A radar signal boundary sample detection method based on matched filtering
Through the cross-correlation function matching filtering processing of radar signals, boundary samples are extracted, which solves the problem of difficulty in extracting boundary samples in radar signal training, and improves the detection ability of neural networks for long-distance weak signals.
Patent Information
- Application Number
- CN202110751839.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-02
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2041-07-02
AI Technical Summary
In radar signal training, it is difficult to extract boundary samples, resulting in the identification algorithm based on good sample training that cannot effectively detect weak signals from a long distance.
Using a method based on matching filtering, the correlation characteristics of the radar signal are used to perform matching filtering processing through the cross-correlation function of the standard sample and the original intermediate frequency data to extract the boundary sample of the radar signal.
This method reduces labor costs, provides a large number of boundary samples, and improves the detection and recognition ability of neural networks to long-distance weak signals.
Smart Images
Figure CN113534086B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent processing of passive detection data, and in particular to the preparation of training samples for neural network training. Background Art
[0002] In pattern recognition based on neural network training, under the condition of using the same features and identifiers, the training data determines the performance of the trained model. Training samples can be simply divided into three categories: good samples (samples that are easily separated from other class samples), boundary samples (i.e. fine precision samples: samples that are close to other class samples), and poor samples (samples that are easily confused with other class samples). The role of training is to generate classification hypersurfaces. From this aspect, boundary samples are samples that are located near the ideal classification hypersurface.
[0003] Through simulation training, it is also proved that: the existence of good samples can make the training pattern class area more compact and the interval between different pattern class areas larger, but too many good samples can easily make the trained pattern class area too small, thereby increasing the risk of misidentification of boundary samples and poor samples; in the training samples, the existence of poor samples may make the trained pattern class area as large as possible, and at the same time, it is easy for different pattern class areas to overlap, increasing the classification error; if boundary samples are used for training, the trained class area can be as large as possible, and the adjacent class areas are as non-overlapping as possible, at this time the classification performance is the best, in other words, the boundary samples mark the boundary of the pattern class area. The number of boundary samples has a much greater impact on neural network training than the total number of training samples.
[0004] In the preparation of radar signal training samples, manual selection is often used to select sample information. In order to ensure the correctness of the samples, the selectors usually select high-quality samples that are easy to judge manually, that is, good samples, and lack boundary samples. In practical applications, the signal receiving system is affected by various factors such as noise and interference. Especially for long-distance weak signals, the signal itself has low energy and large propagation loss, the received signal signal-to-noise ratio is low, and detection and recognition are difficult. The recognition algorithm based on good sample training cannot detect weak signals in the actual environment. Summary of the invention
[0005] In order to solve the problem of difficulty in extracting characteristic boundary samples of radar signals, the present invention proposes a radar signal boundary sample detection method based on matched filtering. By utilizing the fact that the same radar signal has correlation characteristics at different times, based on the standard samples of the radar signal and the matched filtering algorithm, the detection of the original intermediate frequency data is realized to obtain the characteristic boundary samples of the radar signal.
[0006] The present invention is achieved through the following technical solutions:
[0007] Step 1: Obtain a standard sample sequence of radar signals;
[0008] Step 2: Use the standard sample as a template and perform matched filtering on the original intermediate frequency detection data;
[0009] Step 3: According to the correlation peak, determine whether the radar signal exists and extract a new sample sequence;
[0010] Step 4: Traverse backward to find other sample sequences based on the interception time and period value of the previous radar signal;
[0011] The same radar signal has related characteristics at different times. Based on the characteristics, matched filtering is performed through standard samples of the same radar signal, and the original intermediate frequency data is detected to obtain boundary samples of the same radar signal.
[0012] Furthermore, the step 2 uses a cross-correlation function to implement signal matching filtering:
[0013] Among them, x 1 (t) and x 2 (t) is the signal intermediate frequency data, C x1x2 (τ) is x 1 (t) and x 2 (t) is the cross-correlation function of
[0014] The basic principle of matching filtering by detecting signal cross-correlation is as follows:
[0015] Original signal to be tested: the original intermediate frequency detection signal affected by noise, expressed as x 1 (t) = s 1 (t)+w 1 (t);
[0016] Standard signal: It is a standard detection signal selected under high signal-to-noise ratio conditions, and the formula is x 2 (t) = s 2 (t)+w 2 (t);
[0017] w 1 (t)w 2 (t) is additive Gaussian noise with zero mean;
[0018] s 1 (t)s 2 (t) is a radar signal without noise influence;
[0019] Theoretical explanation:
[0020] The cross-correlation function of the two signals is expressed as follows:
[0021]
[0022] Among them C s1s2 (τ) is s 1 (t) and s 2 (t) cross-correlation function; C w1s2 (τ) is w 1 (t) and s 2 (t), C s1w2 (τ) is s 1 (t) and w 2 (t) cross-correlation function; C w1w2 (τ) is w 1 (t) and w 2 (t) is the cross-correlation function of
[0023] Theoretically, the input noise w 1 (t), w 2 (t) are unrelated to each other, but are also related to s 1 (t) and s 2 (t) is irrelevant, then C w1s2 (τ), C s1w2 (τ), C w1w2 (τ) is zero.
