A Non-cooperative Unknown IoT Device Radio Frequency Sensing Method Based on Feedforward Neural Network

Abstract

This invention discloses a radio frequency (RF) sensing method for non-cooperative unknown IoT devices based on a feedforward neural network. This invention focuses on the communication signals emitted by non-cooperative unknown IoT devices. Since such devices inevitably need to emit communication signals to transmit the information they collect, capturing these signals can effectively improve the detection probability. This invention extracts communication signal features through a neural network, eliminating the need to obtain these features from the device supplier, thus significantly improving the feasibility and convenience of deployment. Feedforward neural networks are among the simplest neural network structures, but their representational capabilities are relatively weak, generally making it difficult to complete complex tasks. This invention employs a specific loss function, enabling the trained feedforward neural network to identify non-cooperative unknown IoT devices. Therefore, this invention has the advantage of lower hardware performance requirements compared to similar methods, thereby enhancing its deployment potential and reducing deployment costs.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology. Specifically, it relates to feature processing and pattern recognition of communication signals from unknown devices, which can be used for radio frequency sensing of non-cooperative unknown Internet of Things devices. Background Technology

[0002] With the development of IoT technology, IoT devices are widely used in smart homes, industrial production, smart logistics, and other fields. The application of IoT technology has greatly facilitated people's lives, while empowering traditional industries and creating new economic growth points.

[0003] While the Internet of Things (IoT) technology has brought immense value, some criminals have exploited this opportunity to commit malicious acts, posing new challenges to its applications. Intrusion by non-cooperative, unknown IoT devices is one such threat that urgently needs to be addressed. For example, intrusive IoT devices in smart homes can easily steal family members' privacy and lifestyle habits; intrusive IoT devices in smart business settings can lead to the leakage of trade secrets. These intrusive IoT devices typically attempt to disguise themselves and do not actively respond to the sensing and communication requests of detection devices, hence the term "non-cooperative devices." The existence of such threats has created an urgent need for technologies to detect non-cooperative, unknown IoT devices.

[0004] For such non-cooperative devices, radio frequency (RF) sensing is currently one of the most effective methods. This is because, although these devices do not actively respond to requests from sensing devices, they need to actively transmit communication signals to communicate with the outside world and transmit the information they have collected. Once this communication signal can be sensed and identified, the presence of an intruding IoT device can be determined. Existing signal detection methods for this type of signal are typically based on classical hypothesis testing theory, detecting the presence of the target signal from background noise. However, with the development of IoT technology, the actual deployed IoT systems are becoming increasingly complex, and the space is filled with electromagnetic waves from various device communications. The complex electromagnetic environment makes existing detection methods increasingly inadequate, and the detection results cannot meet the usage requirements. At the same time, to avoid legitimate devices in the scenario being accidentally detected as intruding IoT devices, it is necessary to know the communication protocols or signal characteristics of legitimate devices from their suppliers. This is impractical given the current situation where IoT devices are diverse, come from different suppliers, and suppliers are generally unwilling to disclose their trade secrets.

[0005] In conclusion, introducing feedforward neural network technology to process communication signals emitted by non-cooperative, unknown IoT devices is an effective method to address the challenges of detecting intrusive IoT devices. Developing this method can effectively counter the threat of intrusive IoT devices and promote the safe and efficient development of IoT technology and industry.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing methods for sensing non-cooperative unknown IoT devices by proposing a radio frequency (RF) sensing method for such devices based on a feedforward neural network. By leveraging the robustness of neural networks, it enables operation in complex electromagnetic environments. Furthermore, by extracting the communication signal characteristics emitted by IoT devices using neural networks, and then using these characteristics to determine the legitimacy of the devices, it eliminates the need to obtain information about the communication protocols or signal characteristics of legitimate devices from their suppliers. Finally, semantic space segmentation based on semantic centers enables efficient detection of non-cooperative unknown intrusive IoT devices, providing technical support for the secure and rapid development of IoT technology.

