Intelligent metasurface electromagnetic wave regulation and control system for brain-computer interface

Abstract

The invention belongs to the technical field of brain-computer interfaces, and particularly relates to an intelligent metasurface electromagnetic wave regulation and control system for a brain-computer interface. The system comprises an electroencephalogram / electro-oculogram signal acquisition module, an electroencephalogram signal preprocessing module, a motor imagery electroencephalogram signal decoding module, a blink detection and mode coding module, a mode selection and electromagnetic function mapping module, a programmable logic control module and a dual-polarization reconfigurable metasurface array. Wherein EEG and EOG signals are fused, and an EOG mode selection + EEG fine adjustment hierarchical control logic is constructed; the left and right hand motor imagery is decoded through the multi-scale filter bank convolutional neural network, and the classification performance under limited training samples is improved; a dual-polarization 2-bit reconfigurable reflection metasurface is utilized to realize electromagnetic functions of polarization-independent beam deflection, joint beam deflection and polarization conversion under 45-degree linear polarization incidence, RCS reduction and the like; the security and robustness of the system are improved through a softmax confidence threshold mechanism; the method can be used for various man-machine interaction scenes.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interface technology, specifically relating to an intelligent metasurface electromagnetic wave modulation system for brain-computer interfaces. Background Technology

[0002] Brain-computer interfaces (BCIs) acquire brain activity signals and decode them into control commands that external devices can understand, providing new solutions for applications such as assisted control and human-computer interaction. EEG signals occupy a significant position among various EEG signals due to their advantages of low cost, non-invasiveness, and relatively low technical difficulty. These signals mainly include EEG paradigms such as motor imagery (MI), event-related potentials (P300), and steady-state visual evoked potentials (SSVEP). Acquired EEG signals can be accurately decoded into corresponding external machine control commands using BCI technologies based on P300, SSVEP, or motor imagery, enabling deep interaction between the human brain and machines. Among these, non-invasive motor imagery-based BCIs are widely studied due to their high safety and reproducibility.

[0003] On the other hand, reconfigurable metasurfaces, by integrating tunable elements at the subwavelength scale, can flexibly control the reflection, transmission direction, polarization state, and scattering characteristics of electromagnetic waves. Existing research has proposed 1-bit, 2-bit, and even 3-bit coded metasurfaces based on PIN diodes, varactor diodes, or CMOS switches, achieving functions such as beam scanning, polarization conversion, and radar cross section (RCS) reduction. However, these systems typically rely on external controllers and manual operation, limiting their application to limited, predefined scenarios and making it difficult to establish a direct, real-time mapping relationship with human neural activity.

[0004] Existing work has attempted to introduce human biosignals to drive reconfigurable electromagnetic devices, such as a fusion system of SSVEP visual stimulus encoding and metasurface spatiotemporal encoding, programmable metasurfaces controlled by P300 EEG signals, and gesture-controlled metasurfaces based on surface electromyography. However, there is still a lack of methods for combining non-invasive motor imagery-based BCIs with adjustable metasurfaces, and existing systems suffer from problems such as limited control signal types, limited mapping between multifunctional electromagnetic modulation and human intention, and limited functionality. Furthermore, EOG signals, as a direct reflection of eye movements, have the characteristics of large amplitude and easy detection, making them suitable for mode selection or confirmation commands. However, existing EOG-based control methods mostly use them alone for simple on / off control, rarely integrating them deeply with MI-BCIs to construct a multi-level control structure of "mode selection + fine adjustment," and have not been combined with 2-bit reconfigurable metasurfaces to achieve multifunctional electromagnetic wave modulation. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent metasurface electromagnetic wave control system for brain-computer interfaces, so as to overcome the problems of insufficient control freedom, single control signal, limited support for complex electromagnetic functions and low system robustness in the existing brain-computer interface, and meet the needs of more application scenarios.

[0006] The intelligent metasurface electromagnetic wave modulation system for brain-computer interfaces provided by this invention is based on a dual-polarized reconfigurable metasurface controlled by a mixture of EEG and EEG signals, and controls electromagnetic waves in real time. It includes: (1) fusing non-invasive EEG and EOG signals to construct a hierarchical control logic of "EOG mode selection + EEG fine adjustment"; (2) robustly decoding left and right hand motion imagination through a multi-scale filter bank convolutional neural network to improve classification performance under limited training samples; (3) using a dual-polarized 2-bit reconfigurable reflective metasurface to realize three typical electromagnetic functions: polarization-independent beam deflection, joint beam deflection and polarization conversion under 45° linear polarization incident, and RCS reduction; (4) introducing an implicit "no-command" state through a softmax confidence threshold mechanism to improve the system's security and robustness when the signal quality is poor or the user has no clear intention.

[0007] Specifically, the intelligent metasurface electromagnetic wave modulation system for brain-computer interfaces includes:

[0008] The EEG / EOG signal acquisition module is used to acquire electroencephalography (EEG) signals by using multi-channel electrodes placed in the sensorimotor cortex of the subject, and to acquire electrooculography (EOG) signals by using electrodes placed in the forehead region.

[0009] The EEG signal preprocessing module is used to perform bandpass filtering, sampling, and segmentation processing on the EEG signal to obtain multi-channel time series data for decoding;

[0010] The motor imagery EEG signal decoding module is a classifier based on a multi-scale convolutional neural network. It is used to receive preprocessed EEG time series, extract time-frequency features, and output motor imagery categories and their classification confidence scores, including at least left-hand motor imagery and right-hand motor imagery.

[0011] The blink detection and pattern encoding module is used to detect the amplitude and timing features of the EOG signal within a preset time window, identify three intentions: blinking with the left eye, blinking with the right eye, and blinking with both eyes simultaneously, and encode the recognition results into a pattern selection signal.

[0012] The mode selection and electromagnetic function mapping module is used to select a preset metasurface working mode according to the mode selection signal and the motion imagination category, and generate a phase encoding matrix corresponding to the target working mode.

[0013] The programmable logic control module, which is a field programmable gate array (FPGA), is used to convert the phase encoding matrix into multiple DC bias control signals;

[0014] Dual-polarized reconfigurable metasurface arrays, including 8 arrays arranged in a rectangular array 8 Reconfigurable reflection units; each reflection unit includes: two independent 2-bit reflection phase shifting networks based on 6 PIN diodes, each acting on two orthogonal line polarization components. The two phase shifters are connected to four bias lines. Each reflection unit is controlled by an independent DC voltage to obtain different phase distributions on the array surface, thereby realizing real-time control of the beam direction, polarization state and scattering characteristics of the reflected electromagnetic wave.

[0015] Furthermore:

[0016] In the EEG / EOG signal acquisition module, the EEG electrodes located near the sensorimotor cortex are C3, Cz, C4, FC3, FC4, CP3, and CP4, used to acquire EEG signals related to upper limb motor imagery; the EOG electrodes located in the forehead region are Fp1 and Fp2, used to acquire blink signals in the horizontal and vertical directions; the ground electrode is placed at the earlobe; the sampling rate of the EEG and EOG signals is 1000 Hz, and they are transmitted to the host computer for processing via a wireless transmission module.

