A programmable information metasurface, metasurface-assisted communication and perception integrated method and system

Abstract

The application discloses a programmable information metasurface, a metasurface-assisted communication and sensing integrated method and system, comprising that a transmitting programmable information metasurface generates a beam in a wide beam mode, and a receiving programmable information metasurface adopts a random beam mode to collect electromagnetic echo signals; the collected electromagnetic echo signals are input into a trained first deep neural network model for position recognition, and a position recognition result is output; based on the position recognition result, the transmitting programmable information metasurface adopts a double-beam mode to point to a sensing target and a known communication user, and the receiving programmable information metasurface maintains the random beam mode to collect electromagnetic echo signals; the collected electromagnetic echo signals are input into a trained second deep neural network model for human body posture recognition, and a human body posture recognition result is output; and based on the human body posture recognition result, a corresponding communication action is triggered according to a preset decision mapping strategy.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and environmental sensing technology, specifically relating to a programmable information metasurface, a metasurface-assisted integrated communication and sensing method and system. Background Technology

[0002] The deep integration of wireless communication and environmental sensing is a key technological trend towards next-generation intelligent communication systems. Traditional approaches typically design and deploy sensing and communication systems independently. For example, in indoor or specific areas, dedicated devices such as cameras and infrared sensors are often used for environmental monitoring and human behavior recognition, while data transmission relies on independent communication links such as Wi-Fi and cellular networks. This discrete architecture leads to hardware stacking, high system complexity, high energy consumption, and difficulty in achieving efficient collaboration between different systems. Specifically, existing technologies mainly suffer from the following limitations:

[0003] First, while computer vision-based perception solutions are mature, their inherent limitations restrict their application in many privacy-sensitive or demanding scenarios. Visual information captured by cameras is prone to privacy breaches, and their performance degrades significantly in low light, obstructed, or inclement weather conditions. Furthermore, the complex convolutional neural network models required to process image data are computationally expensive, making them difficult to deploy on resource-constrained edge devices.

[0004] Secondly, although metasurfaces, as an emerging cutting-edge electromagnetic wave manipulation technology, provide a new physical basis for realizing integrated hardware and software wireless sensing, their current applications mostly focus on single-function designs. Some schemes utilize metasurfaces to achieve beamforming to improve communication quality, or only realize basic sensing functions such as target presence detection, velocity measurement, and ranging. These schemes have not yet fully utilized the programmable characteristics of metasurfaces to build a closed-loop system capable of finely identifying complex targets and intelligently driving adaptive adjustments to communication strategies. Summary of the Invention

[0005] Purpose of the invention: To address the problem that existing technologies cannot seamlessly integrate precise environmental perception, intelligent target recognition, and adaptive directional communication on a highly integrated hardware platform, this invention proposes a programmable information metasurface and a metasurface-assisted integrated communication and perception method and system.

[0006] Technical solution: In a first aspect, the present invention proposes a programmable information metasurface, which is composed of several binary RIS units, each of which sequentially includes an input layer, a driving layer and an output layer; two PIN diodes are integrated in the output layer;

[0007] When the feed antenna excites the RIS unit, the current induced in the input layer is coupled to the output layer through the metallized vias; by controlling the DC voltage on the drive layer connected to the output layer through two metal vias, two PIN diodes are sequentially turned on to reverse the current direction on the surface of the output layer, thereby realizing binary modulation of the transmission phase of the RIS unit.

[0008] By applying a random binary DC voltage to each binary RIS unit, a random phase encoding distribution is formed on the aperture of the binary RIS unit. Based on the coordinates and binary state of each binary RIS unit, the spatial distribution of the overall scattering field of the programmable information metasurface is obtained.

[0009] By designing an aperture coding sequence, the programmable information metasurface can synthesize directional beams with specific modulation phases.

[0010] Furthermore, the directional beam includes: a wide beam mode pointing in a single direction, a dual beam mode pointing in two directions, and a random beam mode.

[0011] Secondly, this invention proposes a metasurface-assisted integrated communication and sensing method, comprising:

[0012] The transmitting programmable information metasurface uses a wide beam pattern to generate a beam, while the receiving programmable information metasurface uses a random beam pattern to collect electromagnetic echo signals.

