Cascade channel estimation method and device, communication equipment and computer readable storage medium

Abstract

The invention provides a cascade channel estimation method and device, communication equipment and a computer readable storage medium. The cascade channel comprises a static quasi-state channel between the base station and the intelligent reflecting surface, and a time-varying channel between the terminal equipment and the intelligent reflecting surface; the method comprises the steps of obtaining a first signal through an intelligent reflecting surface; constructing a cascade channel model of the time-varying channel; determining a first frequency domain channel matrix of the cascade channel based on the cascade channel model and a basis expansion model; determining a basis coefficient estimation matrix of the cascade channel based on the first signal and the first frequency domain channel matrix; performing optimization processing on the basis coefficient estimation matrix through a first model to obtain optimized cascade channel characteristics; according to the optimized cascade channel characteristics, obtaining an estimated value of the cascade channel; therefore, high-precision and practical estimation of the high-dimensional time-varying channel in the intelligent reflector-assisted mobile communication system is realized.

Description

Technical Field

[0001] This application relates to the fields of wireless communication and artificial intelligence, and in particular to a cascaded channel estimation method, apparatus, network device, and computer-readable storage medium. Background Technology

[0002] With the continuous development of wireless communication technology, spectrum resources are becoming increasingly scarce, making efficient utilization of existing spectrum resources a research hotspot. Reconfigurable Intelligent Surfaces (RIS), due to their ability to adjust electromagnetic wave propagation paths without active power supply, are widely used to improve the spectral efficiency and coverage performance of communication systems. In RIS-assisted communication systems, the accuracy of channel state information is crucial to system performance.

[0003] In existing technologies, channel estimation methods for RIS-assisted communication scenarios mainly include time-varying and time-invariant approaches. Time-invariant channel estimation methods typically assume the user is stationary and ignore the Doppler effect, thus exhibiting lower computational complexity and better estimation performance. However, in real-world communication environments, user mobility causes the channel to change over time, making time-invariant methods ill-suited for dynamic scenarios. Some solutions attempt to model the entire cascaded channel as a sparse structure and employ algorithms such as compressed sensing for joint estimation; however, these methods suffer from high computational overhead and poor model generalization ability when dealing with high-dimensional reflective units and time-varying characteristics. Furthermore, while introducing deep learning methods for time-varying channel estimation can capture nonlinear features, their accuracy and practicality in real-world applications still need improvement due to complex model design and a lack of targeted optimization mechanisms. Summary of the Invention

[0004] This application provides a cascaded channel estimation method, apparatus, communication device, and computer-readable storage medium, which can significantly reduce computational complexity while achieving high-precision and practical estimation of high-dimensional time-varying channels in intelligent reflector-assisted mobile communication systems.

[0005] The technical solution of this application embodiment is implemented as follows: This application provides a cascaded channel estimation method, wherein the cascaded channel includes a static quasi-state channel between a base station and a smart reflector, and a time-varying channel between a terminal device and the smart reflector; the method includes: The first signal is obtained through a smart reflective surface; Construct a cascaded channel model for the time-varying channel; Based on the cascaded channel model and the base extension model, the first frequency domain channel matrix of the cascaded channel is determined; Based on the first signal and the first frequency domain channel matrix, determine the basis coefficient estimation matrix of the cascaded channel; The basis coefficient estimation matrix is ​​optimized using the first model to obtain the optimized cascaded channel characteristics; Based on the optimized cascaded channel characteristics, the estimated value of the cascaded channel is obtained.

[0006] This application provides a cascaded channel estimation apparatus, wherein the cascaded channel includes a static quasi-state channel between a base station and a smart reflector, and a time-varying channel between a terminal device and the smart reflector; comprising: An acquisition unit is used to acquire a first signal through a smart reflective surface; A processing unit is used to construct a cascaded channel model of the time-varying channel; The determining unit is configured to: determine the first frequency domain channel matrix of the cascaded channel model based on the first signal and the basis extension model; determine the basis coefficient estimation matrix of the cascaded channel based on the first signal and the first frequency domain channel matrix; and optimize the basis coefficient estimation matrix using the first model to obtain the optimized cascaded channel characteristics. The acquisition unit is further configured to acquire an estimated value of the cascaded channel based on the optimized cascaded channel characteristics.

[0007] This application provides a network device, the network device comprising: Memory is used to store executable instructions or computer programs. The processor, when executing computer-executable instructions or computer programs stored in the memory, implements the cascaded channel estimation method provided in the embodiments of this application.

[0008] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions for implementing the cascaded channel estimation method provided in this application when executed by a processor.

[0009] In the technical solution of this application embodiment, a first signal is acquired through an intelligent reflector; a cascaded channel model of a time-varying channel is constructed; based on the cascaded channel model and a basis extension model, a first frequency domain channel matrix of the cascaded channel is determined; based on the first signal and the first frequency domain channel matrix, a basis coefficient estimation matrix of the cascaded channel is determined; the basis coefficient estimation matrix is ​​optimized using the first model to obtain optimized cascaded channel features; and the estimated value of the cascaded channel is obtained based on the optimized cascaded channel features. Thus, firstly, a first signal is acquired through an intelligent reflector; then, a cascaded channel model containing a direct path component and multiple scattered path components is constructed, and the high-dimensional time-varying channel information is reduced to a low-dimensional basis coefficient matrix using a basis extension model, thereby reducing the number of parameters to be estimated. Next, the basis coefficient estimation matrix of the channel is estimated based on the received signal and the basis coefficient matrix, and a trained residual super-resolution convolutional neural network is used to optimize the basis coefficient estimation matrix to improve the channel estimation accuracy. Finally, the optimized features are converted into channel estimates in complex form. In this way, on the one hand, the computational complexity is reduced by using the base extension model, and on the other hand, the estimation accuracy is further improved by using deep learning. Furthermore, the method avoids using ideal channel state information as a label, thus enhancing its practicality. Attached Figure Description

[0010] Figure 1 This is a schematic diagram illustrating an application scenario of an embodiment of this application; Figure 2 This is a flowchart illustrating a cascaded channel estimation method provided in an embodiment of this application. Figure 1 ; Figure 3 This is a schematic diagram of the network structure of the first model provided in the embodiments of this application.

[0011] Figure 4 This is a flowchart illustrating a cascaded channel estimation method provided in an embodiment of this application. Figure 2 ; Figure 5 This is a simulation diagram of the cascaded channel estimation provided in the embodiments of this application; Figure 6 This is a performance comparison diagram of cascaded channel estimation provided in the embodiments of this application; Figure 7 This is a schematic diagram comparing the complexity of cascaded channel estimation provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of the cascaded channel estimation device provided in the embodiments of this application; Figure 9 This is a schematic structural diagram of a communication device provided in an embodiment of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0014] In the following description, some embodiments are referred to, which describe a subset of all possible embodiments. However, it is understood that some embodiments may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0016] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0017] 1) Intelligent Reflector (RIS): A passive device composed of a large number of tunable phase units. By controlling the reflection coefficients of these units, the propagation path of electromagnetic waves is changed, thereby enhancing signal coverage and communication quality. RIS can improve channel conditions without increasing additional transmit power, and is especially suitable for high-bandwidth communication scenarios such as millimeter waves.