[0017] In the EEG signal preprocessing module, a 2-second imagery time window is provided for each motor imagery task. To reduce disturbances at the beginning and end, only a continuous 1-second interval is extracted as input to the motor imagery decoding module. Therefore, each trial is converted into a data matrix of size 8×1000. Subsequently, a bandpass filter of 8-30 Hz is applied to each channel signal, and a 50 Hz notch filter is superimposed to preserve the rhythm associated with motor imagery and suppress low-frequency drift and power line interference. After re-referencing with a common average reference, trials containing obvious blinking or large-amplitude eye movement artifacts are eliminated by combining the electrooculogram information of the Fp1 and Fp2 channels. Finally, the EEG signals of each channel in each trial are normalized according to a preset rule, and scaled according to the maximum absolute amplitude of the channel in that trial, so that the normalized data mainly falls within the range of amplitudes. The numerical range is then input into the multi-scale convolutional decoding network in the form of an 8×1000 matrix.

[0018] The motion image decoding module is a multiscale filterbank convolutional neural network (MSFBCNN), comprising:

[0019] The input layer receives multi-channel EEG data segmented by trial number. The data structure for a single trial is as follows: ,in For the number of channels, The number of sampling points is set, and it is configured to extract a predetermined length of EEG segment from the middle of the time window of each motion visualization task as network input;

[0020] The multi-scale temporal convolutional feature extraction layer contains several parallel one-dimensional temporal convolutional branches. Each branch uses convolutional kernels of different lengths in the temporal dimension and convolutional kernels with a kernel length of 1 in the channel dimension. It performs multi-scale filtering on the same EEG segment and splices the feature maps output by each branch in the filter dimension to obtain a fused temporal feature representation.

[0021] The spatial convolutional layer includes several spatial filters. Its convolutional kernel has a length in the channel dimension equal to the number of EEG channels and a length in the time dimension of 1. It is used to perform spatial filtering on the time features in the channel dimension, compress the channel dimension to 1, and extract the spatial pattern across channels.

[0022] Regularization and nonlinear layers, including batch normalization layers, square nonlinear layers, average pooling layers, logarithmic nonlinear layers, and dropout layers with a predetermined dropout rate, are used to construct robust features related to bandwidth power and suppress network overfitting.

[0023] The classification layer maps the pooled and regularized features to the output class space, outputting the softmax probability and confidence of at least the left-handed and right-handed motor imagery.

[0024] The blink detection and pattern encoding module is used to extract a first time window of duration, starting from the command time, from the EOG signal each time a mode switching command arrives, and to detect the peak amplitude, polarity, and occurrence time of the EOG waveform within this time window, wherein:

[0025] When a single-peak waveform with an amplitude exceeding the first threshold and within a preset time range appears in the EOG channel corresponding only to the Fp1 electrode within the time window, the blink is identified as a left eye blink and encoded as a first mode selection signal.

[0026] When a single-peak waveform with an amplitude exceeding the second threshold and within a preset time range appears in the EOG channel corresponding only to the Fp2 electrode within the time window, the blink is identified as a right eye blink and encoded as a second mode selection signal.

[0027] When the amplitude of the two channels corresponding to the Fp1 and Fp2 electrodes exceeds the third threshold within the time window and the time interval between their peak occurrences is less than the predetermined time threshold, the blink is identified as a binocular blink and encoded as a third mode selection signal.

[0028] The first duration is preferably 400-800 ms.

[0029] The mode selection and electromagnetic function mapping module is configured to enable the system to have at least the following three working modes: (1) In the first working mode, the phase codes of the two orthogonal polarization channels are kept consistent to achieve polarization-independent beam deflection; (2) In the second working mode, under the preset 45° linear polarization incident condition, phase codes for left-hand circular polarization reflection beam deflection +20° and right-hand circular polarization reflection beam deflection +20° are generated according to the motion imagination category to achieve joint beam deflection and polarization conversion functions; (3) In the third working mode, phase codes randomly distributed on the array surface are generated to expand the scattering direction distribution to reduce the radar cross section (RCS).

[0030] The programmable logic control module, with a field-programmable gate array (FPGA) at its core, along with memory, voltage conversion circuitry, and multiple general-purpose input / output (GPIO) interfaces, forms a configurable 256-channel digital bias control platform. The input side connects to a host computer via a standard USB serial interface. The host computer generates the corresponding metasurface operating mode code based on the mode selection signal and phase encoding index output from the EEG decoding module, and sends it to the FPGA control core in the form of serial data frames. The FPGA internally stores multiple sets of phase encoding matrices corresponding to each operating mode. Each matrix encodes a 256-word control word, corresponding to the four bias lines of the 8×8 unit in the metasurface array. After receiving the mode index from the host computer, the FPGA reads the target control word from the code table and refreshes each GPIO output bit by bit, causing the on / off states of all PIN diodes to switch synchronously within one clock cycle, thereby achieving a comprehensive update of the phase state of the metasurface units. In terms of hardware structure, the programmable logic control module provides four 68-pin connectors. In each connector, besides four fixed ground terminals, the remaining 64 pins are used as programmable GPIOs, totaling 256 independent bias outputs. Each GPIO outputs a DC bias voltage of approximately ±2.5 V through voltage conversion and current limiting circuitry, adapting to the on / off requirements of the PIN diode phase shifters in the metasurface unit of this invention, and capable of driving the metasurface array. The module is powered by a wide-range DC input of 9–36 V, with internally integrated voltage regulation and positive / negative power generation circuits. The total power consumption of the board is approximately tens of watts, suitable for long-term stable operation.

[0031] The dual-polarized reconfigurable metasurface array is a two-dimensional array consisting of 2-bit reconfigurable metasurface units operating in the microwave frequency band, and a DC bias network consisting of bias lines of each metasurface unit; wherein:

[0032] The metasurface unit consists of three metal layers and two dielectric layers; the top metal layer is used to receive and radiate electromagnetic waves, the middle metal layer serves as a ground layer, and the bottom metal layer serves as a phase shifter to achieve 2-bit reconfigurable functionality; the two dielectric layers are located between the top metal layer and the middle metal layer, and between the middle metal layer and the bottom metal layer, respectively.

[0033] A radiating patch located above the first dielectric substrate is used to receive and radiate electromagnetic waves;

[0034] The bottom metal layer, located beneath the second dielectric substrate, consists of two identical reflective phase shifters for independently adjusting the reflection phase in the x and y polarization directions. Each reflective phase shifter comprises a transmission line with a pair of vertical stubs, a fan-shaped open-circuit stub, and a horizontal stub. Two diodes, PIN1 and PIN2, are connected between the horizontal stub and the vertical stub of the transmission line, respectively, while another diode, PIN3, is connected to the fan-shaped open-circuit stub via the transmission line.

[0035] An intermediate metal layer is disposed between the first and second dielectric substrates, serving as a shared ground plane for the top metal patch and the bottom reflective phase shifter. This intermediate metal layer separates the radiating portion from the bias lines to reduce the impact of dense bias lines in the array on radiation performance. Two metal vias are designed between the three metal layers for connection: via 1 connects the top layer and phase shifter 1, controlling the reflection phase of the incident wave in the y-polarization direction; via 2 connects the top layer and phase shifter 2, controlling the reflection phase of the incident wave in the x-polarization direction. A circular hole is drilled at the locations of vias 1 and 2 in the intermediate layer ground plane to prevent short circuits between the vias and the ground plane.

