Search

CN-122027073-A - Inverse airspace signal processing method and device based on deep learning

CN122027073ACN 122027073 ACN122027073 ACN 122027073ACN-122027073-A

Abstract

The invention provides a method and a device for processing a reverse-control airspace signal based on deep learning, and belongs to the technical field of airspace signal reverse control. The method comprises the steps of collecting unmanned aerial vehicle control signals, performing online learning on the control signals by adopting an artificial neural network ANN with 32-bit fixed-point operation, analyzing to obtain analysis results comprising the frequency, the frequency point and the speed of the control signals, realizing the artificial neural network ANN through FPGA modularization, generating digital source signals with the same frequency as the unmanned aerial vehicle control signals based on the analysis results, performing FSK modulation on the digital source signals by adopting a DDS algorithm, performing power amplification and directional emission on the modulated signals, and realizing airspace signal reaction. The airspace signal countering technology can improve the resolution precision of complex signals, reduce the consumption of hardware resources and consider the instantaneity and the cost.

Inventors

  • HU SONGHUI
  • LENG HUAHUI
  • Xiao Hengpeng

Assignees

  • 深圳市飞思腾科技有限公司

Dates

Publication Date
20260512
Application Date
20260212

Claims (10)

  1. 1. The inverse airspace signal processing method based on deep learning is characterized by comprising the following steps of: Step 1, acquiring an unmanned aerial vehicle control signal, and performing online learning on the control signal by adopting an artificial neural network ANN of 32-bit fixed-point operation, so as to obtain an analysis result comprising the frequency, the frequency point and the speed of the control signal; step 2, generating a digital source signal with the same frequency as the unmanned aerial vehicle control signal based on the analysis result; step 3, performing FSK modulation on the digital source signal by adopting a DDS algorithm; and 4, carrying out power amplification and directional emission on the modulated signal to realize space domain signal reaction.
  2. 2. The method of claim 1, wherein the acquiring the unmanned aerial vehicle control signal, performing online learning on the control signal by using an artificial neural network ANN with 32-bit fixed point operation, and analyzing to obtain an analysis result including the frequency, the frequency point and the rate of the control signal, includes: An artificial neural network ANN comprising an input layer, a hidden layer and an output layer is realized through FPGA modularization; the input of the input layer is the output error of the unmanned aerial vehicle signal and at least 2 delay values thereof.
  3. 3. The method of claim 1, wherein the FPGA modularization includes control_UNIT, MAC_ UNIT, ACTIVATION _ FUNCTION, LEARNING _UNIT, and two RAM memories; The control_unit is used for coordinating a module to work, the MAC_unit executes multiplication accumulation operation, and the LEARNING _unit updates the artificial neural network ANN synaptic weight through a back propagation algorithm.
  4. 4. The method of claim 3, wherein ACTIVATION _function employs a piecewise approximate sigmoid FUNCTION, satisfying: FA (α) =0 when α is less than or equal to-4; FA (α) = (α+4) 2/32 when-4 < α < 0; when 0≤α≤4, FA (α) =1- (. Alpha. -4) 2/32; FA (α) =1 when α > 4.
  5. 5. The method of claim 1, wherein the acquiring the unmanned aerial vehicle control signal, performing online learning on the control signal by using an artificial neural network ANN with 32-bit fixed point operation, and analyzing to obtain an analysis result including the frequency, the frequency point and the rate of the control signal, further comprises: and acquiring an unmanned aerial vehicle control signal, adopting an artificial neural network ANN of 32-bit fixed-point operation, setting the range of a learning coefficient eta to be 0.07-0.5, performing online learning on the control signal, and analyzing to obtain an analysis result containing the frequency, the frequency point and the speed of the control signal.
  6. 6. A method according to claim 2 or 3, wherein the FPGA is either LATTICE ICE UP5K, XC S200A or XC6SLX 4.
  7. 7. The method of claim 1, wherein the generating a digital source signal having the same frequency as the drone control signal based on the parsing result comprises: based on the analysis result, the amplitude and the phase of the digital source signal are adjusted through the self-adaptive output of the artificial neural network ANN, so that the signal matching degree is more than or equal to 98%, and a digital source signal with the same frequency as the unmanned aerial vehicle control signal is generated.
  8. 8. The method of claim 1, wherein the acquiring the drone control signal comprises: and acquiring unmanned aerial vehicle control signals by adopting an ADC (analog-to-digital converter), dynamically adjusting the sampling rate according to the frequency of the unmanned aerial vehicle signals, and outputting 32-bit fixed-point data.
  9. 9. The method of claim 1, wherein FSK modulating the digital source signal using a DDS algorithm comprises: and adopting a DDS algorithm, setting the frequency offset range of FSK modulation to be 10 kHz-100 kHz, performing FSK modulation on the digital source signal, and analyzing the digital source signal, so that the modulated signal rate is consistent with the unmanned aerial vehicle signal rate.
  10. 10. A reverse airspace signal processing device based on deep learning, comprising: the signal acquisition module is used for acquiring unmanned aerial vehicle control signals, carrying out online learning on the control signals by adopting an artificial neural network ANN with 32-bit fixed-point operation, and analyzing to obtain analysis results comprising the frequency, the frequency point and the speed of the control signals; the signal generation module is used for generating a digital source signal with the same frequency as the unmanned aerial vehicle control signal based on the analysis result; the modulation module is used for performing FSK modulation on the digital source signal by adopting a DDS algorithm; And the amplifying and transmitting module is used for carrying out power amplification and directional transmission on the modulated signals so as to realize space domain signal reaction.