[0013] The collected electromagnetic echo signals are preprocessed and then input into the trained first deep neural network model for position recognition, which outputs the position recognition result.

[0014] Based on the location recognition results, the transmitting programmable information metasurface uses a dual-beam mode to point at the sensing target and the known communication user, while the receiving programmable information metasurface maintains a random beam mode to collect electromagnetic echo signals.

[0015] The collected electromagnetic echo signals are preprocessed and then input into a trained second deep neural network model for human posture recognition, which outputs human posture recognition results.

[0016] Based on the human posture recognition results, a corresponding communication action is triggered according to a preset decision mapping strategy; the decision mapping strategy defines the mapping relationship from a specific human posture recognition result to the corresponding communication command.

[0017] Both the transmitting programmable information metasurface and the receiving programmable information metasurface are programmable information metasurfaces.

[0018] Furthermore, the human posture is explicitly defined as a variety of highly distinguishable static postures.

[0019] Furthermore, both the first and second deep neural network models employ a multilayer perceptron to construct the feature extraction backbone, introduce a Dropout layer to suppress overfitting, and finally provide the classification result through a Softmax output layer.

[0020] Furthermore, the first deep neural network model and the second deep neural network model are trained according to the following steps:

[0021] In the experimental scenario, the programmable information metasurface receives single-tone echo signals through a random beam pattern;

[0022] The received single-tone echo signals are preprocessed and labeled, including location labels and human pose labels, to obtain a dataset;

[0023] Using the dataset, the first deep neural network model and the second deep neural network model are trained and tested.

[0024] Thirdly, this invention proposes a metasurface-assisted integrated communication and sensing system, comprising:

[0025] The programmable information-emitting metasurface is configured to generate a beam using a wide-beam pattern during the location recognition phase; and to point to the sensing target and the known communication user using a dual-beam pattern based on the location recognition result output by the first deep neural network model during the human posture recognition phase.

[0026] The receiving programmable information metasurface is configured to use a random beam pattern to collect electromagnetic echo signals in both the position recognition stage and the human posture recognition stage.

[0027] The first deep neural network model is configured to output the location identification result based on the electromagnetic echo signal collected during the location identification phase;

[0028] The second deep neural network model is configured to output human posture recognition results based on the electromagnetic echo signals collected during the human posture recognition stage.

[0029] The control unit has a pre-stored decision mapping strategy and is configured to trigger corresponding communication actions based on the human posture recognition results and according to the preset decision mapping strategy; the decision mapping strategy defines the mapping relationship from a specific human posture recognition result to the corresponding communication command.

[0030] Both the transmitting programmable information metasurface and the receiving programmable information metasurface are programmable information metasurfaces as described in claim 2.

[0031] Beneficial Effects: This invention successfully designed and implemented an intelligent integrated sensing and communication system based on deep neural networks and programmable information metasurfaces. This system innovatively integrates electromagnetic sensing, artificial intelligence recognition, and dynamic beam communication, forming a complete "sensing-decision-communication" technological closed loop. Compared with existing technologies, this invention has the following advantages:

[0032] (1) First, the present invention utilizes the beam agility of programmable metasurfaces to achieve seamless switching from wide-area scanning to precise directional communication. Second, the designed lightweight multilayer perceptron model can efficiently and accurately identify and classify human posture and spatial position simultaneously, providing a reliable basis for intelligent decision-making. Finally, the system can adaptively execute predefined communication strategies based on the identification results, realizing on-demand allocation of communication resources and adaptive distribution of content.

[0033] (2) Experimental results show that the technical solution proposed in this invention is not only feasible but also possesses excellent performance. It exhibits significant advantages in sensing accuracy, communication efficiency, system power consumption, and security. This fully demonstrates that the technical route of deeply integrating artificial intelligence with metasurface physical layers provides a brand-new hardware architecture and system paradigm for future sixth-generation mobile communications, intelligent IoT, human-computer interaction, and other fields, possessing broad application prospects and significant industrial value. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of an integrated electromagnetic sensing scenario based on metasurfaces for communication and sensing.

