CN113905383B - IFF signal identification method, device and medium based on radio frequency fingerprint - Google Patents
IFF signal identification method, device and medium based on radio frequency fingerprint Download PDFInfo
- Publication number
- CN113905383B CN113905383B CN202110988297.7A CN202110988297A CN113905383B CN 113905383 B CN113905383 B CN 113905383B CN 202110988297 A CN202110988297 A CN 202110988297A CN 113905383 B CN113905383 B CN 113905383B
- Authority
- CN
- China
- Prior art keywords
- constellation
- radio frequency
- iff
- signal
- frequency fingerprint
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000010586 diagram Methods 0.000 claims abstract description 72
- 238000013135 deep learning Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims abstract description 15
- 239000003086 colorant Substances 0.000 claims abstract description 13
- 238000013507 mapping Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 30
- 238000013527 convolutional neural network Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 12
- 238000012886 linear function Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 230000009545 invasion Effects 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 46
- 230000000875 corresponding effect Effects 0.000 description 18
- 238000004891 communication Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 230000004913 activation Effects 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000002346 layers by function Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W12/00—Security arrangements; Authentication; Protecting privacy or anonymity
- H04W12/60—Context-dependent security
- H04W12/69—Identity-dependent
- H04W12/79—Radio fingerprint
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an IFF signal identification method, device and medium based on radio frequency fingerprint, the method comprises the following steps: mapping the I/Q baseband signals obtained after IFF signal processing to a complex plane to draw a constellation diagram, calculating density values corresponding to constellation points in the constellation diagram, and marking the constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram serving as radio frequency fingerprint data; inputting the radio frequency fingerprint data of the IFF signals into a pre-configured deep learning network to identify radio frequency fingerprint characteristics and obtain an identification classification result; and demodulating, decoding and reading the IFF signal to obtain the contained information, and stopping receiving the IFF signal and stopping sending the information to the source of the IFF signal if the information or the identification and classification result is unknown equipment. The invention can effectively extract the radio frequency fingerprint data of the IFF signal, and guide the radio frequency fingerprint data into the depth residual convolution neural network to effectively identify and classify, thereby effectively preventing camouflage invasion.
Description
Technical Field
The present invention relates to the field of signal identification, and in particular, to an IFF signal identification method, apparatus, and medium based on radio frequency fingerprint.
Background
The IFF (individual identity recognition system) is one of important means for confirming the identity of modern aviation wireless communication, and can rapidly and conveniently monitor and communicate with aviation equipment in an airspace on the premise of guaranteeing information safety. The rapid development of modern aviation technology puts forward higher requirements on the utilization rate of the air space, the IFF signal identification can be used as an important source for obtaining the state information of the electronic aviation equipment of the opposite party, and the characteristics of the opposite party aircraft, the travel and other important information can be counted according to analysis, identification and research on the received IFF signal. This information is important in airspace monitoring and relates to the flight safety of the aircraft throughout the airspace.
Because the IFF signals interact in an open communication medium, the wireless signals transmitted by any one of the IFF wireless devices can be received by all users within its coverage area. By the characteristics, the IFF equipment is extremely high in convenience and flexibility in deployment, obvious information potential safety hazards are generated, malicious users can eavesdrop interaction information among legal equipment in an open IFF wireless communication medium, so that information data of legal user identity identification and authentication methods are eavesdrop, the obtained information is cracked by utilizing a high-performance computer, and finally the identity of a legal user can be counterfeited to actively initiate camouflage invasion to a legal IFF wireless network.
Radio frequency fingerprinting (Radio Frequency Fingerprinting, RFF) is considered an effective way to increase the security of wireless information networks. Radio frequency fingerprinting refers to the specific effect of the inherent defects of a wireless network device in the manufacturing process on radio frequency signals. The identity of the wireless device is identified using the unique and unique rf fingerprint characteristic extracted from the rf signal.
Basically, the rf fingerprint is mainly derived from the electronic component tolerances of the rf circuit part of the device itself. Specifically, the tolerance of the element mainly considers two aspects of manufacturing tolerance and drift tolerance. The manufacturing tolerance is mainly introduced in the links of actual production, assembly and debugging of the equipment. Because of non-ideal factors such as manufacturing process, hardware processing precision and the like, a certain gap exists between an actual electrical parameter value and a nominal value when the element leaves a factory. The existence of component tolerances makes certain subtle differences even for the same model of equipment produced by the same manufacturer. Drift tolerance refers to the change in component values during use of the device caused by aging of the device over time, and changes in the surrounding environment such as temperature and humidity during operation. This difference in hardware is sufficient to meet the requirements of the communication standard, but can have a different effect on the electromagnetic wave signal transmitted by the transmitter. These effects are reflected in particular in the electromagnetic wave signals of the transmitters, so that even under the same information stimulus, their output signals exhibit different characteristics and can thus be correlated with the identity of the individual transmitters.
