Intelligent reflecting surface mtc system resource optimization method and system

Abstract

The application relates to the technical field of wireless communication, and discloses an intelligent reflecting surface MEC system resource optimization method and system, which comprises the following steps: receiving a user equipment incident signal through an intelligent reflecting surface array, estimating a signal arrival angle from a signal covariance matrix, and analyzing a Doppler frequency shift; inputting signal parameters into a Kalman filter to track a user motion trajectory, and establishing a motion state model; inputting the motion state and real-time channel state information into a prediction model based on a long short-term memory network to predict a future time slot signal-to-noise ratio; comparing the channel quality prediction value with a rated value, calculating a power compensation amount, generating a phase offset matrix, and forming an enhanced beam at a predicted position; and when it is predicted that a user equipment will generate a calculation task, pre-allocating a calculation resource block and pre-loading context data at an edge server, so that the communication and calculation resources are cooperatively optimized, and the service continuity and resource utilization efficiency in a mobile scene are improved.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology and discloses a method and system for optimizing MEC system resources using an intelligent reflective surface. Background Technology

[0002] With the development of mobile edge computing technology, achieving efficient collaborative optimization of communication and computing resources has become an urgent technical problem. Traditional technologies still have many shortcomings. For example, traditional resource allocation methods are based on the current channel state for reactive allocation. When user equipment moves to the edge of signal coverage, the delay in processes such as channel detection and resource request leads to a high service interruption rate. They also lack forward-looking perception of the user equipment's movement status and cannot configure resources in advance before channel degradation. Existing intelligent reflector technology mainly optimizes communication link quality but lacks collaborative management with edge computing resources. When user equipment generates computationally intensive tasks, there is a timing mismatch between the allocation of communication and computing resources, leading to increased task processing delays. Existing systems have poor adaptability to dynamically changing wireless environments and cannot predict future resource needs based on the user equipment's movement trajectory, making it difficult to guarantee service continuity for mobile users during high-speed movement. Summary of the Invention

[0003] The purpose of this section is to outline some aspects of the embodiments of this application and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents, and such simplifications or omissions should not be construed as limiting the scope of this application.

[0004] To address the aforementioned technical problems, this application provides a method and system for optimizing MEC system resources using intelligent reflective surfaces.

[0005] On the one hand, this application provides a resource optimization method for an MEC system with a smart reflector, including S1 receiving an incident signal from a user equipment, estimating the signal angle of arrival from the covariance matrix of the incident signal, and analyzing the Doppler frequency shift of the incident signal; S2 inputs the signal angle of arrival and Doppler frequency shift into a Kalman filter to track the motion trajectory and speed changes of the user equipment, and establishes a motion state model of the user equipment. The motion state model and real-time channel state information are then input into a signal quality detection model, and the future time slot signal-to-noise ratio is predicted through the signal quality detection model. S3 compares the predicted channel quality value with the channel rating value to obtain the signal attenuation level, calculates the required power compensation, and generates a phase offset matrix for the smart reflector to form an enhanced beam at the predicted position. When S4 predicts that a user device will generate a computing task in a future time slot, it pre-allocates computing resource blocks in the edge server and loads the task-related context data into the memory area.

[0006] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: Estimating the angle of arrival of the signal from the covariance matrix of the incident signal includes: The covariance matrix is ​​constructed by the incident signals received by each element of the intelligent reflective surface array; The covariance matrix is ​​decomposed into eigenvalues, and the eigenvectors are divided into signal subspace and noise subspace. The spatial spectral function is calculated using a spectral peak search algorithm by taking advantage of the orthogonality between the signal subspace and the noise subspace. The angle of arrival of the incident signal is determined by detecting the peak position of the spatial spectral function.

[0007] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The analysis of the Doppler frequency shift of the incident signal includes: The received baseband signal is sampled in the time domain to obtain a signal sequence; The signal sequence is transformed from the time domain to the frequency domain to obtain the frequency domain representation of the signal; The actual offset of the carrier frequency is calculated by detecting the peak offset in the frequency domain representation; The radial velocity of the user equipment is obtained by using the physical relationship between the carrier frequency offset and the motion speed.

[0008] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The process of tracking the motion trajectory and speed changes of user equipment and establishing a motion state model of user equipment includes: constructing a state vector containing position, velocity, and acceleration based on the signal angle of arrival and Doppler frequency shift; The motion state at the next moment is predicted using the kinematic equations through the state transition matrix in the Kalman filter. The measured signal parameters and the predicted state are fused using the observation matrix, and the optimal state estimate is updated using Kalman gain. A motion state model containing the relationship between position and velocity is established by continuously updating the state estimation results.

[0009] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The step of inputting the motion state model and real-time channel state information into the signal quality detection model to predict the future time slot signal-to-noise ratio includes: The motion parameters output by the motion state model and the reference signal received power in the real-time channel state information are used together as input features; By using a prediction model based on a long short-term memory network, the spatiotemporal correlation between motion state and channel quality is learned; Using a trained model, the signal-to-noise ratio (SNR) trend of future time slots can be predicted based on the current motion state and channel conditions. The signal-to-noise ratio data sources include the reference signal received power measured by the cell reference signal, and the signal-to-interference-plus-noise ratio measured by the demodulated reference signal.

[0010] As a preferred embodiment of the MEC system resource optimization method for intelligent reflectors proposed in this application, the step of obtaining the signal attenuation level by comparing the predicted channel quality value with the rated channel value includes: The predicted signal-to-noise ratio is compared with the minimum signal-to-noise ratio threshold required for communication quality. The difference between the predicted signal-to-noise ratio and the minimum signal-to-noise ratio threshold is calculated as a quantitative indicator of the degree of signal attenuation. The signal-to-noise ratio margin required to maintain reliable communication is determined by the signal-to-noise ratio requirements corresponding to the channel coding and modulation methods. The severity of signal attenuation is assessed using the difference and signal-to-noise ratio margin.

[0011] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The calculation of the required power compensation amount and the generation of the phase offset matrix for the smart reflector include: Based on the degree of signal attenuation, the additional power compensation required to achieve the target channel capacity is calculated by back-calculating using Shannon's formula. By combining the channel matrix between the smart reflector and the user equipment, the phase configuration required for beamforming is calculated. By optimizing the algorithm, the intelligent reflective array forms a phase shift matrix that enables energy focusing at the predicted position.

[0012] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The phase offset matrix is ​​calculated using a zero-forcing beam, and forming an enhanced beam at the predicted location includes: Obtain channel state information from the intelligent reflector to the predicted location; Construct a system channel matrix that includes all interfering channels; The phase offset matrix is ​​calculated by using a null-forcing beam to create a main lobe at the predicted location while creating nulls in other directions. The phase shift of each reflector element of the intelligent reflector is configured using the calculated phase shift matrix, forming an enhanced beam with concentrated energy at the predicted position.

[0013] As a preferred embodiment of the MEC system resource optimization method for intelligent reflective surfaces proposed in this application, wherein: The step of pre-allocating computing resource blocks in the edge server and loading task-related context data into the memory region includes: A resource demand prediction model is built from the historical task execution records of user equipment to predict the required computing resource scale based on task type and data processing volume. In the resource manager of the edge server, separate resource isolation areas are divided, and computing resource blocks containing CPU cores and memory are pre-allocated according to the predicted computing resource requirements. Based on the service characteristics and task types of user devices, relevant application context, model parameters and configuration data are prefetched from the edge storage system to the resource isolation area; Establish a resource reservation mapping table to record the binding relationship between pre-allocated resource blocks and corresponding user devices and task types, ensuring that tasks can be scheduled and executed immediately upon arrival.

