CN106199539A - Ground bounce removal method based on prewhitening filter - Google Patents

Abstract

The invention discloses a kind of ground bounce removal method based on prewhitening filter, including: the radar return data containing clutter and noise are carried out process of pulse-compression, and according to range gate data rearrangement；K-th range gate after pulse compression is set to reference unit, and the radar return data of slow time dimension are established as autoregressive process, estimate reference unit Parameters of Autoregressive Models by Burg algorithm based on Third-order cumulants；By the reference unit Parameters of Autoregressive Models structure prewhitening filter estimated；Utilize the prewhitening filter constructed that the data of current detection unit are carried out whitening filtering process；Filtered time-domain signal is carried out moving-target detection process.The clutter existed under target conditions at a slow speed is effectively suppressed by the present invention, originally the clutter collected around at zero-frequency is effectively suppressed, disperseing its power to make clutter in the form that frequency domain representation is approximation white noise, the clutter after power dispersion is obviously reduced for the impact of target detection at a slow speed.

Description

Ground bounce removal method based on prewhitening filter

Technical field

The present invention relates to the Clutter Rejection Technique in Radar Signal Processing, a kind of ground based on prewhitening filter is miscellaneous Ripple suppressing method.

Background technology

Since the radar birth thirties in 20th century, Radar Technology obtains development at a high speed.In recent years, at various radars The theory and application research of reason technology becomes a big hot topic and key problem.When radar carries out target detection, around target There are all passive reflective things beyond the moving target of the required detection of various ambient interferences, also referred to as clutter, i.e. radar Reflection echo, is often mingled with the environment clutter that intensity is bigger in radar target signal, if weak signal target is submerged in the most miscellaneous In ripple, particularly when strong clutter makes receiver transship, the discovery of these targets will become extremely difficult.Even and if target is also It is not in clutter background, it is desirable in sheet of clutter, tell Moving Target Return be soon also one and be difficult to Work.In order to deal with day by day complicated radar operating environment, improve power of test, reduce false alarm rate, open the most very early Having begun to detect the research work of echo signal under various clutter background, Clutter Rejection Technique is one of key technology therein.

For the reconnaissance radar of ground, its major function is to detect the people in enemy's motion in intricately clutter environment The moving targets such as member, vehicle, tank and low latitude helicopter, measure its position, movement velocity and the direction of motion, distinguish its character. A difficult point during clutter recognition in target detection is always ground investigation Radar Targets'Detection at a slow speed, if target spectrum at a slow speed Be positioned near the frequency spectrum of clutter, then the power of target is often less than the power of clutter, it is difficult to carry out effective signal detection.And And owing to the frequency of target at a slow speed is near clutter spectrum broadening, the Clutter Rejection Technique such as traditional MTI, AMTI is owing to detecing on ground The DOPPLER RESOLUTION that mine survey reaches is the highest, also can suppress echo signal itself, it is difficult to have effect while clutter reduction.Closely Nian Lai, it is proposed that many new Clutter Rejection Techniques, as based on wavelet decomposition and medium filtering, clutter based on Clutter-map Technology Suppressing method, decomposes based on EMD and the filtering method etc. of Hilbert conversion, but to the clutter existed under target conditions at a slow speed Inhibition is not highly desirable.

Patent of invention 201210211297.7 discloses a kind of underwater reverberation suppression method based on support vector regression, main Reverberation under water is established as autoregressive process for reverberation under water, and it is carried out whitening filtering, it is achieved disappearing of reverberation interference Remove.Patent of invention 201110409933.2 discloses a kind of self-adapting clutter suppression moving-target signal processing technology and realization side Method, disturbs mainly for land clutter, the signal after pulse pressure is once offseted process, is then fed into narrow band FIR filter group, The output of each bank of filters carries out distance unit CFAR respectively and processes, and obtains testing result through the big output of choosing, but the most special Target velocity situation is made discussion by door.Both at home and abroad relevant ground is investigated radar and be there is the clutter recognition side under target conditions at a slow speed The patent of method is not yet found.

Summary of the invention

It is an object of the invention to provide a kind of ground bounce removal method based on prewhitening filter.

Realize the technical scheme is that a kind of ground bounce removal method based on prewhitening filter, including following step Rapid:

Radar return data containing clutter and noise are carried out process of pulse-compression, and reset according to range gate by step 1 Data；

Step 2, is set to reference unit by the k-th range gate after pulse compression, and by the radar return number of slow time dimension According to being established as autoregressive process, estimate reference unit Parameters of Autoregressive Models by Burg algorithm based on Third-order cumulants；K is With the range gate number that current processing unit is separated by two range gate；

Step 3, constructs prewhitening filter by reference unit Parameters of Autoregressive Models；

Step 4, utilizes the prewhitening filter constructed that the data of current detection unit are carried out whitening filtering process；

Step 5, carries out moving-target detection process, completes ground bounce removal filtered time-domain signal.