[0036] The 2-bit independently adjustable unit can independently control four reflection phase states: 0°, 90°, 180°, and 270°, respectively, in two orthogonal linear polarization directions. Each reflection unit in the array is independently addressed through four bias lines, forming a total of 256 bias control channels in the 8×8 array. By setting different phase encoding matrices, the array can achieve functions such as polarization-independent beam deflection, polarization conversion and beam deflection control of reflected waves under 45° linear polarization incident waves, and RCS reduction.

[0037] The electromagnetic wave control process of the intelligent metasurface electromagnetic wave control system of the present invention is as follows:

[0038] (1) The electrooculogram (EOG) signal of the subject and the electroencephalogram (EEG) signal under the motor imagery task were acquired by the multi-channel EEG electrode and the frontal EOG electrode in the EEG / EOG signal acquisition module, and digitally acquired at the first sampling rate.

[0039] (2) The EEG signal is bandpass filtered, downsampled and segmented by trial through the EEG signal preprocessing module, and the multi-channel EEG segment corresponding to each trial is used as the input sample for decoding the motor imagery signal;

[0040] (3) Input the processed EEG fragments into the multi-scale convolutional neural network model in the motor imagery EEG signal decoding module, extract time-frequency features and output the corresponding motor imagery category and classification confidence;

[0041] (4) Through the blink detection and pattern encoding module, a time window of a preset length is extracted from the EOG signal, the peak amplitude, polarity and occurrence time of the EOG waveform within the time window are calculated, and the left eye blink, the right eye blink or both eyes blink are determined accordingly, and the determination result is encoded into a pattern selection signal.

[0042] (5) Through the mode selection and electromagnetic function mapping module, a preset metasurface working mode is selected according to the mode selection signal and the motion imagination category and its confidence level, and a corresponding phase encoding matrix is ​​generated;

[0043] (6) The phase encoding matrix is ​​converted into multiple DC bias control signals by the programmable logic control module FPGA and loaded onto the bias ports of each reflective unit of the reconfigurable metasurface array to change the working state of each unit diode;

[0044] (7) By using a dual-polarized reconfigurable metasurface array, the working state of the metasurface reflection unit can be adjusted under the corresponding electromagnetic wave incident, so as to realize the functions of beam deflection, polarization conversion and RCS reduction in the preset working mode.

[0045] The preset working modes include at least the following:

[0046] Mode 1: The electrooculogram signal corresponds to the left eye blinking, which is functional mode 1. Under different linear polarization incident conditions, the phase encoding of the two orthogonal polarization channels remains consistent to achieve polarization-independent beam deflection. Specifically, if the motion imagery category is left-hand motion imagery, the corresponding main beam of the reflected wave deflects to the target deflection angle; if the motion imagery category is right-hand motion imagery, the corresponding main beam of the reflected wave deflects to another target deflection angle.

[0047] Second mode: The electrooculogram signal corresponds to the second functional mode of right eye blinking. Under 45° linear polarization incident conditions, if the motion imagery category is left-hand motion imagery, the reflected wave is deflected to the target angle and converted into a left-hand circularly polarized beam; if the motion imagery category is right-hand motion imagery, the reflected wave is deflected to the target angle and converted into a right-hand circularly polarized beam.

[0048] The third mode: The electrooculogram signal corresponds to the third functional mode of binocular blinking. A phase code with a random spatial distribution is generated to control the metasurface to achieve RCS reduction within the target frequency band.

[0049] The motion imagery decoding module, while outputting the softmax probability, compares the maximum class probability with a preset confidence threshold. When the maximum probability is less than the confidence threshold, it does not perform an encoding update operation based on the current motion imagery output. Instead, it maintains the phase encoding matrix corresponding to the previous valid command, thus preserving the current electromagnetic function state when the user has no explicit motion imagery or the signal quality is low. In other words, the output is considered an invalid command. In this case, the mode selection and electromagnetic function mapping module maintains the previous frame's metasurface phase encoding unchanged, achieving an implicit "command-free" state and improving system robustness.

[0050] In summary, this invention acquires motor imagery EEG signals from multi-channel electrodes in the sensorimotor cortex and blinking EEG signals from frontal electrodes. The EEG signals are filtered and segmented before being input into a multi-scale convolutional neural network to classify left and right hand motor imagery. The EEG signals are then subjected to amplitude and temporal threshold detection within a preset time window to identify three blinking patterns: left eye, right eye, and both eyes. Based on the EEG classification results and blinking patterns, control logic maps and generates phase encoding matrices corresponding to the three operating modes. An external field FPGA applies DC control voltage to each unit of an 8×8 dual-polarized 2-bit reconfigurable reflective metasurface array, enabling independent unit-level control. This allows the metasurface to perform polarization-independent beam deflection, polarization conversion and beam deflection control under 45° linear polarization incidence, and RCS reduction. This invention achieves a direct mapping from EEG / EEG signals to multifunctional electromagnetic wave modulation, offering advantages such as high control freedom, fast response speed, universal structure, easy integration, and low loss. It can be applied to various scenarios such as brain-controlled electromagnetic systems, intelligent wireless communication, and immersive human-computer interaction.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) Multimodal control and hierarchical mapping;

[0053] This invention introduces non-invasive EEG and EOG signals simultaneously in a single system for the first time. It enables the selection of three metasurface working modes through a blinking mode, and then uses motion-imagination EEG to finely adjust the electromagnetic functions within the modes, constructing a hierarchical mapping structure that significantly improves the degree of control freedom.

[0054] (2) A dedicated MSFBCNN structure for MI-EEG characteristics;

[0055] This invention employs a multi-scale filter bank convolutional neural network to decode MI-EEG. Through parallel one-dimensional temporal convolution branches and subsequent spatial convolutions, as well as a square-mean-log nonlinear structure, it effectively extracts temporal and spatial features related to frequency band power. Even with limited training samples and noisy backgrounds, it can still maintain high classification accuracy and robustness.

[0056] (3) Dual-polarized 2-bit reconfigurable metasurface multifunctional

[0057] The metasurface unit of this invention can achieve 2-bit discrete phase control in two orthogonal linear polarization directions. Combined with an 8×8 array and 256 bias channels, it can realize various electromagnetic functions such as polarization-independent beam deflection, beam deflection and polarization conversion joint control under 45° linear polarized wave incidence, and random phase RCS reduction on the hardware platform, thereby reducing the complexity of the multi-functional requirements of metasurface hardware.

[0058] (4) Control logic that balances real-time performance and security

[0059] By using EOG-triggered time window detection and a softmax confidence threshold mechanism, this invention ensures that no erroneous metasurface state switching occurs when the user has no clear intention or the signal quality is poor. Instead, the current electromagnetic function remains unchanged, thereby reducing the risk of false triggering and improving the practicality and safety of the overall system.