Machine learning is the mainstream method for solving many artificial intelligence problems at the present stage, and learns data according to the process of simulating human life learning. The deep learning method has good advantage in feature learning capability, and the feature extraction and representation capability is stronger than that of the traditional machine learning method. Deep learning methods have emerged in a number of successful applications in image, speech recognition, autopilot, etc., and recent research has begun to attempt to introduce deep learning methods into wireless device identification methods. The patent CN109684995a discloses a specific radiation source identification method and device based on a depth residual error network, aiming at the characteristics of non-stability and non-linearity of communication signals, a gray image of a spectrum at the time of Hilbert is used as a representation form of signals, a depth residual error network is utilized to extract radio frequency fingerprint characteristics of the radiation source to finish classification identification, but because more fine characteristics are contained in the spectrum image at the time of Hilbert, the training difficulty of deep learning is increased, in addition, direct time-frequency analysis is carried out on received intermediate frequency or radio frequency signals by adopting Hilbert-Huang conversion, the analysis is carried out without unified conversion to a baseband, the Hilbert-Huang conversion mainly comprises two processes of Empirical Mode Decomposition (EMD) and Hilbert conversion, and key parameters of the Empirical Mode Decomposition (EMD) are the total layer number N of decomposition, so as to obtain an Intrinsic Mode Function (IMF) and a residual mode component. Different carrier frequencies (signal frequencies) are sensitive to the selection of the total layer number N of Empirical Mode Decomposition (EMD), even if the same transmitter selects different carrier frequency transmitting signals, the obtained Hilbert-Huang conversion time-frequency result is different by adopting the same EMD decomposition mode. Input into a subsequent deep learning network for classification recognition may not achieve the desired result.
Disclosure of Invention
The invention aims to solve the technical problems: aiming at the problems in the prior art, the method, the device and the medium for identifying the IFF signal based on the radio frequency fingerprint are provided, the radio frequency fingerprint data of the IFF signal can be effectively extracted, and the IFF signal is led into a depth residual convolution neural network to be effectively identified and classified, so that camouflage invasion can be effectively prevented.
In order to solve the technical problems, the invention adopts the following technical scheme:
an IFF signal identification method based on radio frequency fingerprints comprises the following steps:
s1) acquiring an IFF signal, processing the IFF signal to obtain an I/Q baseband signal, mapping the I/Q baseband signal onto a complex plane to draw a constellation diagram, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
s2) inputting the radio frequency fingerprint data of the IFF signals into a pre-configured deep learning network to identify radio frequency fingerprint characteristics and obtain an identification classification result;
s3) demodulating, decoding and reading the IFF signal to obtain the contained information, and stopping receiving the IFF signal and stopping sending information to the source of the IFF signal if the information or the identification and classification result is unknown equipment.
Further, the calculating of the corresponding density value of each constellation point in the constellation in step S1) includes the following steps:
s11) selecting one constellation point in a constellation diagram as a current constellation point, setting a corresponding effective area by taking the current constellation point as a center, and calculating the ratio of the effective area to the number of constellation points in the constellation diagram to be used as a density value of the current constellation point;
s12) returning to step S11) until a density value for each constellation point in the constellation is obtained.
Further, the different colors in step S1) are specifically gray values or color values corresponding to the density values one by one.
Further, in step S11), the effective area is square, and the function expression of the density value of the current constellation point is as follows:
in the above formula, i represents the sequence number of the current constellation point, b represents the square side length centered on the current constellation point, and h (P i )，v(P i ) The values of the ordinate and abscissa of the current constellation point are respectively, N is the number of all constellation points in the constellation region, j is the serial number of other constellation points in the constellation, and h (P j )，v(P j ) The values of the ordinate and abscissa of the other constellation points, respectively.
Further, step S1) is preceded by a step of configuring a deep learning network, specifically including: the method comprises the steps of constructing a depth residual convolution neural network, and inputting a preset signal model as sample data into the depth residual convolution neural network for training and learning, wherein the depth residual convolution neural network comprises 6 structural residual layers and 1 dimension matching layer which are sequentially connected, and the structural residual layers comprise a 1 multiplied by 1 convolution module, 2 residual units and 1 pooling layer which are sequentially cascaded.
Further, the residual unit comprises two cascaded convolution layers, the first convolution layer is activated by using a linear rectification function, the second convolution layer is activated by using a linear function, and short circuit connection is added to add the input feature map and the output feature map subjected to convolution operation, so that the difference of the input and output of the total feature map is forced to be learned by the network in the training process.
Further, the depth residual convolutional neural network performs training learning by using random gradient descent, the value of weight attenuation is set to be 0.0001, the value of momentum is set to be 0.9, and the value of learning rate is set to be 0.1.