[0035] Figure 2 This is a schematic diagram of a reconfigurable smart surface configuration; where, Figure 2 In this context, (a) represents the RIS structure composition; Figure 2 In the figure, (b) represents the transmission coefficient of the RIS unit in binary state;

[0036] Figure 3 To achieve position and attitude sensing using beamforming, Figure 3 In the diagram, (a) represents the aperture code corresponding to different far-field patterns; Figure 3 (b) in the diagram represents different far-field radiation patterns, from left to right: a directional beam pointing to (0°, 0°), a directional beam pointing to (20°, 135°), a dual beam pointing to both (0°, 0°) and (37°, 180°), and a random radiation mode;

[0037] Figure 4 This is a schematic diagram of the echo signal composition in a RIS system.

[0038] Figure 5 A schematic diagram of a DNN model architecture used for classifying location and pose;

[0039] Figure 6 This is a schematic diagram of position and attitude perception, where, Figure 6 In this context, (a) represents the RIS configuration corresponding to a single-tone echo signal; Figure 6 In the diagram, (b) represents the location identification confusion matrix; Figure 6 In the diagram, (c) represents the pose recognition confusion matrix when the human target is in different positions;

[0040] Figure 7 This is a schematic diagram illustrating position and attitude sensing using beamforming; where, Figure 7 (a) in the diagram represents the aperture code of the transmitted RIS and the corresponding dual-beam pattern; Figure 7 In the diagram, (b) represents the attitude recognition confusion matrix when using the dual-beam mode. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description, in conjunction with the accompanying drawings and embodiments, will further illustrate the metasurface-assisted integrated communication and sensing system proposed in this invention.

[0042] Example 1:

[0043] This invention proposes a metasurface-assisted integrated communication and sensing system. Specifically, it is a communication and sensing integrated system based on metasurfaces and deep neural networks, capable of simultaneously sensing and recognizing human posture and position, and achieving real-time communication. Figure 1 As shown, this system achieves simultaneous perception of human posture and position through a metasurface, classifies the perceived information using a deployed deep neural network, and schedules communication based on the classification results. Specifically, it includes: a programmable information metasurface (RIS), a deep neural network model, and a control unit with a pre-stored dynamic decision mapping strategy.

[0044] The programmable information metasurface (RIS) proposed in this embodiment of the invention consists of 324 RIS units of 18×18. Each unit adopts a multi-layer architecture of "input-drive-output" and uses a horn antenna as the power supply to provide initial electromagnetic wave radiation for the entire system.

[0045] Figure 2Image (a) illustrates the transmissive RIS and its feeding structure designed according to an embodiment of the present invention. It employs a multilayer dielectric structure, including two RO4350B dielectric substrates and an F4B adhesive layer between them. The power supply uses a microstrip patch antenna to illuminate the transmissive RIS at a vertical distance of 28 mm. When the feeding antenna excites the transmissive RIS, the current induced in the input layer is coupled to the output layer through metallized vias. To achieve modulation of the unit transmission phase, two PIN diodes are integrated in the output layer. By controlling the DC voltage on the drive layer connected to the output layer through two metal vias, the two PIN diodes can be sequentially turned on to reverse the current direction on the output layer surface, thereby achieving binary modulation of the unit transmission phase.

[0046] Figure 2 (b) in the figure gives the simulated transmission coefficients of the unit structure. In the full-wave simulation, the PIN diode in the on state is modeled as a series circuit of resistor and inductor, and in the off state as a parallel circuit of resistor and capacitor. Due to the reversal mechanism of the current direction on the output layer surface, the unit exhibits opposite transmission phases and similar transmission amplitudes in the two states, thus forming a binary RIS unit.