[0060] (5) Hardware structure versatility and ease of integration;

[0061] The EEG / EOG acquisition module, FPGA control module, and dual-polarized 2-bit metasurface structure used in this invention can all be implemented on existing mature hardware platforms. The system architecture is clear and easy to integrate with existing wireless communication, radar, or sensing systems, providing a unified platform for the expansion of brain-controlled electromagnetic functions and providing innovative solutions for special scenarios such as industrial automation. Attached Figure Description

[0062] Figure 1 This is a block diagram of the overall structure of the intelligent metasurface electromagnetic wave control system of the present invention.

[0063] Figure 2 This is a diagram showing the distribution of EEG / EOG signal acquisition electrodes in this invention.

[0064] Figure 3 This is a schematic diagram of the time series of a single EEG signal acquisition test according to the present invention.

[0065] Figure 4 This is a schematic diagram of the multi-scale filter bank convolutional neural network structure used in the motion imagination decoding module of this invention.

[0066] Figure 5 This is a schematic diagram of the structure and phase-shifting structure of the dual-polarized 2-bit reconfigurable metasurface unit of the present invention.

[0067] Figure 6 For the present invention 8 Diagram of an 8-bit dual-polarized 2-bit reconfigurable metasurface array and a schematic diagram of the bias network connection relationship.

[0068] Figure 7 This is a flowchart of the reconfigurable metasurface signal processing and control based on the hybrid control of EEG and EEG signals of this invention.

[0069] Figure 8 This is a schematic diagram of the phase encoding matrix and an electromagnetic performance test diagram of the present invention in one working mode.

[0070] Figure 9 This is a schematic diagram of the phase encoding matrix and an electromagnetic performance test diagram of the present invention in the second working mode.

[0071] Figure 10 This is a schematic diagram of the phase encoding matrix and an electromagnetic performance test diagram of the present invention in the third operating mode.

[0072] Figure 11 This is a schematic diagram of the latency test of the present invention in a real system.

[0073] The diagram is labeled as follows: 1 is the top metal patch of the metasurface unit; 2 is the first dielectric layer; 3 is the middle metal ground plane; 4 is the second dielectric layer; 5 is the first reflective phase shifter; 6 is the second reflective phase shifter; 7 is the first via connecting the top layer and the first phase shifter; 8 is the second via connecting the top layer and the second phase shifter; 9 is a schematic diagram of the overall structure of the first phase shifter; 10 is the main transmission line with two vertical stubs; 11 is a horizontal stub; 12 is a rectangular patch; 13 are two PIN diodes between the vertical and horizontal stubs; 14 is a fan-shaped open-circuit stub; 15 is a PIN diode between the fan-shaped patch and the transmission line; 16 is the via of the first phase shifter; 17 is the second phase shifter, which has the same structure as the first phase shifter; 18 is a two-dimensional array composed of 2-bit reconfigurable metasurface units; 19 is a phase shift array at the bottom of the metasurface; 20 is a DC bias network composed of bias lines; and 21 is the pin header interface for connecting to the FPGA. Detailed Implementation

[0074] The present invention will be further described below with reference to the accompanying drawings and embodiments; this embodiment provides detailed implementation methods and specific operation processes, but the scope of protection of the present invention is not limited to the following embodiments.

[0075] I. System Overall Structure

[0076] like Figure 1 As shown, the dual-polarized reconfigurable metasurface real-time electromagnetic wave modulation system based on hybrid control of EEG and EEG signals of the present invention generally includes an EOG / EEG signal acquisition part, a mode selection and control logic part, and a 2-bit dual-polarized reconfigurable metasurface electromagnetic modulation part.

[0077] 1. EEG / EOG signal acquisition: EEG signals of motor imagery were acquired using multi-channel electrodes, and blink signals were acquired using forehead EOG electrodes. A 16-channel bioelectrical signal acquisition board (OpenBCI CytonDaisy) with an onboard 32-bit processor and a sampling rate of 1000 Hz was used. Seven channels were used for acquiring motor imagery EEG, two channels were used for acquiring EOG, and the remaining channels were reserved for future expansion. Signal preprocessing was performed by a host computer.

[0078] 2. Control Logic Section: Multi-scale Filter Bank Convolutional Neural Network (MSFBCNN) is used to classify motion imagery EEG, and threshold detection is performed on EOG signals to identify three modes: left eye blink, right eye blink, and both eyes blink; the two are combined to determine and select the corresponding mode and transmit the phase encoding matrix to the FPGA.

[0079] 3. 2-bit Dual-Polarized Reconfigurable Metasurface Electromagnetic Control Section: Composed of an 8×8 dual-polarized 2-bit reconfigurable reflective metasurface array and its bias network, the FPGA converts the two-dimensional phase encoding matrix into multiple DC bias control signals, which are output to the bias ports of the reconfigurable metasurface array to achieve addressing control of the reflection phase state of each unit. This enables various electromagnetic functions such as reflected beam deflection, polarization conversion, and RCS reduction.

[0080] II. EEG / EOG Signal Acquisition and Experimental Paradigm

[0081] like Figure 2 As shown, this invention uses a low-cost OpenBCI CytonDaisy biosignal acquisition board as the front-end hardware, with an onboard 32-bit processor and 16 analog input channels. The sampling rate is set to 1000 Hz, and the signal is wirelessly transmitted to the host computer via a Wi-Fi module. EEG electrodes are arranged in the sensorimotor cortex and its surrounding area according to the international 10-20 system, with a focus on channels C3, C2, C4, FC3, FC4, CP3, and CP4 for upper limb motor imagery detection. Ground electrodes are placed on the earlobe, and EOG electrodes are placed at Fp1 and Fp2 positions to simultaneously sense horizontal and vertical eye movements. This allows for the simultaneous acquisition of motor imagery EEG signals and blink signals without adding additional hardware.

[0082] like Figure 3 As shown, the single-trial procedure in this embodiment includes: 0-2 seconds is the fixation cross phase, where the subject stares at the white cross in the center of the screen; at 2 seconds, an audio cue is played; from 2-3 seconds, a cue image with directional arrows is presented, indicating the left or right hand motor imagery task; from 3-5 seconds is the motor imagery phase, where the subject performs kinesthetic imagery of the corresponding hand movements according to the cue, such as clenching and unclenching a fist; from 5-7 seconds is the feedback phase, where the screen displays the current system decoding result; from 7-8 seconds, the screen goes black for a short rest. To ensure stable EEG data, the subject remains seated during the experiment, minimizing body movement. During the BCI calibration phase, each subject completes several sets of trials to collect training data. This experiment selected three subjects. Taking one subject as an example, seven sessions were designed, each session containing 12 left and right hand motor imagery exercises. The first six sessions (72 exercises / class) were used to train the MSFBCNN model, and the last session (12 exercises / class) was used for offline experimental testing.

[0083] III. EEG Preprocessing and Multi-Scale Convolutional Decoding Network

[0084] 1. EEG signal preprocessing

[0085] The EEG signal preprocessing module performs the following steps on the raw EEG data:

[0086] Data Acquisition and Trial Division: In each motor imagery task, subjects were required to complete a 2-second motor imagery time window according to a pre-defined paradigm. The sampling frequency was 1000 Hz, and EEG signals were simultaneously acquired from eight electrode channels covering the sensorimotor cortex. To mitigate the impact of attentional fluctuations during the initiation and tail phases, a continuous 1-second segment of signal was extracted from each 2-second motor imagery time window as the decoding input. After extraction, each trial corresponded to an 8×1000 data matrix, where 8 represents the number of channels and 1000 represents the number of sampling points within 1 second.