Further, when the depth residual convolutional neural network performs training learning, if the error rate stagnates, the value of the learning rate is divided by 10.
The invention also provides an IFF signal identification device based on the radio frequency fingerprint, which comprises:
the radio frequency fingerprint data processing module is used for processing the IFF signal to obtain an I/Q baseband signal, drawing a constellation diagram according to the I/Q baseband signal, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
the depth residual error neural network classification recognition module is used for constructing a depth residual error convolutional neural network, inputting a preset signal model as sample data into the depth residual error convolutional neural network for training and learning, inputting radio frequency fingerprint data of an IFF signal into the trained depth residual error convolutional neural network for recognizing radio frequency fingerprint characteristics and obtaining a recognition classification result;
and the masquerading attack judging module is used for demodulating, decoding and reading the information contained in the IFF signal, and if the information or the identification and classification result is unknown equipment, confirming that masquerading attack exists, stopping receiving the IFF signal and stopping sending the information to the source of the IFF signal.
The invention also proposes a computer readable storage medium storing a computer program programmed or configured to perform any of the radio frequency fingerprint based IFF signal identification methods.
Compared with the prior art, the invention has the following advantages:
1. the invention maps the I/Q signal to the complex plane to obtain the constellation diagram after the IFF signal is processed, and for the data processing on the constellation diagram, the invention adopts a region calculation mode to calculate the density distribution condition of constellation points on the constellation diagram, and marks the constellation points with different density values on the constellation diagram by using different colors, so that the region with larger density is highlighted, on one hand, the distinction between different IFF signals can be intuitively embodied, and on the other hand, new characteristics are added on the basis of the constellation diagram, and the accuracy of classification recognition after deep learning training is improved.
2. The deep learning network adopts the deep residual convolution neural network, effectively solves the problem of performance degradation of the neural network along with the increase of depth, improves the recognition accuracy, performs network optimization by using random gradient descent, sets a parameter value when the network performance is optimal, and ensures that the deep residual convolution neural network can reach an optimal state.
3. According to the invention, the IFF signal is identified and demodulated and decoded, and under the condition that the demodulation and decoding result is inconsistent with the identification and classification result, the possible camouflage attack is determined, and the corresponding processing is carried out, so that the safety performance is improved.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the invention.
Fig. 2 is a schematic diagram of constellation point density calculation in an embodiment of the present invention.
Fig. 3 is a schematic diagram of a constellation diagram converted into a constellation point density diagram according to an embodiment of the present invention.
Fig. 4 is a constellation point density chart obtained by processing different IFF signals in an embodiment of the present invention.
Fig. 5 is a schematic structural diagram of a structured residual layer of a depth residual convolutional neural network used in an embodiment of the present invention.
Fig. 6 is a schematic diagram of calculation of residual units of a depth residual convolutional neural network employed in an embodiment of the present invention.
Fig. 7 is a schematic structural diagram of a depth residual convolutional neural network constructed in an embodiment of the present invention.
Detailed Description
The invention is further described below in connection with the drawings and the specific preferred embodiments, but the scope of protection of the invention is not limited thereby.
As shown in fig. 1, the invention provides an IFF signal identification method based on radio frequency fingerprint, comprising the following steps:
s1) acquiring an IFF signal, processing the IFF signal to obtain an I/Q baseband signal, wherein the I/Q baseband signal comprises an in-phase component (I component) and a quadrature component (Q component) of the baseband signal, mapping the I/Q baseband signal onto a complex plane to draw a constellation diagram, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
s2) inputting the radio frequency fingerprint data of the IFF signals into a pre-configured deep learning network to identify radio frequency fingerprint characteristics and obtain an identification classification result;
s3) demodulating, decoding and reading the IFF signal to obtain the contained information, and stopping receiving the IFF signal and stopping sending information to the source of the IFF signal if the information or the identification and classification result is unknown equipment.
In this embodiment, the rf fingerprint data may be effectively extracted based on the constellation diagram. In the field of digital communications, in order to more intuitively represent the mapping relationship between sampled signals, sampling points used for decision are usually plotted on a complex plane, and the resulting diagram of this process is called a constellation diagram. The constellation diagram is very useful for researching the relation of the I/Q baseband signals, and provides a convenient way for identifying modulation problems such as amplitude unbalance, amplitude noise, related interference, quadrature error, phase error, modulation error ratio and the like. The constellation of the communication signal includes the characteristic of the stationary part, so that the constellation can be used as the radio frequency fingerprint. Each symbol on the constellation can be represented by a complex number to be considered as a set of points on the complex plane, with the horizontal real axis representing the I component and the vertical imaginary axis representing the Q component. The difference between different IFF signals can be intuitively known according to the distribution condition of constellation points on a constellation diagram.
