Method and device for estimating channel characteristics of frequency band for delay spread estimation in wireless communication system

Abstract

Provided are a method and device for estimating channel characteristics in a wireless communication system. The method comprises the processes of: estimating, on the basis of input data, the channel characteristics of a reception channel by using a plurality of extended convolutional blocks having a parallel connection structure in a convolutional neural network (CNN)-based model; and estimating the level of delay spread of the reception channel on the basis of the estimated channel characteristics.

Description

Method and apparatus for estimating channel characteristics of a frequency band for delay spread estimation in a wireless communication system

[0001] The present disclosure relates to a method and apparatus for estimating channel characteristics in a wireless communication system.

[0002] To meet the increasing demand for wireless data traffic following the 4th generation (4G) system (e.g., LTE (long-term evolution) system), 5th generation (5G) systems (e.g., NR (New Radio) system) have been developed and commercialized. 5G systems can be implemented in the mmWave band. To mitigate path loss and increase the transmission distance of radio waves in the mmWave band, beamforming, massive array multiple input / output (massive MIMO), full-dimensional multiple input / output (Full Dimensional MIMO: FD-MIMO), array antenna, analog beamforming, and large-scale antenna technologies are being discussed for 5G systems.

[0003] Orthogonal Frequency Division Multiplexing (OFDM) is used in wireless communication systems such as 4G and 5G systems. Since OFDM is robust against interference between signals and frequency selective fading, it is suitable for broadband signal transmission. Frequency selective fading is a type of fading phenomenon (e.g., signal attenuation) that occurs in a multipath communication environment. It refers to a phenomenon where different frequency components are attenuated to different degrees, causing the signal to be strongly affected only in specific frequency bands. This frequency selective fading primarily occurs in communication environments where delay spread caused by multipath is severe. Due to this frequency selectivity, the transmitted signal in a wireless communication system may be weak in some frequency bands within the entire frequency band, while being received strongly in other frequency bands.

[0004] In OFDM-based communication systems, overall performance is directly influenced by frequency domain channel estimation performance; therefore, various algorithms are being researched to enhance the overall performance of the communication system by efficiently analyzing frequency domain channel characteristics. For example, to improve frequency domain channel estimation performance, it is important to effectively extract the frequency selectivity characteristics of the receiving channel in the communication device and design an optimal filter accordingly. Furthermore, improving channel estimation performance in wireless communication systems can lead to increased cell coverage and throughput.

[0005] In wireless communication systems, the frequency-selective characteristics of multipath channels are strongly correlated with the delay spread of the receiving channel; therefore, properly estimating the delay spread is necessary to improve reception performance in communication devices, such as base stations and user equipment (UEs), that perform communication through channel estimation.

[0006] One or more aspects of the present disclosure provide a method and apparatus for estimating channel characteristics of a frequency band based on artificial intelligence (AI) in a wireless communication system, and for estimating delay spreading based on the estimated channel characteristics of the frequency band.

[0007] One or more aspects of the present disclosure provide a method and apparatus for efficiently estimating channel characteristics based on AI for a communication device using least square (LS) channel estimation, and for estimating delay spread in a wireless communication system based on said estimated channel characteristics.

[0008] According to an embodiment of the present disclosure, a method is provided to be performed in a communication device in a wireless communication system. The method may include the step of obtaining a channel matrix in the frequency domain estimated using an LS channel estimation method for a receiving channel. The method may include the step of constructing input data for a convolutional neural network (CNN)-based model for estimating channel characteristics of the receiving channel from the channel matrix. The method may include the step of estimating the channel characteristics of the receiving channel using a plurality of extended convolution blocks having a parallel connection structure in the CNN-based model based on the input data. The method may include the step of estimating the delay spread level of the receiving channel based on the estimated channel characteristics.

[0009] The above method may further include a process of performing iterative training of the CNN-based model using the plurality of extended convolution blocks having the parallel connection structure.

[0010] The process of performing the above iterative learning can be terminated when convergence is achieved based on the validation loss of the above CNN-based model.

[0011] The process of estimating the channel characteristics may include a process of extracting the channel characteristics through multiple reception fields of the multiple extended convolution blocks.

[0012] The plurality of extended convolution blocks having the above parallel connection structure can be set based on different expansion rates that indicate the degree of dispersion of the reception field in the frequency domain.

[0013] The process of configuring the input data may further include a process of performing a 1D point-wise convolution operation for resizing the input data based on the number of the multiple expanded convolution blocks.

[0014] The operation mode of the communication device includes a learning mode and an operation mode, and the method may further include: in the learning mode: a process of performing iterative learning of the process of estimating the channel characteristics and the process of estimating the delay spread level; and in the operation mode: when the iterative learning is performed a predetermined number of times, a process of estimating the channel characteristics of the receiving channel using the CNN-based model and estimating the delay spread level based on the estimated channel characteristics.

[0015] The above method may further include a process of performing additional training on the CNN-based model based on actual channel data collected by the communication device as input data in the above operating mode.

[0016] The above delay diffusion level can be divided into multiple levels depending on the degree of delay diffusion.

[0017] The above method may further include a process of concatenating channel characteristic data of the receiving channel output from the plurality of extended convolution blocks, and a process of outputting a probability value that the channel characteristic of the receiving channel belongs to the delay spreading level based on the channel characteristic data.

[0018] According to an aspect of the present disclosure, a communication device in a wireless communication system comprises one or more processors including processing circuitry; and a memory for storing instructions, wherein when the instructions are executed individually or collectively by the one or more processors, the communication device may cause to obtain a channel matrix in the frequency domain estimated using a least square (LS) channel estimation method for a receiving channel. When the instructions are executed individually or collectively by the one or more processors, the communication device may cause to construct input data for a convolutional neural network (CNN)-based model for estimating channel characteristics of the receiving channel from the channel matrix. When the instructions are executed individually or collectively by the one or more processors, the communication device may cause to estimate the channel characteristics of the receiving channel using a plurality of extended convolution blocks having a parallel connection structure in the CNN-based model based on the input data. When the above commands are executed individually or collectively by the one or more processors, the communication device may be caused to estimate the delay spread level of the receiving channel based on the estimated channel characteristics.

[0019] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause iterative training of the CNN-based model using the plurality of extended convolution blocks having the parallel connection structure.

[0020] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause the iterative learning to terminate based on the convergence of the validation loss of the CNN-based model.

[0021] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause the channel characteristics to be extracted through a plurality of receiving fields of the plurality of extended convolution blocks.

[0022] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause different expansion rates to be set for the plurality of expanded convolution blocks, which indicate the degree of dispersion of the reception field in the frequency domain.

[0023] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause to perform a 1D point-wise convolution operation for resizing the input data based on the number of the multiple extended convolution blocks.

[0024] The operation mode of the communication device includes a learning mode and an operation mode, and when the commands are executed individually or collectively by the one or more processors, the communication device may further cause the device to: in the learning mode: perform iterative learning of estimating the channel characteristics and estimating the delay spread level, and in the operation mode: when the iterative learning has been performed a predetermined number of times, to use the CNN-based model to estimate the channel characteristics of the receiving channel and to estimate the delay spread level based on the estimated channel characteristics.