[0087] Bandpass filtering and power frequency suppression: To preserve the μ and β rhythmic components related to motion imagery and suppress low-frequency drift and high-frequency noise, bandpass filtering is applied to the EEG signals of each channel. The passband of the bandpass filter is selected in the range of 4-40 Hz, and combined with a power frequency notch filter to suppress power frequency interference components, thereby improving the signal-to-noise ratio and highlighting the rhythmic activities related to motion imagery.

[0088] Rereference and Normalization: To reduce the impact of the reference electrode position on the waveform morphology, a common average reference method can be used to rereference the multi-channel EEG. Subsequently, amplitude normalization is performed on each channel to ensure that the data from each channel falls within a uniform numerical range, facilitating convolutional network training and convergence. After the above processing, each trial EEG signal is represented as a normalized 8×1000 real-valued matrix.

[0089] 2. Multi-scale filter bank convolutional decoding network structure

[0090] like Figure 4 As shown, based on the preprocessed EEG trial data described above, this invention constructs a multi-scale filter bank convolutional neural network. This network generally consists of three parts: a feature extraction layer, a regularization layer, and a classification layer.

[0091] Feature extraction layer: Treats the EEG data from a single trial as having a size of The two-dimensional input tensor, where For the number of channels, The number of time sampling points is specified. To simultaneously capture rhythmic features at different time scales, the feature extraction layer employs a multi-scale temporal convolution structure: several one-dimensional convolutional branches are arranged in parallel along the time dimension, each using a temporal convolutional kernel of different lengths. In a specific embodiment, the multi-scale temporal convolution module includes four parallel convolutional branches with kernel lengths of 64, 40, 26, and 16 sampling points, respectively, to characterize the μ / β rhythmic variations at different time scales. Each branch performs convolution operations on the 8-channel EEG data along the time axis, followed by batch normalization and a non-linear activation function to enhance the expressive power of the features and the training stability of the network. The feature maps output by each branch are concatenated along the channel dimension to achieve channel-level fusion of multi-scale features, thereby obtaining a comprehensive feature representation containing information from different time scales. If necessary, convolution or linear transformation along the channel direction can be added after the feature extraction layer to achieve weighted combination and spatial filtering of the 8 electrode channels, further improving separability.

[0092] Temporal pooling and regularization layers: To reduce feature dimensionality and enhance robustness to temporal shifts, a one-dimensional pooling operation (average pooling) is applied along the temporal dimension after the multi-scale convolutional output to downsample the responses within local time windows. Following pooling, a regularization layer is applied, including batch normalization and Dropout components, to mitigate overfitting and improve the network's generalization ability across different subjects. After feature extraction and regularization, a low-dimensional and highly discriminative feature tensor is obtained.

[0093] Classification Layer and Confidence Output: The aforementioned feature tensor is flattened and input into a fully connected layer. A Softmax activation function is used to output two categories: the posterior probability distributions of left-handed and right-handed motor imagery. In one implementation, each component of the Softmax output vector represents the confidence level of the current trial belonging to the corresponding category, and the maximum value is taken as the confidence index of the current decoding result. When the maximum confidence level is greater than or equal to a preset threshold, the corresponding category label is output as the EEG command input for subsequent reconstructable metasurface control modes; when the maximum confidence level is lower than the threshold, the current trial is judged as a "no command" state, and no control signal is sent to the metasurface, thereby effectively avoiding erroneous control caused by low-confidence misjudgment.

[0094] In one embodiment of the present invention, MSFBCNN is trained using an offline supervised learning method. First, EEG data of motion imagery completed by three subjects under a standard paradigm are collected, and each trial is labeled as a different motion imagery category according to the experimental labels. All trials are converted into an 8×1000 input matrix according to the aforementioned preprocessing procedure, forming the training set, validation set, and test set. During training, the network parameters are iteratively updated using a cross-entropy loss function and a gradient descent optimization algorithm until the validation set performance converges and reaches the preset accuracy requirement. In the online application stage, newly acquired EEG signals are converted into an 8×1000 input matrix according to the same preprocessing steps and fed into the pre-trained MSFBCNN model to obtain the corresponding category probabilities and confidence levels in real time. After confidence threshold determination, the decoding results are converted into discrete EEG control commands, which are combined with the EOG mode selection results to jointly drive the dual-polarized reconfigurable metasurface to achieve corresponding beam pointing and polarization state control.

[0095] The aforementioned EEG preprocessing and multi-scale convolutional decoding network module enables automatic feature extraction and robust classification of motor imagery EEG signals, providing a reliable upstream neural signal decoding foundation for the real-time electromagnetic wave modulation of this invention.

[0096] IV. Blink Detection and Pattern Selection Coding

[0097] This invention utilizes the EOG signal to achieve three-level working mode selection. When each mode switching command arrives, the blink detection module extracts a preset-length time window of 600 ms from the EOG signal of channels Fp1 and Fp2, starting from the command time, and analyzes the peak amplitude, polarity, and occurrence time of the waveform within the window.

[0098] If only the EOG channel corresponding to Fp1 has a single-peak waveform with an amplitude exceeding the first threshold and within the preset time range within the window, it is determined to be a left eye blink and the selection mode is determined to be mode one.

[0099] If only the channel corresponding to Fp2 has an amplitude exceeding the second threshold and meets the same timing conditions, it is determined to be a right eye blink and is determined to be mode two.

[0100] If both channels Fp1 and Fp2 show peak values ​​exceeding the third threshold within the window, and the time interval between the peak values ​​is less than the preset time threshold, then it is determined that both eyes are blinking simultaneously, and mode three is selected.

[0101] The preferred time window length is 400-800 ms, and the threshold value can be adaptively set based on the EOG amplitude of each subject through pre-experimentation. Building upon mode selection, this invention further introduces a classification confidence threshold to achieve a "command-free" state. The maximum probability output by the motion imagery decoding module is compared with a preset confidence threshold. When the maximum probability is lower than this threshold, the current motion imagery is considered an invalid command, and the system maintains the previous frame's metasurface phase encoding unchanged, thereby avoiding false triggering of electromagnetic functions when the signal-to-noise ratio is low or the subject is not focused.