[0008] The objective of this invention is achieved through the following technical solution: a non-cooperative radio frequency sensing method for unknown IoT devices based on a feedforward neural network, comprising the following steps:

[0009] (1) Deployment of communication signal monitoring equipment: Several communication signal monitoring devices are deployed in scenarios where it is necessary to monitor intrusive IoT devices. The monitored radio frequency signal waveform x(t) is expressed as:

[0010] x(t)=s(t)+n(t), t=1, 2,...,T

[0011] Where T represents the duration of the radio frequency signal, s(t) represents the waveform of the communication signal transmitted by the IoT device, and n(t) represents the received noise;

[0012] (2) Feature extraction of communication signals: Different feature extraction strategies are selected according to different scenarios. All features of the extracted communication signals are expressed in vector form as follows:

[0013] X = [x1, x2, ..., x] G ]

[0014] Where, x g Let G represent the g-th feature extracted from x(t), and let G represent the total number of features extracted from x(t).

[0015] (3) Feature compression representation based on feedforward neural network: The features of communication signals are compressed and represented using a feedforward neural network, as specifically expressed below:

[0016] R i =ReLu(W i ×Ri-1 +b i ), i = 1, 2, ..., I

[0017] Where ReLu() represents a nonlinear activation function, R i R0 represents the output of the i-th layer of the feedforward neural network, where R0 = X and W i b represents the learnable weights of the i-th layer of the feedforward neural network. i Let R represent the learnable bias of the i-th layer of the feedforward neural network, and I represent the number of layers in the feedforward neural network; the output R of the last layer of the feedforward neural network... I Compressed representation as a characteristic of communication signals;

[0018] (4) Semantic space segmentation based on semantic center: During the training phase, M training data from N legitimate IoT devices are used, and various loss functions are introduced to adjust the learnable parameter W. i b i Training is performed; the m-th, m=1,2,...,M training data X m The output of the feedforward neural network is denoted as Let n be the device number to which the m-th training data belongs. m ∈{1, 2, ..., N}, n m Simultaneously, as the label of the m-th training data, the center vector of the device numbered n in the semantic space is denoted as c. n n = 1, 2, ..., N;

[0019] The loss function is expressed as follows:

[0020]

[0021] in:

[0022]

[0023] k and r are hyperparameters; L1 is the classification loss function, Softmax() expresses the Softmax function, and Classifier() represents a classification neural network; L2 is the reconstruction loss function, and Decoder() represents a decoding neural network; the loss function L3 aims to make the distance between the representations of all communication signals belonging to the same device in the training set as close as possible in the semantic space, and the distance between the representations of all communication signals belonging to different devices as far as possible, and this distance can be controlled by the hyperparameter r; the loss function L3 simultaneously constrains the difference and absolute distance between the representations of different classes in the semantic space, and the weight between these two constraints can be controlled by the hyperparameter k;

[0024] (5) Device Classification and Identification of Non-cooperative Unknown Intruding IoT Devices: Collect the radio frequency signals of all legitimate IoT devices in the scene and calculate the semantic center; the formula for calculating the semantic center of a legitimate IoT device numbered n is expressed as:

[0025]

[0026] Where, δ n () is a conditional function; if X m Let δ be a feature extracted from the radio frequency signal emitted by the IoT device numbered n. n (X m ) = 1, otherwise δ n (X m ) = 0; Through this step, the characteristics of communication signals emitted by legitimate IoT devices are identified and classified;

[0027] Each received communication signal waveform from an IoT device x u (t), to compute a low-dimensional compressed representation of signal features. The following formula can be used to determine whether a device is an unknown, intrusive IoT device:

[0028]

[0029] Where distance() expresses the Minkowski distance between vectors, and φ expresses the set distance threshold; when y u When the value is -1, determine the signal waveform x. u (t) indicates an intrusion from an unknown IoT device; otherwise, determine the signal waveform x. u (t) comes from known legitimate IoT devices, whose IDs n = y u .

[0030] Furthermore, the deployment of the communication signal monitoring equipment is specifically as follows: a virtual two-dimensional coordinate system is constructed in the scenario where it is necessary to monitor intrusive IoT devices; any point is selected as the origin of the coordinate system, any direction is selected as the positive x-axis, and its perpendicular direction is selected as the positive y-axis; starting from the origin of the coordinate system, several communication signal monitoring devices are evenly arranged along the positive and negative directions of the x and y axes at equal intervals d; each communication signal monitoring device can monitor the radio frequency signal waveform within a certain range centered on it, and can save and transmit it to a computer for processing and analysis.

[0031] Furthermore, the feature extraction of the communication signal specifically involves directly selecting the radio frequency signal waveform x(t) itself at all times t as the signal feature.

