Apparatus and method for carrying out digital pre-distortion based on conditional generative adversarial network in wireless communication system

Abstract

The present disclosure relates to a communication apparatus and method for efficiently carrying out digital pre-distortion (DPD) on a transmission signal of a wireless communication system according to various operating conditions. The communication apparatus that carries out DPD on a transmission signal of a wireless communication system, according to an embodiment of the present disclosure, may comprise a power amplifier that amplifies the transmission signal. The communication apparatus may comprise a DPD apparatus that is connected to an input terminal of the power amplifier and carries out, on the basis of operating condition data of the communication apparatus, a DPD operation by using fake data generated for the transmission signal. The DPD apparatus may include a generator based on a conditional GAN. The generator may be configured to receive, as inputs, the operating condition data and input / output signals collected from the power amplifier and generate, on the basis of data acquired through repeated training of the conditional GAN using the collected input / output signals and the operating condition data, the fake data learned to compensate for nonlinear distortion at the output of the power amplifier.

Description

Device and method for performing conditional adversarial generative neural network-based digital transposition distortion in a wireless communication system

[0001] The present disclosure relates to an apparatus and method for performing digital pre-distortion in a wireless communication system.

[0002] To meet the increasing demand for wireless data traffic following the 4G system (i.e., LTE (long-term evolution) system), 5G systems supporting high data rates have been developed and commercialized. 5G systems can be implemented in the millimeter wave (mmWave) band. To support such high data rates, wireless communication systems need to support wide bandwidth and high center frequencies, and among the various components for transmitting and receiving wireless signals, the power and dynamic range of radio frequency (RF) components, such as radio frequency front end (RFFE) circuits, need to be increased. Additionally, high output and wide linearity may be required for power amplifiers (PAs) (e.g., high power amplifiers (HPA)) included in the RFFE to amplify the transmitted signal. In ranges where the input signal magnitude is relatively small, the PA can maintain linearity between the input and output signals; however, in ranges where the input signal magnitude is relatively large, it becomes difficult to maintain linearity, resulting in non-linear distortion. Therefore, digital pre-distortion (DPD) technology is utilized to compensate for non-linear distortion in Power Amplifiers (PAs) used in the RF (Radio Frequency) stage of communication devices, such as base stations, repeaters, and UEs (user equipment), that transmit wireless signals in wireless communication systems.

[0003] Whenever the above-mentioned PA operates at high efficiency, it becomes abruptly non-linear and causes distortion in the output signal. In communication devices, the PA is one of the components that require significant power consumption. Furthermore, to achieve high output power and efficiency conditions, PAs such as those in base stations operate near the saturation region; however, operation near the saturation region can cause severe non-linearity in the PA, which can negatively affect communication quality. In communication devices such as base stations, repeaters, and UEs (hereinafter referred to as base stations, etc.), DPD technology can be applied to improve the efficiency and linearity of the power amplifier simultaneously in order to mitigate the non-linearity occurring in the PA.

[0004] With reference to FIGS. 1a to 1d, the general concept of DPD technology will be explained.

[0005] FIG. 1a illustrates the ideal linear characteristics (11) between the input and output of a PA. In FIG. 1a, the input of the PA may be nearly identical to the output of the PA or may differ depending on the scaling factor (K) of the PA. FIG. 1b illustrates an example of non-linear distortion (12) in the operation of an actual PA. In FIG. 1b, the output of the PA can be represented by a function (F) related to its input. Non-linear distortion (12) as in FIG. 1b degrades in-band signal quality and causes strong out-of-band distortion. To compensate for this non-linear distortion, DPD technology pre-distorts the signal to be input to the PA so that the output of the PA becomes linear. FIG. 1c illustrates an example of applying a distortion (13) to the input signal that can compensate for the non-linear distortion (12) of FIG. 1b.

[0006] FIG. 1d briefly illustrates an example in which a DPD block is connected to the input terminal of a PA. In FIG. 1d, the DPD block (14) applies pre-distortion (16) to the front of the PA (15) to cancel out non-linear distortion (17) on the input signal. Then, the non-linear distortion (17) generated in the PA (15) is compensated (removed) by the pre-distortion (16), so that a signal having linear characteristics (18) can be output from the output terminal of the PA (15).

[0007] The general concept of DPD technology described above can easily perform DPD operations to compensate for signal distortion for signals whose signal levels change slowly, as shown in the example of FIG. 1d. Hereinafter, a model / device that performs DPD operations in a communication device will be referred to as a DPD model / device. The above DPD model / device can be implemented in a digital signal processing (DSP) module within the communication device.

[0008] In recent wireless communication environments, such as 5G systems utilizing ultra-high frequency bands like millimeter wave (mmWave) bands, the power level and bandwidth of transmitted signals can change rapidly to match real-time traffic in order to improve energy and / or frequency efficiency. In such communication environments, problems may arise when applying DPD technology because the operating conditions of power amplifiers, such as base stations, change. In such rapidly changing wireless communication environments, it is necessary to update DPD models / devices with coefficient(s) suitable for the characteristics of the signal and power amplifier through fast DPD adaptation.

[0009] In addition, recent wireless communication environments require wide bandwidth for high data rates, and signals with a high peak-to-average power ratio (PAPR) are transmitted to achieve high frequency efficiency. In such wireless communication environments, power amplifiers, such as base stations transmitting wireless signals, may exhibit more complex nonlinearity and memory effects. The memory effect refers to a phenomenon where previous signals influence the current signal and can degrade the performance of DPD models / devices. These complex nonlinearity and memory effects in power amplifiers, such as base stations, increase the complexity of DPD models / devices (e.g., polynomial-based DPD models / devices), increase the time required for computations in base stations to track and compensate for the nonlinearity of the power amplifier, and can cause an increase in hardware resources required in base stations due to the increased complexity of DPD models / devices.

[0010] The present disclosure provides a communication device and method for efficiently performing digital pre-distortion on a transmission signal of a wireless communication system according to various operating conditions.

[0011] The present disclosure provides a communication device and method including a conditional GAN ​​(generative adversarial network)-based DPD device in a wireless communication system.

[0012] The present disclosure provides a DPD device and method that apply a conditional GAN ​​to a transmission signal of a wireless communication system.

[0013] According to an embodiment of the present disclosure, a communication device that performs digital pre-distortion (DPD) on a transmission signal of a wireless communication system may include a power amplifier that amplifies the transmission signal. The communication device may include a DPD device connected to the input terminal of the power amplifier that performs DPD operation using fake data generated for the transmission signal based on operating condition data of the communication device. The DPD device may include a generator based on a conditional generative adversarial network (GAN). The generator may be configured to receive the operating condition data and an input / output signal collected from the power amplifier, and to generate the fake data trained to compensate for non-linear distortion at the output of the power amplifier based on data obtained through repeated training based on the conditional GAN ​​using the collected input / output signal and the operating condition data.

[0014] In one embodiment, the operation mode of the DPD device may include a learning mode for repeated training based on the conditional GAN ​​and an operation mode in which DPD operation is performed on an actual input transmission signal from the communication device.

[0015] In one embodiment, the device may further include a data preprocessing block that performs data preprocessing on the input signal of the DPD device and the operating condition data. When the DPD device operates in the operating mode, the input signal may include an actual input transmission signal. When the DPD device operates in the learning mode, the input signal may include an output signal collected from the power amplifier. The data preprocessing may include at least one of normalization and one-hot encoding.

[0016] In one embodiment, the data preprocessing block may be configured to merge the input signal and the operating condition data into a single input data set.

[0017] In one embodiment, the generator of the DPD device may be configured to merge the input signal and the operating condition data into a single input data set.

[0018] In one embodiment, the operating condition data may include at least one of signal bandwidth, PAPR, output signal power, ambient temperature of the communication device, and beamforming-related information.