[0102] V. Dual-polarized 2-bit reconfigurable metasurface structure and bias network

[0103] 1. Unit structure

[0104] like Figure 5 As shown in the left figure, the metasurface unit of this invention is a dual-polarization independently adjustable 2-bit reflective unit with a center frequency of 3.75 GHz. The overall structure employs a three-layer metal and two-layer dielectric stack. The top metal patch 1 is used to receive and radiate electromagnetic waves. The bottom metal layer consists of two identical reflective phase shifters, namely the first phase shifter 5 and the second phase shifter 6, used to independently adjust the reflection phase in the x and y polarization directions. The middle metal layer 3 serves as a shared ground plane for the top metal patch and the bottom reflective phase shifter, and it separates the radiating portion from the bias lines to reduce the impact of dense bias lines in the array on radiation performance. Each reflective phase shifter consists of a main transmission line 10 with a pair of vertical stubs, a fan-shaped open-circuit stub 14, and a horizontal stub 11 with two rectangular patches 12 at both ends. Two PIN diodes 13 are connected between the horizontal stub and the vertical stub of the transmission line, respectively. Another PIN diode 15 is connected to the fan-shaped open-circuit stub by the transmission line. Two through-holes are designed to connect the three metal layers. The first through-hole 7 connects the top layer and the first phase shifter to control the reflection phase of the y-polarized incident wave; the second through-hole 8 connects the top layer and the second phase shifter to control the reflection phase of the x-polarized incident wave. A 1 mm radius circular hole is drilled at two locations in the intermediate floor layer to prevent short circuits between the through-holes and the floor layer. Figure 5 The second phase shifter 17 in the right figure has the same structure as the first phase shifter. The operating state of the phase shifter is controlled by a PIN diode, and the PIN diode in the on state is equivalent to a resistor. and inductor The series connection. A PIN diode in the off state is equivalent to a capacitor. and inductor The 2-bit reflective phase shifter defines four coded states: '00', '01', '10', and '11'. The first digit, '0' or '1', represents the cutoff or conduction state of the two diodes between the main transmission line and the vertical stub in the phase shifter, and the second digit, '0' or '1', represents the cutoff or conduction state of the diode between the main transmission line and the fan-shaped stub. Codes 'x' and 'y' represent the states of the second and first phase shifters, respectively. Each phase shifter is controlled by two DC bias lines, and the DC bias lines of each unit together form a bias network. Independent control of each unit is achieved by applying a DC control voltage from an external FPGA to the PIN diodes of the phase shifter.

[0105] 2. Arrays and Bias Networks

[0106] like Figure 6 As shown, this invention uses a 2-bit multifunctional reconfigurable metasurface with a center frequency of 3.75 GHz and overall dimensions of 460 × 560 × 4.5 mm. 3 (Length × Width × Height) This embodiment includes: an 8×8 two-dimensional array 18 composed of 2-bit reconfigurable metasurface units and a DC bias network 20 composed of bias lines led out from the bottom phase-shifting array 19 of each unit. Each unit leads out four DC bias lines, forming a total of 256 bias channels.

[0107] In this embodiment:

[0108] The metasurface unit has a side length of 55 mm, the first dielectric layer 2 has a thickness of 3 mm, and the second dielectric layer 4 has a thickness of 1 mm.

[0109] The top metal patch 1 has a side length of 23 mm. The main transmission lines 10 of the bottom phase shifter 5 have lengths of 18 mm and 9.5 mm, and a width of 2.88 mm.

[0110] The two vertical slender branches are 3 mm long and 0.5 mm wide; the horizontal branch 11 is 14 mm long and 2 mm wide; the two square patches 12 have a side length of 6.5 mm.

[0111] The radius of the sector patch 14 is 6 mm.

[0112] A first and a second metal connection hole are provided between the top metal patch 1 and the bottom phase shifters 5 and 6. The first hole has a radius of 1 mm and the second hole has a height of 4.5 mm.

[0113] A circular hole with a radius of 1 mm is opened at the location of the through hole in the intermediate stratum to prevent short circuits;

[0114] The top metal patch, the middle layer, and the bottom phase shifter are made of copper foil with a thickness of 0.02 mm.

[0115] The dielectric material is F4B with a dielectric constant of 2.47 and a loss tangent of 0.002.

[0116] The PIN diodes used, 13 and 15, are MADP-000907-14020, with equivalent parameters of... , , , .

[0117] Through the design of the aforementioned stacked structure, microstrip phase shifter geometry parameters, and array and bias network, this invention achieves independent 2-bit phase modulation of two orthogonally polarized incident waves on a single reflective metasurface, providing a hardware foundation for subsequent beam deflection, polarization conversion, and RCS reduction functions.

[0118] VI. Real-time control of metasurfaces based on EEG / EOG

[0119] The method of this invention includes two stages: offline training and online control, and its process is as follows: Figure 7 As shown.

[0120] 1. Offline phase:

[0121] Install EEG / EOG electrode caps and perform short-time baseline recording to check impedance and noise levels;

[0122] Organizing subjects according to Figure 3 The experimental paradigm was used to complete several sessions of motion imagery tasks, and synchronized EEG and EOG data were recorded.

[0123] The EEG data is filtered, segmented, and truncated. The resulting C×T segments are then input into MSFBCNN for training to obtain a stable left and right hand motion imagery classifier.

[0124] Estimate the EOG peak distribution on offline data and select the corresponding amplitude and time series threshold;

[0125] Based on the phase response of the metasurface unit, phase encoding matrices for three operating modes are designed and preloaded into the FPGA.

[0126] 2. Online Phase:

[0127] When the system enters standby mode, the subjects switch between mode one, mode two and mode three by blinking with their left eye, right eye or both eyes, and determine the corresponding mode to switch to;

[0128] In the selected mode, the participants spontaneously imagined the movements of their left and right hands, and MSFBCNN output the left and right hand categories and their Softmax confidence scores in real time.

[0129] The control logic module selects the signal and motion image category according to the mode and searches for the corresponding phase encoding matrix in the predefined mapping relationship: In mode 1, the phase encoding of the two polarization channels is kept consistent to achieve polarization-independent beam deflection, where the left-hand MI corresponds to the main beam deflection to -10° and the right-hand MI corresponds to the deflection to -20°; In mode 2, under 45° linear polarization incident, the left-hand MI corresponds to the reflected wave in the +20° direction of left-handed circular polarization (LHCP) and the right-hand MI corresponds to the +20° direction of right-handed circular polarization (RHCP); In mode 3, no MI category is required, and only a pre-set random phase distribution is used to achieve RCS reduction in the target frequency band.

[0130] If the softmax probability of the current MI classification is lower than the preset confidence threshold, the encoding of the previous frame will remain unchanged to achieve the "no command" state.

[0131] The FPGA converts the selected phase encoding matrix into 256 DC bias levels and loads them into the array, completing an electromagnetic function update.

[0132] like Figures 8-10 As shown, schematic diagrams of typical phase encoding matrices and corresponding far-field radiation measurement and simulation comparison results are presented in three modes: In the left-eye blinking selection mode one, the left-hand movement imagination and right-hand movement phenomena correspond to a main beam deflection of -10° and -20°, respectively. The simulated and measured main lobe directions are basically consistent, verifying that the metasurface controlled by a combination of electrooculography and electroencephalography in this invention can achieve directional radiation of linearly polarized beams under this encoding.

[0133] In the second mode of right eye blinking selection, the left hand movement imagination and right hand movement phenomenon correspond to the reflected wave being able to be converted into a left-hand circularly polarized wave and a right-hand circularly polarized wave when deflected to +20°, respectively; near the design center frequency of 3.75 GHz, the axial ratio of both circular polarization modes is less than 3 dB, and the simulation and actual results of the main beam deflection angle are basically consistent.

[0134] The binocular blink selection mode 3 shows that the RCS reduction of the randomly encoded reconfigurable metasurface is greater than 10 dB in the 3.7 GHz to 3.85 GHz frequency band.

