequence to the other communication device, the input to the decoder is the latent space representation with noise added to the signal by the time the signal reaches the other communication device, and the output from the decoder is the bit sequence input to the encoder; and a providing step of providing the encoder to the one communication device and providing the decoder to the other communication device.

Images