Description

Inverse airspace signal processing method and device based on deep learning Technical Field The invention relates to the technical field of airspace signal countering, in particular to a countering airspace signal processing method and device based on deep learning, which are suitable for signal suppression scenes of airspace targets such as unmanned aerial vehicles and the like. Background Along with the popularization of unmanned aerial vehicle technology, illegal invasion airspace situations frequently occur, and control needs to be realized through a signal countering technology. The existing countering method generally adopts a process of 'resolving a preset model-simulating signals-modulating emission', but has the defects that firstly, the preset model has poor adaptability to nonlinear and time-varying unmanned aerial vehicle signals, the resolving precision is insufficient, the matching degree of countering signals and target signals is low, secondly, the signal processing adopts a floating point operation architecture, the hardware resource consumption is high, the real-time performance is difficult to meet the dynamic airspace countering requirement, and thirdly, the high-density hardware dependence leads to high equipment cost and poor portability. The application therefore proposes a method and a device for processing inverse spatial domain signals based on deep learning, which at least partially solve the problems. Disclosure of Invention In view of the above, the present application has been made to provide a spatial domain signal cancellation technique that overcomes the above-mentioned problems or at least partially solves the above-mentioned problems, and that can improve the resolution accuracy of complex signals, reduce the consumption of hardware resources, and achieve both real-time performance and cost. In a first aspect, the present invention provides a method for processing inverse spatial signal based on deep learning, including the following steps: Step 1, acquiring an unmanned aerial vehicle control signal, and performing online learning on the control signal by adopting an artificial neural network ANN of 32-bit fixed-point operation, so as to obtain an analysis result containing the frequency, the frequency point and the speed of the control signal; step 2, generating a digital source signal with the same frequency as the unmanned aerial vehicle control signal based on the analysis result; step 3, performing FSK modulation on the digital source signal by adopting a DDS algorithm; and 4, carrying out power amplification and directional emission on the modulated signal to realize space domain signal reaction. In some embodiments of the present invention, the acquiring a control signal of an unmanned aerial vehicle, performing online learning on the control signal by using an artificial neural network ANN of 32-bit fixed point operation, and analyzing to obtain an analysis result including a frequency, a frequency point and a rate of the control signal, including: an artificial neural network ANN comprising an input layer, a hidden layer and an output layer is realized through FPGA modularization, wherein the artificial neural network ANN is a multi-layer sensor; the input of the input layer is the output error of the unmanned aerial vehicle signal and at least 2 delay values thereof. In some embodiments of the present invention, the FPGA modularization includes control_UNIT, MAC_ UNIT, ACTIVATION _ FUNCTION, LEARNING _UNIT, and two RAM memories; The control_unit is used for coordinating a module to work, the MAC_unit executes multiplication accumulation operation, and the LEARNING _unit updates the artificial neural network ANN synaptic weight through a back propagation algorithm. In some embodiments of the present invention, the ACTIVATION _function employs a piecewise approximate sigmoid FUNCTION, which satisfies the following: FA (α) =0 when α is less than or equal to-4; FA (α) = (α+4) 2/32 when-4 < α < 0; when 0≤α≤4, FA (α) =1- (. Alpha. -4) 2/32; FA (α) =1 when α > 4. In some embodiments of the present invention, the collecting the unmanned aerial vehicle control signal, performing online learning on the control signal by using an artificial neural network ANN of 32-bit fixed point operation, and analyzing to obtain an analysis result including the frequency, the frequency point and the rate of the control signal, further includes: and acquiring an unmanned aerial vehicle control signal, adopting an artificial neural network ANN of 32-bit fixed-point operation, setting the range of a learning coefficient eta to be 0.07-0.5, performing online learning on the control signal, and analyzing to obtain an analysis result containing the frequency, the frequency point and the speed of the control signal. In some embodiments of the invention, the FPGA is any one of LATTICE ICE UP5K, XC S200A or XC6SLX 4. In some embodiments of the present invention, the generating, based on the parsi