[0047] To generate a complex electromagnetic sensing field, embodiments of the present invention drive a programmable information metasurface to produce a random receiving mode. Specifically, by applying a random binary DC voltage to each binary RIS unit, a random phase-coded distribution can be formed on the aperture of the binary RIS unit. At this time, the overall scattering field of the programmable information metasurface can be characterized by the following formula. Wherein, the spatial distribution of the scattering field is determined by the coordinates and binary state of each binary RIS unit. Assuming the human target consists of K scattering points, the OFDM echo signal at the m-th subcarrier can be expressed as:

[0048]

[0049] In the formula, I and J are the indices of the transmitting RIS and receiving RIS units, respectively; The reflection coefficient at the k-th scattering point; It is the discrete amplitude and phase of the OFDM signal transmitted at the m-th subcarrier frequency; It is the Green's function. , These represent the transmission coefficients of the transmitting and receiving RIS units, respectively; It is the distance between the feed source and the i-th unit of the transmitting RIS; This represents the frequency of the Mth subcarrier; It is the distance between the k-th scattering point and the i-th emitting RIS unit; It is the distance between the kth scattering point and the jth unit of the receiving RIS; It is the distance between the feed source and the i-th unit of the receiving RIS.

[0050] By precisely designing the aperture coding sequence, this information metasurface can synthesize directional beams with specific modulation phases. Figure 3 Four typical aperture coding examples are shown, corresponding from left to right: a directional beam pointing to (0°, 0°), a directional beam pointing to (20°, 135°), a dual beam pointing to both (0°, 0°) and (37°, 180°), and a random radiation mode. Full-wave simulation results ( Figure 3 (b) in the figure confirms that direction can be generated based on aperture coding. Figure 1 Furthermore, it produces a directional beam with controllable radiation phase. This characteristic enables the programmable information metasurface proposed in this embodiment of the invention to directly perform quadrature phase shift keying modulation on electromagnetic waves through electronic control, laying the physical foundation for subsequent integration of sensing and communication functions.

[0051] The programmable information metasurface designed above enables sensing and data acquisition.

[0052] This invention employs a deep neural network model to achieve high-precision, non-visual perception of human position and posture. The deep neural network model uses electromagnetic echo signals acquired from a programmable information metasurface as input, and through an end-to-end learning method, directly establishes a non-linear mapping relationship from the electromagnetic echo signals to position and posture categories.

[0053] In this embodiment of the invention, the electromagnetic echo signal comprises two parts: an OFDM echo signal received through a fixed beam pattern and a single-tone echo signal received through a random beam pattern. Its signal structure is as follows: Figure 4 As shown.

[0054] To ensure the diversity and comprehensiveness of the constructed dataset, the experiment was conducted indoors, with a 3m x 3m square area defined. Four poses were performed at each of the nine grid locations, resulting in 36 different states. The human poses collected in this dataset were explicitly defined as four highly discriminative static poses: upright, left arm extended, right arm extended, and both arms extended laterally.

[0055] Based on the above process, this embodiment of the invention successfully constructed a large-scale echo signal dataset containing multi-location and multi-pose labels, suitable for training deep neural network models. Through supervised learning, the deep neural network model can extract microwave scattering patterns and time-domain features related to specific poses from the data, thereby achieving accurate identification and classification of unknown echo signals.

[0056] Before inputting the data into the deep neural network model, the raw single-tone echo signals acquired by the programmable information metasurface are preprocessed. After annotation, the entire dataset is divided into training and testing sets according to a predetermined ratio to ensure the scientific validity of the model training and evaluation process.

[0057] Subsequently, deep neural network models for classifying location and pose were constructed and trained, respectively. The two network architectures are similar, as follows: Figure 5 As shown, this deep neural network model uses a multilayer perceptron to construct the feature extraction backbone, introduces a Dropout layer to suppress overfitting, and finally outputs the classification result through a Softmax layer. The input of the deep neural network model is a preprocessed one-dimensional real vector, and the output dimension is consistent with the number of human pose and position categories to be classified. During the training phase, cross-entropy is used as the loss function, and the Adam optimizer is used for backpropagation and iterative parameter updates until the model converges. To objectively evaluate the model performance, a test set is used to systematically validate the trained model. Figure 6 As shown in (b) and (c), the confusion matrix clearly demonstrates the recognition performance of the two models in different categories, effectively verifying the classification accuracy and reliability of the algorithm proposed in this embodiment of the invention.