In general, a constellation diagram uses one color to display constellation points and uses another color as a base color, in this case, features of the constellation diagram are not obvious enough, in a machine learning process by using the constellation diagram, training difficulty is high, and accuracy is affected, for this embodiment, the constellation diagram is further processed, in step S1), a density value corresponding to each constellation point in the constellation diagram is calculated first, including the following steps:
s11) selectingTaking one constellation point in the constellation diagram as a current constellation point, setting a corresponding effective area by taking the current constellation point as a center, calculating the ratio of the number of the constellation points in the effective area and the constellation diagram to obtain the density value of the effective area, and taking the density value of the current constellation point as the density value of the current constellation point, as shown in fig. 2, assuming that N constellation points exist on the constellation diagram in a certain period of time, starting from the upper left corner of the constellation diagram, sequentially marking each constellation point in the left-to-right and top-to-bottom directions, wherein the first constellation point is P 0 The last constellation point is P N-1 With the current constellation point P i For the center, a square with a side length of b is used as an effective area to calculate a constellation point P i Density values of (2);
s12) returning to the step S11) until the density value of each constellation point in the constellation diagram is obtained, traversing all constellation points, and calculating the density map of the constellation diagram.
And then configuring colors corresponding to the density values one by one, respectively marking each constellation point with the corresponding color to obtain a constellation point density map, wherein the colors corresponding to the density values one by one can be gray values corresponding to the density values one by one or can be color values corresponding to the density values one by one, and because the range of the gray values is limited, all the density values can not be represented, and in the embodiment, the color values corresponding to the density values one by one are adopted, and the density values of all the constellation points in the constellation map are represented through the color picture.
The effect of converting the constellation diagram into the constellation point density diagram according to the step S1 is shown in fig. 3, in the actual situation, for three different IFF signals, the constellation point density diagram obtained by processing according to the step S1 is shown in fig. 4, it can be obviously seen that each constellation point in the constellation point density diagram is marked with a color corresponding to the density value thereof, compared with the constellation diagram, the distribution situation of the constellation points can be more intuitively reflected, the extraction of features in machine learning training is more facilitated, and the recognition accuracy is improved.
In step S11) of the present embodiment, the density value of the effective area is a ratio of all constellation points in the effective area to the total number of constellation points in the constellation map, and since the effective area is square in this embodiment, the difference between the abscissas of all constellation points in the effective area and the current constellation point is less than half of the length of the effective area in the horizontal direction, the difference between the ordinates of all constellation points in the effective area and the current constellation point is less than half of the length of the effective area in the vertical direction, and the function expression of the density value of the current constellation point is as follows:
in the above formula, i represents the sequence number of the current constellation point, b represents the square side length centered on the current constellation point, and h (P i )，v(P i ) The values of the ordinate and abscissa of the current constellation point are respectively, N is the number of all constellation points in the constellation region, j is the serial number of other constellation points in the constellation, and h (P j )，v(P j ) The values of the ordinate and abscissa of the other constellation points, respectively.
The g [. Cndot ] function in formula (1) is:
the density value of the current i point constellation point is quantized by calculating the ratio of the constellation point which falls in a square area with the i point constellation point as the center and the side length of b to all constellation points of the constellation diagram area. And giving different colors to the current ith point constellation point according to the calculated density value. The constellation point density map finally shows the effect that the more points and the denser areas are highlighted.
Step S1) of the present embodiment further includes a step of configuring the deep learning network, specifically including: in the embodiment, as shown in fig. 5, the depth residual convolutional neural network integrates functional layers of the convolutional neural network on the basis of a traditional residual module to obtain an improved network structure suitable for signal identification, in order to fully exert the high efficiency of the residual module on information flow in operation, the basic unit of the traditional residual module is changed and overlapped, then a 1×1 liter dimension convolutional module is added again, and the 1×1 convolutional module, 2 residual units and 1 pooling layer are stacked in sequence, thereby forming a new structural residual layer.
In the structured residual layer, the convolution operation is firstly performed by adopting a 1×1 convolution kernel to complete the dimension rising of the number of image channels, so as to adjust the number of image channels, then, through two residual units, the residual unit structure is as shown in fig. 6, the residual unit functions to prevent the gradient from disappearing or gradient exploding in the training and learning process of the network, so that the expected network model cannot be reached (generally, in deep learning, the deeper network structure can obtain more representative characteristics, but the deeper network layer tends to have the problems of gradient disappearing and gradient exploding, the problem is mainly that the smaller the variation is when the network is deeper, the gradient gradually disappears, so that the residual is proposed to replace the original signal learning. And the last step is to carry out the pooling layer operation, and the scheme adopts the maximum pooling layer operation, so that the dimension of the image is reduced and the calculated amount is reduced while the main characteristics of the image are reserved.