[0018] 2) Cascaded Channel: In a RIS-assisted wireless communication system, the signal is sent from the terminal to the RIS and then reflected by the RIS back to the base station. Therefore, the entire link includes two main channel segments: one is the static quasi-state channel between the base station and the RIS, and the other is the time-varying channel between the RIS and the terminal device. The overall channel formed by the combination of these two channels is called the cascaded channel.

[0019] 3) Static quasi-state channel: refers to a channel state that changes slowly or even remains unchanged over a relatively long time scale. In this application, the channel between the base station and the smart reflector is regarded as a static quasi-state channel, which changes slowly and is easy to model and estimate.

[0020] 4) Time-varying channel: refers to a channel state that changes rapidly over a short period of time, usually caused by user movement or environmental changes. In this application, the channel between the terminal device and the smart reflector is a time-varying channel, and its time-varying characteristics need to be handled using a specialized modeling method.

[0021] 5) Basis Expansion Model (BEM): A method that represents time-varying channel parameters as a linear combination of a set of basis functions. By selecting appropriate basis functions, high-dimensional time-varying channel information can be compressed into a low-dimensional basis coefficient matrix, thereby reducing the number of parameters to be estimated and reducing computational complexity.

[0022] 6) Basis Coefficient Estimation Matrix: Based on the basis extension model, this matrix contains estimated coefficient values ​​corresponding to a set of basis functions obtained after processing the received signal. It encapsulates the main dynamic characteristics of the channel and serves as crucial input data for subsequent deep learning model training.

[0023] 7) Residual Super-Resolution Convolutional Neural Network: A convolutional neural network with a residual structure, primarily used to improve the resolution or accuracy of input signals. In this application, this network is used to optimize the basis coefficient estimation matrix, thereby improving the accuracy and robustness of the final channel estimation.

[0024] 8) Minimum Mean Square Error Estimation (LMMSE): A commonly used statistical estimation method that obtains the optimal estimation result by minimizing the mean square error between the estimated value and the true value. In this application, LMMSE estimation is used to generate label data in the training samples, avoiding the use of ideal channel state information and improving the practicality of the system.

[0025] 9) Pilot signal: In an Orthogonal Frequency Division Multiplexing (OFMD) system, a known reference signal used for channel estimation. The pilot signal is embedded in the transmitted signal, and the receiver uses its known information to estimate the current channel state.

[0026] 10) OFDM symbol: A basic time unit in orthogonal frequency division multiplexing (OFDM) technology, containing modulated signals of multiple subcarriers. Each OFDM symbol corresponds to the transmission content within a time period.

[0027] In related technologies, with the rapid development of wireless communication technology, spectrum resources are becoming increasingly scarce. Traditional wireless communication systems mainly rely on increasing the number and power of base stations to improve coverage and communication quality, but this brings high costs and energy consumption problems. Intelligent Reflectors (RIS) can optimize signal propagation paths and improve spectrum utilization efficiency without increasing additional spectrum resources by intelligently adjusting the amplitude and phase of incident electromagnetic waves, thereby effectively alleviating the problem of spectrum resource scarcity. However, this advantage and capability of RIS depends on accurate channel state information, and the passive nature of RIS and the high dimensionality of the reflector unit make obtaining accurate channel state information very challenging.

[0028] Existing methods for acquiring channel state information are categorized into time-varying and non-time-varying channel estimation based on end-user mobility. Non-time-varying channel estimation methods generally exhibit good performance because they do not need to consider the time-varying channel dynamics and Doppler shift caused by user mobility. However, in RIS-assisted wireless communication scenarios, the number of reflecting elements increases exponentially with the increase in the dimensionality of the RIS reflector array, leading to significant overhead. Therefore, the computational cost of non-time-varying channel estimation methods needs to be carefully considered. However, ignoring user mobility is impractical in real-world scenarios, as the time-varying and Doppler shifts caused by user mobility severely impact channel estimation performance. Therefore, research into time-varying channel estimation methods is necessary.

[0029] To address the aforementioned issues, this application proposes a cascaded channel estimation method, apparatus, communication device, and computer storage medium, aiming to resolve the problems of low channel estimation accuracy, excessive computational complexity, and insufficient practicality in existing technologies. By constructing a cascaded channel model of a time-varying channel and combining a radix extension model with deep learning techniques, efficient and accurate estimation of the cascaded channel is achieved, demonstrating good practicality and promising prospects for wider application.

[0030] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of this application.

[0031] like Figure 1 As shown, the communication system 100 may include a terminal device 110, a network device 120, and a smart reflective surface 130. The network device 120 communicates with the terminal device 110 through the smart reflective surface 130.

[0032] It should be understood that the embodiments of this application are only illustrated by way of example with communication system 100, but the embodiments of this application are not limited thereto. That is to say, the technical solutions of the embodiments of this application can be applied to various communication systems, such as: Long Term Evolution (LTE) system, LTE Time Division Duplex (TDD) system, Universal Mobile Telecommunication System (UMTS), Internet of Things (IoT) system, Narrow Band Internet of Things (NB-IoT) system, enhanced Machine-Type Communications (eMTC) system, 5G communication system (also known as New Radio (NR) communication system), or future communication systems, etc.

[0033] exist Figure 1 In the communication system 100 shown, network device 120 can be an access network device that communicates with terminal device 110. The access network device can provide communication coverage for a specific geographical area and can communicate with terminal device 110 (e.g., UE) located within that coverage area.

[0034] Network device 120 may be an evolved Node B (eNB or eNodeB) in a Long Term Evolution (LTE) system, or a Next Generation Radio Access Network (NG RAN) device, or a base station (gNB) in an NR system, or a radio controller in a Cloud Radio Access Network (CRAN), or the network device 120 may be a relay station, access point, vehicle-mounted equipment, wearable device, hub, switch, bridge, router, or network equipment in a future evolved Public Land Mobile Network (PLMN), etc.

[0035] Terminal device 110 can be any terminal device, including but not limited to terminal devices that are connected to network device 120 or other terminal devices via wired or wireless connections.

[0036] For example, the terminal device 110 can refer to an access terminal, user equipment (UE), user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device. The access terminal can be a cellular phone, cordless phone, Session Initiation Protocol (SIP) phone, IoT device, satellite handheld terminal, Wireless Local Loop (WLL) station, Personal Digital Assistant (PDA), handheld device with wireless communication capabilities, computing device or other processing device connected to a wireless modem, vehicle-mounted device, wearable device, terminal device in a 5G network, or terminal device in a future evolved network, etc.

[0037] Terminal device 110 can be used for device-to-device (D2D) communication.

[0038] It should be noted that, Figure 1 This application is merely an example illustrating the system to which this application applies. Of course, the methods shown in the embodiments of this application can also be applied to other systems. Furthermore, the terms "system" and "network" are often used interchangeably herein.