[0014] This application provides a resource optimization system for an intelligent reflective surface MEC system, including: The signal sensing and processing module includes a signal parameter extraction unit and a motion state unit. It receives the incident signal from the user equipment through an intelligent reflective array, estimates the signal angle of arrival from the covariance matrix of the incident signal, and analyzes the Doppler frequency shift of the incident signal. The motion state unit processes the signal angle of arrival and Doppler frequency shift to track the motion trajectory and speed changes of the user equipment and establish a motion state model of the user equipment. The resource prediction and decision-making module includes a channel quality prediction unit and a resource pre-allocation unit. The channel quality prediction unit inputs the motion state model and real-time channel state information into the signal quality detection model to predict the future time slot signal-to-noise ratio. The resource pre-allocation unit obtains the signal attenuation degree by comparing the channel quality prediction value with the channel rating value, calculates the power compensation amount, and generates a smart reflector phase offset matrix configuration instruction. The resource configuration execution module includes a communication configuration unit and a resource allocation unit. The communication configuration unit calculates the phase offset matrix to predict the position and form an enhanced beam according to the phase offset matrix configuration instruction. When the resource pre-allocation unit predicts that the user equipment will generate a computing task in a future time slot, it pre-allocates computing resource blocks in the edge server and loads the task-related context data into the memory area.

[0015] The beneficial effects of this application are as follows: This application achieves accurate prediction of user equipment motion trajectory and channel quality changes by extracting and analyzing signal angle of arrival and Doppler frequency shift, combined with motion state modeling using Kalman filters. This enables the system to pre-configure resources and improve service continuity.

[0016] This application uses comparative analysis of channel quality prediction values ​​and rated values ​​to accurately calculate power compensation requirements and generate an optimized phase offset matrix, enabling the intelligent reflector to form a precise enhanced beam, thereby improving energy efficiency while ensuring communication quality.

[0017] This application predicts the computing task requirements of user devices, completes the allocation of computing resources and preloads context data in advance, realizes deep collaboration between communication resources and computing resources, and improves task processing efficiency and resource utilization.

[0018] This application enables the system to continuously learn from prediction bias through a feedback adaptive correction mechanism, thereby reducing the prediction error rate in long-term operation, enhancing adaptability to dynamic environments, effectively preventing resource deadlock and communication and computing resource conflicts, improving resource turnover efficiency in multi-user scenarios, and enhancing robustness to position prediction errors and improving system energy efficiency through robust zero-forcing beamforming and joint power phase optimization. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained through these drawings without creative effort. Wherein: Figure 1 A schematic diagram of the structure of a resource optimization method for an intelligent reflective surface MEC system provided in this application; Figure 2 A flowchart of a resource optimization system for an intelligent reflective surface MEC system is provided for this application; Figure 3 A flowchart of resource prediction and optimization decision-making for a resource optimization method for an intelligent reflective surface MEC system provided in this application; Figure 4 This application provides a system module interaction and data flow diagram for a resource optimization system of an intelligent reflective surface MEC system; Figure 5 This application provides an example of the state transition matrix for a resource optimization system of an intelligent reflective surface MEC system. Detailed Implementation

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0023] Example 1 like Figure 1 As shown, a resource optimization method for an intelligent reflective surface MEC system aims to maximize system resource utilization and minimize task latency, including: S1 receives the incident signal from the user equipment through the intelligent reflector array, estimates the signal angle of arrival from the covariance matrix of the incident signal through multiple signal classification, and analyzes the Doppler frequency shift of the incident signal through fast Fourier transform. Estimating the angle of arrival of the signal from the covariance matrix of the incident signal includes: The covariance matrix is ​​constructed by the incident signals received by each element of the intelligent reflective surface array; The covariance matrix is ​​decomposed into eigenvalues, and the eigenvectors are divided into signal subspace and noise subspace. The spatial spectral function is calculated using a spectral peak search algorithm by taking advantage of the orthogonality between the signal subspace and the noise subspace. The angle of arrival of the incident signal is determined by detecting the peak position of the spatial spectral function.

[0024] Specifically, multiple elements of the intelligent reflective array simultaneously receive incident signals from user equipment. The signal received by each element is processed by the radio frequency front end, converted into a baseband signal, and then digitally sampled to form a corresponding signal sequence. These signal sequences retain their respective spatial characteristics.

[0025] Due to the different arrival paths of the user signals received by each element of the intelligent reflector, a phase difference is generated between the elements. The covariance matrix formed by these phase differences can be used to decompose the signal into a signal subspace and a noise subspace, which are orthogonal. By searching for the direction that minimizes the orthogonality between the steering vector and the noise subspace, the signal angle of arrival can be determined.

[0026]

[0027] The guide vector reflects the phase difference between array elements when the signal is incident from the θ direction. , where d is the spacing between reflector elements, λ is the carrier signal wavelength, M is the number of reflector elements, and j is the imaginary unit.

[0028] The noise subspace matrix is ​​composed of eigenvectors corresponding to the small eigenvalues ​​after the eigenvalue decomposition of the covariance matrix.

[0029] The superscript H indicates conjugate transpose. Represents the noise subspace matrix The conjugate transpose of .

[0030] Values ​​are obtained by scanning θ. The peak position of the signal is the estimated angle of arrival. When the signal-to-noise ratio is higher than 5dB, the angle estimation error is less than 1°.

[0031] The multi-signal classification algorithm used in this application is applicable to scenarios with single-user signal sources, narrowband signal transmission, intelligent reflector, and receiver channels where the amplitude and phase consistency meets the preset accuracy and the incident signal-to-noise ratio is higher than 5dB. To address Doppler spread caused by high-speed movement, the algorithm suppresses Doppler effects through time-domain filtering and frequency-domain correction preprocessing, ensuring that the angle of arrival estimation accuracy meets the requirements of motion tracking.

[0032] A preferred method for constructing the covariance matrix includes: The covariance matrix of the received signal is calculated by using the signal sequence received by each element of the intelligent reflective array. The matrix reflects the correlation characteristics between the received signals of each element and contains the spatial distribution information of the signal. The received signal is preprocessed for noise reduction using statistical analysis methods to ensure that the covariance matrix can accurately reflect the spatial characteristics of the signal.

[0033] Furthermore, eigenvalue decomposition is performed on the constructed covariance matrix. The eigenvalues ​​are arranged in order of magnitude, and the corresponding eigenvectors are divided into signal subspace and noise subspace according to the distribution characteristics of the eigenvalues. The eigenvectors corresponding to eigenvalues ​​greater than the eigenvalue constitute the signal subspace, and the eigenvectors corresponding to eigenvalues ​​less than the eigenvalue constitute the noise subspace, thus achieving effective separation between the signal space and the noise space.

[0034] A preferred method for calculating the spatial spectrum function is as follows: Based on the orthogonality between the signal subspace and the noise subspace, a spatial spectrum function is constructed using a spectral peak search algorithm. This spatial spectrum function reflects the energy distribution of the signal in different directions of arrival. During the calculation, by traversing all possible angles of arrival, the degree of orthogonality between the signal subspace and the noise subspace in each direction is evaluated, thereby forming a complete spatial spectrum distribution.

[0035] Furthermore, by detecting the peak positions of the spatial spectral function, the main direction of arrival of the incident signal can be determined. The peaks in the spatial spectral function correspond to the actual angle of arrival of the signal, and the height of the peaks reflects the signal intensity in that direction. By identifying the positions of all significant peaks, estimates of the angle of arrival of multiple incident signals can be obtained.