[0135] like Figure 11As shown, to verify the real-time performance of the system, the test platform consists of an EEG cap, a host computer, an FPGA, a signal generator, a transmitting horn antenna, two receiving horn antennas, and an oscilloscope. The signal transmitter generates a continuous wave with a center frequency of 3.75 GHz, which is radiated by the transmitting horn located in the middle, forming a plane wave normally incident on the metasurface. The two receiving horns are fixed at theoretical beam deflection angles of +20° and −20°, respectively, and connected to the oscilloscope for real-time synchronous recording of the time-domain voltage envelope. During the experiment, the user first enters the beam deflection mode by blinking with their left eye, and then executes a right-hand movement imagery, triggering a phase-coded command to deflect the beam to −20°; the waveform amplitude of the channel connected to the right receiving horn antenna increases significantly, while the signal of the channel connected to the left receiving antenna approaches noise. The above response results show that the system can achieve real-time selective switching of the main lobe according to the user's intention. Simultaneously, it verifies that the overall control delay of this system is on the order of 3-5 seconds, meeting the real-time requirements of most brain-controlled communication and interaction scenarios.

[0136] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A smart metasurface electromagnetic wave modulation system for brain-computer interfaces, characterized in that, A dual-polarized reconfigurable metasurface based on hybrid control of EEG and EEG signals enables real-time electromagnetic wave modulation. This includes: fusing non-invasive EEG and EOG signals to construct a hierarchical control logic of "EOG mode selection + EEG fine-tuning"; robustly decoding left and right hand motion imagery using a multi-scale filter bank convolutional neural network to improve classification performance with limited training samples; utilizing a dual-polarized 2-bit reconfigurable reflective metasurface to achieve three typical electromagnetic functions: polarization-independent beam deflection, joint beam deflection and polarization conversion under 45° linear polarization incidence, and RCS reduction; and introducing an implicit "no-command" state through a softmax confidence threshold mechanism to improve system security and robustness when signal quality is poor or the user has no clear intent. Specifically, this includes: The EEG / EOG signal acquisition module is used to acquire motor imagery EEG signals through multi-channel electrodes placed in the sensorimotor cortex of the subject, and to acquire electrooculogram (EOG) signals through electrodes placed in the forehead region. The EEG signal preprocessing module is used to perform bandpass filtering, sampling, and segmentation processing on the EEG signal to obtain multi-channel time series data for decoding; The motor imagery EEG signal decoding module is a classifier based on a multi-scale convolutional neural network. It is used to receive preprocessed EEG time series, extract time-frequency features, and output motor imagery categories and their classification confidence scores, including at least left-hand motor imagery and right-hand motor imagery. The blink detection and pattern encoding module is used to detect the amplitude and timing features of the EOG signal within a preset time window, identify three intentions: blinking with the left eye, blinking with the right eye, and blinking with both eyes simultaneously, and encode the recognition results into a pattern selection signal. The mode selection and electromagnetic function mapping module is used to select a preset metasurface working mode according to the mode selection signal and the motion imagination category, and generate a phase encoding matrix corresponding to the target working mode. The programmable logic control module, which is a field-programmable gate array (FPGA), is used to convert the phase encoding matrix into multiple DC bias control signals; Dual-polarized reconfigurable metasurface arrays, including 8 arrays arranged in a rectangular array 8 Reconfigurable reflection units; each reflection unit includes: two independent 2-bit reflection phase shifting networks based on 6 PIN diodes, each acting on two orthogonal line polarization components. The two phase shifters are connected to four bias lines. Each reflection unit is controlled by an independent DC voltage to obtain different phase distributions on the array surface, thereby realizing real-time control of the beam direction, polarization state and scattering characteristics of the reflected electromagnetic wave.

2. The intelligent metasurface electromagnetic wave control system according to claim 1, characterized in that, In the EEG / EOG signal acquisition module, the electrodes located near the sensorimotor cortex are C3, Cz, C4, FC3, FC4, CP3, and CP4, used to acquire EEG signals related to upper limb motor imagery; the electrodes located in the forehead region are Fp1 and Fp2, used to acquire blink signals in the horizontal and vertical directions; the ground electrode is placed at the earlobe; the sampling rate of the EEG and EOG signals is 1000 Hz, and they are transmitted to the host computer for processing via a wireless transmission module.

3. The intelligent metasurface electromagnetic wave control system according to claim 2, characterized in that, The motion image decoding module is a multi-scale filter bank convolutional neural network (MSFBCNN), comprising: The input layer receives multi-channel EEG data segmented by trial number. The data structure for a single trial is as follows: ,in For the number of channels, The number of sampling points is set, and it is configured to extract a predetermined length of EEG segment from the middle of the time window of each motion visualization task as network input; The multi-scale temporal convolutional feature extraction layer contains several parallel one-dimensional temporal convolutional branches. Each branch uses convolutional kernels of different lengths in the temporal dimension and convolutional kernels with a kernel length of 1 in the channel dimension. It performs multi-scale filtering on the same EEG segment and splices the feature maps output by each branch in the filter dimension to obtain a fused temporal feature representation. The spatial convolutional layer includes several spatial filters. Its convolutional kernel has a length in the channel dimension equal to the number of EEG channels and a length in the time dimension of 1. It is used to perform spatial filtering on the time features in the channel dimension, compress the channel dimension to 1, and extract the spatial pattern across channels. Regularization and nonlinear layers, including batch normalization layers, square nonlinear layers, average pooling layers, logarithmic nonlinear layers, and dropout layers with a predetermined dropout rate, are used to construct robust features related to bandwidth power and suppress network overfitting. The classification layer maps the pooled and regularized features to the output class space, outputting the softmax probability and confidence of at least the left-handed and right-handed motor imagery.

4. The intelligent metasurface electromagnetic wave control system according to claim 3, characterized in that, The blink detection and pattern encoding module is used to extract a first time window of duration, starting from the command time, from the EOG signal each time a mode switching command arrives, and to detect the peak amplitude, polarity, and occurrence time of the EOG waveform within this time window, wherein: When a single-peak waveform with an amplitude exceeding the first threshold and within a preset time range appears in the EOG channel corresponding only to the Fp1 electrode within the time window, the blink is identified as a left eye blink and encoded as a first mode selection signal. When a single-peak waveform with an amplitude exceeding the second threshold and within a preset time range appears in the EOG channel corresponding only to the Fp2 electrode within the time window, the blink is identified as a right eye blink and encoded as a second mode selection signal. When the amplitude of the two channels corresponding to the Fp1 and Fp2 electrodes exceeds the third threshold within the time window and the time interval between the peak occurrences of the two is less than the predetermined time threshold, the blink is identified as a blink of both eyes and encoded as a third mode selection signal. The first duration is preferably 400-800 ms.

5. The intelligent metasurface electromagnetic wave control system according to claim 4, characterized in that, The mode selection and electromagnetic function mapping module is configured to enable the system to have the following three working modes: (1) In the first working mode, the phase codes of the two orthogonal polarization channels are kept consistent to achieve polarization-independent beam deflection; (2) In the second working mode, under the preset 45° linear polarization incident condition, phase codes for left-hand circular polarization reflection beam deflection +20° and right-hand circular polarization reflection beam deflection +20° are generated according to the motion imagination category to achieve joint beam deflection and polarization conversion functions; (3) In the third working mode, phase codes randomly distributed on the array surface are generated to expand the scattering direction distribution to reduce the radar cross section (RCS).