[0025] The above delay diffusion level can be divided into multiple levels depending on the degree of delay diffusion.

[0026] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause the channel characteristic data of the receiving channel output from the plurality of extended convolution blocks to concatenate and output a probability value that the channel characteristic of the receiving channel belongs to the delay spreading level based on the channel characteristic data.

[0027] When the above commands are executed individually or collectively by the one or more processors, the communication device may further cause the communication device to perform additional training of the CNN-based model based on actual channel data collected by the communication device as input data in the operating mode.

[0028] According to an embodiment of the present disclosure, in a storage medium storing at least one computer-readable instruction, the at least one instruction causes a communication device to perform at least one operation when executed individually or collectively by one or more processors (1330) including processing circuitry, and the at least one operation may include an operation of obtaining a channel matrix in the frequency domain estimated using a least square (LS) channel estimation method for a receiving channel. The at least one operation may include an operation of constructing input data for a convolutional neural network (CNN)-based model for estimating channel characteristics of the receiving channel from the channel matrix. The at least one operation may include an operation of estimating the channel characteristics of the receiving channel using a plurality of extended convolution blocks having a parallel connection structure in the CNN-based model based on the input data. The at least one operation may include an operation of estimating the delay spread level of the receiving channel based on the estimated channel characteristics.

[0029] The following embodiments are illustrated in the accompanying drawings, where the same reference numerals in the various drawings indicate corresponding parts. The above embodiments will be better understood through the following description with reference to the following exemplary drawings. The embodiments are illustrated in an exemplary manner in the accompanying drawings:

[0030] Figure 1 shows an example of the basic structure of time-frequency resources of a 5G system;

[0031] Figure 2 shows an example of the frame, subframe, and slot structure of a 5G system.

[0032] Figure 3 shows an example of a bandwidth portion setting in a 5G system;

[0033] FIG. 4 illustrates an example of a method for a communication device to estimate channel characteristics in a wireless communication system according to an embodiment;

[0034] FIG. 5 illustrates an example of a method for a communication device to estimate channel characteristics according to an embodiment of the present disclosure;

[0035] FIG. 6 shows an example of an extended convolution filter used in a CNN-based model for channel characteristic estimation in a communication device according to an embodiment of the present disclosure;

[0036] FIG. 7 illustrates an example of a configuration of a CNN-based model comprising a plurality of extended convolution blocks of a parallel connection structure according to an embodiment of the present disclosure;

[0037] FIG. 8 shows an example of one configuration of an extended convolution block (DCB) included in the CNN-based model of FIG. 7;

[0038] FIG. 9 shows an example of a configuration of a fully connected block (FCB) included in the CNN-based model of FIG. 7;

[0039] FIG. 10 illustrates an example of a method for estimating channel characteristics using a CNN-based model for delay spread estimation in a communication device according to an embodiment of the present disclosure;

[0040] FIGS. 11 and 12 show examples of simulation results of delay spread estimation performance using a CNN-based model according to an embodiment of the present disclosure; and

[0041] FIG. 13 shows an example of the configuration of a communication device in a wireless communication system according to an embodiment of the present disclosure.

[0042] 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.

[0043] 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.

[0044] 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.

[0045] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specified 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.

[0046] 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.

[0047] 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).

[0048] In the present disclosure, a base station (BS) is a network entity capable of performing resource allocation for a terminal and communicating with the terminal 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, wireless access unit, base station controller, a node on a network, or a TRP (transmission reception point). A terminal (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.

[0049] For convenience of explanation below, some terms and names defined in the 3GPP NR standard may be used. However, the present invention is not limited by the above terms and names and can be applied in the same way to systems conforming to other standards.

[0050] Figure 1 shows an example of the basic structure of time-frequency resources of a 5G system.

[0051] In FIG. 1, the horizontal axis represents the time domain, and the vertical axis represents the frequency domain. In the time and frequency domains, the basic unit of a resource is a resource element (RE) (101), which can be defined as one OFDM (Orthogonal Frequency Division Multiplexing) symbol (102) on the time axis and one subcarrier (103) on the frequency axis. In the frequency domain For example, 12 consecutive REs can form a single resource block (RB) (104). In one embodiment, multiple OFDM symbols can form a single subframe (110). In FIG. 1 is the number of OFDM symbols per subframe (110) for setting the subcarrier spacing (μ), and for a more specific description of the resource structure in a 5G system, refer to section 4 of TS 38.211 (5G specification).

[0052] The above RE (101) can be represented as an OFDM symbol index and a subcarrier index. The above RB (104) is in the frequency domain. It can be defined as a number of consecutive subcarriers. The resource block RB (101) can be referred to as a physical resource block (PRB). In a 5G system = 12, and the data rate can be increased in proportion to the number of RBs scheduled to the terminal. In a wireless communication system, the base station maps data in units of RBs and can generally perform scheduling for RBs that constitute one slot for the terminal. That is, in a 5G system, the basic time unit for scheduling is a slot, and the basic frequency unit for scheduling can be an RB.

[0053] Figure 2 shows an example of the frame, subframe, and slot structure of a 5G system.

[0054] Referring to FIG. 2, one frame (Frame, 200) is composed of one or more subframes (Subframe, 201), and one subframe may be composed of one or more slots (Slot, 202). For example, one frame (200) may be defined as 10ms. One subframe (201) may be defined as 1ms, in which case one frame (2-00) may be composed of a total of 10 subframes (201). One slot (202, 203) may be defined as 14 OFDM symbols (i.e., the number of symbols per slot ( )=14). One subframe (201) may be composed of one or more slots (2-02, 2-03), and the number of slots (202, 203) per one subframe (201) may vary depending on the setting value μ (204, 205) for the subcarrier spacing. In the example of FIG. 2, cases where μ=0 (204) and μ=1 (205) are shown as the subcarrier spacing setting value are illustrated. For example, when μ=0 (204), one subframe (201) may be composed of one slot (202), and when μ=1 (205), one subframe (201) may be composed of two slots (203). That is, the number of slots per one subframe ( ) may vary, and accordingly, the number of slots per frame ( ) can change.

[0055] Figure 3 shows an example of a Bandwidth Part (BWP) setting in a 5G system.

[0056] FIG. 3 illustrates an example in which a terminal bandwidth (UE bandwidth) (300) is configured into two bandwidth portions, for example, bandwidth portion #1 (BWP#1) (301) and bandwidth portion #2 (BWP#2) (302). However, the present disclosure is not limited thereto, and a base station may configure one or more BWPs for a terminal and configure the following information for each BWP. For example, "locationAndBandwidth" indicates the location and bandwidth in the frequency domain of the BWP, "subcarrierSpacing" indicates the subcarrier spacing to be used in the BWP, and "cyclicPrefix" indicates whether an extended CP (cyclic prefix) is used for the BWP.

[0057] In a 5G system, the base station can set up, for example, four DL BWPs for the terminal and activate one of the four DL BWPs. Additionally, the base station can set up, for example, four UL BWPs for the terminal and activate one of the four UL BWPs.