As shown in fig. 6, the residual unit in this embodiment includes two sequential convolution layers, and uses a ReLU activation function and Linear activation function, respectively, and adds a short-circuit connection (shortcut connection) to add the input feature map x to the convolved output feature map F (x), so as to obtain the difference of the total feature map output H (x) from the input and output of the network learning forced in the training process. The specificity of the residual units in terms of structure and connections enables them to provide "direct" channels for the flow of information inside the network, and is not limited to just inside the residual units, but throughout the entire network model. For extremely deep network architectures, conventional convolutional layer stack connections often result in truncation problems of the error signal during the back propagation phase of the network training process: the number of layers is too large to cause the disappearance or explosion phenomenon of the gradient signal in the deep-to-shallow layer reflow process, so that the training iteration process is difficult to converge. The short circuit connection form inside the residual error unit provides an 'identity mapping' for the operation of the characteristic data, which opens up a direct channel for the flow of signals inside the network, whether forward or reverse, so that the signals can freely spread among the residual error operation units, thereby avoiding the problem of interception of gradient signals.
Let the input of the first residual unit in the network model be x l Output is y l The mapping formed by convolution operation is F (x l ,W l ) Wherein W is l ＝{W l,k And (2) the I1 is less than or equal to K, and represents the weight parameter of the kth convolution layer in the ith residual unit, and then:
y l ＝h(x l )+F(x l ,W l )
x l+1 ＝f(y l ) (3)
for the residual units of Identity Mapping (Identity Mapping), f (y l )＝y l ，h(x l )＝x l The above formula is rewritten as:
x l+1 ＝x l +F(x l ,W l ) (4)
recursively, the output feature map for the 1+1th residual unit can be expressed as:
x l+2 ＝x l+1 +F(x l+1 ,W l+1 )＝x l +F(x l ,W l )+F(x l+1 ,W l+1 ) (5)
then for any deep residual unit L and any shallow L inside the network model.
The above indicates the feature map x of any deep layer in the residual network L Can be mapped x by shallow features l And a series of residual mappings Σf therebetween, then after trainingThe gradient back propagation phase in the pass uses the chain derivative rule for the error signal generated by the loss function epsilon:
this suggests that the back propagation of the gradient with respect to the l-layer consists of two additive components: first is that it does not contain any weight layerSecond is +.The second is related to the weight layer between L and L>Additive part->Meaning that the information can be directly passed back to the shallow layer i instead of having to go back step by step between the weight layers in between, and in the batch gradient descent algorithm, it is not possible to make +.>Since the value of (2) is always-1, even if the value of the weight parameter of the network model is arbitrarily small, the occurrence of the gradient vanishing phenomenon can be prevented by the residual unit.
In this embodiment, the preset signal models include 19 types of fixed error radio frequency signals, such as error radio frequency signals with quadrature error, phase error, modulation error and other parameter combinations, and one type of radio frequency signal errors with different error parameter values from the 19 types of signals, 20 types of signal models are total, and the 20 types of signal models are input into the depth residual error convolutional neural network for training and learning after the corresponding constellation point density map is obtained according to steps S11) to S13), so as to extract the characteristics of each signal model.
As shown in fig. 7, the depth residual convolutional neural network in this embodiment is implemented by adopting a res net network technology, and comprises 32 total layers (only a convolutional layer, a full-connection layer and a dimension matching layer are calculated), including 6 structured residual operation layers, 1 dimension matching layer and 1 full-connection layer. The convolution operation of the network is mainly concentrated on a structured residual layer and a dimension matching layer, and in order to keep the mapping value domain of the original data sample in the initial section of the model, the dimension matching layer activation function adopts Linear activation, so that the dimension between network nodes can be matched, and the calculation is smoothly carried out. Meanwhile, a sequential combination of a ReLu function and a Linear function is adopted in a residual operation unit of the structured residual layer, so that a proper dynamic range limit is provided for deep feature mapping value fields in a network. In order to prevent overfitting caused by excessive ResNet layers, a random inactivation Dropout strategy is added into the model, and the probability Dropout is 0-1.