[0036] This application utilizes the spatial signal processing capabilities of an intelligent reflective array to achieve accurate estimation of the angle of arrival of incident signals. By employing subspace decomposition technology, it effectively distinguishes between signals and noise, thereby improving the accuracy and reliability of the angle of arrival estimation.

[0037] The analysis of the Doppler frequency shift of the incident signal includes: The received baseband signal is sampled in the time domain to obtain a signal sequence; The signal sequence is transformed from the time domain to the frequency domain to obtain the frequency domain representation of the signal; The actual offset of the carrier frequency is calculated by detecting the peak offset in the frequency domain representation; The radial velocity of the user equipment is obtained by using the physical relationship between the carrier frequency offset and the motion speed.

[0038] S2 inputs the signal angle of arrival and Doppler frequency shift into a Kalman filter to track the motion trajectory and speed changes of the user equipment, and establishes a motion state model of the user equipment. The motion state model and real-time channel state information are then input into a signal quality detection model, and the future time slot signal-to-noise ratio is predicted through the signal quality detection model. The process of tracking the motion trajectory and speed changes of user equipment and establishing a motion state model of user equipment includes: constructing a state vector containing position, velocity, and acceleration based on the signal angle of arrival and Doppler frequency shift; The motion state at the next moment is predicted using the kinematic equations through the state transition matrix in the Kalman filter. The measured signal parameters and the predicted state are fused using the observation matrix, and the optimal state estimate is updated using Kalman gain. A motion state model containing the relationship between position and velocity is established by continuously updating the state estimation results.

[0039] This application constructs a two-dimensional absolute coordinate system with the base station location as the origin. Combining the known coordinates of the base station, the deployment location and angle of the intelligent reflector, the signal arrival angle and Doppler frequency shift observation values ​​are mapped to the absolute coordinate space to realize the calculation of the absolute position of the user equipment. This ensures that the motion trajectory tracking result is the absolute position, so that the intelligent reflector can form an enhanced beam according to the predicted absolute position, avoiding beam pointing deviation caused by relative motion.

[0040] The Kalman filter uses a recursive mechanism of prediction, observation, and correction to weight and fuse the position predicted by the kinematic model with the observed position calculated from the signal angle of arrival and Doppler frequency shift. The weights are dynamically determined by the uncertainties of prediction and observation. As time is updated, the system's estimation of the user's motion state gradually converges.

[0041] In this application, a preferred method for constructing motion state vectors includes: Based on the signal angle of arrival measurement results and Doppler frequency shift analysis data, a state vector describing the motion state of the user equipment is constructed. The state vector includes the position coordinates of the user equipment in three-dimensional space, the motion velocity components, and the acceleration change, forming a motion state description system.

[0042] A preferred state prediction processing method is to predict the motion state of the user equipment using the state transition matrix in the Kalman filter. The state transition matrix is ​​established based on the time interval and the kinematic equation, and the predicted state at the next moment is calculated based on the motion state at the current moment.

[0043] Furthermore, a conversion relationship between the state vector and the measured signal parameters is established through the observation matrix. The actual measured signal angle of arrival and Doppler frequency shift data are compared with the predicted state. The predicted and measured values ​​are weighted and fused using the Kalman gain coefficient. The Kalman gain is dynamically adjusted according to the uncertainty of the prediction and measurement to achieve the optimal combination of the predicted state and the measured data.

[0044] It should be noted that the specific implementation method for tracking the motion trajectory and speed changes of user equipment and establishing a motion state model of user equipment in this application includes: The raw signal parameters obtained in S1—angle of arrival θ and Doppler shift Δf—are transformed into a motion state sequence that can be used by the prediction model. Specifically: The angle of arrival provided by multiple signal classification reflects angular information, while the Doppler frequency shift provided by FFT reflects radial velocity. The two are derived from different physical principles and have different update rates. The Kalman filter describes position, velocity, and acceleration in a unified manner through the state vector and uses the observation matrix H to map these two types of measurements to the same state space.

[0045] Due to signal processing delays or obstructions, there may be temporary gaps in the angle of arrival or Doppler shift. The prediction step of the Kalman filter can maintain trajectory continuity by relying on the recursive state of the kinematic model when there are no observations.

[0046] Optimal state estimation of Kalman filter output It has lower noise and more uniform updates than the original measurements, and can be directly used as the position and velocity input features of the LSTM network, avoiding the overfitting caused by the LSTM learning motion patterns directly from noisy angle sequences.

[0047] State covariance matrix The confidence level of the current estimate is quantified, and the state covariance matrix is ​​used to adjust the confidence interval of the LSTM prediction in S3; the resource reserve margin is dynamically calculated, that is, the larger the prediction error, the more resources are reserved.

[0048] First, a state vector is constructed, and the calculation method for the state vector includes:

[0049] in, Let be the motion state vector of the user equipment at time k, where k is the discrete time slot index and T is the vector transpose.

[0050] The coordinates of the user equipment in a two-dimensional horizontal plane are the final quantities that need to be tracked.

[0051] ( ) represents the velocity component (m / s), derived from the Doppler frequency shift and the rate of change of angle.

[0052] ( ) represents the acceleration component (m / s²), reflecting the motion trend.

[0053] Furthermore, prediction is made based on the motion state of the user equipment, and the calculation method is as follows:

[0054] For the prior state estimate at time k-1 based on time k-1 For the posterior state estimation at time k−1 The prior state covariance matrix The posterior state covariance matrix F is the state transition matrix, constructed from the uniform / uniformly accelerated kinematic equations, and includes the sampling period T, for example, T=0.01 seconds. The state transition matrix establishes the temporal dependence of the state, enabling the system to still predict even without observation.

[0055] P is the state covariance matrix, with the diagonal elements representing the estimated variance of each state variable, and Q is the process noise covariance, reflecting the uncertainty of the motion model.

[0056] like Figure 5 The image shows an example of a preferred form of the state transition matrix;

[0057] Furthermore, the observed data is mapped to the state, and the calculation method is as follows:

[0058]

[0059] in, It is the measurement observation vector from S1, i.e., the angle of arrival at time k estimated by multiple signal classification. The k-time Doppler frequency shift extracted by FFT .

[0060] Observation function Description: If the current motion state is Theoretically, the values ​​of the angle of arrival and radial Doppler shift that should be measured are the observation function. Converting Cartesian coordinates (position, velocity) into quantities corresponding to physical measurements (angle, frequency offset) allows the Kalman filter to compare measured values. Compared with the predicted value H ( To correct the state estimate.

[0061] λ is the arctangent function in the four quadrants, used for angle calculation, where λ is the carrier wavelength.

[0062] It should be noted that, Used to calculate the azimuth angle of user equipment relative to the origin, with a value range of (-π, π]. This represents the distance from the user equipment to the origin.

[0063] in, It is the measurement observation vector from S1, i.e., the angle of arrival at time k estimated by multiple signal classification. The k-time Doppler frequency shift extracted by FFT .

[0064] λ is the arctangent function in the four quadrants, used for angle calculation, where λ is the carrier wavelength.

[0065] Because this function is nonlinear, it is linearized using an extended Kalman filter, and the Jacobian matrix is... Used to calculate Kalman gain.

[0066] Furthermore, the calculation results of data fusion are as follows:

[0067] in, The Kalman gain determines the weights of the predicted and observed values ​​when correcting the state. A larger gain indicates greater confidence in the observation.

[0068] For the observation function in The Jacobian matrix at that location.

[0069] This is the posterior optimal state estimate at time k.

[0070] Let be the posterior covariance matrix at time k.

[0071] I is the identity matrix.

[0072] R is the observation noise covariance matrix, and the diagonal elements are the variance of the angle of arrival measurement and the variance of the Doppler frequency shift measurement.

[0073] Updated and The current optimal motion state estimate and its uncertainty are respectively output to the LSTM prediction of S3 and to the resource pre-allocation of S4.

[0074] Furthermore, based on the data fusion results, the motion state estimate of the user equipment is continuously updated. The update process is implemented through a recursive algorithm. Each new measurement data triggers a correction of the state estimate, making the estimation result continuously approach the real motion state. Through continuous state updates, a state model reflecting the real-time motion characteristics of the user equipment is established.

[0075] A preferred method for motion modeling and correlation analysis includes: Based on the continuously updated state estimation results, a motion state model of the user equipment is constructed. The motion state model not only records the current motion state parameters, but also reveals the motion patterns and trends of the user equipment by analyzing the correlation between position and velocity.

[0076] This application achieves high-precision tracking and modeling of the motion state of user equipment through the Kalman filter algorithm. It combines signal measurement parameters with kinematic principles and reduces the impact of measurement noise on tracking accuracy through recursive processing of prediction and correction.

[0077] The step of inputting the motion state model and real-time channel state information into the signal quality detection model to predict the future time slot signal-to-noise ratio includes: The motion parameters output by the motion state model and the reference signal received power in the real-time channel state information are used together as input features; By using a prediction model based on a long short-term memory network, the spatiotemporal correlation between motion state and channel quality is learned; Traditional linear prediction cannot capture the nonlinear fading changes of the channel as the user moves. Long Short-Term Memory (LSTM) networks, through forget gates, input gates, and output gates, can selectively memorize channel characteristics over long time spans and learn spatiotemporal correlations from the joint sequence of motion parameters (position, velocity) and reference signal received power, thereby predicting the signal-to-noise ratio (SNR) of future short time slots. Specifically, LSTM stands for Long Short-Term Memory network, and the update calculation method is as follows:

[0078] in, The input feature vector includes the position, velocity, and real-time RSRP output from the Kalman filter.

[0079] In its hidden state, it carries channel information from past time slots.

[0080] It is a long-term memory carrier in the cellular state.

[0081] Forgotten Gate; For input gates; This is the activation value of the output gate.

[0082] , , , For each gate weight matrix; , , For each gate bias vector.

[0083] σ is the sigmoid activation function; tanh is the hyperbolic tangent activation function; ⊙ is the Hadamard product, which is the element-wise multiplication.

[0084] t is the LSTM input timing index (time slot). This represents the cell state at the previous moment.

[0085] The three-layer LSTM consists of a first layer of 128 cells, a second layer of 64 cells, followed by dropout (0.2) and a fully connected layer, which outputs the signal-to-noise ratio for the next three time slots.

[0086] Using a trained prediction model, the signal-to-noise ratio (SNR) trend of future time slots is predicted based on the current motion state and channel conditions. The training method for the LSTM prediction model is as follows: 10,000 sets of user motion trajectories and corresponding channel quality data were collected, with 80% used as the training set and 20% as the validation set. Each set of samples contains motion parameters (position, velocity) and RSRP values ​​for 20 consecutive time slots, labeled with the SNR values ​​of the next 3 time slots.

[0087] The position coordinates were normalized to [0,1], the velocity to [-1,1], and the RSRP to the range of [-140dBm, -40dBm].

[0088] The loss function is the mean squared error.

[0089] The optimizer is Adam, with an initial learning rate of 0.001 and a decay of 0.9 per 100 epochs.

[0090] Batch size: 64, training rounds: 200.

[0091] This application employs a long short-term memory network to predict the signal-to-noise ratio for the next 3 to 10 time slots. The total prediction duration is less than the sum of the system frame structure duration and the intelligent reflector phase configuration delay, ensuring that the phase offset matrix calculation and configuration can be completed before the user equipment reaches the predicted position. This avoids beam alignment deviation caused by prediction lag. The prediction confidence interval is dynamically adjusted based on the state covariance matrix output by the Kalman filter, and the beam main lobe width is increased for scenarios with large prediction errors.

[0092] The signal-to-noise ratio data sources include the reference signal received power measured by the cell reference signal, and the signal-to-interference-plus-noise ratio measured by the demodulated reference signal.

[0093] Specifically, the kinematic parameters obtained from the motion state model analysis are fused with the channel state information acquired in real time at the feature level. The kinematic parameters include the three-dimensional spatial coordinates of the user equipment, the instantaneous velocity vector, and the acceleration change trend, which are used to describe the motion dynamics characteristics of the user equipment. The channel state information is mainly characterized by the reference signal received power, which reflects the loss characteristics on the electromagnetic wave propagation path. Through feature alignment and normalization, a fused feature vector with consistent dimensions is constructed.

[0094] In this application, a preferred spatiotemporal correlation modeling employs a deep neural network with a long short-term memory mechanism to establish a spatiotemporal correlation model between motion state and channel quality. This network structure, through the synergistic effect of input gates, forget gates, and output gates, achieves selective memorization and forgetting of historical state information, effectively capturing the long-term dependence of channel quality over time. The hidden layers of the network learn the nonlinear mapping relationship between changes in user equipment location and channel quality through abstract representations of spatial features.

[0095] Furthermore, the signal-to-noise ratio (SNR) sequence prediction utilizes a well-trained prediction model to infer the SNR change sequence for multiple future time slots based on the fused feature vector at the current moment. The prediction model also includes processing factors such as the motion trajectory prediction of user equipment, changes in environmental scattering characteristics, and spatial distribution of interference signals. Through nonlinear transformation of a multi-layer neural network, it outputs a temporally continuous SNR prediction value, forming a complete description of the channel quality evolution trend.

[0096] Furthermore, the accuracy of the signal-to-noise ratio (SNR) prediction results is verified through multi-dimensional measurement data. Received power measurements based on the cell reference signal and signal-to-interference-plus-noise ratio (SNR) measurements based on the demodulated reference signal are collected. These two types of measurement data provide measured comparison benchmarks from the two dimensions of large-scale fading characteristics and instantaneous SNR. By establishing an error feedback mechanism between the predicted and measured values, online calibration and parameter optimization of the prediction model are achieved.

[0097] This application deeply integrates the kinematic characteristics of user equipment with wireless channel measurement information, and achieves accurate prediction of channel quality through a deep learning model. The application of long short-term memory network enables the model to effectively learn the temporal evolution law and spatial variation characteristics of channel quality, and the multi-dimensional verification mechanism ensures the reliability of the prediction results.

[0098] S3 compares the predicted channel quality value with the rated channel value to obtain the signal attenuation level, calculates the required power compensation, and generates a phase offset matrix for the smart reflector. The phase offset matrix is ​​calculated by the zero-forcing beam and forms an enhancement beam at the predicted position. like Figure 3 As shown, obtaining the signal attenuation level by comparing the predicted channel quality value with the rated channel quality includes: The predicted signal-to-noise ratio is compared with the minimum signal-to-noise ratio threshold required for communication quality. The difference between the predicted signal-to-noise ratio and the minimum signal-to-noise ratio threshold is calculated as a quantitative indicator of the degree of signal attenuation. The signal-to-noise ratio margin required to maintain reliable communication is determined by the signal-to-noise ratio requirements corresponding to the channel coding and modulation methods. The severity of signal attenuation is assessed using the difference and signal-to-noise ratio margin.

[0099] The predicted future time slot signal-to-noise ratio (SNR) values ​​are compared and analyzed with the minimum SNR threshold set by the communication system. The minimum SNR threshold is preset based on the basic reliability requirements of the communication system and reflects the minimum signal quality requirements required to ensure basic communication quality.

[0100] Based on the difference between the predicted signal-to-noise ratio and the threshold value, a quantitative evaluation system for the degree of signal attenuation is established. The difference directly reflects the degree of deviation of the channel quality from the minimum requirement. By establishing the mapping relationship between the magnitude of the difference and the attenuation level, the degree of signal attenuation is transformed into a quantifiable graded index.

[0101] Based on the specific signal-to-noise ratio (SNR) requirements of the currently used channel coding schemes and modulation methods, this paper analyzes the SNR margin required to maintain reliable communication. Different coding and modulation combinations have different requirements for channel quality. This paper establishes a correspondence between coding and modulation schemes and SNR requirements.

[0102] By combining the signal-to-noise ratio difference index and the required signal-to-noise ratio margin of the system, the severity of signal attenuation is comprehensively judged. By establishing an evaluation model that comprehensively considers the current attenuation situation and system requirements, the degree of signal attenuation is divided into different levels.

[0103] This application establishes a complete signal attenuation assessment process, enabling precise perception and quantitative analysis of channel quality changes. By employing a multi-factor comprehensive judgment method, it ensures the accuracy and practicality of the assessment results.

[0104] The calculation of the required power compensation amount and the generation of the phase offset matrix for the smart reflector include: Based on the degree of signal attenuation, the additional power compensation required to achieve the target channel capacity is calculated using Shannon's formula. By combining the channel matrix between the smart reflector and the user equipment, the phase configuration required for beamforming is calculated. By optimizing the algorithm, the intelligent reflective array forms a phase shift matrix that enables energy focusing at the predicted position.

[0105] The power compensation amount and the phase offset matrix of the smart reflector are jointly optimized, not calculated independently. Based on the predicted power compensation amount ΔP, the array gain target to be provided by the smart reflector is determined. This gain target is then substituted into the optimization target of zero-forcing beamforming as a constraint, so that the phase offset matrix provides a signal enhancement effect that matches the power compensation amount while satisfying the main lobe energy focusing, thus achieving a one-to-one mapping between power compensation requirements and phase configuration.

[0106] A preferred method for calculating power compensation is to use the signal attenuation assessment results and the relationship between channel capacity and signal-to-noise ratio revealed by Shannon's formula to calculate the additional power compensation required to achieve the target channel capacity. The calculation process comprehensively considers the current channel state, the degree of signal attenuation, and the communication quality indicators required by the system. By combining theoretical calculations with actual conditions, an accurate power compensation value is determined to ensure that the expected communication quality level can be restored after compensation.

[0107] In this application, according to Shannon's formula, channel capacity and signal-to-noise ratio are logarithmically related. If the predicted signal-to-noise ratio is lower than the minimum signal-to-noise ratio required to maintain the target capacity, additional transmit power or RIS array gain is required to compensate. The amount of compensation depends on the difference between the predicted signal-to-noise ratio and the required signal-to-noise ratio, as well as the maximum gain that RIS beamforming can provide. The power compensation calculation method is as follows:

[0108] in, For target channel capacity The minimum signal-to-noise ratio required at (bps) and system transmission bandwidth B (Hz).

[0109] The future signal-to-noise ratio (dB) predicted by LSTM; This is the required power compensation amount.

[0110] The upper limit of the compensation amount is limited by the emission power spectral density specification and the maximum adjustable phase shift range of RIS.

[0111] For example, in a practical implementation, when the predicted signal-to-noise ratio drops from 12dB to 4dB, a compensation of 6dB is calculated. By contributing 5dB array gain through RIS beamforming, and adding a 1dB increase in base station transmit power, the actual restored signal-to-noise ratio reaches 10.5dB, meeting the requirements of video services and avoiding stuttering caused by channel degradation.

[0112] It should be noted that the power compensation amount ΔP and the subsequent phase offset matrix are not determined independently; they jointly affect the final received signal-to-noise ratio. In practical implementation, the base station transmit power and RIS phase can be adjusted jointly through alternating optimization or lookup table methods to minimize total power consumption or meet specific energy efficiency targets. As a simplified implementation, ΔP can be determined first according to the above formula, and then the phase offset matrix can be calculated based on the compensated equivalent channel.

[0113] Furthermore, by combining the channel matrix information between the intelligent reflector and the user equipment, the phase configuration parameters required for beamforming are calculated. By analyzing the characteristic structure of the channel matrix, the spatial characteristics of the signal during wireless propagation are obtained, and the optimal phase adjustment scheme for each reflector is determined accordingly.

[0114] Specifically, the phase offset matrix optimization solution uses an optimization algorithm to solve the phase offset matrix of the intelligent reflector array to ensure that an enhanced beam with energy focusing is formed at the predicted position. The optimization process aims to maximize the received signal power at the target position, while taking into account the parameter constraints of the reflector unit, and finds the optimal phase configuration scheme through iterative calculation.

[0115] This application achieves a collaborative design of power compensation and beamforming by combining theoretical calculations with optimization algorithms. The phase offset matrix obtained by the optimization algorithm further improves the efficiency of signal transmission. The hierarchical processing method ensures the reliability of the system and optimizes its performance.

[0116] The phase offset matrix is ​​calculated using a zero-forcing beam, and forming an enhanced beam at the predicted location includes: Obtain channel state information from the intelligent reflector to the predicted location; Construct a system channel matrix that includes all interfering channels; The phase offset matrix is ​​calculated by using a null-forcing beam to create a main lobe at the predicted location while creating nulls in other directions. The specific calculation method for the phase offset matrix of forced zero beamforming is as follows: Zero-forcing beamforming is a zero-forcing algorithm that constructs a system channel matrix that includes the target direction and other interference directions, calculates the pseudo-inverse of the channel matrix, and obtains a set of beamforming weights. These beamforming weights coherently superimpose the signals in the target direction while destructively superimposing the signals in the interference direction, thereby forming a main lobe at the predicted user location and nulls in the other directions.

[0117]

[0118] in, H is the zero-forcing beamforming vector; H is the system channel matrix, with the first row being the channel vector from the RIS to the intelligent reflector at the user's predicted location to the target user. The remaining lines are channel vectors from RIS to each interference direction.

[0119] e1=[1,0,…,0]T, used to extract the target direction; It is the 2-norm of the vector.

[0120] The optimal beamforming vector has the following elements. phase This refers to the phase shift value of each reflection unit in the RIS.

[0121] The phase offset matrix is ​​calculated using a zero-forcing beam, and forming an enhanced beam at the predicted location includes: Obtain channel state information from the intelligent reflector to the predicted location; Construct a system channel matrix that includes all interfering channels; The phase offset matrix is ​​calculated by using a null-forcing beam to create a main lobe at the predicted location while creating nulls in other directions. The phase shift of each reflector element of the intelligent reflector is configured using the calculated phase shift matrix, forming an enhanced beam with concentrated energy at the predicted position.

[0122] As a specific implementation method, the system channel matrix H is constructed as follows: First, define the interference direction. The interference direction includes the direction of other user equipment currently communicating within the same coverage area as the smart reflector; the direction of frequently occurring strong interference sources obtained based on historical signal angle of arrival statistics; and when the accurate location of other users cannot be obtained, the entire space except for the predicted user direction is discretized into several angle intervals, with the center of each interval serving as a potential interference direction.

[0123] The number of interference directions K is dynamically determined based on the number of smart reflector array elements M and the actual interference environment. Usually, K = min(M-1, 8) is taken, that is, it does not exceed the number of array elements minus 1, and the maximum is 8. In specific implementation, the received signal strength in each direction can be measured by uplink detection signal, and the K directions with the strongest strength can be selected as interference directions.

[0124] Secondly, the channel vectors for each interference direction are measured. For the i-th interference direction (i=1,...,K), the smart reflector estimates the channel response vector from the smart reflector to that direction by receiving the pilot signal transmitted by the user equipment or interference source in that direction. Its dimension is M×1, where M is the number of reflection units. The channel vector corresponding to the predicted user direction is denoted as... , H is an (M+1)×M matrix, where the first row is the target channel and the remaining rows are the interference channels.

[0125] To address location prediction errors, the target channel vector can be locally averaged or a robust zero-forcing criterion can be adopted before calculating the zero-forcing beam, which involves adding a virtual perturbation term to the system channel matrix to broaden the main lobe width.

[0126] By using a phase offset matrix, the signal power measured at the predicted location is increased. Even if there is a slight deviation in the predicted location, the main lobe can still cover the user's actual location, while reducing interference to other users on the same frequency.

[0127] The phase shift of each reflector element of the intelligent reflector is configured using the calculated phase shift matrix, forming an enhanced beam with concentrated energy at the predicted position.

[0128] Each reflector unit of the intelligent reflector only supports discrete phase adjustment. This application discretizes the continuous phase value obtained by zero-forcing beamforming calculation according to the number of phase quantization bits supported by the actual hardware. Under the premise of ensuring the beam main lobe gain and interference suppression performance, the continuous phase to discrete phase conversion is completed, so that the calculated phase offset matrix can be directly configured in the real hardware, meeting the engineering feasibility requirements.

[0129] Acquire channel state information between the smart reflector and the predicted location of the user equipment. This information includes the amplitude response and phase characteristics of the signal propagation path, measured using pilot or reference signals. The channel state information accurately reflects the propagation characteristics of electromagnetic waves between the smart reflector and the target location.

[0130] Specifically, based on the acquired channel state information, a system channel matrix containing the target channel and all interfering channels is constructed. The channel matrix is ​​used to describe the channel characteristics from each reflector of the smart reflector to the predicted location and potential interference direction of the user equipment.

[0131] Furthermore, by calculating the phase offset matrix of the smart reflector through zero-forcing beam, and by solving the pseudo-inverse of the system channel matrix, the optimal phase configuration scheme is obtained. A preferred method for implementing zero-forcing beam includes forming nulls in the interference direction while enhancing the signal in the target direction, thereby maximizing the signal-to-interference-to-noise ratio at the target location.

[0132] The specific beamforming method configures the phase shift parameters of each reflector of the intelligent reflector through the calculated phase offset matrix, controls the phase offset of each reflector, and enables the reflected signals to coherently superimpose at the predicted position to form an enhanced beam with concentrated energy. This achieves signal cancellation in the specified interference direction and effectively suppresses co-channel interference.

[0133] This application achieves precise spatial beam control through zero-forcing beamforming, and innovatively combines interference suppression with signal enhancement. By utilizing the spatial degrees of freedom of the intelligent reflector, it effectively suppresses interference while enhancing the target signal, thereby improving the system's communication quality and spectral efficiency.

[0134] When it is predicted that a user device will generate a computing task in a future time slot, S4 pre-allocates computing resource blocks in the edge server and loads task-related context data into the memory area. The allocation amount of the computing resource blocks is determined by a linear regression model based on the user device's historical task execution data.

[0135] The step of pre-allocating computing resource blocks in the edge server and loading task-related context data into the memory region includes: A resource demand prediction model is built from the historical task execution records of user equipment to predict the required computing resource scale based on task type and data processing volume. A preferred method for implementing a resource demand forecasting model includes: The CPU and memory requirements of different users' computing tasks (such as video analysis and object recognition) are approximately linearly related to the amount of data and the type of task. By collecting historical task records, a multiple linear regression model is established, and the covariance of the Kalman filter output is used to dynamically adjust the reserved margin to avoid resource shortages or waste due to prediction bias.

[0136] Methods for calculating resource demand forecasting include:

[0137] The calculation method for reserved resources is as follows:

[0138] Where dataSize is the amount of input data for the task (MB).

[0139] taskType is the task type encoding (e.g., 1=video transcoding, 2=image recognition).

[0140] The deadline is the maximum allowed processing delay (ms).

[0141] , , , The regression coefficients are updated every 100 task samples.

[0142] The standard deviation of the predicted residuals is statistically derived from historical prediction errors.

[0143] The risk coefficient can be set to the default value of 2, which corresponds to approximately 95% confidence level, or it can be set according to the specific implementation.

[0144] R represents the predicted computing resource requirements; For the final pre-allocated resource amount In the resource manager of the edge server, separate resource isolation areas are divided, and computing resource blocks containing CPU cores and memory are pre-allocated according to the predicted computing resource requirements. Based on the service characteristics and task types of user devices, relevant application context, model parameters and configuration data are prefetched from the edge storage system to the resource isolation area; Establish a resource reservation mapping table to record the binding relationship between pre-allocated resource blocks and corresponding user devices and task types, ensuring that tasks can be scheduled and executed immediately upon arrival.

[0145] As a specific implementation method, the resource demand forecasting model adopts multiple linear regression, and the method for determining its input feature vector and model parameters is as follows: For each historical task record, construct a feature vector X=[x1,x2,x3], where: x1 represents the amount of input data for the task, in MB, ranging from 0.1 to 1000, and is the actual measured value.

[0146] x2 represents the task type encoding, taskType, which uses integer encoding: 1=video transcoding, 2=image recognition, 3=natural language processing, 4=augmented reality rendering, 5=other. For mixed tasks, the encoding of the primary task type is used; if it cannot be classified, the default encoding is 5.

[0147] x3 is the maximum allowable processing delay deadline, in milliseconds (ms), ranging from 10 to 5000. It is carried by the user device in the task request; if not carried, it defaults to 500ms.

[0148] The initial coefficients were obtained through offline training: at least 1000 historical task records were collected, each containing a feature vector and the actual number of CPU cores and memory consumed. The initial values ​​were obtained by fitting using the least squares method. Example initial coefficients are as follows: .

[0149] During online operation, a coefficient update is triggered every N=100 new, completed task samples. The update method uses the sliding window least squares method, fitting only the most recent 500 samples to adapt to changes in user task patterns.

[0150] This application does not independently optimize communication beams and allocate computing resources. Instead, it takes maximizing the overall system service quality, minimizing total latency, and maximizing resource utilization as joint optimization objectives. It simultaneously inputs channel prediction results, motion state information, and computing task requirements into the resource decision model, so that communication resource enhancement and computing resource pre-allocation form a coordinated match: when the predicted channel quality deteriorates, the reserved margin of computing resources is appropriately increased; when the predicted computing tasks are intensive, the communication beam enhancement intensity is increased, thereby achieving deep coordinated optimization of communication and computing resources.

[0151] Based on historical task execution records of user devices, a computing resource demand prediction model is established. This model analyzes the correlation between historical task types, data processing volume, and actual resource consumption, and uses linear regression to establish a quantitative relationship between task characteristics and resource demands. Based on the characteristic parameters of the current predicted task, the expected computing resource scale demand is calculated through the regression model, providing a quantitative basis for resource pre-allocation.

[0152] Specifically, independent resource isolation zones are defined in the resource manager of the edge server. Based on the computing resource requirements output by the predictive model, computing resource blocks containing a specific number of CPU cores and memory capacity are pre-allocated to ensure the exclusivity and security of the pre-allocated resources and avoid resource competition and interference between different user devices.

[0153] Furthermore, based on the user device's business characteristics and task type, relevant application context data is proactively pre-fetched from the edge storage system. The pre-loaded content includes model parameters, configuration files, and initialization data required for task execution. Furthermore, the data is preloaded into the allocated resource isolation area to ensure that the full data environment required for execution can be quickly obtained when the task arrives.

[0154] Specifically, the method for establishing a resource reservation mapping table includes: the system records the binding relationship between each pre-allocated computing resource block and the corresponding user device and task type. The mapping table maintains the resource allocation status, validity period information and priority identifier. When the predicted task actually arrives, the system queries the mapping table to realize the immediate scheduling and execution of the task, effectively eliminating the delay in the resource allocation process.

[0155] S4 also includes a resource reservation lifecycle management mechanism. Each pre-allocated computing resource block is associated with a validity period, which is determined by the predicted task arrival time window. If no corresponding task is detected within the validity period, the system sends a status query command to the user device. If any of the following conditions are met: the validity period expires and no task arrives, the user device leaves the current edge server coverage area, or the user device explicitly cancels the task, the reserved resources are released immediately. If the task is still being transmitted or the predicted arrival time is delayed, the system dynamically extends the validity period.

[0156] This application achieves accurate prediction and rapid response of computing resources by establishing a complete resource pre-allocation system. A regression prediction model based on historical data ensures the accuracy of resource allocation, a resource isolation mechanism guarantees the stability of service quality, context preloading technology effectively reduces task execution latency, and resource mapping management achieves efficient matching of resources and tasks. In mobile scenarios, by jointly predicting user movement trajectories, channel quality changes, and computing task requirements, it achieves forward-looking and collaborative configuration of communication beams and computing resources to reduce the probability of service interruption and task processing latency.

[0157] Example 2 like Figure 2 As shown, a resource optimization system for an intelligent reflective surface MEC system includes a signal sensing and processing module, a resource prediction and decision-making module, and a resource allocation execution module. The signal sensing and processing module includes a signal parameter extraction unit and a motion state unit. The resource prediction and decision-making module includes a channel quality prediction unit and a resource pre-allocation unit; The resource configuration execution module includes a communication configuration unit and a resource allocation unit. like Figure 4 As shown, the system module interaction and data flow diagram of a smart reflective surface MEC system resource optimization system includes: The signal sensing and processing module includes a signal parameter extraction unit and a motion state unit. It receives the incident signal from the user equipment through an intelligent reflective array, estimates the signal angle of arrival from the covariance matrix of the incident signal, and analyzes the Doppler frequency shift of the incident signal. The motion state unit processes the signal angle of arrival and Doppler frequency shift through a Kalman filter, tracks the motion trajectory and speed changes of the user equipment, and establishes a motion state model of the user equipment. The resource prediction and decision-making module includes a channel quality prediction unit and a resource pre-allocation unit. The motion state model and real-time channel state information are input into the signal quality detection model to predict the future time slot signal-to-noise ratio. The resource pre-allocation unit compares the predicted channel quality value with the channel rating to obtain the signal attenuation level, calculates the power compensation amount, and generates a smart reflector phase offset matrix configuration instruction. The resource collaborative arbitration unit is used to detect resource conflicts between communication compensation and resource pre-allocation calculation, make joint allocation decisions based on service priorities, and perform dynamic resource lending when resources are scarce. The resource configuration execution module includes a communication configuration unit and a resource allocation unit. The communication configuration unit calculates the phase offset matrix according to the phase offset matrix configuration instruction and forms an enhanced beam at the predicted position. When the resource pre-allocation unit predicts that the user equipment will generate computing tasks in future time slots, it pre-allocates computing resource blocks in the edge server and loads the task-related context data into the memory area. The resource lifecycle management unit is responsible for setting the validity period for each pre-allocated resource block, monitoring the resource usage status, and automatically releasing the resources when the validity period expires or the user moves out of the coverage area.

[0158] Example 3 Taking a vehicle on a city road as an example, the user equipment speed is 72 km / h (20 m / s). The intelligent reflective surface is deployed on a roadside lamppost and contains 64 reflective units. System parameters: carrier frequency 3.5 GHz, bandwidth 20 MHz, sampling period T = 10 ms. Simulation operation is as follows: At time t=0, the user is located 50m away from RIS, with an arrival angle of 30° and an estimated angle error of <1°.

[0159] After five updates, the position tracking error of the Kalman filter converged to 0.3m.

[0160] The LSTM model takes the motion parameters and RSRP of the most recent 10 time slots as input and predicts the SNR for the next 30ms. The root mean square error between the predicted SNR and the actual SNR is 1.2dB, while the error of the traditional linear prediction is 3.5dB.

[0161] When the predicted SNR is 5dB below the minimum threshold, the power compensation ΔPcomp = 6dB is calculated, and a zero-forcing phase offset matrix is ​​generated. Simulations show that the main lobe gain of the beam formed at the predicted position is 12dB higher than that of the conventional random phase beam, and the interference direction suppression ratio is 28dB.

[0162] Meanwhile, based on the user's historical task patterns, a video analysis task is generated every 2 seconds, with a data volume of 2MB. The linear regression model prediction requires 1.5 CPU cores and 512MB of memory. Resource pre-allocation is completed 200ms before the task arrives, reducing the actual task execution latency from 150ms to 30ms.

[0163] Comparative experiment: Without this method, the probability of handover interruption is 12%; with this method, the interruption probability drops to 1.5%. Resource utilization increases from 55% to 82%.

[0164] Three comparison schemes were set up in the MATLAB and Python co-simulation platform: Scheme A (this application): Kalman spectroscopy, LSTM, zero-forcing beamforming, resource pre-allocation; Option B (No Prediction): RIS beamforming based solely on instantaneous CSI + MEC real-time allocation; Option C (Communication Prediction Only): LSTM channel prediction + beamforming, without MEC pre-allocation; Option D (prediction calculation only): linear regression resource pre-allocation, no channel prediction beam adjustment; The simulation results are shown in Table 1 below: Table 1 Comparison of Implementation Results of Schemes

[0165] This application significantly outperforms other comparative schemes in terms of communication quality and computational latency. For example, the interruption probability is reduced by 65% ​​compared to scheme C and by 87% compared to scheme D.

[0166] It is important to note that the constructions and arrangements of this application shown in several different exemplary embodiments are merely illustrative. Although only two embodiments are described in detail in this disclosure, those who consult this disclosure will readily understand that many modifications are possible without substantially departing from the novel teachings and advantages of the subject matter described in this application. These modifications may include, for example, changes in the size, dimensions, structure, shape, and proportions of various elements, as well as parameter values ​​(e.g., temperature, pressure, etc.), installation arrangements, the use of materials, colors, orientations, etc. For example, an element shown as integrally formed may be composed of multiple parts or elements, the position of elements may be inverted or otherwise altered, and the nature or number or position of discrete elements may be changed or altered. Therefore, all such modifications are intended to be included within the scope of this application. The order or sequence of any process or method steps may be changed or rearranged by alternative embodiments. Any "apparatus plus function" clause is intended to cover, and not only structurally equivalent but also equivalent structures, the structures performing the functions described herein. Other substitutions, modifications, alterations, and omissions may be made in the design, operation, and arrangement of the exemplary embodiments without departing from the scope of this application. Therefore, this application is not limited to a particular embodiment, but extends to various modifications that still fall within the scope of the appended claims.

[0167] Furthermore, in order to provide a concise description of exemplary embodiments, not all features of actual embodiments (i.e., those features that are not relevant to the best mode of performing this application as currently considered, or those features that are not relevant to implementing this application) may be omitted.