[0032] Furthermore, the feature extraction of the communication signal specifically involves: in some scenarios, if the characteristics of legitimate IoT devices and intrusive IoT devices are known, more efficient features can be extracted from x(t) using these characteristics, including: for devices with a stable bias in the radio frequency signal level, calculating the unbiased estimate of the first-order statistic mean X1 of x(t) as a feature; for devices whose AC power of the radio frequency signal is significantly different from other devices, calculating the unbiased estimate of the second-order statistic variance X2 of x(t) as a feature.

[0033] Furthermore, the feature extraction of the communication signal specifically involves: considering the sensing scenario of non-cooperative IoT devices, since most IoT devices adopt the MQTT protocol or conform to the OPC-UA standard, the packet structure and communication timing of the device's communication data packets have special characteristics, and consequently, the communication signal also has special characteristics. Specific features are extracted from these characteristics as communication signal features; the expression for the specific features is:

[0034]

[0035] Where x represents the calculated proprietary feature, elemax[·] represents taking the maximum value of all elements in a vector or matrix, ⊙ represents the element-wise multiplication of a vector or matrix; l(t) is a waveform designed according to the OPC-UA standard or MQTT protocol.

[0036] Furthermore, the specific design method for l(t) is as follows:

[0037] Take the fixed content specified in the OPC-UA standard or MQTT protocol, process this content according to the Internet protocol stack to obtain the corresponding baseband signal, and then modulate the baseband signal to obtain the modulated waveform l′(t). Next, in order to make the dimension of l′(t) the same as x(t) and satisfy the operation rules of vector element-wise multiplication, copy l′(t) into a column vector l′(t). The l′(t) obtained at this time can be used as l(t).

[0038] For legitimate IoT devices, the communication signal is measured directly when the legitimate device transmits a communication signal that conforms to the OPC-UA standard or MQTT protocol, and this measurement is taken as l(t).

[0039] Furthermore, the classification loss function makes the input X mm Able to be correctly classified into its corresponding device n m This ensures smooth classification; the reconstruction loss function makes the output of the feedforward neural network... It can be restored to its input X m Thus ensuring Sufficient information about X was retained. m The feature information. In the loss function L3, This constrains the differences in representations between different classes within the semantic space, ensuring that representations of the same type cluster together as much as possible, while representations of different types are kept as far apart as possible; in high-dimensional spaces that are difficult for humans to understand and imagine, regarding the distance between representations... It is a constraint method that conforms to the characteristics of the data; This is a direct constraint on the distance between representations within the semantic space, ensuring that the distance between representations of the same type is less than r, and the distance between representations of different types is greater than r. Applying a direct constraint to absolute distance is a more intuitive form of constraint. The function acts as a smoother, keeping the gradient calculation results within a reasonable range during the training process.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] (1) This invention focuses on the communication signals emitted by non-cooperative, unknown IoT devices. Since such devices inevitably need to emit communication signals to transmit the information they have collected to the outside world, capturing their communication signals can effectively improve their detection probability. In addition, this invention extracts the features of their communication signals through neural networks, eliminating the need to obtain these features from device suppliers, thus effectively improving the feasibility and convenience of deploying this invention.

[0042] (2) Feedforward neural networks are among the most concise neural network structures, but their representational capabilities are relatively weak, making it generally difficult to complete complex tasks. This invention employs a specific loss function, enabling the trained feedforward neural network to identify non-cooperative unknown IoT devices. Therefore, this invention has the advantage of lower hardware performance requirements compared to similar methods, thereby enhancing its deployment potential and reducing deployment costs. Attached Figure Description

[0043] Figure 1 This is the overall flowchart of the present invention.

[0044] Figure 2 This is a schematic diagram illustrating the use of the unique characteristics of communication signals from Internet of Things (IoT) devices in this invention. Detailed Implementation

[0045] The following is for reference only. Figure 1 The technical solution of the present invention will be further described in detail below.