[0019] In one embodiment, the DPD device may further include a discriminator operating in the learning mode. The discriminator may be configured to alternately receive the fake data output from the generator or the input signal collected from the power amplifier as input data, and to determine whether the input data is fake or real data through repeated training according to the operating conditions based on the operating condition data.

[0020] In one embodiment, the operating power applied to the discriminator in the operating mode may be cut off or the operation of the discriminator may be switched to an off state.

[0021] In one embodiment, the repeated training may be performed until one of the predetermined criteria is satisfied. The predetermined criteria may include a first criterion for performing the repeated training a predetermined number of times, a second criterion for performing the repeated training until the difference between the fake data and the real data reaches a predetermined range, and a third criterion for performing the repeated training until a predetermined communication quality based on the operating condition data is satisfied.

[0022] In one embodiment, the communication device may include a DAC connected between the output terminal of the DPD device and the input terminal of the power amplifier to convert the output signal of the DPD device into an analog signal. The communication device may include an ADC that converts the output signal collected from the power amplifier into a digital signal. The communication device may further include a processor that controls the operation of the DPD device based on at least one of the predetermined criteria.

[0023] According to an embodiment of the present disclosure, a method for performing digital pre-distortion (DPD) on a transmission signal in a communication device of a wireless communication system may include a step of receiving operating condition data of the communication device and an input / output signal collected from a power amplifier for DPD operation in a DPD device including a conditional GAN-based generator, and verifying acquired data through repeated training based on the conditional GAN ​​using the collected input / output signal and the conditional operating data. The method may include a step of performing DPD operation using generated fake data for the transmission signal based on the verified data and the operating condition data, so that non-linear distortion at the output of the power amplifier is compensated.

[0024] According to an embodiment of the present disclosure, a communication device for performing digital pre-distortion (DPD) on a transmission signal of a wireless communication system comprises a power amplifier for amplifying the transmission signal, one or more processors including a processing circuitry of a DPD device including a conditional GAN-based generator, and a memory for storing instructions. When the instructions are executed individually or collectively by the one or more processors, the generator of the DPD device receives operating condition data of the communication device and an input / output signal collected from the power amplifier, and can cause the generator to generate the fake data trained to compensate for non-linear distortion at the output of the power amplifier based on data obtained through repeated training of the conditional GAN ​​using the collected input / output signal and the operating condition data.

[0025] According to an embodiment of the present disclosure, in a storage medium storing at least one computer-readable command as an embodiment, the at least one command causes the DPD device to perform at least one operation when executed individually or collectively by one or more processors in a communication device including a DPD device that performs digital pre-distortion (DPD) at the input terminal of a power amplifier that amplifies a transmission signal, and the at least one operation may include an operation of receiving operating condition data of the communication device and an input / output signal collected from the power amplifier, and an operation of generating fake data trained to compensate for non-linear distortion at the output of the power amplifier based on data obtained through repeated training based on a conditional GAN ​​using the collected input / output signal and the operating condition data.

[0026] FIGS. 1A, 1B, 1C, and 1D are drawings for illustrating the general concept of DPD technology in a wireless communication system.

[0027] FIG. 2 is a diagram briefly illustrating an example configuration of a communication device including an artificial neural network-based DPD device in a wireless communication system.

[0028] FIG. 3a is a diagram briefly showing the structure of a GAN model, which is a generative AI model.

[0029] FIG. 3b is a diagram briefly showing the structure of a conditional GAN ​​model applied to an embodiment of the present disclosure,

[0030] FIG. 4a is a diagram showing an example of a configuration of a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure.

[0031] FIGS. 4b, FIGS. 4c, FIGS. 4d, FIGS. 4e, FIGS. 4f, FIGS. 4g, FIGS. 4h, FIGS. 4i, FIGS. 4j, FIGS. 4k, FIGS. 4l, FIGS. 4m and FIGS. 4n are drawings for illustrating examples of operations in the learning mode and operation mode of a conditional GAN-based DPD device according to an embodiment of the present disclosure.

[0032] FIG. 5 is a diagram showing an example of a configuration of a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure.

[0033] FIG. 6 is a diagram illustrating an example of a method performed in a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure, and

[0034] FIG. 7 is a drawing showing an example of a configuration of a communication device in a wireless communication system according to an embodiment of the present disclosure.

[0035] The operating principles of the present disclosure will be described in detail below with reference to the attached drawings. In describing the present disclosure below, specific descriptions of related known functions or configurations will be omitted if it is determined that such detailed descriptions would unnecessarily obscure the essence of the present disclosure. Furthermore, the terms described below are defined in consideration of their functions in the present disclosure, and these may vary depending on the intentions or practices of the user or operator. Therefore, their definitions should be based on the content throughout this specification.

[0036] The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure is complete and to fully inform those skilled in the art of the scope of the invention, and the present disclosure is defined only by the scope of the claims. Throughout the specification, like reference numerals refer to like components.

[0037] At this point, it will be understood that each block of the process flow diagrams and combinations of the flow diagrams can be executed by computer program instructions.

[0038] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specific logical function(s). It should also be noted that in some alternative execution examples, the functions mentioned in the blocks may occur out of order. For instance, two blocks described in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order according to their corresponding functions.

[0039] In this embodiment, the term "part" refers to a software or hardware component such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit), and the "part" performs certain roles. However, the meaning of "part" is not limited to software or hardware. The "part" may be configured to reside in an addressable storage medium or configured to run one or more processors. Thus, as an example, the "part" includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within the components and "parts" may be combined into a smaller number of components and "parts" or further separated into additional components and "parts." In addition, the components and 'parts' may be implemented to utilize one or more CPUs within the device or secure multimedia card. Also, in the embodiments, 'parts' may include one or more processors.

[0040] In the present disclosure, each of the phrases such as “A / B”, “A or B”, “A and / or B”, “at least one of A and B”, “at least one of A or B”, “A, B or C”, “at least one of A, B and C”, and “at least one of A, B, or C” may include any one of the items listed together in the corresponding phrase, or all possible combinations thereof. Terms such as “first”, “second”, or “first” or “second” may be used simply to distinguish a component from another component and do not limit the components in any other aspect (e.g., importance or order).

[0041] Terms used in the following description to identify connection nodes, terms referring to network entities, terms referring to messages, terms referring to interfaces between network entities, terms referring to various identification information, etc., are examples provided for the convenience of explanation. Accordingly, the present disclosure is not limited to the terms described below, and other terms referring to objects having equivalent technical meanings may be used.

[0042] In the present disclosure, a base station (BS) is a network entity capable of performing resource allocation for terminals and communicating with terminals through a wireless network, and may be at least one of an eNode B, Node B, gNB, RAN (Radio Access Network), AN (Access Network), RAN node, IAB (Integrated Access / Backhaul) node, a wireless access unit, a base station controller, a node on a network, or a TRP (transmission reception point). A terminal (user equipment: UE) may be at least one of a terminal, MS (Mobile Station), cellular phone, smartphone, computer, or a multimedia system capable of performing communication functions.

[0043] The DPD model commonly used in communication devices such as base stations is the generalized memory polynomial (GMP) model, and artificial-intelligence neural network (ANN) models are being introduced instead of the traditional GMP model. However, DPD methods based on GMP models can have very high implementation complexity because the amount of computation increases exponentially depending on the precision of the GMP model, and DPD methods based on ANN models require a long time for training the ANN, and it can be difficult to determine the optimal timing for the ANN's hyperparameters. In addition, various DPD models are currently being studied, including Feed Forward Neural Networks (FFNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and combinations of these artificial neural networks.

[0044] FIG. 2 is a diagram briefly illustrating an example configuration of a communication device including an artificial neural network-based DPD device in a wireless communication system.

[0045] The communication device of FIG. 2 includes a DPD device (220) using an artificial neural network connected to a power amplifier (PA) (210) to compensate for non-linear distortion in the output of the PA (210).