6. The intelligent metasurface electromagnetic wave control system according to claim 5, characterized in that, The programmable logic control module, with a field-programmable gate array as its core, combined with memory, voltage conversion circuits and multiple general purpose input / output interfaces (GPIO), constitutes a configurable 256-channel digital bias control platform. The input side connects to the host computer via a USB serial interface. The host computer generates the corresponding metasurface working mode code based on the mode selection signal and phase encoding index output by the EEG signal decoding module, and sends it to the FPGA control core in the form of serial port data frames. The FPGA internally stores multiple sets of phase encoding matrices corresponding to each working mode. Each matrix is ​​encoded as a control word of length 256, corresponding to the four bias lines of the 8×8 unit in the metasurface array. After receiving the mode index sent by the host computer, the FPGA reads the target control word from the code table and refreshes each GPIO output bit by bit, so that the on / off state of all PIN diodes switches synchronously within one clock cycle, thereby realizing the overall update of the phase state of the metasurface unit. In terms of hardware structure, the programmable logic control module provides four 68-pin connectors. In addition to four fixed ground terminals, the remaining 64 pins of each connector are used as programmable GPIOs, for a total of 256 independent bias outputs. Each GPIO outputs a DC bias voltage of ±2.5 V through voltage conversion and current limiting circuits to adapt to the on / off requirements of the PIN diode phase shifters in the metasurface unit and drive the metasurface array.

7. The intelligent metasurface electromagnetic wave control system according to claim 6, characterized in that, The dual-polarized reconfigurable metasurface array is a two-dimensional array consisting of 2-bit reconfigurable metasurface units operating in the microwave frequency band, and a DC bias network consisting of bias lines of each metasurface unit; wherein: The metasurface unit consists of three metal layers and two dielectric layers; the top metal layer is used to receive and radiate electromagnetic waves, the middle metal layer serves as a ground layer, and the bottom metal layer serves as a phase shifter to achieve 2-bit reconfigurable functionality; the two dielectric layers are located between the top metal layer and the middle metal layer, and between the middle metal layer and the bottom metal layer, respectively. A radiating patch located above the first dielectric substrate is used to receive and radiate electromagnetic waves; The bottom metal layer located below the second dielectric substrate consists of two identical reflective phase shifters for independently adjusting the reflection phase in the x and y polarization directions. The reflective phase shifter consists of a transmission line with a pair of vertical stubs, a fan-shaped open-circuit stub, and a horizontal stub. Two diodes, PIN1 and PIN2, are connected between the horizontal stub and the vertical stub of the transmission line, respectively, and another diode, PIN3, is connected to the fan-shaped open-circuit stub by the transmission line. An intermediate metal layer is disposed between the first dielectric substrate and the second dielectric substrate as a ground plane shared by the top metal patch and the bottom reflective phase shifter. The intermediate metal layer separates the radiating portion from the bias lines to reduce the impact of dense bias lines in the array on the radiation performance. Two metal vias are designed between the three metal layers for connection. The first via 1 is used to connect the top layer and the first phase shifter 1 to control the reflection phase of the incident wave in the y-polarization direction. The second via 2 is used to connect the top layer and the second phase shifter 2 to control the reflection phase of the incident wave in the x-polarization direction. A circular hole is drilled in the intermediate layer ground plane at the positions of the first via 1 and the second via 2 to avoid short circuit between the via and the ground plane. The 2-bit independent adjustable unit can independently adjust four reflection phase states of 0°, 90°, 180° and 270° in two orthogonal linear polarization directions. Each reflection unit in the array is independently addressed through four offset lines, and the entire 8×8 array forms 256 offset control channels. By setting different phase encoding matrices, the array can realize polarization-independent beam deflection, polarization conversion and beam deflection control of reflected waves under 45° linear polarization wave incident, and RCS reduction function.

8. The intelligent metasurface electromagnetic wave control system according to any one of claims 1-7, characterized in that, The process for electromagnetic wave manipulation is as follows: (1) The electrooculogram (EOG) signal of the subject and the electroencephalogram (EEG) signal under the motor imagery task were acquired by the multi-channel EEG electrode and the frontal EOG electrode in the EEG / EOG signal acquisition module, and digitally acquired at the first sampling rate. (2) The EEG signal is bandpass filtered, downsampled and segmented by trial through the EEG signal preprocessing module, and the multi-channel EEG segment corresponding to each trial is used as the input sample for decoding the motor imagery signal; (3) Input the processed EEG fragments into the multi-scale convolutional neural network model in the motor imagery EEG signal decoding module, extract time-frequency features and output the corresponding motor imagery category and classification confidence; (4) Through the blink detection and pattern encoding module, a time window of a preset length is extracted from the EOG signal, the peak amplitude, polarity and occurrence time of the EOG waveform within the time window are calculated, and the left eye blink, the right eye blink or both eyes blink are determined accordingly, and the determination result is encoded into a pattern selection signal. (5) Through the mode selection and electromagnetic function mapping module, a preset metasurface working mode is selected according to the mode selection signal and the motion imagination category and its confidence level, and a corresponding phase encoding matrix is ​​generated; (6) The phase encoding matrix is ​​converted into multiple DC bias control signals by the programmable logic control module FPGA and loaded onto the bias ports of each reflective unit of the reconfigurable metasurface array to change the working state of each unit diode; (7) By using a dual-polarized reconfigurable metasurface array, the working state of the metasurface reflection unit can be adjusted under the corresponding electromagnetic wave incident, thereby realizing the beam deflection, polarization conversion and RCS reduction functions in the preset working mode.

9. The intelligent metasurface electromagnetic wave control system according to claim 8, characterized in that, The preset working modes include: First mode: The electrooculogram signal corresponds to the left eye blinking, which is the first functional mode. Under different linear polarization incident conditions, the phase coding of the two orthogonal polarization channels remains consistent to achieve polarization-independent beam deflection. Among them, if the motion imagery category is left-hand motion imagery, the corresponding main beam of the reflected wave is deflected to the target deflection angle; if the motion imagery category is right-hand motion imagery, the corresponding main beam of the reflected wave is deflected to another target deflection angle. Second mode: The electrooculogram signal corresponds to the second functional mode of right eye blinking; under the condition of 45° linear polarization incident, if the motion imagery category is left-hand motion imagery, the reflected wave is deflected to the target angle and converted into a left-hand circularly polarized beam; if the motion imagery category is right-hand motion imagery, the reflected wave is deflected to the target angle and converted into a right-hand circularly polarized beam. The third mode: the electrooculogram signal corresponds to the third function mode of blinking; a phase code with random spatial distribution is generated to control the metasurface to achieve RCS reduction in the target frequency band.

10. The intelligent metasurface electromagnetic wave control system according to claim 8, characterized in that, The motion imagery decoding module outputs the softmax probability and compares the maximum class probability with a preset confidence threshold. When the maximum probability is less than the threshold, it does not perform the encoding update operation based on the output of the current motion imagery, but instead maintains the phase encoding matrix corresponding to the last valid command, thereby maintaining the current electromagnetic function state unchanged when the user has no clear motion imagery or the signal quality is low.

Images