[0058] Finally, the trained deep neural network model is deployed in the integrated system to achieve an end-to-end real-time perception and decision-making closed loop: the echo signal collected in real time from the metasurface is preprocessed and input into the deep neural network model, the deep neural network model outputs the classification result, and executes the corresponding communication protocol or resource scheduling strategy accordingly, thus forming a complete closed loop from physical perception to intelligent decision-making.

[0059] During the sensing process, dynamic optimization of beamforming is crucial for improving system performance. In the position recognition phase, the transmitting programmable information metasurface generates a wide beam to cover the entire sensing area; while in the attitude recognition phase, the far-field beam of the transmitting programmable information metasurface can be adjusted to a directional beam pointing towards the target to enhance illumination intensity. Considering the simultaneous presence of communication users and sensing targets in the system, the transmitting programmable information metasurface can also generate directional dual beams pointing towards the user and the target respectively, thereby simultaneously enhancing the signal reception quality at both locations.

[0060] To achieve end-to-end perception optimization, this embodiment of the invention employs a two-stage beam switching strategy. First, in the location recognition phase, the transmitting programmable information metasurface generates a wide beam, while the receiving programmable information metasurface uses a random beam mode to achieve initial target localization. After obtaining the target's location information, the system enters the attitude recognition phase. At this time, the transmitting programmable information metasurface switches to a dual-beam mode, precisely pointing to the perceived target and the known communication user respectively, while the receiving programmable information metasurface maintains random beam reception. Through this dynamic adjustment mechanism, regardless of the target's location, the transmitting programmable information metasurface can achieve continuous directional tracking.

[0061] Based on the target location, the transmitting programmable information metasurface invokes the corresponding dual-beam encoding, pointing towards the user and the target respectively. It is worth noting that when the target and user positions are close in angle, the beam pattern may appear as a single beam due to the beamwidth limitation of the programmable information metasurface. This reflects the spatial resolution characteristics that the system needs to consider in actual deployment.

[0062] Experimental data shows that, compared to the wide-beam mode, the dual-beam mode improves signal strength by an average of 5.71 dB at the user's location and by an average of 8.56 dB at the target's location. This significant improvement in signal strength directly leads to enhanced sensing performance. To evaluate the improved sensing performance, a dedicated attitude recognition sub-network was trained for each target location in the corresponding dual-beam mode. Test results show that the attitude recognition accuracy in the dual-beam mode is on average 11.7% higher than that in the wide-beam mode, fully validating the effectiveness of the beamforming strategy proposed in this embodiment of the invention. Figure 7 As shown, this embodiment of the invention transmits a dual-beam pattern of RIS (as indicated). Figure 7 (a) in the figure achieves precise energy projection on the target area, and the attitude recognition results based on this ( Figure 7 (b) shows a higher classification accuracy, demonstrating the enhancement effect of beamforming on sensing performance. This beamforming-based sensing enhancement method provides a reliable technical path for realizing high-performance integrated sensing and communication systems.

[0063] After achieving high-precision attitude recognition through beam tracking, the system enters the communication decision and execution phase. To achieve intelligent communication, a fixed decision mapping strategy is pre-stored within the system's control unit. This strategy defines the mapping relationship from a specific attitude classification result to the corresponding communication command. For example, the system maps the four identified human attitudes (such as standing, left arm extended, right arm extended, and both arms extended laterally) to four different types of pre-stored digital content (such as images, commands, or data packets).

[0064] Once the attitude recognition neural network outputs a classification result, the control unit immediately queries the decision mapping strategy and triggers the corresponding communication action. The system then sends the mapped digital content to the target user via an established directional beam pointing towards the user. This process intuitively demonstrates how the system transforms attitude information from the physical world into communication commands, achieving "what you see is what you get" intelligent interaction.

[0065] In summary, this invention constitutes a complete "perception-decision-communication" workflow, which greatly improves spectrum and energy efficiency through intelligent beam management; seamless connection based on behavior recognition enables truly natural human-computer interaction; and the directional transmission feature also enhances communication security at the physical layer.