[0046] Referring to FIG. 2, the DPD device (220) can generate / output a pre-distorted signal by inputting not only an input signal x(t) (21) containing an information signal but also operating condition parameters (22) obtained through sensors (230). The operating condition parameters (22) can be divided into three types, for example, first to third operating condition parameters (22a, 22b, 22c). The first operating condition parameters (22a) are signal-dependent parameters and may include, for example, signal average power, signal bandwidth, and / or PAPR. The second operating condition parameters (22b) are environment-dependent parameters and may include, for example, ambient temperature. The third operating condition parameters (22c) are transmitter setting-dependent parameters and may include, for example, beam azimuth angle, beam elevation angle, etc. in a communication device using an array antenna such as a base station using beamforming. In the example of FIG. 2, the DPD device (220) includes the operating condition parameters (22) together in the input to perform training for DPD operation, and then generates a DPD model coefficient set so that the DPD actuator (221) can select DPD model coefficient(s) that match the operating conditions and use them for DPD operation. The DPD actuator (221) performs the operation of pre-distorting the input signal using the DPD model coefficient(s) selected according to the operating conditions, converting the pre-distorted digital signal into an analog signal, and transmitting it to the PA (210). The DPD device (220) of FIG. 2 may be configured, for example, using a convolution filter function.However, in the example of FIG. 2, as operating condition parameters (22) are added to the DPD device (220), the complexity of the artificial neural network in the DPD device (220) increases, and the hardware complexity of the communication device increases, which leads to an increase in the operating cost and / or power consumption of the communication device, such as a base station.

[0047] In this disclosure, the terms DPD model and DPD device may be used interchangeably. In one embodiment, the DPD device may be implemented as a field programmable gate array (FPGA) or at least one processor performing digital signal processing (DSP) that operates according to the algorithm of the DPD model. In this disclosure, the DPD model may be implemented as a conditional generative adversarial network (conditional GAN) based DPD model. This disclosure proposes a method of applying a conditional GAN ​​to a DPD model for digital pre-distortion in communication devices such as base stations, repeaters, UEs, etc. (hereinafter, base stations, etc.) that utilize power amplifiers (PAs). The above conditional GAN ​​can be understood as a so-called generative AI model.

[0048] FIG. 3a is a diagram briefly illustrating the structure of a generative AI model, specifically a GAN model. As is well known, the GAN model is a generative AI model proposed by Ian Goodfellow, and the GAN model operates in such a way that two neural networks, including a generator (310a) and a discriminator (320a), compete with each other to learn / train. The generator (310a) and the discriminator (320a) can be referred to as the generator and the discriminator, respectively.

[0049] The generator (310a) receives a latent space vector (Z) (301) as input and real data (X RealBy learning the characteristics of )(303), real data (X Real Fake data similar to )(303)(X Fake It can generate / output )(302). The latent space is a space that represents the core features of the data by compressing high-dimensional vectors into low-dimensional vectors, and can be used for training the GAN model. The discriminator (320a) is real data (X Real )(303) is received as input, and that input data is real data(X Real ) or fake data(X Fake It can determine whether it is ). In addition, the discriminator (320a) can determine whether it is fake data (X) from the generator (310a). Fake )(302) is received as input, and that input data is real data(X Real ) or fake data(X Fake It can determine whether ). The generator (310a) can determine whether real data (X) through repeated training. Real The goal is to generate data having a feature distribution substantially identical to that of ). The discriminator (320a) also, through repeated training, [identifies] the type of input data (i.e., real data (X). Real ) or fake data(X Fake It can more accurately distinguish )). Through this iterative training, fake data (X) with feature distributions similar to real data (XReal) is identified. Fake It can generate / output )(302). The generator (310a) and the discriminator (320a) can perform learning / training simultaneously or alternately, and due to the competitive learning / training structure between the two neural networks of the generator (310a) and the discriminator (320a), the GAN can have a superior data generation ability compared to existing artificial neural network structures.

[0050] FIG. 3b is a diagram briefly illustrating the structure of a conditional GAN ​​model applied to an embodiment of the present disclosure.

[0051] The GAN model exemplified in Fig. 3a is an unsupervised learning data generation model and therefore does not require operating conditions; however, in the conditional GAN ​​model of Fig. 3b, learning / training can be performed by adding operating condition data representing specific conditions to the input data set of the conditional GAN ​​model to generate data that meets specific conditions, for example, when a communication device operates. The unsupervised learning method is a machine learning algorithm that can find patterns and correlations in the input data set without providing the output data, which is the result of the learning algorithm, in advance. The unsupervised learning method is a learning method that aligns the probability distribution (i.e., feature distribution) of the characteristics possessed by real data and fake data.

[0052] Referring to FIG. 3b, the conditional GAN ​​model may include two neural networks, a generator (310b) and a discriminator (320b). The generator (310b) and the discriminator (320b) may be referred to as the generator and the discriminator, respectively.

[0053] The conditional GAN ​​model of FIG. 3b is distinguished from the GAN model of FIG. 3a in that operational condition data (305, 306) is added to the input data of the generator (310b) and the discriminator (320b). The operational condition data (305, 306) may include at least one of parameters representing various operational situations (e.g., signal bandwidth, PAPR, output signal power, ambient temperature of the communication device, beamforming-related information (e.g., beamforming angle), etc.) in the case of a communication device such as a base station. In one embodiment, the same operational condition data (305, 306) may be input to the generator (310b) and the discriminator (320b). In another embodiment, distinct operational condition data (305, 306) may be input to the generator (310b) and the discriminator (320b).

[0054] Referring to FIG. 3b, the generator (310b) is real data (X Real )(303) and operating condition data (305) are received as input, and a characteristic distribution according to the operating condition is learned with the help of a discriminator (320b), and through the learning, real data (X) according to the operating condition Real Fake data (X) with a feature distribution similar to )(303) Fake )(302) can be generated / output. In one embodiment, when training for DPD operation is performed at a communication device such as a base station, the data (Z)(301) may use, for example, the input / output signal of the PA collected from the PA for training. In one embodiment, when actual DPD operation is performed at a communication device such as a base station after training is completed, the data (Z)(301) may use, for example, the input transmission signal (input TX signal) to be actually transmitted.

[0055] Also, referring to FIG. 3b, the discriminator (320b) is real data (X Real )(303) and operating condition data (306) are received as input, and the real data (X Real )(303) real data (X) according to the operating conditions Real ) or fake data(X Fake It can determine whether it is ). In addition, the discriminator (320b) can determine whether it is fake data (X) from the generator (310b). Fake Receive )(302) as input, and that fake data(X Fake )(302) is real data (X) according to the operating conditions Real ) or fake data(X Fake It can determine whether ). The generator (310b) obtains real data (X) according to the operating conditions through repeated training. Real The goal is to generate data having a feature distribution substantially identical to that of ). The discriminator (320b) also, through repeated training, [identifies] the type of input data (i.e., real data (X).Real ) or fake data(X Fake It is possible to more accurately identify )). And by conducting this iterative training, fake data (X) according to the relevant operating conditions Fake ) is real data(X Real It can be trained to resemble the feature distribution that the generator (310b) possesses. The generator (310b) and the discriminator (320b) can perform learning / training simultaneously or alternately, and due to the competitive learning / training structure between the two neural networks under various operating conditions, it can have a superior data generation ability compared to GAN.