[0168] It should be understood that numerous specific implementation decisions can be made during the development of any practical implementation, such as in any engineering or design project. Such development efforts may be complex and time-consuming, but for those of ordinary skill in the art who benefit from this disclosure, the development effort will be a routine task in design, manufacturing, and production without requiring extensive experimentation.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application, and all such modifications and substitutions should be covered within the scope of the claims of this application.

Claims

1. A resource optimization method for an intelligent reflective surface MEC system, characterized in that, include: S1 receives the incident signal from the user equipment, estimates the signal angle of arrival from the covariance matrix of the incident signal, and analyzes the Doppler frequency shift of the incident signal. S2 inputs the signal angle of arrival and Doppler frequency shift into a Kalman filter to track the motion trajectory and speed changes of the user equipment, and establishes a motion state model of the user equipment. The motion state model and real-time channel state information are then input into a signal quality detection model, and the future time slot signal-to-noise ratio is predicted through the signal quality detection model. S3 compares the predicted channel quality value with the channel rating value to obtain the signal attenuation level, calculates the required power compensation, and generates a phase offset matrix for the smart reflector to form an enhanced beam at the predicted position. When S4 predicts that a user device will generate a computing task in a future time slot, it pre-allocates computing resource blocks in the edge server and loads the task-related context data into the memory area.

2. The method for optimizing MEC system resources using a smart reflective surface as described in claim 1, characterized in that: The estimated angle of arrival of the incident signal in the covariance matrix includes: The covariance matrix is ​​constructed by the incident signals received by each element of the intelligent reflective surface array; The covariance matrix is ​​decomposed into eigenvalues, and the eigenvectors are divided into signal subspace and noise subspace. The spatial spectral function is calculated using a spectral peak search algorithm by taking advantage of the orthogonality between the signal subspace and the noise subspace. The angle of arrival of the incident signal is determined by detecting the peak position of the spatial spectral function.

3. The MEC system resource optimization method for a smart reflective surface as described in claim 2, characterized in that: The analysis of the Doppler frequency shift of the incident signal includes: The received baseband signal is sampled in the time domain to obtain a signal sequence; The signal sequence is transformed from the time domain to the frequency domain to obtain the frequency domain representation of the signal; The actual offset of the carrier frequency is calculated by detecting the peak offset in the frequency domain representation; The radial velocity of the user equipment is obtained by using the physical relationship between the carrier frequency offset and the motion speed.

4. The method for optimizing MEC system resources using a smart reflective surface as described in claim 3, characterized in that: The process of tracking the motion trajectory and speed changes of user equipment and establishing a motion state model of user equipment includes: constructing a state vector containing position, velocity, and acceleration based on the signal angle of arrival and Doppler frequency shift; The motion state at the next moment is predicted using the kinematic equations through the state transition matrix in the Kalman filter. The measured signal parameters and the predicted state are fused using the observation matrix, and the optimal state estimate is updated using Kalman gain. A motion state model containing the relationship between position and velocity is established by continuously updating the state estimation results.

5. The method for optimizing MEC system resources using a smart reflective surface as described in claim 4, characterized in that: The motion state model and the real-time channel state information input signal quality detection model predict the future time slot signal-to-noise ratio, including: The motion parameters of the motion state model and the reference signal received power in the real-time channel state information are used together as input features; By using a prediction model based on a long short-term memory network, the spatiotemporal correlation between motion state and channel quality is learned; Using a trained model, the signal-to-noise ratio (SNR) trend of future time slots can be predicted based on the current motion state and channel conditions. The signal-to-noise ratio data sources include the reference signal received power measured by the cell reference signal, and the signal-to-interference-plus-noise ratio measured by the demodulated reference signal.

6. The method for optimizing MEC system resources using a smart reflective surface as described in claim 5, characterized in that: The step of obtaining the signal attenuation level by comparing the predicted channel quality value with the channel rating value includes: The predicted signal-to-noise ratio is compared with the minimum signal-to-noise ratio threshold required for communication quality. The difference between the predicted signal-to-noise ratio and the minimum signal-to-noise ratio threshold is calculated as a quantitative indicator of the degree of signal attenuation. The signal-to-noise ratio margin required to maintain reliable communication is determined by the signal-to-noise ratio requirements corresponding to the channel coding and modulation methods. The severity of signal attenuation is assessed using the difference and signal-to-noise ratio margin.

7. The method for optimizing MEC system resources using a smart reflective surface as described in claim 6, characterized in that: The calculation of the required power compensation amount and the generation of the phase offset matrix for the smart reflector include: Based on the degree of signal attenuation, the additional power compensation required to achieve the target channel capacity is calculated by back-calculating using Shannon's formula. By combining the channel matrix between the smart reflector and the user equipment, the phase configuration required for beamforming is calculated. By optimizing the algorithm, the intelligent reflective array forms a phase shift matrix that enables energy focusing at the predicted position.

8. The MEC system resource optimization method for a smart reflective surface as described in claim 7, characterized in that: The phase offset matrix is ​​calculated using a zero-forcing beam, and forming an enhanced beam at the predicted location includes: Obtain channel state information from the intelligent reflector to the predicted location; Construct a system channel matrix that includes all interfering channels; The phase offset matrix is ​​calculated by using a null-forcing beam to create a main lobe at the predicted location while creating nulls in other directions. The phase shift of each reflector element of the intelligent reflector is configured using the calculated phase shift matrix, forming an enhanced beam with concentrated energy at the predicted position.

9. The method for optimizing MEC system resources using a smart reflective surface as described in claim 1, characterized in that: The step of pre-allocating computing resource blocks in the edge server and loading task-related context data into the memory region includes: A resource demand prediction model is built from the historical task execution records of user equipment to predict the required computing resource scale based on task type and data processing volume. In the resource manager of the edge server, separate resource isolation areas are divided, and computing resource blocks containing CPU cores and memory are pre-allocated according to the predicted computing resource requirements. Based on the service characteristics and task types of user devices, relevant application context, model parameters and configuration data are prefetched from the edge storage system to the resource isolation area; Establish a resource reservation mapping table to record the binding relationship between pre-allocated resource blocks and corresponding user devices and task types, ensuring that tasks can be scheduled and executed immediately upon arrival.

10. A resource optimization system for an intelligent reflective surface MEC system, characterized in that... Includes a resource optimization method for an MEC system with a smart reflective surface as described in any one of claims 1-9; wherein: The signal sensing and processing module includes a signal parameter extraction unit and a motion state unit. It receives the incident signal from the user equipment through an intelligent reflective array, estimates the signal angle of arrival from the covariance matrix of the incident signal, and analyzes the Doppler frequency shift of the incident signal. The motion state unit processes the signal angle of arrival and Doppler frequency shift to track the motion trajectory and speed changes of the user equipment and establish a motion state model of the user equipment. The resource prediction and decision-making module includes a channel quality prediction unit and a resource pre-allocation unit. The channel quality prediction unit inputs the motion state model and real-time channel state information into the signal quality detection model to predict the future time slot signal-to-noise ratio. The resource pre-allocation unit obtains the signal attenuation degree by comparing the channel quality prediction value with the channel rating value, calculates the power compensation amount, and generates a smart reflector phase offset matrix configuration instruction. The resource configuration execution module includes a communication configuration unit and a resource allocation unit. The communication configuration unit calculates the phase offset matrix to predict the position and form an enhanced beam according to the phase offset matrix configuration instruction. When the resource pre-allocation unit predicts that the user equipment will generate a computing task in a future time slot, it pre-allocates computing resource blocks in the edge server and loads the task-related context data into the memory area.

Images