[0046] Existing methods based on classical hypothesis testing theory for detecting non-cooperative, unknown intrusion IoT devices face numerous challenges, including complex electromagnetic environments and unknown communication protocols of legitimate IoT devices, resulting in low detection rates and failing to meet practical application requirements. Furthermore, in real-world applications, IoT devices often have poor computational performance, making it difficult to run complex neural networks. To address these challenges, this invention proposes a feedforward neural network-based radio frequency sensing method for non-cooperative, unknown IoT devices. The implementation steps of this invention are as follows:

[0047] (1) Deployment of Communication Signal Monitoring Equipment. In scenarios requiring monitoring of intrusive IoT devices, a sensor array is uniformly deployed. Specifically, a virtual two-dimensional coordinate system is constructed within this scenario. A point is chosen as the origin, a direction is chosen as the positive x-axis, and its perpendicular direction is chosen as the positive y-axis. Starting from the origin, several communication signal monitoring devices are uniformly deployed along the positive and negative x and y axes at equal intervals d. Each communication signal monitoring device can monitor the radio frequency signal waveform within a certain range centered on it, and can save and transmit it to a computer for processing and analysis. The monitored radio frequency signal waveform can be expressed as:

[0048] x(t)=s(t)+n(t), t=1, 2,...,T

[0049] Here, x(t) is a column vector with a dimension equal to the number of communication signal monitoring devices, T represents the duration of the radio frequency signal, and s(t) represents the waveform of the communication signal transmitted by the IoT device. At different times t, s(t) may originate from either a legitimate IoT device or an intrusive IoT device. When no IoT device is communicating in the scene, s(t) = 0. n(t) is the received noise, such as additive white Gaussian noise.

[0050] (2) Feature extraction of communication signals can be carried out by selecting different strategies according to different scenarios.

[0051] In general scenarios, the radio frequency signal waveform x(t) at all times t can be directly selected as the signal feature.

[0052] In some scenarios, if the characteristics of legitimate and intrusive IoT devices are known, these characteristics can be used to extract more efficient features from x(t). For example, if the RF signal levels of some devices exhibit a stable bias, the unbiased estimator of the first-order statistic mean X1 of x(t) can be calculated; if the AC power of the RF signal of some devices differs significantly from that of other devices, the unbiased estimator of the second-order statistic variance X2 of x(t) can be calculated, as shown in the following expression:

[0053]

[0054] in,[·] .2 The expression replaces each element of a vector or matrix with its square in turn.

[0055] Specifically, in the non-cooperative IoT device sensing scenario considered in this invention: because most IoT devices use the MQTT protocol or conform to the OPC-UA standard, the packet structure and communication timing of the device's communication data packets have unique characteristics, and consequently, their communication signals also have unique characteristics, from which a class of efficient proprietary features can be extracted. Appendix Figure 2 This illustration demonstrates how the present invention utilizes the unique characteristics of communication signals from Internet of Things (IoT) devices. Specifically, the communication data packets specified by the MQTT protocol contain fixed content. For example, the first byte of the packet header field expresses the packet name, the packet protocol name field contains the fixed four letters 'MQTT', and the packet version field contains a finite number of MQTT version numbers. Devices conforming to the OPC-UA standard also have uniform fixed content in their communication signals, as specified by the standard, such as: OpenConnection, Hello, and Acknowledge. Based on this fixed content, the method for calculating the unique characteristics is as follows:

[0056]

[0057] Where x represents the calculated proprietary feature, elemax[·] represents taking the maximum value of all elements in the vector or matrix, and ⊙ represents the element-wise multiplication of the vector or matrix. l(t) is a waveform designed according to the OPC-UA standard or MQTT protocol. The specific design method is as follows:

[0058] The process involves taking the fixed content specified in the OPC-UA standard or MQTT protocol, processing it according to the TCP / IP or other Internet protocol stack, obtaining the baseband signal corresponding to this fixed content, and then obtaining the modulated waveform l′(t) according to the common modulation method of IoT devices. Next, to ensure that the dimension of l′(t) is the same as x(t) and to satisfy the element-wise multiplication rule of vectors, l′(t) is copied into a column vector l′(t) (where each element is l′(t), and the dimension is the number of communication signal monitoring devices). The resulting l′(t) can then be used as l(t). Specifically, for legitimate IoT devices, the communication signal can be directly measured when the legitimate device transmits a communication signal conforming to the OPC-UA standard or MQTT protocol, and this measurement can be used as l(t).