[0058] The above examples are not limited to the above, and various parameters related to the BWP may be configured for the terminal in addition to the above configuration information. The above information may be transmitted by the base station to the terminal via upper layer signaling, for example, Radio Resource Control (RRC) signaling. According to the embodiment, at least one BWP among a single BWP or a plurality of configured BWPs may be activated. Information regarding the activation status of the BWP configured according to the embodiment may be transmitted semi-statically from the base station to the terminal via RRC signaling or dynamically via Downlink Control Information (DCI).

[0059] The settings for the BWP supported by the above 5G can be used for various purposes.

[0060] According to some embodiments, in exemplary cases where the bandwidth supported by the terminal is smaller than the system bandwidth, this can be supported through the BWP setting. For example, by setting the frequency location of the BWP to the terminal, the terminal can transmit and receive data at a specific frequency location within the system bandwidth.

[0061] Additionally, according to some embodiments, a base station may set multiple BWPs for a terminal for the purpose of supporting different numerologies. For example, to support data transmission and reception using both a 15 kHz subcarrier interval and a 30 kHz subcarrier interval for a terminal, two bandwidth portions may be set to subcarrier intervals of 15 kHz and 30 kHz, respectively. Different bandwidth portions may be frequency division multiplexed, and when data transmission and reception is to be performed with a specific subcarrier interval, the BWP set for that subcarrier interval may be activated.

[0062] Recently, neural network-based algorithms based on learning large amounts of data are being implemented in various technology fields, including communication systems. For example, neural network-based algorithms and AI-based techniques suitable for the architecture of next-generation mmWave or THz band communication systems, such as 5G or 6G systems, are being proposed and researched. Recently, various AI-based optimization techniques are attracting attention to effectively solve the problem of receiving channel estimation in wireless communication systems.

[0063] Below, we will describe a method for estimating delay spread using AI-based channel characteristics of a frequency band in a communication device that performs channel estimation in a wireless communication system.

[0064] Technology related to channel estimation methods used in communication devices such as base stations and terminals includes the least squares (LS) estimation method and the minimum mean square error (MMSE) estimation method.

[0065] The aforementioned LS estimation method estimates channel characteristics by minimizing the difference between a known reference signal (or pilot signal) for channel estimation between a transmitting device and a receiving device and the received signal. This LS estimation method is widely used in real-time communication systems because it is relatively simple to implement in communication devices and has low computational complexity. However, the performance of the LS estimation method may degrade in communication environments where the influence of delay spreading or noise is significant. The aforementioned MMSE estimation method minimizes channel estimation errors by considering both noise and the statistical characteristics of the channel. While the MMSE estimation method guarantees stable channel estimation performance in noisy communication environments, it has high computational complexity.

[0066] According to an embodiment, a method for estimating channel characteristics of a frequency band based on AI is provided to reduce the impact of delay spreading during channel estimation in a communication device, such as a base station, that uses an LS estimation method for low complexity as a channel estimation method. According to the method according to the embodiment of the present disclosure, it is possible to minimize the reduction in communication performance due to delay spreading while using an LS estimation method of low complexity.

[0067] An example of an AI model that can be used for estimating channel characteristics in the frequency band is a convolutional neural network (CNN) based model. For example, CNN-based models demonstrate excellent performance in the field of computer vision. CNNs can process input data in the frequency domain through various types of operations in the network, extract features from the input data, and ultimately generate results for classification or prediction. In an example case where channel characteristics are estimated using an AI model, the input data may include, for example, an I (in-phase) channel signal and a Q (quadrature) channel signal (hereinafter referred to as IQ signals) in the frequency domain.

[0068] CNNs are neural networks primarily designed for two-dimensional data, such as images, and can form a hierarchical structure by combining multiple layers. CNNs utilize two elements—receptive fields and convolution filters—to process various input data and extract useful features from it. The convolution filters consist of matrices that detect or estimate specific patterns in the input data and can be used to learn local patterns within the input data. CNNs can extract features from input data by using multiple convolution filters across multiple layers. The receptive field refers to the area of ​​data that can be recognized by the convolution filter at a single step. As the CNN passes through multiple layers, the area of ​​data recognizable by the convolution filter (e.g., the receptive field) can progressively expand. By utilizing multiple convolution filters, CNNs can progressively and accurately detect or estimate specific patterns in the input data.

[0069] In technologies related to CNN-based models, such as ResNet (residual network) and VGGNet (visual geometry group network), features of input data can be extracted and learned by setting the size of the convolution filters of all layers to be the same and stacking the convolution layers deeply. Specifically, in CNN-based models, as input data passes through the layers, features ranging from low-level to high-level are extracted by the convolution operations of the convolution filters, and CNN-based models can learn model parameters in a direction that minimizes a pre-set loss function using the extracted features.

[0070] Due to the nature of CNN structures, which extract data features by performing convolution operations, the range of data processed by the convolution filter (reception field) widens as input data passes through multiple layers; therefore, stacking relatively deep layers in CNN-based models generally demonstrates superior performance.

[0071] Figure 4 is a diagram showing an example of a method for estimating channel characteristics using a CNN-based model in a communication device that uses an LS channel estimation method in a wireless communication system, and estimating the delay spread of a frequency band according to the estimated channel characteristics.

[0072] Referring to FIG. 4, the communication device performs channel estimation using an LS channel estimation method (410). To explain the LS channel estimation method, if we assume that in the channel estimation process using a reference signal (or pilot signal), the channel matrix or channel vector to be estimated is H, the symbol vector of the transmitted signal (e.g., reference signal or pilot signal) is x, the noise signal vector is n, and the received signal vector is y, then y can be expressed as shown in [Equation 1] below.

[0073] [Mathematical Formula 1]

[0074]

[0075] Channel estimation is the process of estimating the channel matrix H, which represents the channel characteristics, in the above [Equation 1]. In the LS channel estimation method, the channel matrix H that minimizes the square of the error (y-Hx) between the signal transmitted through the channel (Hx) and the signal received by the communication device (y) can be estimated in the manner shown in [Equation 2] below. In [Equation 2] below, is the norm operation of a vector.

[0076] [Mathematical Formula 2]

[0077]

[0078] For example, in the case where the communication device is a base station, the base station can estimate the initial channel matrix H of the receiving channel, where the uplink signal transmitted by the terminal is received, in the manner described above. For example, the uplink signal may be a reference signal for channel estimation. In a communication system, the channel matrix has complex values ​​for each subcarrier in the frequency domain and can be represented as an I / Q signal containing amplitude and phase information for each subcarrier. The channel matrix reflects the frequency-selective characteristics caused by multipath fading between the transmitter and the receiver.

[0079] The communication device performs (initial) channel estimation using an LS channel estimation method, then analyzes the estimated channel characteristics (e.g., spatial characteristics of channel data) using a CNN-based model with deep layers (420), and can use the spatial characteristic information of the channel from the analysis to estimate the delay spread of the channel (430). The communication device can obtain a compensated / modified channel matrix by applying a filter, etc., to the channel matrix to compensate for the estimated delay spread so as to reduce the influence of interference / noise, etc., on the channel matrix estimated for the receiving channel.

[0080] There are two main methods for expanding the receptive field of a CNN structure: stacking convolution layers deeply and increasing the size of convolution filters. The example in Figure 4 illustrates a method of stacking convolution layers deeply by connecting multiple convolution blocks (420-1, …, 420-n) in series. As shown in Figure 4, performance improvement in channel characteristic estimation can be expected by expanding the range of data processed by convolution filters (receptive field) through the method of stacking convolution layers deeply. However, in the examples where convolution layers are stacked deeply or the size of convolution filters is increased, the number of parameters that must be processed by the communication device increases, thereby increasing the computational complexity (e.g., amount of computation) for channel characteristic estimation. In next-generation wireless communication systems where real-time signal processing and the resulting stability are essential requirements, such an increase in computational complexity can lead to a decrease in the performance of the overall communication system.

[0081] According to an embodiment of the present disclosure, a CNN-based model with a new structure is provided. The CNN-based model according to an embodiment of the present disclosure can improve channel characteristic estimation performance without using a structure that expands the receiving field by connecting a plurality of convolution blocks in series, as shown in the example of FIG. 4. The CNN-based model according to an embodiment of the present disclosure reduces the computational complexity of a CNN-based model that estimates channel characteristics of a frequency band in a communication device using an LS channel estimation method. One or more embodiments of the present disclosure to be described below propose a CNN-based model using convolution (filter) blocks of a parallel connection structure suitable for the delay spread estimation problem of a multipath channel, and can provide improved channel characteristic estimation performance with less computational complexity compared to technologies related to CNN-based models using convolution (filter) blocks of a serial connection structure.

[0082] According to an embodiment of the present disclosure, in an exemplary case where a communication device that performs channel estimation using an LS channel estimation method in an orthogonal frequency division multiplexing (OFDM)-based wireless communication system estimates channel characteristics for delay spread estimation, one or more embodiments of the present disclosure may provide a method suitable for extracting channel characteristics in a frequency band in real time based on AI in the communication device.

[0083] FIG. 5 illustrates an example of a method for estimating channel characteristics of a frequency band using a CNN model for delay spread estimation in a communication device using an LS channel estimation method according to an embodiment of the present disclosure.

[0084] Referring to FIG. 5, in operation (510), the method according to the embodiment may include a data configuration operation in which a channel matrix in the frequency domain, initially estimated by an LS channel estimation method, is configured into the format of input data for an AI model according to the embodiment of the present disclosure. In operation (520), the method according to the embodiment may include an operation of training an AI model to estimate channel characteristics using the configured input data. In operation (530), the method according to the embodiment may include an operation of estimating the delay spread level of a receiving channel using the channel characteristics estimated through the AI ​​model. According to the embodiment, the AI ​​model (521) is a pre-trained AI model (521) that has completed learning for convenience of explanation. The operations 510 to 530 may be repeated a predetermined number of times by a communication device such as a base station, for example, for the estimation of channel characteristics. According to the embodiment, the AI ​​model (521) may use a CNN-based model (522) that includes a plurality of dilated convolution filters in a parallel connection structure. According to the embodiment, the pre-trained AI model (521) and the CNN-based model (522) that have completed learning can be understood as the same.

[0085] Referring specifically to the method of FIG. 5, in the operation 510 above, the communication device can reconstruct the channel matrix initially estimated by the LS channel estimation method, considering the pre-training operation of the AI ​​model proposed in the present disclosure, to match the input data format of the AI ​​model. For example, the channel matrix can be represented as an I / Q signal containing amplitude and phase information in each subcarrier, having complex values ​​for each subcarrier in the frequency domain. In the operation 510 above, the communication device obtains a channel matrix by separating and matrixing the I channel component and the Q channel component of the channel coefficients for each subcarrier, and can construct input data for the pre-training of the AI ​​model according to an embodiment of the present disclosure through resizing operations, such as considering the number of convolution filters (or convolution blocks) in the AI ​​model from the obtained channel matrix.

[0086] According to an embodiment, the input data may further include information regarding the delayed diffusion level estimated during previous training in the iterative training process of the AI ​​model, and may be used to improve the accuracy of estimating channel characteristics in the AI ​​model. The information regarding the delayed diffusion level estimated during previous training may be used as ground truth information to improve the estimation accuracy of the AI ​​model. For example, in iterative training for estimation by delayed diffusion level as described in [Table 1] below, the ground truth information may be included in the input data, and iterative training may be performed so that the estimation result of the AI ​​model's channel characteristics matches the ground truth information.

[0087] In the above 520 operation, the communication device can iteratively train an AI model using multiple extended convolution filters of a parallel connection structure to minimize the value of a loss function suitable for delay spread estimation using the input data configured in the above 510 operation. The communication device can train the AI ​​model, i.e., a CNN-based model, for a preset number of training iterations (e.g., the number of epochs, where an epoch refers to the process of the AI ​​model learning the entire given input data once). Additionally, the communication device can terminate iterative training in the example case where the validation loss for verifying the training of the AI ​​model converges, and use the AI ​​model with the updated optimal model weight values ​​as the pre-trained AI model (521).

[0088] In the above 530 operation, the communication device can estimate the delay spread of the receiving channel in the frequency band using the channel characteristics of the receiving channel estimated using a pre-trained AI model (521). In one embodiment, the communication device can estimate the delay spread of the receiving channel after the training of the AI ​​model is completed. In one embodiment, the communication device can also repeatedly estimate the delay spread of the receiving channel during the iterative training process of the AI ​​model. However, the present disclosure is not limited thereto, and according to the embodiment, the delay spread of the receiving channel may be estimated by including additional information, not limited to but including SNR (signal-to-noise ratio) information, in the channel matrix estimated by the above LS channel estimation method. In one embodiment, the delay spread estimation may be performed by estimating the delay spread level for the receiving channel as a probability value. The delay spread level may be divided into multiple levels according to the degree of delay spread.

[0089] According to an embodiment of the method illustrated in FIG. 5, the channel matrix initially estimated through the LS channel estimation method may not accurately estimate the received channel due to interference / noise effects caused by delay spreading in the communication environment. Therefore, the communication device can estimate the channel characteristics of the received channel by applying a pre-trained AI model (521) through iterative learning in the learning mode to a CNN-based model (522) in the operation mode, and estimate delay spreading using the estimated channel characteristics. The communication device can estimate delay spreading according to the channel characteristics in the operation mode and obtain a compensated / modified channel matrix by applying a filter, etc., to the estimated channel matrix to compensate for the estimated delay spreading.

[0090] According to an embodiment, the method of FIG. 5 illustrates a method of training an AI model through iterative learning using input data in a learning mode, but the present disclosure is not limited thereto, and according to an embodiment, it is also possible to perform additional learning on a CNN-based model (522) using channel data of a newly collected I / Q signal format as input data while the communication device performs actual signal transmission and reception in an operating mode.

[0091] According to an embodiment of the present disclosure, the communication device may operate by being divided into a learning mode and an operation mode.

[0092] In one embodiment, in a learning mode, the communication device assumes delay spreading due to multipath in various communication environments and performs channel estimation by an LS channel estimation method using pre-given / set communication parameters, and the AI ​​model can perform iterative training to estimate channel characteristics by receiving input data of a defined format from the estimated channel matrix. In one embodiment, the iterative training may be performed until the number of training iterations of the AI ​​model reaches a defined number (e.g., number of epochs). In one embodiment, the iterative training may be performed until a defined channel quality considering delay spreading is reached.

[0093] In one embodiment, when the communication device completes the training of the AI ​​model in the operation mode and obtains a pre-trained AI model (521), the communication device can use the pre-trained AI model (521) as a CNN-based model (522) in an actual communication environment. In the operation mode, the communication device performs channel estimation by the LS channel estimation method and can estimate the channel characteristics of the receiving channel using a CNN-based model (522) that utilizes a plurality of extended convolution filters of a parallel connection structure. Subsequently, the communication device can estimate the delay spread according to the channel characteristics in the operation mode and obtain a compensated / modified channel matrix by applying a filter, etc., to the estimated channel matrix to compensate for the estimated delay spread.

[0094] FIG. 6 shows an example of an extended convolution filter used in a CNN-based model for channel characteristic estimation in a communication device according to an embodiment of the present disclosure.

[0095] According to an embodiment of the present disclosure, a supervised learning-based CNN model specialized in high-dimensional data feature extraction may be used for channel-selective feature estimation in the frequency domain. According to an embodiment of the present disclosure, the CNN-based model may analyze the spatial features of the channel matrix estimated for the received channel in a plurality of reception fields and reflect them in the delay spread estimation of the received channel.

[0096] Referring to FIG. 6, the channel matrix estimated by the LS channel estimation method in the communication device has complex values ​​for each subcarrier in the frequency domain, as in the example of reference numeral 610, and can be represented as an I / Q signal containing amplitude and phase information for each subcarrier. The convolution filter (620) used in the conventional CNN-based model is exemplified as having a local structured reception field containing, for example, a fixed number of consecutive subcarriers in the frequency domain (e.g., three consecutive subcarriers in FIG. 6). The extended convolution filter (630) used in the CNN-based model according to the embodiment of the present disclosure may have a reception field containing, for example, a fixed number of dispersed subcarriers in the frequency domain (e.g., three dispersed subcarriers in FIG. 6). The extended convolution filter (630) is an example where the spacing between subcarriers (e.g., degree of dispersion in the frequency domain) is 3 (e.g., dilation rate = 3). By using the extended convolution filter (630) as in the example of FIG. 6, there is an advantage in that the input data in the frequency domain can be reflected in the training of the CNN model over a wide range while maintaining the same number of operations as the existing convolution filter (620).

[0097] The CNN-based model according to an embodiment of the present disclosure has a structure utilizing a plurality of extended convolution filters, and the extension rates of the plurality of extended convolution filters can be set to different values ​​for delayed spread estimation. By utilizing extended convolution filters in such a CNN-based model, it is more efficient for channel-selective characteristic estimation, which requires analyzing not only the relationships between adjacent frequency components but also the relationships between distant frequency components during channel-specific estimation.

[0098] FIG. 7 illustrates an example configuration of a CNN-based model comprising a plurality of extended convolution blocks of a parallel connection structure according to an embodiment of the present disclosure. A CNN-based model such as the example in FIG. 7 can be included in a communication device, such as a base station, that uses an LS channel estimation method and used to estimate channel characteristics of a frequency band for delay spread estimation.

[0099] Referring to FIG. 7, a CNN-based model according to an embodiment of the present disclosure includes a convolution filter block (720) comprising a plurality of dilated convolution blocks (DCBs) of a parallel connection structure (DCB 1 720-1, DCB 2 720-2, ..., and DCB N 720-N). In one embodiment, the CNN-based model may further include at least one of a preprocessing block (710), a concatenated block (730), a fully connected block (FCB) (740), and a delayed diffusion estimation block (750). For example, at least one of the preprocessing block (710), the concatenated block (730), the fully connected block (FCB) (740), and the delayed diffusion estimation block (750) is coupled with the convolution filter block (720).

[0100] According to the embodiment illustrated in FIG. 7, a plurality of blocks (710 to 750) may be implemented to include a memory storing a program of a CNN-based model and a processor that controls the overall operation of the communication device according to the CNN-based model. In one embodiment, the communication device may further include a channel estimation block that performs channel estimation according to an LS channel estimation method. In one embodiment, the communication device may further include a transceiver for transmitting and receiving wireless signals. In one embodiment, the CNN-based model of the plurality of blocks (710 to 750) may be implemented using a field programmable gate array (FPGA). In one embodiment, the communication device may operate by distinguishing between a learning mode and an operation mode according to the method of FIG. 5.

[0101] Referring to FIG. 7, the preprocessing block (710) can receive a frequency domain channel matrix estimated by the LS channel estimation method and perform a resizing operation based on the number of DCBs (720-1, …, 720-N) included in the convolution filter block (720). The resizing operation may utilize a 1D pointwise convolution operation, which is useful for reducing or expanding the channel axis in the input data. According to an embodiment, the resizing operation may utilize a known 1D pointwise convolution operation, which is useful for reducing or expanding the channel axis in the input data. Through the resizing operation, the number of DCBs (720-1, …, 720-N) (or the number of convolution filters) that extract channel features of the received channel can be adjusted. The number of the above DCBs (720-1, …, 720-N) (or the number of convolution filters) can be adjusted or optimized as needed in the communication device (or CNN-based model).

[0102] The convolution filter block (720) receives the resized channel matrix data as input data. The convolution filter block (720) includes N DCBs (720-1, …, 720-N) in a parallel connection structure, and the DCBs (720-1, …, 720-N) include N expansion rates (dil1, …dil) to estimate channel characteristics over a wide range in the frequency domain by considering frequency-selective characteristics. N ) can be set. In one embodiment, N expansion rates may all be set differently from each other. In one embodiment, at least some of the N expansion rates may be set differently from each other. In one embodiment, at least some of the N expansion rates may be set identically. In one embodiment, each DCB of the DCBs (720-1, …, 720-N) may include at least one convolution filter.

[0103] The above N DCBs (720-1, …, 720-N) are configured to extract channel characteristics through multiple reception fields by connecting N DCBs (720-1, …720-N) in parallel, rather than using a structure that expands the reception field by deeply connecting multiple convolution blocks in series as in the example of FIG. 4. By using a convolution filter block (720) with a parallel connection structure in this way, the computational complexity of the CNN-based model can be reduced compared to the serial connection structure of FIG. 4, and the channel characteristic estimation performance can be improved.

[0104] The above-mentioned connecting block (730) connects / connects channel characteristic data of the receiving channel extracted / outputted from a plurality of DCBs (720-1, …, 720-N) and outputs it to a fully connected block (FCB) (740).