[0039] To facilitate understanding of the technical solutions of the embodiments of this application, the technical solutions of this application are described in detail below through specific embodiments. The above-mentioned related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. The embodiments of this application include at least some of the following contents.

[0040] See Figure 2 , Figure 2 This is a flowchart illustrating the cascaded channel estimation method provided in the embodiments of this application, which will be described in detail below.

[0041] Step 201: Obtain the first signal through the intelligent reflective surface.

[0042] In this embodiment of the application, the network device obtains the first signal through a smart reflective surface, and the network device may refer to a base station.

[0043] In this embodiment, the intelligent reflector is a passive device composed of a large number of adjustable phase units. The intelligent reflector can dynamically adjust the amplitude and phase of the incident electromagnetic wave, thereby optimizing the wireless channel. These reflector units can be controlled independently, allowing the intelligent reflector to adaptively change the signal propagation path according to changes in the communication environment. The reflector units are the basic components of the intelligent reflector, and each reflector unit has the ability to adjust signal reflection characteristics. The number of reflector units determines the spatial resolution and channel control capability of the intelligent reflector; generally, more reflector units mean stronger signal enhancement and higher spatial freedom.

[0044] In practical applications, smart reflectors are widely deployed between base stations and users to improve signal coverage, increase transmission rates, and reduce the impact of multipath interference. For example, in urban environments where tall buildings cause severe signal obstruction, smart reflectors can enhance signal strength and improve communication quality by adjusting the reflection path.

[0045] In this embodiment, the first signal refers to the signal received in the OFDM system, which typically contains information about multiple subcarriers and is used in the subsequent channel estimation process. The first signal can be transmitted to the network device through the reflection path of the smart reflector and is collected at the receiving end. For example, in a 5G communication system, when a user terminal is in a high-speed moving state, the signal sent by the terminal reaches the base station after being reflected by the smart reflector. At this time, the first signal received by the base station is the signal on the reflection path. The first signal may contain multipath components, so further modeling and processing are required.

[0046] In practical implementation, smart reflectors, acting as an intermediate medium, can flexibly adjust the signal reflection path, thereby improving communication link quality and increasing the accuracy of channel estimation. Furthermore, since smart reflectors themselves do not have transmission capabilities, their deployment cost is relatively low, making them suitable for large-scale deployment in complex environments.

[0047] In some embodiments, the smart reflective surface includes a plurality of reflective elements; acquiring a first signal through the smart reflective surface includes: When the m-th reflective unit of the smart reflector is turned on, the first signal is acquired on the m-th OFDM symbol; wherein, the first signal is the m-th frequency domain received signal, and m is greater than or equal to 1.

[0048] Here, "conduction" refers to the state in which a reflecting unit is operational, i.e., activated to reflect signals. When a reflecting unit is in conduction, it can adjust the signal in a specific direction or frequency, thereby affecting the signal propagation path and reception performance. The conduction state is usually dynamically adjusted by the control system based on channel feedback information to achieve optimal signal enhancement. OFDM is a digital modulation technique that divides the data stream into multiple parallel subcarriers for transmission, with each subcarrier corresponding to an OFDM symbol. The OFDM symbol serves as the basic time unit during transmission, carrying data and reference signals required for channel estimation.

[0049] By controlling the conduction state of the m-th reflection unit and acquiring the frequency domain received signal on the corresponding m-th OFDM symbol, the channel information affected by the m-th reflection unit can be effectively extracted. Combining the sequential conduction of reflection units with the timing characteristics of OFDM symbols helps achieve more accurate channel modeling and estimation. In specific implementation, assuming the system is configured with M reflection units, each reflection unit can be turned on sequentially within M OFDM symbol periods, and the corresponding frequency domain received signal can be recorded respectively. In this method, the system can effectively reduce computational complexity and ensure that the contribution of each reflection unit can be independently evaluated, thus providing high-quality data input for subsequent deep learning model training. In this way, time-varying channel information can be efficiently acquired without increasing additional hardware overhead, while providing structured and low-dimensional training samples for the deep learning network, thereby improving channel estimation accuracy and reducing computational complexity.

[0050] For example, the RIS reflector array is set to ON / OFF mode, and the transmitted signal is... The signal turns on the m-th reflection unit, so the m-th frequency domain received signal can be represented by the following formula (1).

[0051] (1) in, To send a signal for the m-th OFDM, This refers to the concatenated channel corresponding to the m-th RIS unit during the m-th OFDM symbol period. This represents the noise on the m-th OFDM symbol.

[0052] Here, when the m-th reflection unit is activated, the terminal sends a known signal. The signal received by the receiving end It only includes the process from the transmitter → the m-th RIS unit → the receiver. Through this time-division multiplexing method, the M coupled cascaded channels are decoupled, allowing the received data for each channel to be acquired unit by unit and symbol by symbol. This is the foundation for all subsequent processing. Furthermore, It can be expressed by the following formulas (2) and (3).

[0053] (2) (3) in, It is the channel corresponding to the Nth OFDM. The result of the N-point FFT transformation, It is the equivalent frequency domain response after taking the diagonal. Because The off-diagonal elements approach 0 after the FFT, so only the diagonal elements are retained.

[0054] Step 202: Construct a cascaded channel model for the time-varying channel.

[0055] In this embodiment, the cascaded channel includes a static quasi-state channel between the base station and the smart reflector, and a time-varying channel between the terminal device and the smart reflector.

[0056] Here, a static quasi-state channel refers to a channel state that changes slowly or even remains unchanged over a relatively long timescale. This channel state is applicable to the communication path between the base station and the smart reflector. A time-varying channel, on the other hand, refers to a channel state that changes rapidly over a short period of time. This channel state is usually caused by user movement or environmental changes. This channel state is applicable to the communication path between the smart reflector and the terminal device.

[0057] In some embodiments, constructing a cascaded channel model for a time-varying channel includes: Based on the time-varying channel corresponding to the m-th reflecting unit, a cascaded channel model is constructed as an expression containing a direct path component and multiple scattered path components. The direct path component has a complex exponential term with a constant amplitude but a phase that changes linearly with time, and the rate of linear change is determined by the normalized Doppler frequency offset. The scattered path components are represented as the superposition of multiple paths with different time delays and time-varying complex amplitudes.

[0058] Here, we consider an N-subcarrier OFDM system assisted by RIS, with M reflection units deployed on the RIS and a single antenna for transmission and reception. Without considering the channel between the user and the base station, the cascaded channel corresponding to the m-th RIS reflection unit can be represented by formula (4).

[0059] (4) In the formula, The direct trajectory LOS component, For the scattering path component, Let L be the normalized Doppler frequency offset, L be the number of paths in the cascaded channel, and τ be the normalized delay of the l-th path. , This is a cyclic prefix. Here, N represents the number of subcarriers (or sampling points) in the effective data portion of an OFDM symbol. This portion carries the actual information (data or pilot signals). This represents the length of the cyclic prefix. The cyclic prefix is ​​a sequence of signals that is copied from the end of the valid data portion of an OFDM symbol and added to the beginning of that symbol. Represents the total length of a complete OFDM symbol.