[0024] It can be concluded that judging whether the detection signals containing noise are cross-correlated can determine whether the radar signals they contain are cross-correlated, that is: C x1x2 (τ) = C s1s2 (τ).
[0025] From the above process, it can be concluded that whether the radar signals are correlated can be directly determined by cross-correlating the original test signal containing noise with the standard signal, thereby determining whether they are the same radar signal.
[0026] In the verification of the above technical solution, it was found that the present invention has the following advantages: the target matching filtering method can reduce labor costs, provide a large number of boundary samples for neural network training, and improve the detection and recognition ability of the neural network for long-distance weak signals. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 A boundary sample labeling method according to an embodiment of the present invention;
[0028] Figure 2 Main lobe high signal-to-noise ratio signal spectrum;
[0029] Figure 3 Low signal-to-noise ratio signal spectrum;
[0030] Figure 4 Cross-correlation calculation results; DETAILED DESCRIPTION
[0031] In order to make the technical solutions and effects of the present invention more clear and explicit, the technical solutions in the embodiments of the present invention will be further elaborated in detail below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0032] Perform standard sample extraction, matched filtering and sample cycle detection on the intermediate frequency data of a radar signal actually detected:
[0033] (1) First, analyze the intermediate frequency data through spectrum display to obtain the signal parameters and modulation mode: carrier frequency agility, staggered repetition period, pulse compression system, signal bandwidth of 3.7MHz, and signal pulse width of 29.3us;
[0034] (2) Select the complete signal cycle data at a certain moment with a signal-to-noise ratio > 15 dB, and record it as the standard sample x of the radar signal s (t) = {p 1 , p 2 , p 3 ......}, the signal is Figure 2 As shown;
[0035] (3) Read the original intermediate frequency data such as Figure 3 As shown, its signal-to-noise ratio is 7dB, using matched filtering technology, with radar signal standard sample x s (t) is the matching template, and the original intermediate frequency detection data is matched and filtered. The data of the original intermediate frequency detection data is X(t), and a section of signal x is taken in sequence from sequence number 1 i (t) traverses. Calculate the relevant situation, the relevant algorithm is as follows: Among them, the correlation function C represents the two signals x i (t) and x s (t) A measure of the degree of association between
[0036] (4) At t s At the moment when the signal envelope is completely aligned with the standard pulse, the correlation peak is greater than the detection threshold, and the result after matched filtering is as follows Figure 4 As shown, at this time, the correlation quantity C is greater than the detection threshold Th. The detection threshold is generally set to 8 and can be adjusted according to the actual situation. The target is judged to be the same target and the starting position t at this moment is recorded. s , get the value from the starting position t s Start to offset the intermediate frequency data within the range of one period T plus an appropriate tolerance δ and store it as a new sample sequence x n (t) = {p 1 ′,p 2 ′,p 3 '......}; extract the original intermediate frequency data at that moment;
[0037] (5) It takes a long time to find the matching position during the first match. When the subsequent data is matched continuously, it is only necessary to start the relevant match from the position that is delayed by one period T minus the tolerance δ value after the last matching point, so that the position of the subsequent matching point can be quickly determined. The same method of steps 3 and 4 is used in sequence to obtain a new sample sequence.
Claims
1. A radar signal boundary sample detection method based on matched filtering, characterized in that: Step 1: Obtain a standard sample sequence of intermediate frequency data of radar signals; Step 2: Use the standard sample as a template and the original intermediate frequency detection data to perform matched filtering using a cross-correlation function: Among them, x1(t) and x2(t) are the intermediate frequency data of the signal, C x1x2 (τ) is the cross-correlation function of x1(t) and x2(t), T is a radar signal period; Step 3: According to the correlation peak, determine whether the radar signal exists and extract a new sample sequence; Step 4: Traverse backward to find other sample sequences based on the interception time and period value of the previous radar signal; The radar signal has correlation characteristics at different times. Based on the characteristics, matched filtering is performed through standard samples of the same radar signal, and the original intermediate frequency data is detected to obtain boundary samples of the same radar signal.