In this embodiment, the following parameter optimization is performed when the depth residual convolutional neural network is configured, so as to ensure that the depth residual convolutional neural network of this embodiment can achieve an optimal effect:
using color augmentation (enhancing the color data), generating sample training data for enhancing the color data, including image brightness, saturation and contrast change, and collecting data to prepare training a deep learning model, wherein certain serious conditions of classification data are frequently encountered when the data are collected, and the data set is too small to easily cause over fitting of the model, so that a data enhancing technology is adopted;
using batch normalization (batch normalization) after each convolution layer and before activating the function, the training of the deep learning network model is difficult because the network structure contains many hidden layers, each layer parameter changes and optimizes along with the training, so the input distribution of the hidden layers always changes, and each hidden layer faces the problem of related coefficient deviation. Along with the training, the input of each layer is not independent and distributed due to the fact that data of the previous layer is required to be adapted to new input distribution and falls into a saturation region, so that learning efficiency is low and even gradient disappears;
configuring a depth residual convolution neural network, training and learning by using SGD (random gradient descent), setting a weight decay value to be 0.0001, and setting a momentum value to be 0.9, and enabling a trained model to meet the requirement through an empirical value;
setting the learning rate (learning rate) to be 0.1 so as to adapt to the depth residual convolutional neural network in the embodiment, and dividing the learning rate by 10 if the error rate stagnates when the depth residual convolutional neural network performs training learning;
in the machine learning model, if the parameters of the model are too many and the training samples are too few, the trained model is easy to generate the phenomenon of over fitting. The specific expression is as follows: the model has smaller loss function on training data and higher prediction accuracy; however, the loss function is larger on the test data, and the prediction accuracy is lower. Dropout can effectively relieve the occurrence of overfitting, has an integrated learning effect, and when Dropout propagates forwards, the activation value of a certain neuron stops working with a certain probability p, so that the model generalization is stronger, and certain local features are not relied on.
According to the method described above, the present invention further provides an IFF signal identification device based on radio frequency fingerprint, including:
the radio frequency fingerprint data processing module is used for processing the IFF signal to obtain an I/Q baseband signal, drawing a constellation diagram according to the I/Q baseband signal, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
the depth residual error neural network classification recognition module is used for constructing a depth residual error convolutional neural network, inputting a preset signal model as sample data into the depth residual error convolutional neural network for training and learning, inputting radio frequency fingerprint data of an IFF signal into the trained depth residual error convolutional neural network for recognizing radio frequency fingerprint characteristics and obtaining a recognition classification result;
and the masquerading attack judging module is used for demodulating, decoding and reading the information contained in the IFF signal, and if the information or the identification and classification result is unknown equipment, confirming that masquerading attack exists, stopping receiving the IFF signal and stopping sending the information to the source of the IFF signal.
The invention also proposes a computer readable storage medium storing a computer program programmed or configured to perform any of the radio frequency fingerprint based IFF signal identification methods.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. While the invention has been described with reference to preferred embodiments, it is not intended to be limiting. Therefore, any simple modification, equivalent variation and modification of the above embodiments according to the technical substance of the present invention shall fall within the scope of the technical solution of the present invention.
Claims (8)
1. The IFF signal identification method based on the radio frequency fingerprint is characterized by comprising the following steps of:
the configuration of the deep learning network specifically comprises the following steps: constructing a depth residual convolution neural network, and inputting a preset signal model as sample data into the depth residual convolution neural network for training and learning;
s1) acquiring an IFF signal, processing the IFF signal to obtain an I/Q baseband signal, mapping the I/Q baseband signal onto a complex plane to draw a constellation diagram, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
s2) inputting radio frequency fingerprint data of an IFF signal into a preconfigured deep learning network to identify radio frequency fingerprint features and obtain an identification classification result, wherein the deep residual convolution neural network comprises 6 structural residual layers and 1 dimension matching layer which are sequentially connected, the structural residual layers comprise 1X 1 convolution modules, 2 residual units and 1 pooling layer which are sequentially cascaded, the dimension matching layer is activated by adopting a linear function, the residual units comprise two cascaded convolution layers, the first convolution layer is activated by using a linear rectification function, the second convolution layer is activated by using a linear function, short circuit connection is added to add an input feature map and an output feature map subjected to convolution operation, and therefore the difference of network learning input and output is forced in the training process by obtaining total feature map output;
s3) demodulating, decoding and reading the IFF signal to obtain the contained information, and stopping receiving the IFF signal and stopping sending information to the source of the IFF signal if the information is inconsistent with the identification and classification result.
2. The method for identifying an IFF signal based on radio frequency fingerprints according to claim 1, wherein calculating the density value corresponding to each constellation point in the constellation in step S1) includes the steps of:
s11) selecting one constellation point in a constellation diagram as a current constellation point, setting a corresponding effective area by taking the current constellation point as a center, and calculating the ratio of the effective area to the number of constellation points in the constellation diagram to be used as a density value of the current constellation point;
s12) returning to step S11) until a density value for each constellation point in the constellation is obtained.
3. The method for identifying IFF signals based on radio frequency fingerprints as recited in claim 1, wherein the different colors in step S1) are gray values or color values corresponding to the density values one by one.
4. The method for identifying an IFF signal based on radio frequency fingerprints according to claim 2, wherein in step S11), the effective area is square, and the function expression of the density value of the current constellation point is as follows:
in the above formula, i represents the sequence number of the current constellation point, b represents the square side length centered on the current constellation point, and h (P i )，v(P i ) The values of the ordinate and abscissa of the current constellation point are respectively, N is the number of all constellation points in the constellation region, j is the serial number of other constellation points in the constellation, and h (P j )，v(P j ) The values of the ordinate and abscissa of the other constellation points, respectively.