[0060] Here, the direct path component refers to the path component of a signal that propagates directly from the transmitter to the receiver in a communication system. The direct path component is typically the shortest and most stable propagation mode. Its amplitude remains constant, but its phase changes linearly with time and user movement. This linear phase change is primarily caused by the Doppler effect resulting from user motion. The mathematical form of the direct path component is usually represented by a complex exponential term, the frequency of which is determined by the normalized Doppler frequency offset. Modeling the direct path component allows for a more accurate reflection of the characteristics of stable paths in the channel.

[0061] Normalized Doppler frequency offset refers to the parameter obtained by normalizing the actual measured Doppler frequency shift according to the maximum possible frequency shift. Normalized Doppler frequency offset describes the impact of user movement on the channel; a smaller value indicates a slower movement speed and a smaller impact on channel changes. In the embodiments of this application, the normalized Doppler frequency offset determines the phase change rate of the direct path component. Through this determination, the cascaded channel model can dynamically adapt to the user's movement status.

[0062] Scattered path components refer to the multiple indirect propagation paths formed when a signal is reflected, refracted, or diffracted by obstacles in the environment (such as buildings, trees, etc.) during propagation. Each scattered path component has a different time delay and time-varying complex amplitude. Therefore, in modeling, the modeling system superimposes all scattered path components to comprehensively reflect the total contribution of all indirect propagation paths. The existence of scattered path components increases the complexity of the channel, but also provides more information dimensions, which helps to improve the accuracy of channel estimation.

[0063] Time-varying complex amplitude refers to the characteristic of the amplitude and phase of a signal in a scattered path component changing over time. Due to the influence of user movement and environmental changes, the signal strength and phase of the scattered path components may fluctuate rapidly, making time-varying complex amplitude a more reasonable way to describe these components. Using time-varying complex amplitude to describe these scattered path components allows the model to better capture the dynamic characteristics of the channel.

[0064] Different path delay superposition refers to summing signals from multiple paths with different propagation time delays to form a complete cascaded channel model response. The time delay of each scattering path component depends on the propagation distance, and superposition is usually achieved by summing the complex signals of each scattering path component. By superimposing these path signals, the system can obtain the overall transmission characteristics of the cascaded channel model at a certain moment, providing basic data for subsequent channel estimation.

[0065] In practical implementation, the direct path component and the scattered path component together constitute the main components of the cascaded channel model. The direct path component represents the characteristics of a stable path, while the scattered path component reflects the dynamic changes of complex propagation paths in the environment. The combination of the direct and scattered path components gives the cascaded channel model both stability and flexibility, enabling it to better adapt to various communication scenarios.

[0066] In practical implementation, the normalized Doppler frequency offset, as a key parameter, directly affects the phase change rate of the direct path component. By appropriately setting the normalized Doppler frequency offset, time-varying channel behavior under different moving speeds can be effectively simulated, thereby improving the model's applicability and predictive ability. There is a close correlation between the time-varying complex amplitude and the superposition of different time-delay paths. Due to the different propagation environment and path length of each scattered path component, the time delay and signal attenuation of these components will also differ. Therefore, the system must accurately reproduce the dynamic changes of each scattered path component through time-varying complex amplitude modeling, and integrate these components into a unified cascaded channel model response through superposition operations.

[0067] In actual implementation, the entire process includes the following steps: First, measurement data of the time-varying channel is acquired based on the m-th reflecting unit; second, the direct path component is extracted and modeled, and the phase change characteristics of the direct path component are determined using the normalized Doppler frequency offset; next, multiple scattered path components are identified and modeled, and the different time delays and time-varying complex amplitudes corresponding to the multiple scattered path components are considered; finally, all scattered path components are superimposed to construct a cascaded channel model.

[0068] Step 203: Based on the cascaded channel model and the base extension model, determine the first frequency domain channel matrix of the cascaded channel.

[0069] In this embodiment, the Basis Extended Model (BEM) is a method that represents time-varying channel parameters as a linear combination of a set of basis functions. By selecting a suitable set of basis functions, high-dimensional time-varying channel information can be compressed into a low-dimensional basis coefficient matrix. The selection of basis functions is crucial to the performance of the model; commonly used basis functions include sine basis, polynomial basis, etc.

[0070] In this embodiment, the first frequency domain channel matrix refers to the channel response matrix represented in the frequency domain, which is typically obtained by performing a Fourier transform on the time domain channel response. For example, in a millimeter-wave communication system, the channel response may contain multiple scattering paths, each corresponding to a different time delay and Doppler frequency offset. The basis extension model maps these scattering paths into the frequency domain, thereby facilitating subsequent processing.

[0071] In practical applications, the base extension model can significantly reduce the number of parameters required for channel estimation. This reduction in the number of parameters can lower computational complexity and improve the real-time performance and practicality of the system.

[0072] In some embodiments, the cascaded channel model is linearly combined using multiple basis functions and a basis extension model to obtain the first frequency domain channel matrix of the cascaded channel model; wherein, the basis extension model is used to reduce the dimensionality of the time-varying channel to a basis coefficient matrix.

[0073] Here, the cascaded channel is modeled using a basis extension model based on basis functions, yielding the cascaded channel function in the time domain. It can be expressed by formula (5).

[0074] (5) Where Q is the number of basis functions. This is the nth element corresponding to the qth basic coefficient. The basis coefficients are the basis functions corresponding to the q-th basis function of the u-th symbol on the l-th path. This represents the modeling error.

[0075] Here, to reduce the complexity of channel estimation, a basis extension model is used to model this time-varying channel. Specifically, the value of each path in the time-domain channel response at each time step is approximated as a linear combination of the values ​​of a pre-selected set of basis functions (such as complex exponential functions, polynomial functions, etc.) at that time step. The weights of this linear combination are called basis coefficients. In this way, a large number of channel parameters that originally needed to be estimated individually at each time step are compressed into a much smaller number of basis coefficients. Ignoring modeling errors, the frequency-domain concatenated channel matrix over the entire OFDM symbol period can be expressed as a weighted sum of these basis coefficient matrices and a fixed matrix composed of basis functions and Fourier transform matrices.

[0076] Basis functions are a set of mathematical functions used to describe channel characteristics. They are typically chosen as orthogonal or non-orthogonal functions that approximate actual channel variations, such as polynomial functions, Fourier functions, and wavelet functions. Basis functions can capture the channel's variation over time and frequency, thus providing a more accurate model of channel state information.

[0077] Basis extension model is a mathematical model that compresses high-dimensional time-varying channel state information into low-dimensional parameters. The basic idea of ​​the basis extension model is to use a finite set of basis functions to approximate the original channel response, representing the original channel as a linear combination of these basis functions. In this way, the first frequency domain channel matrix in the cascaded channel model, which originally needed to be estimated, is transformed into a low-dimensional basis coefficient matrix, thus significantly reducing computational complexity and storage requirements.

[0078] Linear combination processing refers to the process of summing multiple basis functions according to certain weights to reconstruct the original channel response. Linear combination processing obtains a frequency domain expression that approximates the true channel by multiplying each basis function by its corresponding basis coefficient and then summing the results. Linear combination processing makes channel modeling more flexible and efficient while maintaining modeling accuracy.

[0079] Time-varying channel dimensionality reduction refers to mapping the originally time-varying and high-dimensional channel information to a low-dimensional basis coefficient space through a basis expansion model. Time-varying channel dimensionality reduction not only simplifies the channel estimation process but also reduces the input dimensionality of deep learning networks, helping to improve model training efficiency and generalization ability.

[0080] After ignoring the modeling error, the first frequency domain channel matrix after modeling is obtained, which can be represented by the following equation (6).

[0081] (6) In the formula, ; ; .

[0082] Here, `diag()` represents creating a diagonal matrix whose elements on the main diagonal are equal to the vectors in order. The elements of the matrix are 0, while all off-diagonal elements of the matrix are 0.

[0083] Step 204: Based on the first signal and the first frequency domain channel matrix, determine the basis coefficient estimation matrix of the cascaded channel.

[0084] In this embodiment, the basis coefficient estimation matrix is ​​a set of estimated coefficient values ​​corresponding to basis functions obtained by processing the received signal based on the basis extension model. The basis coefficient estimation matrix contains the main dynamic characteristics of the channel and is important input data for subsequent deep learning model training. Specifically, the basis coefficient estimation matrix can be solved using the known pilot signal and the received signal through the least squares method or other estimation methods. For example, in an OFDM system, the pilot signal is embedded in the transmitted signal. The receiver uses the known information of the pilot signal to estimate the current channel state and combines it with the first frequency domain channel matrix to estimate the basis coefficients.

[0085] In some embodiments, obtaining the basis coefficient estimation matrix of the cascaded channel based on the first signal and the first frequency domain channel matrix includes: obtaining the first basis coefficient vector according to the basis extension model and the first signal; obtaining the pilot signal of the first signal; and obtaining the basis coefficient estimation matrix of the cascaded channel based on the first basis coefficient vector and the pilot signal using the least squares method.

[0086] Here, the basis extension model is a mathematical modeling method that maps high-dimensional time-varying channel state information to a low-dimensional parameter space. By selecting a suitable set of basis functions to represent the changing characteristics of the channel, the basis extension model simplifies the originally complex time-varying channel modeling into a problem of estimating basis coefficients. The selection of basis functions should reflect the main changing characteristics of the channel, such as Doppler shift and delay spread. The advantage of the basis extension model is that it can significantly reduce the number of parameters to be estimated, thereby reducing computational complexity and improving estimation efficiency.

[0087] The first signal is modeled using the base extension model, and the expression of the received signal after modeling using the base extension model can be represented by the following equation (7).

[0088] (7) In the formula, ; ; ; .

[0089] here, This is represented as a basis extension model. The received pilot signal is obtained from the received signal. The pilot signal is a known reference signal used in a communication system to assist in channel estimation, typically inserted periodically into the data stream by the terminal device. The base station uses these known pilot signals to compare with the received signal to extract channel state information. In this application, the pilot signal serves as an important input for basis coefficient estimation, used to construct the observation data required for the least squares method. The process of acquiring the pilot signal includes frequency offset compensation of the received signal and extraction of pilot symbols at preset positions. Due to the known nature and stability of the pilot signal, its role in channel estimation is similar to that of an anchor point, providing a reliable benchmark for subsequent algorithms.

[0090] The basic coefficient estimation matrix of the m-th OFDM symbol obtained by the least squares method can be represented by formula (8).

[0091] (8) In the formula, This refers to the pilot signal on the m-th OFDM signal after ideal frequency offset compensation at the receiver. The basic coefficient matrix The submatrix at the upper pilot position. The basis coefficient estimation matrix is ​​a low-dimensional parameter matrix obtained by processing the received signal and using the basis extension model method. This matrix is ​​used to represent the state information of the time-varying channel. The basis coefficient estimation matrix extracts pilot signals under multiple OFDM symbols and calculates the basis coefficients corresponding to each basis function using the least squares method, thereby compressing and modeling the high-dimensional channel state information. The basis coefficient estimation matrix can effectively reduce the number of parameters to be estimated, reduce computational complexity, and retain the main characteristic information of the channel.

[0092] In this application, the basis coefficient estimation matrix can be solved by expressing the relationship between the basis extension model and the pilot signal as a system of linear equations, and then obtaining the optimal basis coefficient estimate by minimizing the sum of squared errors. For example, the basis coefficient estimation is solved using the least squares method. The basis coefficient estimation matrix obtained by the least squares method can accurately reflect the changing trend of the actual channel, providing high-quality training samples for subsequent deep learning network training.

[0093] Step 205: Optimize the basis coefficient estimation matrix using the first model to obtain the optimized cascaded channel characteristics.

[0094] In some embodiments, the first model is a trained residual super-resolution convolutional neural network.

[0095] Here, the first model includes a residual super-resolution convolutional neural network, see [link to relevant documentation]. Figure 3 , Figure 3 This is a schematic diagram of the structure of a first model provided in an embodiment of this application. It includes multiple convolutional layers and residual connections. Each layer may include a Conv+ReLU activation function. The first model can effectively extract features from the input signal and perform nonlinear modeling. The residual super-resolution convolutional neural network (RCN) is a deep learning model. Its structural design draws inspiration from residual networks. The RCN alleviates the vanishing gradient problem by introducing skip connections, accelerating the training process and improving model performance. The RCN consists of multiple convolutional layers, activation function layers, and pooling layers. It can automatically learn multi-scale features of the channel, thereby achieving high-precision channel estimation. For example, the first model is used to optimize the basis coefficient estimation matrix, thereby improving the accuracy and robustness of the final channel estimation.

[0096] In some embodiments, a training sample set is constructed, which includes a first sample and a label corresponding to the first sample; wherein, the first sample is a vector composed of the real and imaginary parts of the basis coefficient estimation matrix, and the label corresponding to the first sample is obtained by performing linear minimum mean square error estimation on the cascaded channel; based on the training sample set, a residual super-resolution convolutional neural network is trained with the goal of minimizing the mean square error between the neural network output and the label.

[0097] Here, the residual super-resolution convolutional neural network, through its residual learning structure, can effectively learn and compensate for errors in the initial estimation, outputting optimized results that are closer to the real channel. Network training requires constructing a sample set. Each training sample consists of a first sample and its corresponding label, for example, represented as an "input-label" pair. The first sample, or input sample, is the basis coefficient estimation matrix obtained using the least squares method. This matrix is ​​split into real and imaginary parts and then concatenated to form a two-channel input data. The label corresponding to the first sample is the training label, obtained by estimating the channel using a linear minimum mean square error (LMMSE) estimator, and the resulting channel estimation matrix is ​​used as the label. Similarly, this label matrix is ​​split into real and imaginary parts and then concatenated.