[0105] The above fully connected block (FCB) (740) outputs a probability value that the channel characteristics of a receiving channel belong to a preset delay spread level using a softmax function used in a multi-classification model, which is a function that converts input data (e.g., channel characteristic data) into probability values. According to an embodiment, it may also be possible to use other known functions used in a multi-classification model that perform an operation equivalent to the softmax function.

[0106] The above delay spread estimation block (750) estimates the delay spread level to which the estimated channel characteristics for the receiving channel belong based on the probability value output from the FCB (740). The above delay spread level may be divided into four levels according to the degree of delay spread, as in the labeling of [Table 1] below, for example. However, the present disclosure is not limited thereto, and the number of levels may differ from four. A communication device to which the CNN-based model of FIG. 7 is applied may obtain a compensated / modified channel matrix in which the influence of noise, etc. is reduced by applying a filter, etc. to the channel matrix to compensate for the delay spread according to the estimated delay spread level. In an example embodiment where the communication device is a base station, the base station may perform LS channel estimation for each of the uplink channels of a plurality of terminals using the CNN-based model of FIG. 7, and may estimate the delay spread for each terminal based on the channel characteristics estimated with low complexity for each terminal's channel.

[0107] [Table 1]

[0108]

[0109] In the learning mode of a CNN-based model using extended convolution filters of a parallel connection structure according to an embodiment of the present disclosure, the received signal received by a communication device in the frequency domain of a received channel estimated according to an LS channel estimation method is Y, and the delay spreading level is In this case, the communication device includes the received signal Y in the input data of the CNN-based model, and the delay spread level estimated as a result of channel characteristic estimation learning using the received signal Y Training data can be configured by including the output. The convolution filter block (720) of the CNN-based model can incorporate all I / Q channel components of the input data into the training, and an example of the configuration of training data for training the CNN-based model can be expressed as [Equation 3] below.

[0110] [Mathematical Formula 3]

[0111]

[0112] FIG. 8 shows an example of a configuration of an extended convolution block (DCB) included in the CNN-based model of FIG. 7, and shows an example of a configuration of the i-th DCB (720-i) of a plurality of DCBs (720-1, …720-i, …, 720-N) (where 0 ≤ i ≤ n, i is an integer) included in the convolution filter block (720) of FIG. 7.

[0113] In one embodiment, a plurality of DCBs (720-1, …720-i, …, 720-N) may have the same configuration as the i-th DCB (720-i) of FIG. 8. In one embodiment, at least some of the blocks among the plurality of DCBs (720-1, …720-i, …, 720-N) may have a configuration different from that of the i-th DCB (720-i). Depending on the embodiment, having a different configuration may be understood as a case where the number of repetitions of the connection structure of dilated Cove1D, batch norm, and ReLU within the DCB is different. For example, DCB (720-1) may have a structure where the connection structure of dilated Cove1D, batch norm, and ReLU is repeated twice as in the example of FIG. 8, and DCB (720-2) may have a structure where the connection structure of dilated Cove1D, batch norm, and ReLU is repeated two or more times. The number of repetitions of the connection structure of dilated Cove1D, batch norm, and ReLU in each DCB can use a predetermined structure or be set variably depending on the iterative learning.

[0114] Referring to FIG. 8, the DCB may include dilated convolution layers (dilated Conv1D) (801, 804) expanded into multiple blocks, batch normalization (batch norm) (802, 805), and rectified linear units (ReLU) (803, 806). Each of the dilated convolution layers (dilated Conv1D) (801, 804) is a convolution filter that performs convolution operations for pattern learning to estimate channel characteristics on input 1D data. Each of the dilated convolution layers (dilated Conv1D) (801, 804) operates based on a dilation rate, as in the dilated convolution filter (630) of FIG. 6. Each of the batch norm (802, 805) is a block that normalizes the distribution of input data for the estimation of channel characteristics. Each of the Rectifying Linear Unit (ReLU) (803, 806) is a block that outputs 0 when the input value is less than 0 and outputs the input value as is when it is greater than 0, in order to process non-linear functions such as channel characteristics with low complexity. In FIG. 8, the input data of the DCB is combined with the output data of the Rectifying Linear Unit (ReLU) (806) to become the output data of the DCB. The example in FIG. 8 shows an example in which a structure in which the dilated Cove1D, batch norm, and ReLU are sequentially connected is repeated twice. In one embodiment, the connection structure of the dilated Cove1D, batch norm, and ReLU described above may be repeated one or multiple times.

[0115] Figure 9 shows an example of a configuration of a fully connected block (FCB) included in the CNN-based model of Figure 7.

[0116] Referring to FIG. 9, the FCB (740) may include a plurality of blocks, including dense blocks (901, 903, 905, 907), rectified linear units (ReLU) (902, 904, 906), and softmax (908). The rectified linear units (902, 904, 906) may be optionally included. Each of the dense blocks (901, 903, 905, 907) performs an analysis to separate input data (e.g., channel characteristic data) by delay spread level, and each of the rectified linear units (ReLU) (902, 904, 906) is a block that outputs 0 in cases where the input value is less than 0 and outputs the input value as is in cases where the input value is greater than 0 in order to process non-linear functions such as channel characteristics with low complexity. The softmax (908) receives analyzed channel characteristic data and outputs a probability value that falls within a preset delay spread level. The example in FIG. 9 shows an example where a structure in which Dense and ReLU are sequentially connected is repeated. In one embodiment, the connection structure of Dense and ReLU may be one or multiple repetitions may be repeated. And the softmax (908) is included at the end of the FCB (740).

[0117] FIG. 10 illustrates an example of a method for estimating channel characteristics using a CNN-based model for delay spread estimation in a communication device according to an embodiment of the present disclosure. Here, the CNN-based model may be the CNN-based model illustrated according to the exemplary embodiment of FIG. 7.

[0118] Referring to FIG. 10, in operation 1001, the method may include the process of obtaining a channel matrix in the frequency domain estimated by an LS channel estimation method for a receiving channel. For example, a communication device may obtain a channel matrix in the frequency domain estimated by an LS channel estimation method for a receiving channel.

[0119] 1002 In operation, the above method may include the process of constructing input data for a CNN-based model. For example, a communication device may construct input data for the CNN-based model of FIG. 7 to estimate channel characteristics of a receiving channel from a channel matrix.

[0120] 1003 In operation, the above method may include a process of estimating channel characteristics of a receiving channel using a plurality of extended convolution blocks (DCBs) of a parallel connection structure in the CNN-based model based on input data. For example, a communication device may estimate channel characteristics of a receiving channel using a plurality of extended convolution blocks (DCBs) of a parallel connection structure in the CNN-based model based on input data.

[0121] 1004 In operation, the method may include a process of estimating the delay spread level of the receiving channel based on the estimated channel characteristics. For example, a communication device may estimate the delay spread level of the receiving channel based on the estimated channel characteristics.

[0122] For example, a communication device using the CNN-based model of FIG. 7 can calculate the probability value of the delay spread of a receiving channel, distinguish the delay spread level according to a preset range, and then configure an input data set in the 1002 operation so as to estimate the corresponding delay spread level. The delay spread level estimated in the 1004 operation is a parameter that can be set according to the communication scenario and can be set to four levels as in the example of [Table 1]. In the present disclosure, the delay spread level is not limited to the example of [Table 1], and the degree of distinction of the delay spread level may be increased or decreased.

[0123] The loss function of the CNN-based model according to an embodiment of the present disclosure may utilize, for example, a categorical cross-entropy function used in multi-classification problems. The loss function It can be expressed as shown in [Mathematical Formula 4] below.

[0124] [Mathematical Formula 4]

[0125]

[0126] In the above [Mathematical Formula 4], is labeled It represents the k-th element of the one-hot encoded vector, and represents the k-th element of the final output vector of the CNN-based model. One-hot encoding refers to the process of converting categorical data into numerical data, such as binary data. In this disclosure, since training is performed to predict a total of four delayed diffusion levels as shown in the example of [Table 1], It can be set to. Model parameters can be updated until the training loss (i.e., validation loss) of the validation data for validating the training of a CNN-based model converges.

[0127] According to an embodiment, the method may further include a process of configuring parameters for communication between a terminal and a base station based on the estimated delay spreading level, and a process of the terminal performing communication based on the configured parameters.

[0128] FIGS. 11 and 12 show examples of simulation results of delay spread estimation performance using a CNN-based model according to an embodiment of the present disclosure.

[0129] The simulation environment of Figures 11 and 12 is the result of conducting simulations by configuring training and test data sets according to the normalized delay profiles of the TDL (tapped delay line)-A, TDL-B, TDL-C, TDL-D, and TDL-E channels defined in Table 7.7.2 of Rel 16, 3GPP standard document TR 38.901, using the PUSCH (physical uplink shared channel) 12 RB (resource block) environment in the 3GPP NR specification as an example.

[0130] In FIG. 11, reference number 1110 represents the accuracy of delay spread estimation when using a CNN-based model according to the exemplary embodiment shown in FIG. 7, and reference number 1120 represents the accuracy of delay spread estimation when using the CNN-based model-related technology of FIG. 4.

[0131] For example, FIG. 11 shows the results of a performance comparison between the ResNet (Residual Network) structure in the conventional CNN-based model of FIG. 4 and the ResNet structure in the CNN-based model according to the embodiment of the present disclosure shown in FIG. 7. Here, accuracy refers to the probability of accurately determining the delay spread level of the corresponding received channel when the received LS channel estimated signal is input into the CNN-based model. In FIG. 11, it can be seen that when using extended convolution blocks of a parallel connection structure, as in the CNN-based model of FIG. 7 according to the embodiment of the present disclosure, overall better performance is achieved across all SNR ranges compared to when using convolution blocks of a serial connection structure in the technology related to the CNN-based model of FIG. 4. For example, excellent performance is shown in the [-10, 0 dB] range where the SNR intensity is relatively weak, which demonstrates that the proposed neural network structure is suitable for the channel characteristic estimation problem in situations with severe noise signals. In other words, it can be seen that the accuracy of delay spread estimation according to the embodiment of the present disclosure is improved.

[0132] Referring to FIG. 12, reference numbers 1210 and 1211 respectively represent the success and failure of accurate detection of a delay spread level when using a CNN-based model according to an embodiment of the present disclosure shown in FIG. 7, and reference numbers 1220 and 1221 respectively represent the success and failure of detection of a delay spread level in an example of a related technical method for detecting a delay spread level in the frequency domain without using an AI model. In FIG. 12, reference numbers 1230 and 1231 respectively represent the success and failure of detection of a delay spread level in an example of a related technical method for detecting a delay spread level in the time domain without using an AI model. It can be seen that the delay spread level detection performance of the CNN-based model according to an embodiment of the present disclosure is improved compared to related technical methods that do not use a CNN-based model.

[0133] [Table 2]

[0134]

[0135] Table 2 above is a table comparing the computational complexity of the CNN-based model of FIG. 4 and the CNN-based model of FIG. 7 according to an embodiment of the present disclosure. In Table 2, the number of parameters refers to the total number of optimized model parameters of the trained neural network, and FLOPs (Floating point Operations Per Second) refers to the amount of floating-point operations that must be performed per second for inference after training.

[0136] As summarized in [Table 2] above, the CNN-based model according to the embodiment of the present disclosure illustrated in FIG. 7 can achieve excellent performance with a small amount of computation, and through this, it can be confirmed that the neural network structure proposed in the present disclosure is a structure suitable for wireless communication systems where real-time processing is important.

[0137] A method performed in a communication device using an LS channel estimation method in a wireless communication system as one embodiment may include: a process of obtaining a channel matrix in the frequency domain estimated by the LS channel estimation method for a receiving channel; a process of constructing input data for a convolutional neural network (CNN)-based model for estimating channel characteristics of the receiving channel from the channel matrix; a process of estimating channel characteristics of the receiving channel using a plurality of extended convolution blocks of a parallel connection structure in the CNN-based model based on the input data; and a process of estimating the delay spread level of the receiving channel based on the estimated channel characteristics.

[0138] In one embodiment, the method may further include a process of iteratively training the CNN-based model using the plurality of extended convolution blocks of the parallel connection structure.

[0139] In one embodiment, the iterative learning in the above method may be terminated when the validation loss of the CNN-based model converges.

[0140] In one embodiment, the method can extract the channel characteristics through a plurality of receiving fields of the plurality of extended convolution blocks.

[0141] In one embodiment, different expansion rates representing the degree of dispersion of the reception field in the frequency domain may be set for the plurality of expanded convolution blocks in the above method.

[0142] In one embodiment, the process of configuring the input data in the above method may further include the process of performing a 1D point-wise convolution operation for resizing the input data based on the number of the plurality of extended convolution blocks.

[0143] In one embodiment, the operation mode of the communication device in the above method includes a learning mode and an operation mode, and the method may further include a process of iteratively learning the estimation of the channel characteristics and the estimation of the delay spread level in the learning mode, and when the iterative learning is performed a predetermined number of times, a process of estimating the channel characteristics of the receiving channel using a CNN-based model learned in the operation mode and estimating the delay spread level based on the estimated channel characteristics.

[0144] In one embodiment, the method may further include a process of performing additional training on the CNN-based model using actual channel data collected from the communication device as input data in the operating mode.

[0145] In one embodiment, the delay diffusion level in the above method may be divided into multiple levels according to the degree of delay diffusion.

[0146] In one embodiment, the method may further include a process of concatenating channel characteristic data of the receiving channel output from the plurality of extended convolution blocks, and a process of outputting a probability value that the channel characteristic of the receiving channel belongs to the delay spreading level based on the channel characteristic data.

[0147] In one embodiment, a communication device may include one or more processors including processing circuitry and a memory for storing instructions, and when the instructions are executed individually or collectively by the one or more processors, the communication device may obtain a frequency domain channel matrix estimated using an LS channel estimation method for a receiving channel, construct input data for a convolutional neural network (CNN) based model for estimating channel characteristics of the receiving channel from the channel matrix, estimate the channel characteristics of the receiving channel using a plurality of extended convolution blocks having a parallel connection structure in the CNN based model based on the input data, and cause the delay spread level of the receiving channel to be estimated based on the estimated channel characteristics.