[0059] In summary, all the features of the communication signal extracted in this step can be uniformly expressed in vector form as follows:

[0060] X = [x1, x2, ..., x] G ]

[0061] Where, x g Let X represent the g-th feature extracted from x(t), and G represent the total number of features extracted from x(t). In X, proprietary features extracted from the characteristics of IoT devices can effectively improve the detection of intrusive IoT devices.

[0062] (3) Feature compression representation based on feedforward neural network. Since the features extracted in step (2) have high dimensionality, their representation of the communication signal itself is not concise enough. Therefore, feature compression representation is used to extract an efficient representation of the communication signal to classify the source of the radio frequency signal. The feedforward neural network is used to compress and represent the features of the communication signal, as specifically expressed below:

[0063] R i =ReLu(W i ×R i-1 +b i ), i = 1, 2, ..., I

[0064] Where ReLu() represents a nonlinear activation function, R i This represents the output of the i-th layer of the feedforward neural network, specifically, R0 = X, W i b represents the learnable weights of the i-th layer of the feedforward neural network. i Let R represent the learnable bias of the i-th layer of the feedforward neural network, and I represent the layer number of the feedforward neural network. The output R of the last layer of the feedforward neural network... I This method utilizes compressed representations of communication signal characteristics. By inputting X into a feedforward neural network, the features of the communication signal can be represented more precisely. Furthermore, this method can also extract the characteristics of communication signals emitted by legitimate IoT devices. This allows for the identification of legitimate IoT devices even when information about their communication signals cannot be obtained from the manufacturer, thus preventing the misidentification of legitimate IoT devices as intrusive ones.

[0065] (4) Semantic space segmentation based on semantic center. In order to enable R to serve as an efficient representation of communication signal features, during the training phase, M training data from N legitimate IoT devices are used, and various loss functions are introduced to adjust the learnable parameter W. i b i Training is performed. The m-th, m=1,2,...,M training data points X are... m The output of the feedforward neural network is denoted as Let n be the device number to which the m-th training data belongs. m ∈{1,2,...,N},n mSimultaneously, as the label of the m-th training data, the center vector of the device numbered n in the semantic space is denoted as c. n Let n = 1, 2, ..., N. The introduced loss function is specifically expressed as follows:

[0066]

[0067] in:

[0068]

[0069] k and r are hyperparameters.

[0070] L1 is called the "classification loss function", Softmax() expresses the softmax function, and Classifier() represents a classification neural network. The classification loss function makes the input X... m Able to be correctly classified into its corresponding device n m This ensures the smooth progress of the classification process.

[0071] L2 is called the "reconstruction loss function". Decoder() represents a decoding neural network, and the reconstruction loss function makes the output of the feedforward neural network... It can be restored to its input X m Thus ensuring Sufficient information about X was retained. m The characteristic information.

[0072] For L3, overall, L3 serves the following purpose: to maximize the distance between representations of all communication signals belonging to the same device in the training set within the semantic space, and to maximize the distance between representations of all communication signals belonging to different devices, with this distance controllable by the hyperparameter *r*. Compared to existing techniques, the L3 loss function simultaneously constrains both the differences in representations between different classes within the semantic space and their absolute distances, balancing the advantages of both, and the weights between these two constraints can be controlled by the hyperparameter *k*. Specifically, the various parts of L3 include: This constrains the differences in representations between different classes within the semantic space, ensuring that similar representations cluster together as much as possible, while dissimilar representations are kept as far apart as possible. In high-dimensional spaces that are difficult for humans to understand and imagine, the distance between representations... It is a constraint method that conforms to the characteristics of the data; This is a direct constraint on the distance between representations in the semantic space, such that the distance between similar representations is less than r, and the distance between dissimilar representations is greater than r. Applying constraints directly to absolute distance is a more intuitive way of constraining relationships; furthermore, The function acts as a smoother, keeping the gradient calculation results within a reasonable range during the training process.

[0073] The learnable parameters W in the feedforward neural network are trained using the loss function described above. i b i This makes the output of the feedforward neural network... Enough information about input X has been retained. m The feature information, and The dimension is significantly lower than that of the input X. m ,therefore, It can serve as a low-dimensional compressed representation of signal features. Furthermore, due to the low-dimensional compressed representation of each type of signal... They all cluster at the center of their category c n The semantic space segmentation based on the semantic center is completed in the vicinity.