[0098] It is important to note that this application uses LMMSE, a practically achievable estimation result with superior performance to least squares, as the learning target, rather than the theoretically perfect but unobtainable ideal channel state information. This enhances its practicality. Linear Least Mean Square Error (LMMSE) estimation is a channel estimation method based on statistical optimality criteria. Given known noise and channel statistical characteristics, LMMSE provides a channel estimation result with low mean square error. Compared to the traditional least squares method, LMMSE considers prior channel knowledge and the impact of noise, providing more accurate and stable estimation results in real-world communication environments. The labels generated by LMMSE are closer to the true channel state, avoiding the impracticality and infeasibility issues associated with using ideal CSI.

[0099] By using a large number of training samples constructed above, the parameters of the residual super-resolution convolutional neural network are trained with the optimization objective of minimizing the mean squared error between the network output and the LMMSE label. After training, the optimal network model parameters are saved.

[0100] Here, a residual super-resolution convolutional neural network is constructed, and the basis coefficients are estimated to build training samples for the neural network, as shown in formula (9).

[0101] (9) In the formula, J is the number of training samples. For the j-th training sample, Let be the training label of the j-th training sample, which can be further represented by the following formula (10).

[0102] ; (10) In the formula, This is an operation to extract the real part of a complex number. For the operation of taking the imaginary part of a complex number, where, ; .

[0103] Here, the real and imaginary parts of the basis coefficient estimation matrix represent the two dimensions of complex channel state information. Concatenating the real and imaginary parts of the basis coefficient estimation matrix into a vector and using it as training sample input allows for a more comprehensive description of channel characteristics. This method of concatenating the real and imaginary parts of the basis coefficient estimation matrix into a vector as training sample input enables the residual super-resolution convolutional neural network to better learn the nonlinear variation patterns of the channel, thereby improving the accuracy of channel estimation.

[0104] During training, the loss function of the residual super-resolution convolutional neural network (RSN) adopts the mean squared error (MSE) form, which measures the difference between the RSN output and the training labels. The parameters of the RSN are continuously adjusted through backpropagation, allowing the output to gradually approach the target value, ultimately achieving optimal channel estimation. Since linear minimum mean squared error estimation as a label has high reliability, the RSN can achieve fast convergence with limited training data and obtain stable and accurate estimation results. Finally, a first model with optimal weight and threshold parameters is obtained through offline training. .

[0105] Through the steps described above, this embodiment implements an efficient channel estimation method based on deep learning. By reducing the parameter dimensionality through a base extension model, obtaining high-quality labels by combining linear minimum mean square error estimation, and then training with a residual super-resolution convolutional neural network, the accuracy of channel estimation is not only improved, but the computational complexity is also significantly reduced, thus enhancing the practicality of the method.

[0106] In some embodiments, online testing is performed based on the trained first model to obtain the basis coefficient estimation matrix under all OFDM symbols and feed it into the trained first model to obtain the neural network training output value and obtain the optimized cascaded channel features. In actual communication, the following steps are performed during online channel estimation: Following the same method as in the training phase, obtain the basis coefficient estimation matrices corresponding to all OFDM symbols in the current communication environment. Process these basis coefficient estimation matrices in the same format as the training samples (split into real and imaginary parts and concatenate them), and use them as input to the pre-trained residual super-resolution convolutional neural network model. The neural network model processes the input and outputs optimized channel feature data (also in the form of real and imaginary parts for two channels).

[0107] Step 206: Obtain the estimated value of the cascaded channel based on the optimized cascaded channel characteristics.

[0108] After optimizing the basis coefficient estimation matrix, the final cascaded channel estimate can be obtained based on the optimized cascaded channel characteristics. This involves recombining the two channels of data output from the network into a complex channel matrix according to their corresponding positions; this is the final high-precision channel estimate. The cascaded channel estimate reflects the channel state of the entire communication link and can be used for subsequent signal detection, modulation, and demodulation operations. For example, in a mobile communication system, terminal devices can adjust their receiving strategies based on the cascaded channel estimate to improve communication quality. Furthermore, the cascaded channel estimate can be fed back to the base station to help optimize resource allocation and scheduling strategies. In practical implementation, combining the basis extension model with deep learning techniques can achieve efficient and accurate channel estimation, thereby improving the overall performance of the communication system and the user experience.

[0109] The cascaded channel estimation method provided in this application achieves efficient and accurate channel estimation by constructing a cascaded channel model that includes a static quasi-state channel and a time-varying channel, and combining a base extension model with deep learning techniques. This method reduces computational complexity by decreasing the number of parameters to be estimated; improves practicality by using non-ideal channel state information as training labels; and enhances the accuracy and robustness of channel estimation by introducing a first model.

[0110] In some embodiments, obtaining an estimate of the cascaded channel based on the optimized cascaded channel characteristics includes: reorganizing the cascaded channel characteristics based on the correspondence between the real and imaginary parts to obtain an estimate of the cascaded channel, wherein the estimate is a matrix in complex form.

[0111] Here, the two channels of data output from the network are recombined into a complex channel matrix according to their corresponding positions, which is the final high-precision channel estimate. The resulting complex matrix is ​​the estimate of the cascaded channel. This estimate not only includes the amplitude information of the cascaded channel but also retains its phase information, thus reflecting the true state of the cascaded channel more comprehensively. By recombining the cascaded channel features based on the correspondence between their real and imaginary parts, the accuracy of the cascaded channel estimation can be improved without increasing computational complexity. Compared to traditional real-number representation methods, the complex matrix can better capture the dynamic changes of the cascaded channel. Therefore, recombining the cascaded channel features based on the correspondence between their real and imaginary parts helps to adapt to the time-varying effects caused by user movement. Furthermore, this method can effectively reduce the error between the training labels and the actual cascaded channel, further improving the generalization ability of the deep learning model.

[0112] See Figure 4 , Figure 4 This is a flowchart illustrating a cascaded channel estimation method provided in an embodiment of this application. Figure 2 This paper proposes a time-varying channel estimation method based on deep learning with intelligent reflector assistance. The overall process of the method is as follows: Step 401, build a time-varying channel model. Step 402, model using a basis extension model. Step 403, estimate the basis coefficients. Step 404, build a residual super-resolution convolutional neural network and construct training samples for offline training of the residual super-resolution network. Step 405, train the network offline. Step 406, test online. Details are explained below.

[0113] Step 1: Consider a RIS-assisted OFDM system with N subcarriers. The RIS has M reflector elements and a single transmit / receive antenna. The channel between the user and the base station is not considered. The model is as follows: Figure 1 As shown. Therefore, the cascaded channel corresponding to the m-th RIS reflection unit is shown in formula (4).

[0114] Step 2: Set the RIS reflector array to ON / OFF mode and send the Xm signal to turn on the m-th reflector unit. Therefore, the received m-th frequency domain signal is given by formula (1). Further representation of the cascaded channel is given by formulas (2) and (3). Step 3: Model the cascaded channel using the basis extension model based on the basis functions, and express the model as Equation (5).