[0056] The embodiments of the present disclosure provide specific methods for applying a conditional GAN ​​as a generative AI model to a DPD model. The DPD model to which the conditional GAN ​​is applied may be included in various communication devices, such as base stations and UEs, that amplify and output transmission signals through a PA. For convenience, the embodiments of the present disclosure exemplify cases where the DPD model to which the conditional GAN ​​is applied is applied to the transmitting device of a base station; however, the communication devices to which the embodiments of the present disclosure can be applied are not limited to base stations. For example, the DPD model to which the conditional GAN ​​of the present disclosure is applied may also be applied to repeaters or UEs that utilize a PA. In one embodiment, when the above conditional GAN ​​is applied to a DPD model, if parameters representing various operating conditions (e.g., bandwidth, output power, ambient temperature, beamforming-related information (e.g., beam width, beam angle, etc.)) are provided as operating condition parameters of the conditional GAN ​​to train the DPD model in a communication device such as a base station, more efficient DPD operation can be performed. In one embodiment, the above operating condition parameters may include at least one of the operating condition parameters (22) in the example of FIG. 2. The above operating condition parameters may be referred to as operating condition information. By using a DPD model to which the above conditional GAN ​​is applied, a DPD model capable of responding to various operating conditions can be provided through a single DPD model. In one embodiment, the above conditional GAN ​​applied DPD model may be implemented in an FPGA or at least one processor performing a DSP. In one embodiment, the above conditional GAN ​​applied DPD model may be implemented within the RFIC (radio frequency integrated circuits) of a communication device. Hereinafter, the above conditional GAN ​​applied DPD model / device is briefly referred to as a DPD model, DPD It will be referred to as a device or DPD block.

[0057] In one embodiment, the input terminal of the PA in a communication device to which the DPD device is applied may be connected to the output terminal of a DAC (digital-to-analog converter). The output terminal of the DPD device may be connected to the input terminal of the DAC. The DPD device may include a generator and a discriminator as described in the example of FIG. 3b. In one embodiment, data preprocessing may be performed on the input transmission signal (or output signal collected from the PA) input to the input terminal of the DPD device and the operating condition parameters. The data preprocessing may include, for example, at least one of normalization and one-hot encoding. Additionally, the data preprocessing may include absolute value processing for the IQ signal (i.e., I (in-phase) signal and Q (quadrature) signal) in the communication, and time domain processing for the IQ signal. The reason for performing absolute value processing and time domain processing is to generate various basis vector components for removing signal distortion components, and known methods may be used for absolute value processing and time domain processing. The above data preprocessing may be performed in at least one processor that performs DSP. In one embodiment, the data preprocessing may be performed in at least one processor connected to a DPD device or in a block for data preprocessing within the DPD device.

[0058] In one embodiment, the generator in the DPD device receives the operating condition parameters and the output signal collected from the PA and pre-processed data, and the operating condition parameters, and can generate / output fake data trained through repeated training so that non-linear distortion in the output signal of the PA is compensated under the corresponding operating condition. After the repeated training is completed in the DPD device, the generator receives the operating condition parameters and the actual input transmission signal (input TX signal) to be transmitted, and can output the input transmission signal by pre-distorting it so that non-linear distortion in the output signal of the PA is compensated under the corresponding operating condition.

[0059] In one embodiment, the discriminator in the DPD device receives the fake data output from the generator and / or the input signal (i.e., real data) collected from the PA and preprocessed along with the data preprocessed operating condition parameters, and can determine whether the input data is fake data or real data through repeated training according to the operating condition. By conducting such repeated training, the fake data can be learned to resemble the characteristic distribution of the real data according to the operating condition.

[0060] In one embodiment, when actual DPD operation is performed after repeated training is completed in the discriminator and generator of the DPD device, the discriminator may be in an off state where it does not operate. In one embodiment, if the operation mode in which repeated training for accurate DPD operation is performed for the discriminator and generator of the DPD device is called the learning mode, and the operation mode in which DPD operation is performed on an actual input transmission signal is called the operation mode, the learning mode may be performed during the production process of the communication device, and the operation mode may be performed during actual operation after the product is shipped from the communication device.

[0061] Accordingly, the discriminator may be optionally included in the DPD device included in the communication device. For example, the discriminator may be included in the DPD device only for repeated learning / training during the production process of the communication device, and after the repeated learning / training is completed, the discriminator may be separated / removed from the DPD device when the product of the communication device is shipped. After the repeated learning / training is completed, the collected data may be stored as information, such as a look-up table, in the communication device and used in the DPD device.

[0062] In one embodiment, the DPD device includes both a discriminator and a generator, and it may be possible to switch the operation of the discriminator to an off state in the operating mode. For example, if the discriminator is configured with hardware such as an FPGA, the operation of the discriminator may be switched to an off state by cutting off the power supplied to the discriminator.

[0063] FIG. 4a is a diagram showing an example of a configuration of a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure. FIG. 4a shows an example of a configuration required in a learning mode of a DPD device to which a conditional GAN ​​is applied.

[0064] As previously described, the communication device of FIG. 4a may be a base station, repeater, or UE utilizing a PA. Additionally, the DPD device (40) applying the conditional GAN ​​according to the present disclosure to the communication device of FIG. 4a may be implemented by including at least one processor performing an FPGA or DSP. In one embodiment, the communication device may include a transceiver that transmits and receives wireless signals, as shown in the example of FIG. 7 described later, and a controller that controls the overall operation of the communication device. In one embodiment, some components (DAC, ADC (analog-to-digital converter), PA, etc.) connected to the DPD device (40) may be included in the transceiver, and at least one processor performing an FPGA or DSP may be included in the controller. In one embodiment, a data preprocessing block (or data preprocessor) that performs data preprocessing required for the DPD operation of the present disclosure may be included in the controller.

[0065] The communication device of FIG. 4a may include a data preprocessing block (430), a DPD device (40), a DAC (440) that converts a digital signal into an analog signal, a PA (450), and an ADC (460) that converts an analog signal into a digital signal. The DPD device (40) is a generative AI model in which the conditional GAN ​​described in FIG. 3b is applied to the DPD model. The DPD device (40) can train the DPD model by receiving input from operation condition data (402) representing various operation conditions (e.g., signal bandwidth, output signal power, ambient temperature of the communication device, beamforming-related information (e.g., beamforming angle, etc.).

[0066] Additionally, the DPD device (40) may include a generator (410) and a discriminator (420) that constitute a conditional GAN. The DPD device (40) may operate in an operation mode divided into a learning mode in which repeated training for accurate DPD operation is performed and an operation mode in which DPD operation is performed on an actual input transmission signal (401).

[0067] In the above learning mode, both the generator (410) and the discriminator (420) operate for repeated training, and in the above operation mode, the generator (410) operates, but the discriminator (420) does not operate or may be separated / removed / blocked from the DPD device (40). In the above learning mode, the generator (410) of the DPD device (40) receives input signals and operation condition data (403) that are collected from operation condition data (402) and PA (450) and preprocessed in the data preprocessing block (430), and through repeated training, receives fake data (X) trained to compensate for non-linear distortion in the output signal of the PA (450) under the corresponding operation condition. F You can generate / output )(404).

[0068] In one embodiment, the data (407) input to the generator (410) during the training process of the generator (410) may be data obtained by preprocessing the output signal (PA output signal) of the PA (450). The output (404) of the generator (410) generates a fake input data set (41) together with the preprocessed operating condition data (403). The real input data set (42) consists of the preprocessed PA input signal (406) and the preprocessed operating condition data (403). During the training process of the generator (410), the discriminator (420) may set labels as the correct answers for the real input data set (42) and the fake input data set (41). For example, the label of the real input data set (42) may be set to "0" and the label of the fake input data set (41) may be set to "1". Alternatively, the label values ​​may be set in reverse. The generator (410) can use the cross-entropy value as a loss value by using the output result of the discriminator (420) and the labels of the real input data set (42) and the fake input data set (41). Additionally, the generator (410) can use a value such as the Mean Square Error between the output result of the generator (410) and the PA input signal as a loss value. The generator (410) can update the coefficients of the generator (410) using these loss values.

[0069] After repeated training is completed in the DPD device (40), the generator (410) in the above operating mode receives operating condition data (402) and an input transmission signal (401) to be actually transmitted, and can output the input transmission signal (401) with pre-distortion so that non-linear distortion in the output signal of the PA (450) is compensated under the corresponding operating condition.

[0070] In the above learning mode, the discriminator (420) of the DPD device (40) uses fake data (X) from the generator (410) together with the data preprocessed operating condition data (403). FInput signal (i.e., real data (X)) collected from the input / output signals of )(404) and / or PA(450) and preprocessed in the data preprocessing block (430) R )) receives input alternately / sequentially / in parallel, and depending on the corresponding operating condition, the input data is fake data (X F ) whether or real data(X R It is possible to determine whether it is ) through repeated training. By conducting such repeated training, fake data (X) according to the relevant operating conditions F )(404) is real data(X R It can be learned to have a feature distribution similar to )(406).

[0071] In one embodiment, during the training process of the discriminator (420), a fake input data set (41) or a real input data set (42) may be input as input data to the discriminator (410). The fake input data set (41) is fake data (X F It may include operating condition data (403) preprocessed through )(404) and the data preprocessing block (430). The real input data set (42) is real data (X RIt may include )(406) and operation condition data (403). In one embodiment, the fake input data set (41) and the real input data set (42) may be input alternately / sequentially / in parallel. The discriminator (420) may receive the fake input data set (41) and / or the real input data set (42) and perform learning. During the training process of the discriminator (420), the DPD device (40) (or generator (410)) may set labels as the correct answers for the real input data set (42) and the fake input data set (41). For example, the label of the real input data set may be set to "1" and the label of the fake input data set may be set to "0". Alternatively, the label values ​​may be set in reverse. In one embodiment, the label setting in the discriminator (420) may be performed in reverse of the label setting in the generator (410). This is for a more accurate discrimination operation in the discriminator (420).

[0072] In one embodiment, fake data (X) input to the discriminator (420) in learning mode F A label (e.g., "0") may be added to ) as an answer key indicating that the data is fake data. In addition, real data (X) input to the discriminator (420) R A label (e.g., "1") indicating that the data is real data may be added to the discriminator (420). The label may be added for a more accurate discrimination operation in the discriminator (420).

[0073] The data preprocessing performed in the above data preprocessing block (430) may include, for example, normalization and one-hot encoding. The normalization is intended to scale the data by converting it into a certain range, and when the data is normalized into a certain range, the learning / training efficiency of the DPD device (40) can be improved. The one-hot encoding converts categorical data into numerical data such as binary data, thereby improving the learning / training efficiency of the DPD device (40). Additionally, the data preprocessing may further include absolute value processing for the IQ signal and time domain processing for the IQ signal when the format of the IQ signal (i.e., I signal and Q signal) of the input transmission signal (401) at the current time is I(n) and Q(n). For example, if the above IQ signals are I(n) and Q(n), then |I(n)+jQ(n)|, |I(n)+jQ(n)| 2 , |I(n)+jQ(n)| 3Absolute value processing can be performed by adding absolute value items such as [etc.]. Here, |I(n)+jQ(n)| is an absolute value operation representing the magnitude of I(n)+jQ(n). Additionally, the above data preprocessing can be performed by adding items for preprocessing in the time domain to the IQ signal, such as I(n-1), Q(n-1), I(n-2), Q(n-2)… or I(n+1), Q(n+1), I(n+2), Q(n+2)…. Here, n represents the current time point, n-1, n-2, … etc. represent past time points, and n+1, n+2, … etc. represent future predicted time points; I(n-1), Q(n-1), I(n-2), Q(n-2) etc. represent lagging items indicating the IQ signal at past time points, and I(n+1), Q(n+1), I(n+2), Q(n+2) etc. represent leading items indicating the IQ signal at predicted time points.

[0074] In one embodiment, the repeated training performed in the generator (410) and the discriminator (420) may be performed alternately or in parallel. In one embodiment, the repeated training may be performed, for example, a predetermined number of times or real data (X R Fake data (X) with a feature distribution similar to )(404) F )(406) can be performed in various ways, such as until it is generated / output, or until it satisfies the specified communication quality according to the operating conditions in the communication device.

[0075] FIGS. 4b to 4n are drawings for illustrating examples of operations in the learning mode and operation mode of a conditional GAN-based DPD device according to an embodiment of the present disclosure.

[0076] Specifically, FIGS. 4b to 4m are drawings for illustrating examples of training operations of a discriminator and a generator in a learning mode of a conditional GAN-based DPD device according to an embodiment of the present disclosure.

[0077] It should be noted that the examples of FIGS. 4b through 4m are intended to aid in understanding the present disclosure and that the present disclosure is not limited to the following examples.

[0078] First, the examples in FIGS. 4b to 4m assume the situations described in 1) to 3) below, for instance. It assumes that the operating conditions of the DPD device include only two frequency conditions (3.3 GHz and 3.6 GHz) and two temperature conditions (30 degrees and 60 degrees), and that one-hot encoding is performed as data preprocessing.

[0079] 1) Assume the case where the frequency is 3.3 GHz, the temperature is 30 degrees, the input signals of the PA are 1.0+1.0 j, 2.0+2.0 j, 3.0+3.0 j, and the output signals of the PA are 10.0+10.0 j, 20.0+20.0 j, 30.0+30.0 j.

[0080] 2) Assume that in addition to I(n) and Q(n), |I(n) + jQ(n)|, |I(n) + jQ(n)|3, I(n-1), and Q(n-1) are added to the input of the generator (410).

[0081] 3) It is assumed that in the data preprocessing block (430), the data signal is normalized and the operating condition is one-hot encoded.

[0082] According to the assumptions of 1) to 3) above, the data preprocessing block (430) can perform normalization and I / Q signal separation, absolute value processing and time domain processing as shown in the example of [Table 1].

[0083] [Table 1]

[0084]

[0085] In the above two frequency conditions of 3.3 GHz and 3.6 GHz, and temperature conditions of 30 degrees and 60 degrees, the communication device can collect input / output signals (input signal, output signal) from the PA (450), and the above frequency conditions and temperature conditions can be collected / confirmed as operating condition data.

[0086] [Table 2]

[0087]

[0088] FIG. 4c illustrates an operation in which a data preprocessing block (430) receives a PA output signal and operating condition data as inputs under the above assumption, and outputs the preprocessed data and the preprocessed operating condition data (i.e., condition parameters) to a generator (410). The condition parameters can be distinguished, for example, according to a combination of frequency and temperature, and "0101" can represent the operating conditions in which the DPD device operates, such that the frequency is 3.3 GHz and the temperature is 30 degrees.

[0089] FIG. 4b illustrates an example of a training method in a discriminator (420) when a DPD device operates in a learning mode according to an embodiment of the present disclosure, and FIG. 4d to 4i illustrate examples of operations of a generator (410) and a discriminator (420) in the training method of FIG. 4b. The exemplary operations of FIG. 4d to 4i will be described below with reference to FIG. 4b.

[0090] In the D1 and D2 processes of FIG. 4b, the DPD device, with the coefficients of the generator (410) fixed, the generator (410) fake data (X F ) can be output. At this time, the generator (410) can output fake data (X) in an I / Q signal format (I(n), Q(n)) as in the example of FIG. 4d. FIt can output ). FIG. 4e receives preprocessed data of the PA output signal through absolute value processing and time domain processing, and receives, for example, "0101" as the preprocessed operating condition data (i.e., condition parameters) to produce fake data (X F This shows an example of a configuration of a generator (410) that outputs ) in an I / Q signal format (I(n), Q(n)).

[0091] In process D3 of Fig. 4b, the DPD device fake data (X F A fake input data set including a condition parameter (e.g., "0101") can be configured. In the D4 process, the DPD device can set / assign label 0 to the fake input data set. In the D5 process, the fake input data set is input to a discriminator (420), and the discriminator (420) can output a probabilistic prediction result regarding whether the input data is real data or fake data. FIGS. 4f and 4g illustrate an example configuration of a discriminator (420) that receives a fake input data set in an I / Q signal format (I(n), Q(n)) and condition parameters for frequency and temperature, such as "0101", and outputs the probabilistic prediction result.

[0092] In process D6 of FIG. 4b, the PA input signal is normalized through preprocessing in the data preprocessing block (430), and in process D7, the discriminator (420) receives a real input data set as in the example of FIG. 4h. The real input data set is the normalized PA input signal (X R It may include ) and condition parameters (e.g., "0101"). For example, label 1 may be set in the real input data set. In the D9 process, the discriminator (420) may output a probabilistic prediction result regarding whether the input data is real data or fake data.

[0093] In the D10 process of FIG. 4b, the DPD device (or discriminator (420)) can check the loss value (Loss_D) and update the discriminator (420) by comparing the discriminator output and label for each of the fake input data sets and real input data sets that are input alternately / sequentially / in parallel as in the example of FIG. 4i. The loss value (Loss_D) can be obtained using the cross-entropy value for the label and discriminator output as in the example of [Table 3] below.

[0094] [Table 3]

[0095]

[0096] FIG. 4j illustrates an example of a training method in a generator (420) when a DPD device operates in a learning mode according to an embodiment of the present disclosure, and FIG. 4k to 4m illustrate examples of operations of the generator (410) and the discriminator (420) in the training method of FIG. 4j. The exemplary operations of FIG. 4k to 4m will be described below with reference to FIG. 4j.

[0097] In the G1 process of FIG. 4j, with the coefficients of the discriminator (420) fixed, the generator (410) produces fake data (X) as in the example of FIG. 4k. F Can output ). Fake data(X F It is also possible to reuse the output result during the training of the above discriminator (420).

[0098] In the G2 process, labels can be set / assigned to the fake input data set, just as they were during the training of the discriminator (420). At this time, the label values ​​can be set / assigned in the opposite way to those during the training of the discriminator (420). In the G3 process, the output result of the discriminator (420) can be checked. For example, assume that the label for the fake input data set is "1" and the fake output is 0.4.

[0099] Assuming that in the G4 process, the PA input signal is 1.0+1.0j, 2.0+2.0j, or 3.0+3.0j as in the example of FIG. 4L, the output result of the discriminator (420) can be verified using the preprocessed PA input signal (I(n), Q(n) in the I / Q signal format). Here, the real input data set input to the discriminator (420) includes the preprocessed PA input signal and the preprocessed operating condition data (i.e., condition parameters).

[0100] In the G5 process, the loss value can be determined by comparing the output result (discriminator output) of the discriminator (420) and the label for each of the fake input data set and the real input data set. (It is assumed that the loss value is indicated as Loss_G1.) Here, as shown in the example in [Table 4] below, the label of the real input data set can be set to "0" and the label of the fake input data set can be set to "1". The above loss value (Loss_G1) can be obtained using the cross-entropy value for the label and the discriminator output.

[0101] [Table 4]

[0102]

[0103] In the G6 process, the DPD device (or generator (420)) outputs fake data (X) from the generator (420) as in the example of FIG. 4m. F The loss value (Loss_G2) can be determined by comparing the output (X) of the generator (410) with the normalized PA input signal. For example, the output (X) of the generator (410) F A value such as the mean squared error of the normalized PA input signal and the loss value (Loss_G2) may also be used. In the G7 process, the DPD device (or generator (420)) can calculate the total loss value by weighting and summing Loss_G1, which is the loss value identified by the discriminator (420), and Loss_G2, which is the loss value identified by the generator (410), and update the coefficients of the generator (410).

[0104] In the manner described in FIGS. 4b through 4m above, the generator (410) and the discriminator (420) can perform training alternately / sequentially / in parallel. The training can be performed a predetermined number of times or by performing performance checks until the target performance is reached in the DPD device.

[0105] FIG. 4n is a diagram illustrating an example of operation in the operating mode of a conditional GAN-based DPD device according to an embodiment of the present disclosure. The operation of FIG. 4n illustrates an operation performed in the operating mode after the training of the DPD device is performed by the learning mode operation of FIG. 4b to FIG. 4m. It should be noted that the example of FIG. 4n is intended to aid in understanding the present disclosure and is not intended to limit the present disclosure.

[0106] Referring to FIG. 4n, a PA input signal and operating condition data for actual transmission are input to a data preprocessing block (430), and the data preprocessing block (430) can output a preprocessed PA input signal and preprocessed operating condition data through absolute value processing and time domain processing. A generator (410) in which coefficients are updated through the above-mentioned training receives the preprocessed PA input signal and preprocessed operating condition data, and can output the preprocessed PA input signal in an I / Q signal format (I(n), Q(n)) by pre-distorting the preprocessed PA input signal so that non-linear distortion in the output of the PA (450) is compensated under the corresponding operating conditions.

[0107] FIG. 5 is a diagram showing an example of a configuration of a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure. FIG. 5 shows an example of a configuration required in an operating mode of a DPD device to which a conditional GAN ​​is applied, wherein the configuration of FIG. 5 shows an example in which a discriminator is separated / excluded from the DPD device (50).

[0108] Referring to FIG. 5, the data preprocessing block (530), generator (510), DAC (540), PA (550), and ADC (560) may correspond to the data preprocessing block (430), generator (410), DAC (440), PA (450), and ADC (460) in FIG. 4a.

[0109] In the DPD device (50) of FIG. 5, after repeated training such as the example of FIG. 4a is completed, when actual DPD operation is performed, the discriminator can be separated / removed when the product is shipped. In one embodiment, if the discriminator is included in the DPD device (50) even in the operating mode, the operation of the discriminator in the operating mode can be switched to an off state. In this case, the power supplied to the discriminator can be cut off. For example, if the discriminator is composed of hardware such as an FPGA, the operation of the discriminator can be switched to an off state by cutting off the power supplied to the discriminator.

[0110] In the above operating mode, the data preprocessing block (530) receives the input transmission signal (501) and the operating condition data (502), and can perform data preprocessing such as normalization and one-hot encoding on the operating condition data (502) and the input transmission signal (501). The generator (510) of the DPD device (50) receives the preprocessed operating condition data and the data input transmission signal, and can output the input transmission signal after pre-distorting it so that non-linear distortion in the output signal of the PA (550) is compensated under the corresponding operating condition. The pre-distorted input transmission signal is converted into an analog signal through the DAC (540), and the PA (550) amplifies and outputs the analog signal. The amplified signal can be transmitted as a wireless signal through an antenna not shown.

[0111] According to the configuration in which the discriminator of the DPD device (50) is separated / removed in operation mode as in the example of Fig. 5, the hardware complexity of the communication device can be reduced and production costs can be reduced.

[0112] FIG. 6 is a diagram illustrating an example of a method performed in a communication device including a conditional GAN-based DPD device in a wireless communication system according to an embodiment of the present disclosure. The method of FIG. 6 may be performed in the communication device of FIG. 4a or FIG. 5.

[0113] Referring to FIG. 6, in the process of 601, an unillustrated controller or processor of the communication device may collect input / output signals of a power amplifier (PA) having the aforementioned operating condition(s) for training and prepare for learning / training of the DPD device in operating mode.

[0114] In the embodiment of FIG. 6, it is assumed that the output signal of the PA is collected and that the output signal of the PA is converted into a digital signal through an ADC.

[0115] In the 602 process, the data preprocessing block (or data preprocessor) performs data preprocessing (e.g., normalization, one-hot encoding) on ​​the input / output signals of the PA and the operating condition data, and can transmit the preprocessed operating condition data and the preprocessed output signals of the PA to the generator of the DPD device. For example, normalization may be performed on the output signals of the PA, and normalization or one-hot encoding may be performed on the operating condition data.

[0116] In the 603 process, the data preprocessing block or DPD device (or generator) can merge the real input data, which has been preprocessed from the output signal of the PA, and the operating condition data into a single input data set (i.e., real input data set).

[0117] When conditional GAN-based training based on the learning mode of the DPD device starts in process 604, in process 605, the generator provides fake data (X) trained to compensate for non-linear distortion in the output signal of the PA under the corresponding operating conditions based on the operating condition data. F It can generate / output )(i.e., fake output data).

[0118] In the 606 process, the DPD device (or discriminator) can merge the fake output data and the operating condition data into a single input data set (i.e., a fake input data set).

[0119] In the process of 607, the DPD device (or discriminator) can set labels as the correct answers for the real input data set and the fake input data set. For example, the label of the real input data set can be set to "1" and the label of the fake input data set can be set to "0". Alternatively, the label values ​​can be set in reverse.

[0120] In the 608 process, the DPD device (or discriminator) can train the discriminator according to the method of the conditional GAN ​​exemplified in Fig. 3b.

[0121] In the 609 process, the DPD device (or discriminator / generator) can set the labels of the fake input dataset to the labels of the real input dataset. The 609 process is intended to proceed with training in a direction where the generator can deceive the discriminator according to the method of conditional GAN.

[0122] In the 610 process, the DPD device (or generator) can train the generator according to the method of the conditional GAN ​​exemplified in Fig. 3b.

[0123] In process 611, it is checked whether the conditional GAN-based training according to the learning mode of the DPD device has ended or been completed. The repeated training in processes 604 through 611 can be performed in various ways, such as being performed a predetermined number of times, until fake data with a feature distribution similar to real data is generated or output, or until a predetermined communication quality according to the operating conditions of the communication device is satisfied. Whether the repeated training in processes 604 through 611 has started or ended can be determined by the DPD device or by an unillustrated controller or processor of the communication device. In one embodiment, the controller or processor can control the operation of the DPD device.

[0124] When the learning mode of the DPD device is completed / terminated during the 612 process, the discriminator in the DPD device may be hardware-separated / removed for operation in the operating mode, or the power supplied to the discriminator may be cut off, or the operation of the discriminator may be switched to an off state. As an example of one embodiment, after providing the device of FIG. 4a for the learning mode and the device of FIG. 5 for the operating mode, it is also possible to provide / store / copy the data learned through the device of FIG. 4a (e.g., count information for a generator) to the device of FIG. 5.

[0125] In the process of 613, the data preprocessing block in the operating mode of the DPD device receives the input transmission signal and operating condition data, and can perform data preprocessing such as normalization and one-hot encoding on the operating condition data and the input transmission signal. The generator of the DPD device receives the preprocessed operating condition data and the input transmission signal, and can output the input transmission signal with pre-distortion so that non-linear distortion in the output signal of the PA is compensated (i.e., for linearization) under the corresponding operating condition.

[0126] In the embodiment of FIG. 6, repeated training of the generator and discriminator of the DPD device can be performed alternately or in parallel. In the example of FIG. 6, the training of the generator is exemplified as being performed after the training of the discriminator, but conversely, the training of the discriminator may be performed after the training of the generator.

[0127] According to the embodiments of the present disclosure described above, a DPD model / device can be implemented using a single conditional GAN-based AI model under various operating conditions (bandwidth, output power, temperature, etc.). Furthermore, since the present disclosure has a structure in which a discriminator is added to the DPD device compared to existing DPD devices utilizing artificial neural networks, learning proceeds at a higher complexity, allowing for the expectation of improved linearization effects. Additionally, since the discriminator can be separated in the operating mode after the learning mode is completed in the DPD device, the hardware burden on the communication device can be improved.

[0128] FIG. 7 is a drawing showing an example of a configuration of a communication device in a wireless communication system according to an embodiment of the present disclosure. The communication device of FIG. 7 may be one of a base station, a repeater, or a UE described in the embodiments of FIG. 3b to FIG. 6 described above.

[0129] The communication device of FIG. 7 may include a controller (730), a transceiver (710), and a memory (720). The controller (730), transceiver (710), and memory (720) of the communication device may operate according to the DPD model / operation described in the embodiments of FIG. 3b through FIG. 6. However, the components of the communication device are not limited to the examples described above. For example, the communication device may include more components or fewer components than the components described above. In addition, the controller (730), transceiver (710), and memory (720) may be implemented in the form of a single chip. The transceiver (710) is a collective term for the receiver and the transmitter of the communication device and can transmit and receive signals with a counterpart communication device. At this time, the signals transmitted and received may include at least one of control information, such as setting information, and data. The transceiver (710) receives a signal and outputs it to the controller (730), and can transmit the signal output from the controller (730). Additionally, the transceiver (710) of FIG. 7 may include an RF transmitter that up-converts and amplifies the frequency of the transmitted signal, and an RF receiver that low-noise amplifies the received signal and down-converts the frequency. Additionally, the transceiver (710) receives a signal and outputs it to the controller (730), and can transmit the signal output from the controller (730) to a counterpart communication device via a network. The memory (720) can store programs and data required for the DPD model / operation of the communication device according to at least one of the embodiments of FIG. 3b to FIG. 6. Additionally, the memory (720) can store control information or data included in the signal obtained from the communication device. The memory (720) may be composed of a storage medium or a combination of storage media such as ROM, RAM, hard disk, CD-ROM, and DVD.Additionally, the controller (730) can control a series of processes so that the communication device can operate according to at least one or a combination of the embodiments of FIGS. 3b through 6. The controller (730) may include at least one processor.

[0130] In one embodiment, a communication device for performing digital pre-distortion (DPD) on a transmission signal of a wireless communication system comprises a power amplifier (450) for amplifying the transmission signal, one or more processors (730) including a processing circuitry of a DPD device (40) including a conditional generative adversarial network (GAN)-based generator (310b, 410), and a memory (720) for storing instructions, wherein when the instructions are executed individually or collectively by the one or more processors (730), the generator (310b, 410) in the DPD device (40) receives operating condition data (305, 403) of the communication device and an input / output signal (301, 405, 406) collected from the power amplifier (450), and the collected input / output signal (301, 405, 406) and the operating condition data (305, Using 403), based on the data obtained through repeated training based on the conditional GAN, it is possible to cause the generation of the fake data (302, 404) trained to compensate for non-linear distortion in the output of the power amplifier (450).

[0131] In one embodiment, the operation mode of the DPD device (40, 50) may include a learning mode for repeated training based on the conditional GAN ​​and an operation mode in which DPD operation is performed on an actual input transmission signal from the communication device.

[0132] In one embodiment, when the commands are executed individually or collectively by the one or more processors (730), the communication device further causes the DPD device (40, 50) to perform data preprocessing on the input signal and the operating condition data, and when the DPD device (40, 50) is operating in the operating mode, the input signal includes an actual input transmission signal, and when the DPD device (40, 50) is operating in the learning mode, the input signal includes an output signal collected from the power amplifier, and the data preprocessing may include at least one of normalization and one-hot encoding.

[0133] In one embodiment, the DPD device (40) further includes the conditional GAN-based discriminator (320b, 420) operating in the learning mode, and when the commands are executed individually or collectively by the one or more processors (730), the discriminator (320b, 420) in the DPD device (40) receives the fake data (302, 404) output from the generator (310b, 410) or the input signal collected from the power amplifier (450) alternately as input data, and can cause the discriminator to determine whether the input data is fake or real data through repeated training according to the operating conditions based on the operating condition data.

[0134] In one embodiment, a storage medium storing at least one instruction readable by a computer, wherein the at least one instruction, when executed individually or collectively by one or more processors (730) in a communication device including a DPD device that performs digital pre-distortion (DPD) at the input terminal of a power amplifier that amplifies a transmission signal, causes the DPD device to perform at least one operation, and the at least one operation may include an operation of receiving operating condition data of the communication device and an input / output signal collected from the power amplifier, and an operation of generating fake data trained to compensate for non-linear distortion at the output of the power amplifier based on data obtained through repeated training based on a conditional GAN ​​using the collected input / output signal and the operating condition data.

[0135] Methods according to the embodiments described in the claims or specification of the present disclosure may be implemented in the form of hardware, software, or a combination of hardware and software.