5. The method for identifying IFF signals based on radio frequency fingerprints according to claim 1, wherein the depth residual convolutional neural network performs training learning using random gradient descent, sets a value of weight decay to 0.0001, a value of momentum to 0.9, and a value of learning rate to 0.1.
6. The method for identifying an IFF signal based on a radio frequency fingerprint according to claim 1, wherein when the depth residual convolutional neural network performs training learning, if the error rate stagnates, the value of the learning rate is divided by 10.
7. An IFF signal identification device based on radio frequency fingerprint, comprising:
the radio frequency fingerprint data processing module is used for processing the IFF signal to obtain an I/Q baseband signal, drawing a constellation diagram according to the I/Q baseband signal, calculating density values corresponding to constellation points in the constellation diagram, marking constellation points with different density values on the constellation diagram by using different colors to obtain a constellation point density diagram, and taking the constellation point density diagram as radio frequency fingerprint data of the IFF signal;
the depth residual error neural network classification recognition module is used for constructing a depth residual error convolutional neural network, a preset signal model is used as sample data to be input into the depth residual error convolutional neural network for training and learning, radio frequency fingerprint data of an IFF signal is input into the trained depth residual error convolutional neural network for recognizing radio frequency fingerprint characteristics and obtaining recognition classification results, the depth residual error convolutional neural network comprises 6 structural residual error layers and 1 dimension matching layer which are sequentially connected, the structural residual error layers comprise 1X 1 convolutional modules, 2 residual error units and 1 pooling layer which are sequentially cascaded, the dimension matching layer is activated by adopting a linear function, the residual error units comprise two cascaded convolutional layers, a first convolutional layer is activated by adopting a linear rectification function, a second convolutional layer is activated by adopting a linear function, and short circuit connection is added to add an input characteristic map and an output characteristic map which is subjected to convolution operation, so that the difference between network learning input and output is forced in the training process of the total characteristic map output is obtained;
and the masquerading attack judging module is used for demodulating, decoding and reading the information contained in the IFF signal, and if the information is authentication equipment but the identification and classification result is unknown equipment, confirming that masquerading attack exists, stopping receiving the IFF signal and stopping sending the information to the source of the IFF signal.
8. A computer readable storage medium storing a computer program programmed or configured to perform the radio frequency fingerprint based IFF signal identification method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110988297.7A CN113905383B (en) | 2021-08-26 | 2021-08-26 | IFF signal identification method, device and medium based on radio frequency fingerprint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110988297.7A CN113905383B (en) | 2021-08-26 | 2021-08-26 | IFF signal identification method, device and medium based on radio frequency fingerprint |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113905383A CN113905383A (en) | 2022-01-07 |
| CN113905383B true CN113905383B (en) | 2024-02-06 |
Family
ID=79188031
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110988297.7A Active CN113905383B (en) | 2021-08-26 | 2021-08-26 | IFF signal identification method, device and medium based on radio frequency fingerprint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113905383B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117676588A (en) * | 2022-08-31 | 2024-03-08 | 华为技术有限公司 | Equipment identification method, device and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015180465A1 (en) * | 2014-05-30 | 2015-12-03 | 江苏中兴微通信息科技有限公司 | Signalling field of wireless mimo communication system and communication method thereof |
| CN105357014A (en) * | 2015-11-25 | 2016-02-24 | 东南大学 | Wireless equipment radio frequency fingerprint feature extraction method based on differential constellation track diagram |
| CN109085613A (en) * | 2018-06-12 | 2018-12-25 | 东南大学 | The recognition methods of satellite cheating interference and device based on constellation track |
| CN109684995A (en) * | 2018-12-22 | 2019-04-26 | 中国人民解放军战略支援部队信息工程大学 | Specific Emitter Identification method and device based on depth residual error network |
| CN110309854A (en) * | 2019-05-21 | 2019-10-08 | 北京邮电大学 | A kind of signal modulation mode recognition methods and device |
| CN112312457A (en) * | 2020-10-29 | 2021-02-02 | 上海电机学院 | Communication radiation source individual identification method based on complex depth residual error network |
| CN112398553A (en) * | 2020-11-03 | 2021-02-23 | 上海电机学院 | Communication radiation source individual identification method based on differential equipotential sphere diagram |