[0115] Step 4: The system ignores the modeling error and obtains the modeled frequency domain channel matrix as shown in formula (6).

[0116] Step 5: The system obtains the expression of the received signal after modeling using the base extension model as formula (7).

[0117] Step 6: The communication equipment extracts the pilot signal based on the received wireless signal and calculates the estimated channel basis coefficients of the m-th OFDM symbol using the least squares method. See formula (8) for details.

[0118] Step 7: Construct a residual super-resolution convolutional neural network, the network structure is as follows: Figure 3 As shown, the training samples for the neural network are constructed using the basis coefficients estimation, see formulas (9) and (10). This system uses linear minimum mean square error (LMMSE) estimation to obtain the samples.

[0119] Step 8: Randomly initialize the parameters in the neural network, and use the constructed training sample set to train the parameter set using the mean squared error loss function.

[0120] Step 9: Obtain the optimal network model with the best weight and threshold parameters.

[0121] Step 10: The system performs online testing based on the network model obtained in step 9, extracts the basis coefficient estimates from all OFDM symbols, and inputs these estimates into the network model to obtain the training output values ​​of the neural network.

[0122] Step 11: Based on the network output values ​​obtained in step 10, the system transforms these output values ​​to the complex domain to obtain the final neural network channel estimate.

[0123] See Figure 5 , Figure 5 This is a simulation diagram of the cascaded channel estimation provided in the embodiments of this application. The performance of the cascaded estimation was verified through simulation. Among the important parameters in the simulation, the subcarrier length N=128, the number of basis functions Q=4, a 5-path Ricean channel is considered, the normalized Doppler frequency offset is 0.044, the Adam optimizer is used for the neural network, and the mean square error function is used as the loss function. Figure 4 The MSE performance is shown for different training sample numbers V, from Figure 5As can be seen, the number of training samples affects the accuracy of this application. The more training samples, the higher the estimation performance of the cascaded channel of this application. The performance tends to plateau when the number of samples is around 1000. This is because the residual network used in this application converges faster and can extract sample features more effectively. As the number of training samples increases, the estimation performance of this application also increases. However, this is accompanied by an increase in the training time and computational complexity of the neural network used. Therefore, it is necessary to reasonably balance the number of training samples.

[0124] See Figure 6 , Figure 6 This is a schematic diagram showing the performance comparison of cascaded channel estimation provided in the embodiments of this application. Figure 6 Performance curves of embodiments of this application and other prior art are given, using 2000 training samples. Figure 6 It can be seen that the network proposed in this application is affected by the training labels. When the training labels are least squares (LS) estimation, the performance is poor, especially under high signal-to-noise ratio (SNR) conditions. In contrast, the network proposed in this application, whose training labels are estimates obtained using linear minimum mean square error (LMSE), performs better. However, when the training labels are ideal CSI, although the performance of methods corresponding to ideal CSI is better than that proposed in this application, ideal CSI is unavailable in practice, significantly reducing the practicality of the cascaded channel estimation method. The LS channel estimation method performs the worst compared to other methods such as the DNN method and the cascaded channel estimation method proposed in this application because it is affected by time-varying channels, severely impacting channel estimation performance. Furthermore, both the DNN method and the cascaded channel estimation method proposed in this application utilize deep learning for channel estimation; however, the cascaded channel estimation method proposed in this application has a faster convergence speed, better performance with the same training samples, and is more practical.

[0125] See Figure 7 , Figure 7 This is a schematic diagram comparing the complexity of cascaded channel estimation provided in the embodiments of this application. Figure 7 The computational complexity curves for different methods are presented. Since this application utilizes a base-extended model to reduce the number of parameters to be estimated, the computational complexity of this application is also reduced. Figure 7 As can be seen, as the number of reflection units increases, the computational complexity of this application increases less than that of other methods.

[0126] As described above, this application proposes a cascaded channel estimation method. It transforms the high-dimensional time-varying channel into a low-dimensional basis coefficient estimate using a basis extension model, thereby reducing the number of parameters to be estimated. Efficient training and prediction are performed using a residual super-resolution convolutional neural network. Simultaneously, LMMSE estimation is used as the training label to avoid reliance on ideal channel state information, thus improving the method's practicality. Compared with existing technologies, the application of the basis extension model significantly reduces the channel estimation complexity. Secondly, the design of the residual super-resolution convolutional neural network improves the estimation accuracy. Finally, the selection method of training labels enhances the method's practicality. In summary, this application proposes a time-varying channel estimation method suitable for RIS-assisted mobile communication systems. This method can overcome the shortcomings of traditional methods in terms of time-varying nature, complexity, and practicality. The proposed time-varying channel estimation method has good application prospects and promotional value.

[0127] Figure 8 This is a schematic diagram of the structure of the cascaded channel estimation device provided in the embodiments of this application. Figure 1 ,like Figure 8 As shown, the cascaded channel includes a static quasi-state channel between the base station and the smart reflector, and a time-varying channel between the terminal device and the smart reflector; the cascaded channel estimation device 800 includes: Acquisition unit 801 is used to acquire the first signal through the smart reflective surface.

[0128] Processing unit 802 is used to construct a cascaded channel model of the time-varying channel.

[0129] The determining unit 803 is used to determine the first frequency domain channel matrix of the cascaded channel model based on the first signal and the base extension model; determine the basis coefficient estimation matrix of the cascaded channel based on the first signal and the first frequency domain channel matrix; and optimize the basis coefficient estimation matrix through the first model to obtain the optimized cascaded channel characteristics.

[0130] The acquisition unit 801 is further configured to acquire an estimated value of the cascaded channel based on the optimized cascaded channel characteristics.

[0131] In some embodiments, the smart reflective surface includes a plurality of reflective units; the acquisition unit 801 is further configured to acquire the first signal on the m-th OFDM symbol when the m-th reflective unit of the smart reflective surface is turned on; wherein the first signal is the m-th frequency domain received signal, and m is greater than or equal to 1.

[0132] In some embodiments, the processing unit 802 is further configured to construct the cascaded channel model as an expression containing a direct path component and multiple scattered path components based on the time-varying channel corresponding to the m-th reflecting unit; wherein, the direct path component has a complex exponential term with a constant amplitude but a phase that changes linearly with time, and the rate of linear change is determined by the normalized Doppler frequency offset; the scattered path component is represented as the superposition of multiple paths with different time delays and time-varying complex amplitudes.

[0133] In some embodiments, the determining unit 803 is further configured to perform linear combination processing on the cascaded channel model using multiple basis functions and the basis extension model to obtain a first frequency domain channel matrix of the cascaded channel model; wherein the basis extension model is used to reduce the dimensionality of the time-varying channel to a basis coefficient matrix.

[0134] In some embodiments, the determining unit 803 is further configured to obtain a first basis coefficient vector based on the basis extension model and the first signal; In some embodiments, the acquisition unit 801 is further configured to acquire the pilot signal of the first signal; In some embodiments, the determining unit 803 is further configured to obtain the basis coefficient estimation matrix of the cascaded channel based on the first basis coefficient vector and the pilot signal using the least squares method.