[0136] When implemented in software, a computer-readable storage medium may be provided for storing one or more programs (software modules). One or more programs stored in the computer-readable storage medium are configured for execution by one or more processors within a communication / electronic device. One or more programs include instructions that cause the electronic device to execute methods according to the embodiments described in the claims or specification of the present disclosure.

[0137] Such programs (software modules, software) may be stored in random access memory, non-volatile memory including flash memory, ROM (Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), magnetic disc storage devices, CD-ROM (Compact Disc-ROM), Digital Versatile Discs (DVDs), or other forms of optical storage devices, magnetic cassettes. Alternatively, they may be stored in memory composed of some or all of these. Additionally, each constituent memory may include multiple units.

[0138] Additionally, the above program may be stored on an attachable storage device that can be accessed via a communication network such as the Internet, Intranet, Local Area Network (LAN), Wide LAN (WLAN), or Storage Area Network (SAN), or a combination thereof. Such a storage device may be connected to a device performing an embodiment of the present disclosure through an external port. Additionally, a separate storage device on a communication network may be connected to a device performing an embodiment of the present disclosure.

[0139] In the specific embodiments of the present disclosure described above, the components included in the invention are expressed in a singular or plural form according to the specific embodiments presented. However, the singular or plural expression is selected to suit the situation presented for convenience of explanation, and the present disclosure is not limited to singular or plural components; even if a component is expressed in the plural form, it may be composed of a singular form, or even if a component is expressed in the singular form, it may be composed of a plural form.

[0140] Meanwhile, although specific embodiments have been described in the detailed description of the present disclosure, it is understood that various modifications are possible within the scope of the present disclosure. Therefore, the scope of the present disclosure should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof.

Claims

1. A communication device that performs digital pre-distortion (DPD) on a transmission signal of a wireless communication system, A power amplifier (450) that amplifies the above transmission signal; and It includes a DPD device (40) connected to the input terminal of the power amplifier and performing a DPD operation using fake data (302, 404) generated for the transmission signal based on the operation condition data (305, 403) of the communication device, and The above DPD device (40) includes a conditional generative adversarial network (GAN) based generator (310b, 410), and A communication device configured such that the generator (310b, 410) receives the operating condition data (305, 403) and input / output signals (301, 405, 406) collected from the power amplifier (450), and generates the fake data (302, 404) trained to compensate for non-linear distortion in the output of the power amplifier (450) based on data obtained through repeated training based on the conditional GAN ​​using the collected input / output signals (301, 405, 406) and the operating condition data (305, 403).

2. In Paragraph 1, The operation mode of the above DPD device (40, 50) includes a learning mode for repeated training based on the conditional GAN ​​and an operation mode in which DPD operation is performed on an actual input transmission signal in the communication device.

3. In Paragraph 1 or 2, It further includes a data preprocessing block (430) that performs data preprocessing on the input signal of the DPD device (40, 50) and the operating condition data, and When the above DPD device (40, 50) operates in the above operating mode, the input signal includes an actual input transmission signal, and When the above DPD device (40, 50) operates in the learning mode, the input signal includes an output signal collected from the power amplifier (450, 550), and A communication device in which the above data preprocessing includes at least one of normalization and one-hot encoding.

4. In any one of paragraphs 1 to 3, The above data preprocessing block (430) is a communication device configured to merge the input signal and the operating condition data into a single input data set.

5. In any one of paragraphs 1 to 3, The generator (310b, 410, 510) of the above DPD device (40, 50) is a communication device configured to merge the input signal and the operating condition data into a single input data set.

6. In any one of paragraphs 1 to 5, The above operating condition data includes at least one of signal bandwidth, PAPR (peak-to-average power ratio), output signal power, ambient temperature of the communication device, and beamforming-related information.

7. In any one of paragraphs 1 through 6, The above DPD device (40) further includes a discriminator (320b, 420) operating in the learning mode, and A communication device configured such that the discriminator (320b, 420) alternately receives the fake data (302, 404) output from the generator (310b, 410) or the input signal collected from the power amplifier (450) as input data, and determines whether the input data is fake or real data through repeated training according to the operating conditions based on the operating condition data.

8. In any one of paragraphs 1 through 7, A communication device in which the operating power applied to the discriminator (320b, 420) in the above operating mode is cut off or the operation of the discriminator (320b, 420) is switched to an off state.

9. In any one of paragraphs 1 through 8, The above repeated training is performed until one of the defined criteria is satisfied, and The above-determined standards are, A first criterion for performing the above repeated training a predetermined number of times, A second criterion for performing the above repeated training until the difference between the above fake data and real data reaches a predetermined range, and A communication device comprising a third criterion that performs the above repeated training until a predetermined communication quality based on the above operating condition data is satisfied.

10. In any one of paragraphs 1 through 9, A digital-to-analog converter (DAC) (440, 540) connected between the output terminal of the DPD device (40, 50) and the input terminal of the power amplifier (450, 550) to convert the output signal of the DPD device (40, 50) into an analog signal; An analog-to-digital converter (ADC) (460, 560) that converts the output signal collected from the power amplifier into a digital signal; and A communication device further comprising a processor (730) that controls the operation of the DPD device (40, 50) based on at least one of the above-mentioned criteria.

11. A communication device that performs digital pre-distortion (DPD) on a transmission signal of a wireless communication system, A power amplifier (450) that amplifies the above transmission signal; One or more processors (730) comprising a processing circuitry of a DPD device (40) including a conditional generative adversarial network (GAN) based generator (310b, 410); and The device includes a memory (720) for storing instructions, and when the instructions are executed individually or collectively by one or more processors (730), the generator (310b, 410) in the DPD device (40), The operating condition data (305, 403) of the communication device and the input / output signals (301, 405, 406) collected from the power amplifier (450) are received, and A communication device that causes the generation of the fake data (302, 404) trained to compensate for non-linear distortion in the output of the power amplifier (450), based on data obtained through repeated training based on the conditional GAN ​​using the collected input / output signals (301, 405, 406) and the operating condition data (305, 403).

12. In Paragraph 11, The operation mode of the above DPD device (40, 50) includes a learning mode for repeated training based on the conditional GAN ​​and an operation mode in which DPD operation is performed on an actual input transmission signal in the communication device.

13. In Paragraph 11 or 12, When the above commands are executed individually or collectively by the one or more processors (730), the communication device, Further causing data preprocessing to be performed on the input signals of the DPD devices (40, 50) and the operating condition data, When the above DPD device (40, 50) operates in the above operating mode, the input signal includes an actual input transmission signal, and When the above DPD device (40, 50) operates in the learning mode, the input signal includes an output signal collected from the power amplifier, and A communication device in which the above data preprocessing includes at least one of normalization and one-hot encoding.

14. In any one of paragraphs 11 to 12, The above DPD device (40) further includes the conditional GAN-based discriminator (320b, 420) operating in the learning mode, and When the above commands are executed individually or collectively by the one or more processors (730), the discriminator (320b, 420) in the DPD device (40), The fake data (302, 404) output from the generator (310b, 410) or the input signal collected from the power amplifier (450) is alternately received as input data, and A communication device that causes the input data to be determined through repeated training whether it is fake or real data according to operating conditions based on the above operating condition data.

15. In a storage medium storing at least one instruction readable by a computer, The above at least one command is: In a communication device comprising a DPD device that performs digital pre-distortion (DPD) at the input terminal of a power amplifier that amplifies a transmission signal, when executed individually or collectively by one or more processors (730), the DPD device is caused to perform at least one operation. The above at least one operation is: Operation of receiving operating condition data of the communication device and input / output signals collected from the power amplifier; and A storage medium comprising an operation to generate trained fake data such that non-linear distortion in the output of the power amplifier is compensated based on data obtained through repeated training based on a conditional generative adversarial network (GAN) using the collected input / output signals and the operating condition data.

Images