[0148] FIG. 13 is a diagram showing an example of the configuration of a communication device in a wireless communication system according to an embodiment of the present disclosure. The embodiments of FIG. 5 to 12 can be applied to the base station of FIG. 13.

[0149] The base station of FIG. 13 may include a processor (1330), a transceiver (1310), and a memory (1320). The processor (1330), transceiver (1310), and memory (1320) of the communication device of FIG. 13 may operate according to at least one of the embodiments of FIG. 5 to FIG. 12. However, the components of the base station are not limited to the examples described above. For example, the base station may include more components or fewer components than the components described above. In addition, the processor (1330), transceiver (1310), and memory (1320) may be implemented in the form of a single chip. The transceiver (1310) is a collective term for the receiver and the transmitter of the base station and can transmit and receive signals with a terminal or another base station. At this time, the signal transmitted and received may include at least one of control information and data. According to an embodiment, the transceiver (1310) may include a wired or wireless transceiver and may include various configurations for transmitting and receiving signals. The transceiver (1310) receives a signal and outputs it to a processor (1330), and can transmit the signal output from the processor (1330).

[0150] Additionally, the transceiver (1310) receives a communication signal and outputs it to the processor (1330), and can transmit the signal output from the processor (1330) to another base station through a network. The memory (1320) can store programs and data necessary for the operation of a base station according to at least one of the embodiments of FIGS. 5 to 12. Additionally, the memory (1320) can store control information or data included in a signal obtained from the base station. The memory (1320) can be composed of a storage medium or a combination of storage media such as ROM, RAM, a hard disk, CD-ROM, and DVD. Additionally, the processor (1330) can control a series of processes to enable the base station to operate according to at least one of the embodiments of FIGS. 5 to 12.

[0151] According to one or more embodiments, the methods and / or operations described above may be implemented in the form of hardware, software, or a combination of hardware and software.

[0152] In an example case implemented by software, a computer-readable storage medium may be provided for storing one or more programs (software modules). One or more programs stored on the computer-readable storage medium are configured for execution by one or more processors within an 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 this disclosure.

[0153] Such programs (software modules, software) may be stored in random access memory, non-volatile memory including flash memory, ROM (Read Only Memory), Electrically Erasable Programmable Read Only Memory (EEPROM), magnetic disc storage devices, Compact Disc-ROM (CD-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.

[0154] 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.

[0155] 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.

[0156] 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. In a communication device in a wireless communication system, One or more processors (1330) including processing circuitry; and The communication device includes a memory (1320) for storing instructions, and when the instructions are executed individually or collectively by one or more processors (1330), the communication device A channel matrix in the frequency domain estimated using the LS (least square) channel estimation method for the receiving channel is obtained, and Input data for a convolutional neural network (CNN)-based model for estimating the channel characteristics of the receiving channel from the above channel matrix is ​​constructed, and Based on the above input data, the channel characteristics of the receiving channel are estimated using the plurality of extended convolution blocks (720) having a parallel connection structure in the above CNN-based model, and A communication device that causes the delay spread level of the receiving channel to be estimated based on the above-mentioned estimated channel characteristics.

2. In Paragraph 1, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes iterative learning of the CNN-based model using the plurality of extended convolution blocks (720) having the parallel connection structure.

3. In Paragraph 1 or 2, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes the iterative learning to terminate based on the convergence of the validation loss of the above CNN-based model.

4. In any one of paragraphs 1 to 3, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes the channel characteristics to be extracted through multiple receiving fields of the multiple extended convolution blocks (720).

5. In any one of paragraphs 1 to 4, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes different expansion rates to be set for the plurality of expanded convolution blocks (720) that indicate the degree of dispersion of the receiving field in the frequency domain.

6. In any one of paragraphs 1 to 5, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes to perform a 1D point-based convolution (1D pointwise convolution) operation for resizing the input data based on the number of the above-mentioned multiple expanded convolution blocks (720).

7. In any one of paragraphs 1 through 6, The operation mode of the above communication device includes a learning mode and an operation mode, and When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, In the above learning mode: Estimating the above channel characteristics, performing iterative learning of estimating the above delay diffusion level, and In the above operating mode: A communication device that, when the above iterative learning is performed a predetermined number of times, estimates the channel characteristics of the receiving channel using the above CNN-based model and further causes the delay spreading level to be estimated based on the estimated channel characteristics.

8. In any one of paragraphs 1 through 7, The above delay spreading level is a communication device classified into multiple levels according to the degree of delay spreading.

9. In any one of paragraphs 1 through 8, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, Channel characteristic data of the receiving channel output from the above-mentioned plurality of extended convolution blocks (720) is concatenated, and A communication device that further causes the channel characteristic of the receiving channel to output a probability value that belongs to the delay spreading level based on the channel characteristic data.

10. In any one of paragraphs 1 through 9, When the above commands are executed individually or collectively by the one or more processors (1330), the communication device, A communication device that further causes the CNN-based model to perform additional training based on actual channel data collected by the communication device as input data in the above operating mode.

11. A method performed in a communication device in a wireless communication system, A process (1001) of obtaining a channel matrix in the frequency domain estimated using the LS (least square) channel estimation method for a receiving channel; A process (1002) for constructing input data for a convolutional neural network (CNN) based model for estimating the channel characteristics of the receiving channel from the channel matrix; A process (1003) for estimating the channel characteristics of the receiving channel using a plurality of extended convolution blocks (720) having a parallel connection structure in the CNN-based model based on the above input data; and A method comprising a process (1004) of estimating the delay spread level of the receiving channel based on the estimated channel characteristics above.

12. In Paragraph 11, The method further includes a process of performing iterative training of the CNN-based model using the plurality of extended convolution blocks having the parallel connection structure described above, and The process of performing the above iterative learning is a method that terminates based on the convergence of the validation loss of the above CNN-based model.

13. In Paragraph 11 or 12, The process of estimating the above channel characteristics (1003) is, A method comprising the process of extracting the channel characteristics through multiple receiving fields of the multiple extended convolution blocks.

14. In any one of paragraphs 11 through 14, A method in which the plurality of extended convolution blocks having the above parallel connection structure are set based on different extension rates indicating the degree of dispersion of the reception field in the frequency domain.

15. In a storage medium storing at least one instruction readable by a computer, When the above at least one instruction is executed individually or collectively by one or more processors (1330) including processing circuitry, it causes the communication device to perform at least one operation, and The above at least one operation is: Operation (1001) of obtaining a channel matrix in the estimated frequency domain using the LS (least square) channel estimation method for the receiving channel; Operation (1002) for constructing input data for a convolutional neural network (CNN) based model for estimating channel characteristics of the receiving channel from the channel matrix; Based on the above input data, an operation (1003) of estimating the channel characteristics of the receiving channel using a plurality of extended convolution blocks (720) having a parallel connection structure in the above CNN-based model; and A storage medium comprising an operation (1004) for estimating the delay spread level of the receiving channel based on the above-mentioned estimated channel characteristics.

Images