[0074] (5) Device Classification and Identification of Non-Cooperative Unknown Intruding IoT Devices. First, collect the radio frequency signals of all legitimate IoT devices in the scene and calculate their semantic centers. Specifically, the formula for calculating the semantic center of a legitimate IoT device numbered n is expressed as:

[0075]

[0076] Where, δ n () is a conditional function; if X m Let δ be a feature extracted from the radio frequency signal emitted by the IoT device numbered n. n (X m ) = 1, otherwise δ n (X m ) = 0. Through this step, the characteristics of the communication signals emitted by legitimate IoT devices are identified and categorized, thus eliminating the need to obtain the communication signal characteristics of the legitimate IoT devices from the device manufacturers. Based on this, each received communication signal waveform x from the IoT device... u (t), by calculating its low-dimensional compressed representation of signal features according to the above steps. Whether a device is an unknown, intrusive IoT device can be determined using a formula, which is expressed as follows:

[0077]

[0078] Here, `distance()` represents the Minkowski distance between vectors, and `φ` represents the set distance threshold. This threshold can be set based on the training data, where `y` is the distance between vectors. u The value of y is then used as a reference, combined with experience, to set the value. When y u When the value is -1, determine the signal waveform x. u (t) Originates from an unknown intruding IoT device. Otherwise, the signal waveform x u(t) comes from known legitimate IoT devices, whose IDs n = y u This enables radio frequency sensing of non-cooperative, unknown IoT devices.

[0079] The above description is merely a preferred embodiment of the present invention. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall still fall within the protection scope of the technical solutions of the present invention.

Claims

1. A non-cooperative radio frequency sensing method for unknown IoT devices based on feedforward neural networks, characterized in that, Includes the following steps: (1) Deployment of communication signal monitoring equipment: Several communication signal monitoring devices are deployed in scenarios where it is necessary to monitor intrusive IoT devices. The monitored radio frequency signal waveform x(t) is expressed as: x(t)=s(t)+n(t),t=1,2,…,T Where T represents the duration of the radio frequency signal, s(t) represents the waveform of the communication signal transmitted by the IoT device, and n(t) represents the received noise; (2) Feature extraction of communication signals: Different feature extraction strategies are selected according to different scenarios. All features of the extracted communication signals are expressed in vector form as follows: X=[x1,x2,…,x G ] Where, x g Let G represent the g-th feature extracted from x(t), and let G represent the total number of features extracted from x(t). (3) Feature compression representation based on feedforward neural network: The features of communication signals are compressed and represented using a feedforward neural network, as specifically expressed below: R i =ReLu(W i ×R i-1 +b i ),i=1,2,…,I Where ReLu() represents a nonlinear activation function, R i R0 represents the output of the i-th layer of the feedforward neural network, where R0 = X and W i The learnable weights of the i-th layer of the feedforward neural network, b i Let R represent the learnable bias of the i-th layer of the feedforward neural network, and I represent the number of layers in the feedforward neural network; the output R of the last layer of the feedforward neural network... I Compressed representation as a characteristic of communication signals; (4) Semantic space segmentation based on semantic center: During the training phase, M training data from N legitimate IoT devices are used, and various loss functions are introduced to adjust the learnable parameter W. i b i Perform training; take the m-th, m=1,2,…,M training data X m The output of the feedforward neural network is denoted as Let n be the device number to which the m-th training data belongs. m ∈{1,2,…,N},n m Simultaneously, as the label of the m-th training data, the center vector of the device numbered n in the semantic space is denoted as c. n n = 1, 2, ..., N; The loss function is expressed as follows: in: k and r are hyperparameters; L1 is the classification loss function, Softmax() expresses the Softmax function, and Classifier() represents a classification neural network; L2 is the reconstruction loss function, and Decoder() represents a decoding neural network; the loss function L3 aims to make the distance between the representations of all communication signals belonging to the same device in the training set as close as possible in the semantic space, and the distance between the representations of all communication signals belonging to different devices as far as possible, and this distance can be controlled by the hyperparameter r; the loss function L3 simultaneously constrains the difference and absolute distance between the representations of different classes in the semantic space, and the weight between these two constraints can be controlled by the hyperparameter k; (5) Device Classification and Identification of Non-cooperative Unknown Intruding IoT Devices: Collect the radio frequency signals of all legitimate IoT devices in the scene and calculate the semantic center; the formula for calculating the semantic center of a legitimate IoT device numbered n is expressed as: Where, δ n () is a conditional function; if X m Let δ be a feature extracted from the radio frequency signal emitted by the IoT device numbered n. n (X m ) = 1, otherwise δ n (X m ) = 0; Through this step, the characteristics of communication signals emitted by legitimate IoT devices are identified and classified; Each received communication signal waveform from an IoT device x u (t), to compute a low-dimensional compressed representation of signal features. The following formula can be used to determine whether a device is an unknown, intrusive IoT device: Where distance() expresses the Minkowski distance between vectors, and φ expresses the set distance threshold; when y u When the value is -1, determine the signal waveform x. u (t) indicates an intrusion from an unknown IoT device; otherwise, determine the signal waveform x. u (t) comes from known legitimate IoT devices, whose IDs n = y u .

2. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, The deployment of the communication signal monitoring equipment is as follows: a virtual two-dimensional coordinate system is constructed in the scenario where intrusion into IoT devices needs to be monitored; any point is selected as the origin of the coordinate system, any direction is selected as the positive x-axis, and its perpendicular direction is selected as the positive y-axis; starting from the origin of the coordinate system, several communication signal monitoring devices are evenly arranged along the positive and negative x and y axes at equal intervals d; each communication signal monitoring device can monitor the radio frequency signal waveform within a certain range centered on it, and can save and transmit it to a computer for processing and analysis.

3. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, The feature extraction of the communication signal specifically involves directly selecting the radio frequency signal waveform x(t) itself at all times t as the signal feature.

4. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, The feature extraction of the communication signal specifically involves: in some scenarios, if the characteristics of legitimate IoT devices and intrusive IoT devices are known, more efficient features can be extracted from x(t) using these characteristics, including: for devices with a stable bias in the radio frequency signal level, the unbiased estimator of the first-order statistic mean X1 of x(t) is calculated as a feature; for devices with significantly different AC power of radio frequency signals from other devices, the unbiased estimator of the second-order statistic variance X2 of x(t) is calculated as a feature.

5. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, The feature extraction of the communication signal specifically involves: considering the sensing scenario of non-cooperative IoT devices, since most IoT devices use the MQTT protocol or conform to the OPC-UA standard, the packet structure and communication timing of the device's communication data packets have special characteristics, and thus the communication signal also has special characteristics. Specific features are extracted from these characteristics as communication signal features; the expression for the specific features is: Where x represents the calculated proprietary feature, elemax[·] represents taking the maximum value of all elements in a vector or matrix, ⊙ represents the element-wise multiplication of a vector or matrix; l(t) is a waveform designed according to the OPC-UA standard or MQTT protocol.

6. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 5, characterized in that, The specific design method for l(t) is as follows: Take the fixed content specified in the OPC-UA standard or MQTT protocol, process this content according to the Internet protocol stack to obtain the corresponding baseband signal, and then modulate the baseband signal to obtain the modulated waveform l′(t). Next, in order to make the dimension of l′(t) the same as x(t) and satisfy the operation rules of vector element-wise multiplication, copy l′(t) into a column vector l′(t). The l′(t) obtained at this time can be used as l(t). For legitimate IoT devices, the communication signal is measured directly when the legitimate device transmits a communication signal that conforms to the OPC-UA standard or MQTT protocol, and this measurement is taken as l(t).

7. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, The classification loss function makes the input X m Able to be correctly classified into its corresponding device n m This ensures smooth classification; the reconstruction loss function makes the output of the feedforward neural network... It can be restored to its input X m Thus ensuring Sufficient information about X was retained. m The characteristic information.

8. The radio frequency sensing method for non-cooperative unknown IoT devices based on feedforward neural networks according to claim 1, characterized in that, In loss function L3, This constrains the differences in representations between different classes within the semantic space, ensuring that representations of the same type cluster together as much as possible, while representations of different types are kept as far apart as possible; in high-dimensional spaces that are difficult for humans to understand and imagine, regarding the distance between representations... It is a constraint method that conforms to the characteristics of the data; This is a direct constraint on the distance between representations within the semantic space, ensuring that the distance between representations of the same type is less than r, and the distance between representations of different types is greater than r. Applying a direct constraint to absolute distance is a more intuitive form of constraint. The function acts as a smoother, keeping the gradient calculation results within a reasonable range during the training process.

Images