| CN112507855A (en) * | 2020-12-04 | 2021-03-16 | 国网浙江省电力有限公司武义县供电公司 | Communication radiation source individual identification method based on instantaneous envelope equipotential sphere diagram |
| CN112953871A (en) * | 2021-02-23 | 2021-06-11 | 北京邮电大学 | New signal modulation format identification method based on neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2437352B (en) * | 2006-04-21 | 2011-06-29 | Tektronix Int Sales Gmbh | Constellation patterns |
-
2021
- 2021-08-26 CN CN202110988297.7A patent/CN113905383B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015180465A1 (en) * | 2014-05-30 | 2015-12-03 | 江苏中兴微通信息科技有限公司 | Signalling field of wireless mimo communication system and communication method thereof |
| CN105357014A (en) * | 2015-11-25 | 2016-02-24 | 东南大学 | Wireless equipment radio frequency fingerprint feature extraction method based on differential constellation track diagram |
| CN109085613A (en) * | 2018-06-12 | 2018-12-25 | 东南大学 | The recognition methods of satellite cheating interference and device based on constellation track |
| CN109684995A (en) * | 2018-12-22 | 2019-04-26 | 中国人民解放军战略支援部队信息工程大学 | Specific Emitter Identification method and device based on depth residual error network |
| CN110309854A (en) * | 2019-05-21 | 2019-10-08 | 北京邮电大学 | A kind of signal modulation mode recognition methods and device |
| CN112312457A (en) * | 2020-10-29 | 2021-02-02 | 上海电机学院 | Communication radiation source individual identification method based on complex depth residual error network |
| CN112398553A (en) * | 2020-11-03 | 2021-02-23 | 上海电机学院 | Communication radiation source individual identification method based on differential equipotential sphere diagram |
| CN112507855A (en) * | 2020-12-04 | 2021-03-16 | 国网浙江省电力有限公司武义县供电公司 | Communication radiation source individual identification method based on instantaneous envelope equipotential sphere diagram |
| CN112953871A (en) * | 2021-02-23 | 2021-06-11 | 北京邮电大学 | New signal modulation format identification method based on neural network |
Non-Patent Citations (5)
| Title |
|---|
| Fast Deep Learning for Automatic Modulation Classification;Sharan Ramjee等;Asilomar Conference on Signals, Systems and Computers;正文第8页，图6-7 * |
| 一种基于轮廓特征的射频指纹识别方法;崔正阳;胡爱群;彭林宁;;信息网络安全(第10期);全文 * |
| 基于星座图聚类分析的QAM信号调制识别算法及其DSP实现;王希维;;电子元器件应用(第06期);全文 * |
| 基于残差神经网络的辐射源个体识别;张宁;;航天电子对抗(第04期);全文 * |
| 基于深度学习的通信信号调制识别研究;罗兵兵;中国优秀硕士学位论文全文数据库;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113905383A (en) | 2022-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108764077B (en) | Digital signal modulation classification method based on convolutional neural network | |
| He et al. | Cooperative specific emitter identification via multiple distorted receivers | |
| CN112702294B (en) | Modulation recognition method for multi-level feature extraction based on deep learning | |
| Sa et al. | Specific emitter identification techniques for the internet of things | |
| CN112312457B (en) | Communication radiation source individual identification method based on complex depth residual error network | |
| Zhao et al. | Classification of small UAVs based on auxiliary classifier Wasserstein GANs | |
| CN110166387A (en) | A kind of method and system based on convolutional neural networks identification signal modulation system | |
| CN108566260A (en) | It is a kind of based on the concealed communication method for disturbing point multiple access | |
| Pospíšil et al. | Wireless device authentication through transmitter imperfections—Measurement and classification | |
| CN113905383B (en) | IFF signal identification method, device and medium based on radio frequency fingerprint | |
| CN113225282A (en) | Communication signal modulation identification method based on BP neural network | |
| Huang et al. | Deep learning of radio frequency fingerprints from limited samples by masked autoencoding | |
| Tamura et al. | Wireless devices identification with light-weight convolutional neural network operating on quadrant IQ transition image | |
| CN109076038B (en) | Method for estimating parameters of a signal contained in a frequency band | |
| Gul et al. | Fine-grained augmentation for RF fingerprinting under impaired channels | |
| Ying et al. | Differential complex-valued convolutional neural network-based individual recognition of communication radiation sources | |
| KR102347174B1 (en) | Ensemble based radio frequency fingerprinting apparatus and method of identifying emitter using the same | |
| CN114417914A (en) | Radio frequency fingerprint extraction and identification method based on multi-channel convolutional neural network | |
| Wang et al. | Specific emitter identification based on deep adversarial domain adaptation | |
| Wang et al. | Specific emitter identification based on fractal and wavelet theories | |
| Hreshee | Automatic recognition of the digital modulation types using the artificial neural networks. | |
| Wang et al. | A radio frequency fingerprinting identification method based on energy entropy and color moments of the bispectrum | |
| Qu et al. | Specific Emitter Identification Based on Two Residual Networks | |
| CN114201990A (en) | Communication interference identification method based on decision tree | |
| Zhou et al. | Specific emitter identification via feature extraction in Hilbert-Huang transform domain |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |