Counter-airspace signal processing method and device based on deep learning
By employing deep learning-based 32-bit fixed-point arithmetic and FPGA modular technology, high-precision UAV signal analysis and countermeasures were achieved, reducing hardware resource consumption and equipment costs while improving portability.
Patent Information
- Application Number
- CN202610208981.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for countering drone signals suffer from poor adaptability of preset models, high hardware resource consumption, high cost, and poor portability.
Online learning is performed using a 32-bit fixed-point artificial neural network (ANN) based on deep learning, combined with modular implementation using FPGA, to generate digital source signals that match the UAV control signals, and then FSK modulation and directional transmission are performed using the DDS algorithm.
It improves signal resolution accuracy to 98%, reduces hardware resource consumption, meets real-time requirements, reduces equipment cost by 30%, and enhances portability.
Smart Images
Figure CN122027073A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of airspace signal countermeasures technology, specifically to a deep learning-based method and apparatus for countering airspace signal processing, applicable to signal suppression scenarios involving airspace targets such as drones. Background Technology
[0002] With the popularization of drone technology, illegal intrusions into airspace have become frequent, necessitating the use of signal countermeasures technology for control.
[0003] Existing countermeasure methods typically employ a process of "preset model analysis - signal imitation - modulation and transmission," but these methods suffer from the following drawbacks: First, the preset model has poor adaptability to nonlinear and time-varying UAV signals, resulting in insufficient analysis accuracy and low matching degree between the countermeasure signal and the target signal. Second, the signal processing uses a floating-point arithmetic architecture, consuming significant hardware resources and failing to meet the real-time requirements of dynamic airspace countermeasures. Third, the high-density hardware dependency leads to high equipment costs and poor portability. Therefore, this application proposes a deep learning-based countermeasure airspace signal processing method and apparatus that at least partially solves the aforementioned problems. Summary of the Invention
[0004] In view of the aforementioned problems, this application is made in order to provide a spatial signal countermeasure technology that overcomes or at least partially solves the aforementioned problems, improves the accuracy of complex signal analysis, reduces hardware resource consumption, and balances real-time performance and cost.
[0005] In a first aspect, the present invention provides a deep learning-based method for countering spatial signal processing, comprising the following steps: Step 1: Acquire UAV control signals, and use a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online, and parse the results to obtain the frequency, frequency point, and rate of the control signals; the artificial neural network (ANN) is implemented modularly using FPGA; Step 2: Based on the analysis results, generate a digital source signal with the same frequency as the UAV control signal; Step 3: Perform FSK modulation on the digital source signal using the DDS algorithm; Step 4: Amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure.
[0006] In some embodiments of the present invention, the acquisition of UAV control signals, and the online learning of the control signals using a 32-bit fixed-point arithmetic artificial neural network (ANN) to obtain analytical results containing the frequency, frequency point, and rate of the control signals, include: An artificial neural network (ANN) consisting of an input layer, a hidden layer, and an output layer is implemented using FPGA modularization. This ANN is a multilayer perceptron. The input to the input layer is the output error of the UAV signal and at least two delay values.
[0007] In some embodiments of the present invention, the FPGA module includes: CONTROL_UNIT, MAC_UNIT, ACTIVATION_FUNCTION, LEARNING_UNIT, and two RAM memories; The CONTROL_UNIT is used to coordinate the operation of the module, the MAC_UNIT performs multiplication and accumulation operations, and the LEARNING_UNIT updates the synaptic weights of the artificial neural network (ANN) through the backpropagation algorithm.
[0008] In some embodiments of the present invention, the ACTIVATION_FUNCTION employs a piecewise approximation of the sigmoid function, satisfying: When α≤-4, FA(α)=0; When -4<α<0, FA(α)=(α+4)² / 32; When 0≤α≤4, FA(α)=1-(α-4)² / 32; When α>4, FA(α)=1.
[0009] In some embodiments of the present invention, the step of acquiring UAV control signals, using a 32-bit fixed-point arithmetic artificial neural network (ANN) to perform online learning on the control signals, and parsing to obtain an analytical result containing the frequency, frequency point, and rate of the control signals, further includes: The acquisition of UAV control signals employs a 32-bit fixed-point artificial neural network (ANN), with the learning coefficient η set to a range of 0.07 to 0.5. The control signals are then learned online, and the analysis results containing the frequency, frequency point, and rate of the control signals are obtained.
[0010] In some embodiments of the present invention, the FPGA is any one of LATTICE ICE40UP5K, XC3S200A or XC6SLX4.
[0011] In some embodiments of the present invention, generating a digital source signal with the same frequency as the UAV control signal based on the parsing result includes: Based on the analysis results, the amplitude and phase of the digital source signal are adjusted by the adaptive output of the artificial neural network (ANN) to achieve a signal matching degree of ≥98%, and a digital source signal with the same frequency as the UAV control signal is generated.
[0012] In some embodiments of the present invention, the acquisition of UAV control signals includes: An ADC converter is used to acquire UAV control signals, and the sampling rate is dynamically adjusted according to the UAV signal frequency to output 32-bit fixed-point data.
[0013] In some embodiments of the present invention, the step of using the DDS algorithm to perform FSK modulation on the digital source signal includes: The DDS algorithm is used and the frequency offset range of FSK modulation is set to 10kHz~100kHz. The digital source signal is then subjected to FSK modulation and parsing to make the modulated signal rate consistent with the UAV signal rate.
[0014] Secondly, the present invention also provides a deep learning-based counter-spatial signal processing apparatus for implementing the above method, comprising: The signal acquisition module is used to acquire UAV control signals. It employs a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online and parse them to obtain analytical results containing the carrier frequency, modulation frequency offset, and symbol rate of the control signals. The signal acquisition module includes an antenna and an ADC converter. The antenna is an omnidirectional antenna or a multi-antenna array (portable devices use omnidirectional antennas, and vehicle-mounted devices use multi-antenna arrays). The ADC converter outputs 32-bit fixed-point data (format 8:24). The artificial neural network (ANN) is implemented modularly using an FPGA.
[0015] The signal generation module is used to generate a digital source signal with the same carrier frequency as the UAV control signal based on the analysis results. A modulation module is used to perform FSK modulation on the digital source signal using the DDS algorithm; The amplification and transmission module is used to amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure. The amplification and transmission module includes a power amplifier and a directional antenna. The output power of the power amplifier is in the range of 0.5W to 10W, and the beam angle of the directional antenna is in the range of 30° to 120°.
[0016] In some embodiments of the present invention, the signal acquisition module includes an antenna and an ADC converter, the amplification and transmission module includes a power amplifier and a directional antenna, and the ADC converter outputs 32-bit fixed-point data.
[0017] Thirdly, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement a deep learning-based counter-space signal processing method. The processor is an ARM Cortex-A series, Intel Core i series, or FPGA built-in processor (such as the LATTICE ICE40UP5K built-in processor); the memory capacity is ≥1GB, supporting 32-bit fixed-point data storage; the communication interface is used to connect the signal acquisition module, modulation module, and amplification and transmission module, supporting 32-bit fixed-point data transmission with a transmission delay ≤10μs.
[0018] Fourthly, the present invention also provides a computer-readable storage medium, characterized in that it stores a computer program, which, when executed by a processor, implements a deep learning-based method for countering spatial signal processing. The computer-readable storage medium is flash memory, hard disk, or SD card, and the storage format supports binary file storage and has data power-loss protection function; the computer program running environment is Linux, Windows, or FPGA embedded operating system, and supports parallel execution of 32-bit fixed-point arithmetic and neural network algorithms.
[0019] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: Firstly, this application improves the accuracy of complex signal analysis to over 98% through the online learning capability of ANN, significantly increasing the countermeasure success rate. Secondly, through the modular combination of FPGA and fixed-point arithmetic architecture, the DSP unit occupancy is ≤50%, the logic unit occupancy is ≤40%, and the signal processing latency is ≤29μs, meeting the requirements for real-time countermeasures. Thirdly, it is adapted to low-density FPGAs, reducing equipment costs by more than 30% and improving portability, making it widely applicable to portable and vehicle-mounted countermeasure scenarios. Fourthly, the segmented activation function simplifies hardware implementation and further reduces resource consumption. Attached Figure Description
[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating the steps of a deep learning-based counter-spatial signal processing method provided in one embodiment of this application. Figure 2 This is a schematic diagram of the FPGA modular structure of a deep learning-based counter-space signal processing method and device. Figure 3This is a schematic diagram of the control structure of a neural network for a deep learning-based method and device for countering spatial signal processing. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0023] The following will further describe a deep learning-based method and apparatus for countering spatial signal processing in this exemplary embodiment.
[0024] like Figure 1 As shown, an embodiment of this application provides a deep learning-based method for countering spatial signal processing, comprising: Step 1: Acquire UAV control signals, and use a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online, and parse the results to obtain the frequency, frequency point, and rate of the control signals; the artificial neural network (ANN) is implemented modularly using FPGA; An artificial neural network (ANN) comprising an input layer, hidden layer, and output layer is implemented modularly using an FPGA; this ANN is a multilayer perceptron; the input of the input layer consists of the output error of the UAV signal and at least two delay values. Signal acquisition employs an ADC converter, outputting 32-bit fixed-point data, with the sampling rate dynamically adjusted according to the UAV signal frequency.
[0025] like Figure 2 As shown, the FPGA module includes: CONTROL_UNIT, MAC_UNIT, ACTIVATION_FUNCTION, LEARNING_UNIT, and two RAM memories; The CONTROL_UNIT is used to coordinate the operation of the module, the MAC_UNIT performs multiplication and accumulation operations, and the LEARNING_UNIT updates the synaptic weights of the artificial neural network (ANN) through the backpropagation algorithm.
[0026] Figure 2The meanings of the labels in the code are as follows: CONTROL_UNIT (coordination module), MAC_UNIT (multiplication and accumulation operation module), ACTIVATION_FUNCTION (activation function module), LEARNING_UNIT (weight update module), RAM1 (intermediate variable storage module), RAM2 (weight parameter storage module). The modules are connected through a data bus to achieve data interaction. Figure 3 The meanings of each label are as follows: Input - input layer, hidden - hidden layer, output - output layer, error calculation unit (calculates output error e(t)), the input layer receives the error and delay value, after processing by the hidden layer, the output layer outputs the signal parameters, and at the same time updates the weights through error feedback.
[0027] Reference Figure 2 and Figure 3 As shown, the ACTIVATION_FUNCTION uses a piecewise approximation of the sigmoid function, satisfying: ; That is, when α≤-4, FA(α)=0; when -4<α<0, FA(α)=(α+4)² / 32; when 0≤α≤4, FA(α)=1-(α-4)² / 32; when α>4, FA(α)=1.
[0028] The FPGA is any one of LATTICE ICE40UP5K, XC3S200A or XC6SLX4.
[0029] Step 2: Based on the analysis results, generate a digital source signal with the same frequency as the UAV control signal; including: based on the analysis results, adjusting the amplitude and phase of the digital source signal through the adaptive output of the artificial neural network (ANN) to make the signal matching degree ≥98%, and generating a digital source signal with the same frequency as the UAV control signal.
[0030] Step 3: Apply FSK modulation to the digital source signal using the DDS algorithm; including: applying the DDS algorithm and setting the frequency offset range of FSK modulation to 10kHz~100kHz, applying FSK modulation rate and parsing to the digital source signal, resulting in a UAV signal rate consistent with the obtained rate.
[0031] Step 4: Amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure.
[0032] Furthermore, the step of acquiring the UAV control signal and using a 32-bit fixed-point artificial neural network (ANN) to learn the control signal online and obtain an analytical result containing the frequency, frequency point, and rate of the control signal further includes: acquiring the UAV control signal and using a 32-bit fixed-point artificial neural network (ANN) with a learning coefficient η ranging from 0.07 to 0.5 to learn the control signal online and obtain an analytical result containing the frequency, frequency point, and rate of the control signal.
[0033] It should be noted that in this application, frequency refers to carrier frequency, frequency point refers to modulation frequency deviation, and rate refers to symbol rate.
[0034] This invention provides a method and apparatus for countering spatial signal processing based on digital signal sources. The core of this method lies in combining 32-bit fixed-point ANN with FPGA modularity to optimize the signal learning and parsing process.
[0035] like Figure 2 As shown, the FPGA modular structure includes CONTROL_UNIT, MAC_UNIT, ACTIVATION_FUNCTION, LEARNING_UNIT and two RAM memories: CONTROL_UNIT is responsible for receiving signal triggers from the acquisition module and coordinating the timing of each module; Figure 2 The diagram illustrates the connections between CONTROL_UNIT, MAC_UNIT, ACTIVATION_FUNCTION, LEARNING_UNIT, and the RAM memory. MAC_UNIT performs multiplication and accumulation operations on ANNs, adapting to 32-bit fixed-point data processing; ACTIVATION_FUNCTION uses a piecewise approximation of the sigmoid function to reduce hardware complexity; LEARNING_UNIT updates synaptic weights through a backpropagation algorithm, enabling online learning; and the RAM memory stores intermediate variables and weight parameters.
[0036] like Figure 3 As shown, in the ANN control logic, the input layer receives the output error and its delay value of the UAV signal. After the hidden layer is activated, the output layer parses the frequency, frequency point and rate parameters. At the same time, the weights are updated through error feedback to ensure adaptive parsing of time-varying signals.
[0037] Specifically, in some embodiments of the application, the artificial neural network model uses a multilayer perceptron (MLP) as the signal parsing network, and its mathematical model is as follows: (1) Input layer, used to receive the output error sequence of the UAV control signal: ;in The error between the desired signal and the actual signal. The reference signal (such as an ideal carrier wave). To collect signals.
[0038] (2) Hidden layer output: The output of the j-th hidden neuron is: ;in, The weights from the input layer to the hidden layer. σ is the bias, and σ(·) is the activation function.
[0039] (3) Output of the output layer: The output of the kth neuron in the output layer is: ;in, The weights from the hidden layer to the output layer. For bias, output The corresponding signal parameters obtained from the analysis (including frequency, frequency point, and rate).
[0040] (4) Activation function: A piecewise approximation of the Sigmoid function (FA) can be used: This function only requires addition, multiplication, and comparison operations in the FPGA, without exponentiation operations, which significantly saves logic resources.
[0041] The network employs an online learning approach, updating weights through error backpropagation. Specifically, it may include the following loss function: E The weight update rules are as follows: Output layer weights can be updated using the following formula: Output layer weight update: Hidden layer weight update: η is the learning rate, which ranges from 0.07 to 0.5 and is adaptively adjusted according to the dynamic characteristics of the signal.
[0042] The gradient of each layer is calculated using the chain rule, and the specific expression is as follows: Output layer gradient: ,in, ; Hidden layer gradient: , ; Parameter mapping in FPGA implementation specifically includes: Data format: All signals and weights are represented using 32-bit fixed-point numbers in the format 8:24 (8 integers, 24 decimals).
[0043] Computation latency: The delay for a single control cycle is: Clock cycle, in At that time, the latency is less than 29μs. Resource usage: ≤50% DSP, ≤40% LUT, ≤6% BRAM on ICE40UP5K.
[0044] The signal parameter analysis output is obtained through the following decoding formula: ; Among them, =2.4GH, =2.4835GH (ISM band).
[0045] Output u k After decoding, the corresponding values are as follows: u1 corresponds to the carrier frequency. u2 corresponds to the modulation frequency offset ∆ u2 corresponds to symbol rate R s .
[0046] For example, a portable drone countermeasure device has the following hardware configuration: the FPGA is a LATTICEICE40UP5K, the ANN is configured with 3 inputs (signal errors e, e1, e2), 3 hidden neurons, and 1 output neuron; the learning coefficient η=0.5; the signal acquisition module uses a 2.4GHz omnidirectional antenna and a 12-bit ADC converter; the modulation module uses an AD9850 DDS chip; and the amplification and transmission module uses a 1W power amplifier and a directional antenna.
[0047] In step 1, signal acquisition: the antenna receives the 2.4GHz communication signal from the illegal drone, the ADC converter converts it into 32-bit fixed-point data (8:24 format), and transmits it to the FPGA's CONTROL_UNIT; In step 2, the learning analysis shows that CONTROL_UNIT triggers MAC_UNIT to perform multiplication and accumulation operations on the input data (corresponding formula). ), ACTIVATION_FUNCTION processes the hidden layer output through a piecewise function. When the input α=2, FA(α)=1-(2-4)² / 32=0.875; LEARNING_UNIT updates the weights w through the backpropagation algorithm. ji and v kj The output layer analysis yielded a UAV signal frequency of 2450MHz, a frequency point of ±10MHz, and a data rate of 1Mbps. In step 3, signal generation: the signal generation module generates a 2450MHz digital source signal based on the analysis results, with the amplitude adjusted to 0.5Vpp and the phase consistent with the target signal; In step 4, modulation and transmission: the DDS chip performs FSK modulation on the digital source signal with a frequency offset of 50kHz and a modulation rate of 1Mbps; the power amplifier amplifies the signal to 1W, and the directional antenna transmits at a beam angle of 120° to suppress the UAV's communication link, forcing the UAV to return or land.
[0048] For example, in a vehicle-mounted UAV countermeasure system, the hardware configuration includes: an XC3S200A FPGA, an ANN with 6 inputs, 10 hidden neurons, and 1 output neuron; a learning coefficient η = 0.07; a signal acquisition module using a multi-antenna array and a 16-bit ADC converter; and an amplification and transmission module using a 10W power amplifier and an omnidirectional antenna.
[0049] A multi-antenna array collects UAV signals from different directions, an ADC converter outputs 32-bit fixed-point data, CONTROL_UNIT synchronously receives and stores the data in RAM; MAC_UNIT processes multi-channel data in parallel, ACTIVATION_FUNCTION performs activation operations, and LEARNING_UNIT adaptively updates weights, parsing the frequency, frequency point, and rate of the multi-target signals; a signal generation module generates multiple sets of digital source signals corresponding to the frequencies, which are modulated and synthesized using the DDS algorithm; a 10W power amplifier amplifies the signals, which are then transmitted via an omnidirectional antenna, enabling simultaneous countermeasures against multiple UAVs over a wide airspace.
[0050] The technical advantages of this invention include: the online learning capability of ANN improves the accuracy of complex signal analysis to over 98%, significantly increasing the countermeasure success rate; the modular FPGA combined with a fixed-point arithmetic architecture reduces DSP unit usage to ≤50%, logic unit usage to ≤40%, and signal processing latency to ≤29μs, meeting real-time countermeasure requirements; it is compatible with low-density FPGAs, reducing equipment costs by over 30%, improving portability, and enabling widespread application in portable and vehicle-mounted countermeasure scenarios; and the segmented activation function simplifies hardware implementation, further reducing resource consumption.
[0051] In other embodiments of this application, a deep learning-based counter-spatial signal processing apparatus is also disclosed. This apparatus is used to implement the above-described method and includes: The signal acquisition module is used to acquire UAV control signals. It uses a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online and parse them to obtain analytical results containing the frequency, frequency point, and rate of the control signals. The artificial neural network (ANN) is implemented modularly using an FPGA. The signal generation module is used to generate a digital source signal with the same frequency as the UAV control signal based on the analysis results. The modulation module is used to perform FSK modulation on the digital source signal using the DDS algorithm; The amplification and transmission module is used to amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure.
[0052] Furthermore, the signal acquisition module includes an antenna and an ADC converter, the amplification and transmission module includes a power amplifier and a directional antenna, and the ADC converter outputs 32-bit fixed-point data.
[0053] An electronic device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the above-described deep learning-based counter-space signal processing method.
[0054] A computer-readable storage medium storing a computer program that, when executed by a processor, implements a deep learning-based method for countering spatial signal processing.
[0055] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0056] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.
[0057] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0058] The above provides a detailed description of a deep learning-based method and apparatus for countering spatial signal processing. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and its core ideas. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A deep learning-based method for countering spatial signal processing, characterized in that, include: Step 1: Collect UAV control signals, and use a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online, and parse the results to obtain the frequency, frequency point and rate of the control signals; Step 2: Based on the analysis results, generate a digital source signal with the same frequency as the UAV control signal; Step 3: Perform FSK modulation on the digital source signal using the DDS algorithm; Step 4: Amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure.
2. The method according to claim 1, characterized in that, The process of acquiring UAV control signals involves using a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online, and parsing them to obtain analytical results containing the frequency, frequency point, and rate of the control signals, including: Artificial Neural Networks (ANNs) including input, hidden, and output layers are implemented using FPGA modularization. The input to the input layer is the output error of the UAV signal and at least two delay values.
3. The method according to claim 1, characterized in that, The FPGA module includes: CONTROL_UNIT, MAC_UNIT, ACTIVATION_FUNCTION, LEARNING_UNIT, and two RAM memories; The CONTROL_UNIT is used to coordinate the operation of the module, the MAC_UNIT performs multiplication and accumulation operations, and the LEARNING_UNIT updates the synaptic weights of the artificial neural network (ANN) through the backpropagation algorithm.
4. The method according to claim 3, characterized in that, The ACTIVATION_FUNCTION uses a piecewise approximation of the sigmoid function, satisfying: When α≤-4, FA(α)=0; When -4<α<0, FA(α)=(α+4)² / 32; When 0≤α≤4, FA(α)=1-(α-4)² / 32; When α>4, FA(α)=1.
5. The method according to claim 1, characterized in that, The process of acquiring UAV control signals, employing a 32-bit fixed-point artificial neural network (ANN) for online learning of the control signals, and parsing the signals to obtain analytical results containing the frequency, frequency point, and rate of the control signals, also includes: The acquisition of UAV control signals employs a 32-bit fixed-point artificial neural network (ANN), with the learning coefficient η set to a range of 0.07 to 0.
5. The control signals are then learned online, and the analysis results containing the frequency, frequency point, and rate of the control signals are obtained.
6. The method according to claim 2 or 3, characterized in that, The FPGA is any one of LATTICE ICE40UP5K, XC3S200A or XC6SLX4.
7. The method according to claim 1, characterized in that, The step of generating a digital source signal with the same frequency as the UAV control signal based on the analysis result includes: Based on the analysis results, the amplitude and phase of the digital source signal are adjusted by the adaptive output of the artificial neural network (ANN) to achieve a signal matching degree of ≥98%, and a digital source signal with the same frequency as the UAV control signal is generated.
8. The method according to claim 1, characterized in that, The acquisition of UAV control signals includes: An ADC converter is used to acquire UAV control signals, and the sampling rate is dynamically adjusted according to the UAV signal frequency to output 32-bit fixed-point data.
9. The method according to claim 1, characterized in that, The step of using the DDS algorithm to perform FSK modulation on the digital source signal includes: The DDS algorithm is used and the frequency offset range of FSK modulation is set to 10kHz~100kHz. The digital source signal is then subjected to FSK modulation and parsing to make the modulated signal rate consistent with the signal rate of the UAV.
10. A deep learning-based counter-spatial signal processing device, characterized in that, include: The signal acquisition module is used to acquire UAV control signals. It uses a 32-bit fixed-point artificial neural network (ANN) to learn the control signals online and parse them to obtain analytical results containing the frequency, frequency point, and rate of the control signals. The signal generation module is used to generate a digital source signal with the same frequency as the UAV control signal based on the analysis results. The modulation module is used to perform FSK modulation on the digital source signal using the DDS algorithm; The amplification and transmission module is used to amplify the modulated signal and transmit it in a directional manner to achieve spatial signal countermeasure.