[0135] In some embodiments, the first model is a trained residual super-resolution convolutional neural network; the processing unit 802 is further configured to construct a training sample set, the training sample set including a first sample and a label corresponding to the first sample; wherein, the first sample is a vector composed of the real and imaginary parts of the basis coefficient estimation matrix, and the label corresponding to the first sample is obtained after performing linear minimum mean square error estimation on the cascaded channel; based on the training sample set, the residual super-resolution convolutional neural network is trained with the goal of minimizing the mean square error between the neural network output and the label.

[0136] In some embodiments, the acquisition unit 801 is further configured to reorganize the cascaded channel features based on the correspondence between the real and imaginary parts to obtain an estimated value of the cascaded channel, wherein the estimated value is a matrix in complex form.

[0137] Those skilled in the art should understand that Figure 8 The implementation functions of each unit in the cascaded channel estimation device shown can be understood by referring to the relevant descriptions of the aforementioned method. Figure 8 The functions of each unit in the cascaded channel estimation device shown can be implemented by a program running on a processor or by specific logic circuits.

[0138] Figure 9 This is a schematic structural diagram of a communication device 900 provided in an embodiment of this application. The communication device can be a terminal device or a network device. Figure 9 The communication device 900 shown includes a processor 910, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0139] Optionally, such as Figure 9 As shown, the communication device 900 may further include a memory 920. The processor 910 can retrieve and run computer programs from the memory 920 to implement the methods described in this embodiment.

[0140] The memory 920 can be a separate device independent of the processor 910, or it can be integrated into the processor 910.

[0141] Optionally, such as Figure 9 As shown, the communication device 900 may also include a transceiver 930, which the processor 910 can control to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0142] The transceiver 930 may include a transmitter and a receiver. The transceiver 930 may further include antennas, and the number of antennas may be one or more.

[0143] Optionally, the communication device 900 may specifically be a network device in the embodiments of this application, and the communication device 900 may implement the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0144] Optionally, the communication device 900 may specifically be a mobile terminal / terminal device in the embodiments of this application, and the communication device 900 may implement the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0145] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0146] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0147] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0148] This application also provides a computer-readable storage medium for storing computer programs.

[0149] Optionally, the computer-readable storage medium can be applied to the network device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0150] Optionally, the computer-readable storage medium can be applied to the mobile terminal / terminal device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0151] This application also provides a computer program product, including computer program instructions.

[0152] Optionally, the computer program product can be applied to the network device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0153] Optionally, the computer program product can be applied to the mobile terminal / terminal device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0154] This application also provides a computer program.

[0155] Optionally, the computer program can be applied to the network device in the embodiments of this application. When the computer program is run on the computer, it causes the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0156] Optionally, the computer program can be applied to the mobile terminal / terminal device in the embodiments of this application. When the computer program is run on a computer, it causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0157] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0160] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0162] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A cascaded channel estimation method, characterized in that, The cascaded channel includes a static quasi-state channel between the base station and the smart reflector, and a time-varying channel between the terminal device and the smart reflector; the method includes: The first signal is obtained through a smart reflective surface; Construct a cascaded channel model for the time-varying channel; Based on the cascaded channel model and the base extension model, the first frequency domain channel matrix of the cascaded channel is determined; Based on the first signal and the first frequency domain channel matrix, determine the basis coefficient estimation matrix of the cascaded channel; The basis coefficient estimation matrix is ​​optimized using the first model to obtain the optimized cascaded channel characteristics; Based on the optimized cascaded channel characteristics, the estimated value of the cascaded channel is obtained.

2. The method according to claim 1, characterized in that, The intelligent reflective surface includes multiple reflective units; the acquisition of the first signal through the intelligent reflective surface includes: When the m-th reflective unit of the intelligent reflective surface is turned on, the first signal is acquired on the m-th OFDM symbol; wherein, the first signal is the m-th frequency domain received signal, and m is greater than or equal to 1.

3. The method according to claim 2, characterized in that... The construction of the cascaded channel model for the time-varying channel includes: Based on the time-varying channel corresponding to the m-th reflecting unit, the cascaded channel model is constructed as an expression containing a direct path component and multiple scattered path components; wherein, the direct path component has a complex exponential term with a constant amplitude but a phase that changes linearly with time, and the rate of linear change is determined by the normalized Doppler frequency offset; the scattered path component is represented as the superposition of multiple paths with different time delays and time-varying complex amplitudes.

4. The method according to claim 2, characterized in that, The determination of the first frequency domain channel matrix of the cascaded channel based on the cascaded channel model and the base extension model includes: The cascaded channel model is linearly combined using multiple basis functions and the basis extension model to obtain the first frequency domain channel matrix of the cascaded channel model; wherein, the basis extension model is used to reduce the dimensionality of the time-varying channel to a basis coefficient matrix.

5. The method according to claim 1, characterized in that, The step of obtaining the basis coefficient estimation matrix of the cascaded channel based on the first signal and the first frequency domain channel matrix includes: Based on the basis extension model and the first signal, the first basis coefficient vector is obtained; Obtain the pilot signal of the first signal; The basis coefficient estimation matrix of the cascaded channel is obtained by using the least squares method based on the first basis coefficient vector and the pilot signal.

6. The method according to claim 5, characterized in that, The first model is a trained residual super-resolution convolutional neural network; the method further includes: A training sample set is constructed, which includes a first sample and a label corresponding to the first sample; wherein, the first sample is a vector composed of the real and imaginary parts of the basis coefficient estimation matrix, and the label corresponding to the first sample is obtained by performing linear minimum mean square error estimation on the cascaded channel; Based on the training sample set, the residual super-resolution convolutional neural network is trained with the goal of minimizing the mean square error between the neural network output and the label.

7. The method according to any one of claims 1 to 6, characterized in that, The step of obtaining an estimate of the cascaded channel based on the optimized cascaded channel characteristics includes: The cascaded channel features are reorganized based on the correspondence between their real and imaginary parts to obtain an estimated value of the cascaded channel, which is a matrix in complex form.

8. A cascaded channel estimation device, characterized in that, The cascaded channel includes a static quasi-state channel between the base station and the smart reflector, and a time-varying channel between the terminal device and the smart reflector; the device includes: An acquisition unit is used to acquire a first signal through a smart reflective surface; A processing unit is used to construct a cascaded channel model of the time-varying channel; The determining unit is configured to: determine the first frequency domain channel matrix of the cascaded channel model based on the first signal and the basis extension model; determine the basis coefficient estimation matrix of the cascaded channel based on the first signal and the first frequency domain channel matrix; and optimize the basis coefficient estimation matrix using the first model to obtain the optimized cascaded channel characteristics. The acquisition unit is further configured to acquire an estimated value of the cascaded channel based on the optimized cascaded channel characteristics.

9. A communication device, characterized in that, include: The processor and the memory used to store computer programs that can run on the processor. When the processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 7.

10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images