Claims

1. A programmable information metasurface, characterized in that: It is composed of several binary RIS units, each of which includes an input layer, a driver layer, and an output layer in sequence; two PIN diodes are integrated in the output layer. When the feed antenna excites the RIS unit, the current induced in the input layer is coupled to the output layer through the metallized vias; by controlling the DC voltage on the drive layer connected to the output layer through two metal vias, two PIN diodes are sequentially turned on to reverse the current direction on the surface of the output layer, thereby realizing binary modulation of the transmission phase of the RIS unit. By applying a random binary DC voltage to each binary RIS unit, a random phase encoding distribution is formed on the aperture of the binary RIS unit. Based on the coordinates and binary state of each binary RIS unit, the spatial distribution of the overall scattering field of the programmable information metasurface is obtained. By designing an aperture coding sequence, the programmable information metasurface can synthesize directional beams with specific modulation phases.

2. A programmable information metasurface according to claim 1, wherein: The directional beam includes: a wide beam pattern pointing in a single direction, a dual beam pattern pointing in two directions, and a random beam pattern.

3. A metasurface-aided communication and sensing integrated method, characterized in that: include: The transmitting programmable information metasurface uses a wide beam pattern to generate a beam, while the receiving programmable information metasurface uses a random beam pattern to collect electromagnetic echo signals. The collected electromagnetic echo signals are preprocessed and then input into the trained first deep neural network model for position recognition, which outputs the position recognition result. Based on the location recognition results, the transmitting programmable information metasurface uses a dual-beam mode to point at the sensing target and the known communication user, while the receiving programmable information metasurface maintains a random beam mode to collect electromagnetic echo signals. The collected electromagnetic echo signals are preprocessed and then input into a trained second deep neural network model for human posture recognition, which outputs human posture recognition results. Based on the human posture recognition results, a corresponding communication action is triggered according to a preset decision mapping strategy; the decision mapping strategy defines the mapping relationship from a specific human posture recognition result to the corresponding communication command. Both the transmitting programmable information metasurface and the receiving programmable information metasurface are programmable information metasurfaces as described in claim 2.

4. The metasurface-aided communication and sensing integrated method of claim 3, wherein: The human posture is clearly defined as a variety of highly distinguishable static postures.

5. The metasurface-aided communication and sensing integrated method of claim 3, wherein: Both the first and second deep neural network models use a multilayer perceptron to construct the feature extraction backbone, introduce a Dropout layer to suppress overfitting, and finally give the classification result through a Softmax output layer.

6. The metasurface-aided communication and sensing integrated method of claim 3, wherein: The first deep neural network model and the second deep neural network model are trained according to the following steps: In the experimental scenario, the programmable information metasurface receives single-tone echo signals through a random beam pattern; The received single-tone echo signals are preprocessed and labeled, including location labels and human pose labels, to obtain a dataset; Using the dataset, the first deep neural network model and the second deep neural network model are trained and tested.

7. A metasurface-assisted communication and sensing integrated system, characterized in that: include: The programmable information-emitting metasurface is configured to generate a beam using a wide-beam pattern during the location identification phase. In the human posture recognition stage, based on the position recognition result output by the first deep neural network model, a dual-beam mode is used to point to the sensing target and the known communication user. The receiving programmable information metasurface is configured to use a random beam pattern to collect electromagnetic echo signals in both the position recognition stage and the human posture recognition stage. The first deep neural network model is configured to output the location identification result based on the electromagnetic echo signal collected during the location identification phase; The second deep neural network model is configured to output human posture recognition results based on the electromagnetic echo signals collected during the human posture recognition stage. The control unit has a pre-stored decision mapping strategy and is configured to trigger corresponding communication actions based on the human posture recognition results and according to the preset decision mapping strategy; the decision mapping strategy defines the mapping relationship from a specific human posture recognition result to the corresponding communication command. Both the transmitting programmable information metasurface and the receiving programmable information metasurface are programmable information metasurfaces as described in claim 2.

8. The metasurface-assisted communication and sensing integrated system of claim 7, wherein: The human posture is clearly defined as a variety of highly distinguishable static postures.

Images