Satellite-borne ADS-B overlapping signal separation method based on complex neural network

Abstract

The method comprises the following steps: S1, envelope extraction is performed on the received satellite ADS-B overlapping signal to obtain an envelope signal; S2, Hilbert transform is performed on the envelope signal to obtain overlapping IQ complex data; S3, the IQ complex data is input into a complex neural network for overlapping signal separation to obtain separated IQ signals; S4, the data bits of the separated IQ signals are subjected to PPM decoding and then CRC checking; and S5, the separated correct signals are stored according to the checking result. The envelope signal is extracted by using an empirical formula, the calculation amount is reduced, and the separation efficiency is improved; the overlapping signal is separated by using a complex neural network, the separation accuracy is high, the separated signal response time is fast, and the engineering realizability is strong, so that the problems of high algorithm complexity, long time consumption and low separation efficiency of the traditional separation method are solved.

Description

Technical Field

[0001] This invention relates to a method for separating overlapping signals in spaceborne ADS-B based on complex neural networks, belonging to the field of digital signal processing technology. Background Technology

[0002] With the development of aerospace technology, Automatic Dependent Surveillance-Broadcast (ADS-B) technology has become a key technology in the field of aerial surveillance.

[0003] Figure 1 This is a schematic diagram of the structure of a spaceborne ADS-B signal. Figure 1 As shown, ADS-B technology automatically broadcasts information such as aircraft type, airline code, position, speed, altitude, and flight path of the aircraft in a fixed format to achieve airspace monitoring in a broadcast manner.

[0004] Spaceborne ADS-B systems, such as those that mount ADS-B receivers on low-Earth orbit satellites, can overcome the limited coverage of ground-based receivers, enabling global coverage and surveillance. However, spaceborne ADS-B systems face severe collision and overlap problems, which significantly impact signal decoding, leading to erroneous decoding or even loss of critical information.

[0005] Traditional ADS-B separation algorithms are too time-consuming and struggle to separate multiple overlapping signals in a short period of time, thus failing to effectively solve the aforementioned problems. Summary of the Invention

[0006] The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a spaceborne ADS-B overlapping signal separation method based on complex neural networks. This method uses empirical formulas to extract envelope signals, reducing computational load and improving separation efficiency. By employing complex neural networks for overlapping signal separation, the method achieves high separation accuracy, fast signal response time, and strong engineering feasibility, thereby solving the problems of high algorithm complexity, long processing time, and low separation efficiency of traditional separation methods.

[0007] The technical problem to be solved by the present invention is achieved through the following technical solution:

[0008] This invention provides a method for separating overlapping signals in spaceborne ADS-B based on complex neural networks, the method comprising the following steps:

[0009] S1: Extract the envelope of the received spaceborne ADS-B overlapping signal to obtain the envelope signal;

[0010] S2: Perform Hilbert transform on the envelope signal to obtain overlapping IQ complex data;

[0011] S3: Input the IQ complex data into a complex neural network to separate the overlapping signals and obtain the separated IQ signals;

[0012] S4: Perform PPM decoding on the data bits of the separated IQ signal, and then perform CRC verification;

[0013] S5: Store the correctly separated signal based on the verification result.

[0014] To reduce computational load and improve separation efficiency, in step S1, the two orthogonal baseband signals I and Q of the spaceborne ADS-B overlapping signal are acquired, and modulus calculation is performed on I and Q to obtain the envelope signal. An empirical formula is then used to estimate the envelope.

[0015] Preferably, the empirical formula is:

[0016] ,

[0017] in, yes and The maximum value in, yes and The minimum value in.

[0018] Preferably, in the complex neural network, real numbers are used to simulate complex number operations, and the calculation formula is as follows:

[0019]

[0020] in, and It is a real matrix. and For real number vectors, the real and imaginary parts of the convolution operation are represented using matrix notation. Complex convolution is represented as:

[0021]

[0022] The input layer is normalized, and the input of each intermediate layer is normalized before the activation function so that the output follows a normal distribution with a mean of 0 and a variance of 1.

[0023] Preferably, the constructed complex neural network is trained before inputting the IQ complex data into the complex neural network.

[0024] Preferably, the PPM decoding utilizes the statistical characteristics of all sampling points to sum and subtract the sampling points of the preceding and following chips. If the result is greater than 0, the code element is 1; if the result is less than 0, the code element is 0.

[0025] Preferably, the position of the 24-bit CRC check code is determined based on the value of the DF segment of the spaceborne ADS-B signal, and CRC verification is performed.

[0026] In summary, this invention uses empirical formulas to extract envelope signals, reducing computational load and improving separation efficiency. It employs complex neural networks for overlapping signal separation, achieving high separation accuracy, fast signal response time, and strong engineering feasibility. This solves the problems of high algorithm complexity, long processing time, and low separation efficiency of traditional separation methods.

[0027] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the structure of a spaceborne ADS-B signal;

[0029] Figure 2 This is a schematic diagram of the method flow for the spaceborne ADS-B overlapping signal separation method based on complex neural networks according to the present invention;

[0030] Figure 3 This is a schematic diagram of the overlapping IQ signals of the present invention;

[0031] Figure 4 This is a schematic diagram of the IQ signal separated by a complex neural network according to the present invention. Detailed Implementation

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] Numerous specific details are set forth in the following description to provide a full understanding of the invention. However, the invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0034] Deep learning is currently the intelligent learning method that is closest to the human brain. By building a model that is most similar to the hierarchical structure of the human brain, it can break the limitations of shallow learning, accurately represent the intrinsic connections of complex functions, and thus complete the layer-by-layer extraction of data from low-level features to high-level features.

[0035] This invention provides a method for separating overlapping signals of spaceborne ADS-B based on complex neural networks. It uses a complex neural network specifically designed for separating overlapping signals of spaceborne ADS-B to separate the overlapping signals, solving the problem of decoding overlapping signals for ADS-B and greatly improving engineering efficiency.

[0036] Figure 2 This is a schematic flowchart of the spaceborne ADS-B overlapping signal separation method based on complex neural networks according to the present invention. Figure 2 As shown, in this invention, the spaceborne ADS-B overlapping signal separation method based on complex neural networks includes the following steps:

[0037] S1: Extract the envelope of the received spaceborne ADS-B overlapping signal to obtain the envelope signal;

[0038] S2: Perform Hilbert transform on the envelope signal to obtain overlapping IQ complex data;

[0039] S3: Input the IQ complex data into a complex neural network to separate the overlapping signals and obtain the separated IQ signals;

[0040] S4: Perform PPM decoding on the data bits of the separated IQ signal, and then perform CRC verification;

[0041] S5: Store the correctly separated signal based on the verification result.

[0042] For example, a satellite-borne ADS-B overlapping signal can be received with a sampling rate of 10MHz and a signal bandwidth of 2MHz. The satellite-borne ADS-B overlapping signal includes two orthogonal baseband signals, I and Q.

[0043] The following section provides a further introduction to the spaceborne ADS-B overlapping signal separation method based on complex neural networks.

[0044] The above-mentioned envelope extraction of the received spaceborne ADS-B overlapping signal specifically includes: obtaining the two orthogonal baseband signals I and Q of the spaceborne ADS-B overlapping signal, and performing modulus calculation on I and Q: The envelope signal is obtained, and the envelope is estimated using empirical formulas.

[0045] Estimation using empirical formulas eliminates the need for complex square root operations, significantly improving computational efficiency when high precision requirements are not critical. Furthermore, since this invention targets overlapping ADS-B signals on spaceborne networks, which employ pulse position coding, the accuracy of separation primarily depends on the position between pulse envelopes. Empirical formulas have a limited impact on increasing the error rate of the separated data, and subsequent steps utilize the decoded information for CRC verification, ensuring the correctness of the separated signal.

[0046] The empirical formula originates from simplified calculations of finding roots of real numbers in the industry. The difference lies in the Alpha and Beta coefficients. Depending on different criteria, such as minimizing the root mean square error or minimizing the peak error, the values ​​of these two coefficients will vary. Preferably, this invention selects Alpha=1 and Beta=0.4, with the criterion being the minimum root mean square error. This invention does not limit the specific coefficients of the empirical formula; those skilled in the art can choose according to their needs.

[0047] According to one embodiment of the present invention, when the empirical formula is hour, ,in, yes and The maximum value in, yes and The minimum value in.

[0048] Figure 3 This is a schematic diagram of the overlapping IQ signals of the present invention. Performing a Hilbert transform on the above envelope signal yields the following result: Figure 3 The overlapping IQ complex data shown are used to calculate the instantaneous properties of the envelope signal through Hilbert transform, where the instantaneous amplitude is the amplitude of the Hilbert transform; and the instantaneous frequency is the rate of change of the instantaneous phase angle over time.

[0049] It should be added that current common deep learning methods are based on real numbers, but complex numbers have a richer feature representation capability than real numbers. In the neural network of this invention, the complex numerical value is first obtained, and then the real part A and imaginary part B of the complex number are represented as logically different real-valued entities, and complex number operations are simulated internally using real-valued algorithms.

[0050] Specifically, the IQ signal is processed by envelope extraction to obtain the envelope signal of the real number. Then, the envelope signal of the real number is transformed into complex data through Hilbert transform. Finally, the real and imaginary parts of this complex data are input into the constructed complex neural network for separation.

[0051] The filter is in complex form. IQ complex data is input into a complex neural network for overlapping signal separation. In the complex neural network, real numbers are used to simulate complex number operations. The calculation formula is as follows:

[0052]

[0053] in, and It is a real matrix. and Let W be a complex weight matrix, W = A + iB, and h be a complex input vector, h = x + iy. Matrix notation is used to represent the real and imaginary parts of the convolution operation. Complex convolution is represented as:

[0054]

[0055] The input layer is normalized, and the input of each intermediate layer (before the activation function) is also normalized to ensure that the output follows a normal distribution with a mean of 0 and a variance of 1, thus avoiding the problem of internal covariate shift. The specific process is as follows: First, the output of the hidden layer is normalized on a batch; then, it is scaled and shifted; finally, it is activated by the ReLU function.

[0056] It should be noted that before inputting IQ complex number data into the complex neural network, the constructed complex neural network must first be trained. Here is some additional information about the training dataset:

[0057] Considering the relative time delays of different overlapping source signals, overlapping signal datasets are constructed under different signal power ratios and signal-to-noise ratios. The time delays range from 0 μs to 120 μs, with a step size of 2 μs; the power differences between overlapping signals range from 1 dB to 10 dB, with a step size of 1 dB; and the signal-to-noise ratio ranges from 5 dB to 25 dB, with a step size of 5 dB. Each signal stored in the dataset can be viewed as a stitched combination of real and imaginary parts. The signal storage format within the dataset consists of the actual signal in the first half and the signal after Hilbert transform in the second half. Each signal group contains two source signals and one overlapping signal.

[0058] After training the complex neural network for overlapping signal separation, the complex data of the spaceborne ADS-B overlapping signal after envelope extraction and Hilbert transform is directly input into the network to obtain the separated signal, namely the separated IQ signal.

[0059] Figure 4 This is a schematic diagram of the IQ signal separated by a complex neural network according to the present invention. Figure 4 After obtaining the separated signals, the separated IQ signals are decoded separately, and the data bits are PPM decoded to obtain the message.

[0060] Code element correspondence Figure 1 The bits on the sample are encoded into pulses by PPM. In other words, the position between the pulses represents the information of the bits. The data bits are decoded by the position information of the sampling points to obtain the CRC check bits.

[0061] Specifically, PPM decoding makes full use of the values ​​of all sampling points, naming the preceding chip as... The subsequent chip was named ,if It is a high pulse It's a low pulse, this symbol is 1, if It is a low pulse It's a high pulse; this symbol is 0. After sampling the chip, the sampling point is set as follows: , , , , , , , , , Make full use of the values ​​of the sampling points to perform a summation operation on the chip:

[0062]

[0063]

[0064]

[0065] if If it is greater than 0, then this symbol is 1. If it is less than 0, then this symbol is 0.

[0066] In other words, by utilizing the statistical characteristics of the sampling points, the sampling points of the preceding and following chips are summed and subtracted. If the result is greater than zero, the symbol is 1; otherwise, it is 0. Then, the position of the 24-bit CRC check code is determined based on the value of the DF segment, and CRC verification is performed on the signal.

[0067] Table 1 is a schematic diagram of the bit structure of the satellite-borne ADS-B signal data.

[0068] Table 1

[0069]

[0070] As shown in Table 1, for the received satellite ADS-B signal, the DF segment refers to the downlink format. The value of DF in the satellite ADS-B signal is fixed and equals 17 in decimal.

[0071] If the CRC check returns true, then store the separation signal segment; otherwise, discard the separation signal segment.

[0072] For example, CRC check is cyclic redundancy check, which is a type of channel coding. It increases the redundancy of the signal and improves the fault tolerance of the signal. After the information code of length a, a check code of length b is added. Let the total length be L, so the encoding method is also called (L, a) encoding, which is (112, 88) (L=112, a=88) encoding.

[0073] Known and , This is the generator polynomial of the check code, and this polynomial is known:

[0074]

[0075] Converted to binary, this is 1111111111111010000001001. During encoding, the 24-bit checksum is derived as follows:

[0076]

[0077] Where mod means division modulo 2, and the result is the remainder, then the final expression is:

[0078]

[0079] The correctness of the separated ADS-B signal is verified using a CRC checksum. Can be If it is divisible, it means that the separated IQ signals are not erroneous.

[0080] In summary, this invention uses empirical formulas to extract envelope signals, reducing computational load and improving separation efficiency. It employs complex neural networks for overlapping signal separation, achieving high separation accuracy, fast signal response time, and strong engineering feasibility. This solves the problems of high algorithm complexity, long processing time, and low separation efficiency of traditional separation methods.

[0081] The present invention can be well implemented according to the above embodiments. It is worth noting that, based on the above design principles, even if some non-substantial modifications or refinements are made to the structure disclosed in the present invention to solve the same technical problem, the essence of the technical solution adopted is still the same as that of the present invention, and therefore it should also be within the protection scope of the present invention.

Claims

1. A method for separating overlapping signals in spaceborne ADS-B based on complex neural networks, characterized in that, The method includes the following steps: S1: Extract the envelope of the received spaceborne ADS-B overlapping signal to obtain the envelope signal; S2: Perform Hilbert transform on the envelope signal to obtain overlapping IQ complex data; S3: Input the IQ complex data into a complex neural network to separate the overlapping signals and obtain the separated IQ signals; S4: Perform PPM decoding on the data bits of the separated IQ signal, and then perform CRC verification; S5: Store the correctly separated signal based on the verification result; In the complex neural network, real numbers are used to simulate complex number operations, and the calculation formula is as follows: in, and It is a real matrix. and Let W be a complex weight matrix, W = A + iB, and h be a complex input vector, h = x + iy. Matrix notation is used to represent the real and imaginary parts of the convolution operation. Complex convolution is represented as: The input layer is normalized, and the input of each intermediate layer is normalized before the activation function so that the output follows a normal distribution with a mean of 0 and a variance of 1.

2. The method as described in claim 1, characterized in that, In step S1, the two orthogonal baseband signals I and Q of the spaceborne ADS-B overlapping signal are acquired, and the envelope signal is obtained by performing modulus calculation on I and Q. The envelope is then estimated using an empirical formula.

3. The method as described in claim 2, characterized in that, The empirical formula is: ， in, yes and The maximum value in, yes and The minimum value in.

4. The method as described in claim 1, characterized in that, Before inputting the IQ complex number data into the complex neural network, the constructed complex neural network is trained.

5. The method as described in claim 1, characterized in that, The PPM decoding utilizes the statistical characteristics of all sampling points to sum and subtract the sampling points of the preceding and following chips. If the result is greater than 0, the code symbol is 1. If the result is less than 0, the code element is 0.

6. The method as described in claim 1, characterized in that, The position of the 24-bit CRC check code is determined based on the value of the DF segment of the onboard ADS-B signal, and CRC verification is performed.

Images