The present invention is compared with existing Clutter Rejection Technique, and its remarkable advantage is:

(1) clutter existed under target conditions at a slow speed can be suppressed by the present invention, will originally collect around at zero-frequency Clutter effectively suppress, disperse its power make clutter frequency domain representation be approximation white noise form, the clutter after power dispersion Impact for target detection at a slow speed is obviously reduced；

(2) technology such as existing MTI, AMTI can also result in bigger loss while clutter reduction to echo signal, The filtering expense that the present invention causes for echo signal is less；

(3) the Burg algorithm used in the present invention is a kind of Parametric Analysis method, utilizes forward and backward forecast error average Power minimum criteria estimates that reflection coefficient, recycling Levinson recursion are sought the ginseng of autoregression model by low order to high-order Number, it is to avoid calculating autocorrelation matrix and solve Yule-Walker equation, and the degree of accuracy of parameter estimation is compared other and is estimated Method such as auto-covariance method etc. has certain advantage in terms of precision and operand；

(4) for the reality that ground investigation radar resolution is the highest, clutter is configured to Rayleigh Clutter by the present invention, and Estimate whitening filter coefficients based on Third-order cumulants, promoted relative to traditional prewhitening filter filter effect；

(5) combining in Clutter Rejection Technique at existing time-frequency, its filtering generally requires more artificial judgement, this Invention is self-adaptation type wave filter, will not introduce artificial disturbance.

Below in conjunction with the accompanying drawings the present invention is described in further detail.

Accompanying drawing explanation

Fig. 1 is present invention ground bounce removal based on prewhitening filter method flow diagram.

Fig. 2 is conventional MTD target range door Doppler slice figure.

Fig. 3 is the MTD Doppler slice figure after clutter recognition of the present invention.

Fig. 4 is the target detection effect distance-Doppler measurements of the chest, waist and hips figure of MTD after clutter recognition of the present invention.

Detailed description of the invention

The present invention is directed to targets detected by battlefield scout radar detection at a slow speed target Doppler frequency spectrum be located on or near clutter spectrum Region is difficult to a kind of method that the situation of target detection proposes effective suppression land clutter.

In conjunction with Fig. 1, present invention ground bounce removal based on prewhitening filter method comprises the following steps:

Radar return data containing clutter and noise are carried out process of pulse-compression, and reset according to range gate by step 1 Data；

Step 2, is set to reference unit by the k-th range gate after pulse compression, and by the radar return number of slow time dimension According to being established as autoregressive process, estimate reference unit Parameters of Autoregressive Models by Burg algorithm based on Third-order cumulants；K is It is the K+2 range gate with the range gate number that current detection unit is separated by two range gate, i.e. current detection unit；

Step 3, by the reference unit Parameters of Autoregressive Models structure prewhitening filter estimated；

Step 4, utilizes the prewhitening filter constructed that the data of current detection unit are carried out whitening filtering process；

Step 5, carries out moving-target detection process, completes ground bounce removal filtered time-domain signal.

Further, step 2 estimates reference distance by Burg algorithm based on Third-order cumulants and steepest descent method The detailed process of unit Parameters of Autoregressive Models is:

Step 2-1, initializes the iterative parameter needed for Burg algorithm:

$K_m\left(0\right)=e_{m-1}^b\left(0\right)=e_{m-1}^f\left(0\right)=0$

$e_0^f\left(n\right)=e_0^b\left(n\right)=y\left(n\right)$

Wherein K_mFor reflection coefficient, y (n) is the echo-signal of current time, and n represents current time,Before being m rank To error power,It is m rank backward error power；

Step 2-2, calculates the cost function based on Third-order cumulants that each Burg algorithm iteration process is corresponding:

$P_{cum}=E\{\underset{i=-p}{\overset{p}{\Σ}}y(n+i)\[|e_m^f\left(n\right){|}^2+|e_m^b\left(n\right){|}^2\]\}$

Wherein, p is the exponent number of prewhitening filter, and compromising according to actual operation amount and effect in this detailed description of the invention takes rank Number is 5 rank；

Step 2-3, carries out recursion by steepest descent method, definition iterative process intermediate parameters:

$d_p\left(n\right)=(1-\mathrm{\α})d_p(n-1)+\left|S_y\left(n\right)\right|\[|e_m^f\left(n\right){|}^2+|e_m^b\left(n\right){|}^2\]$

$S_y\left(n\right)=\underset{i=-p}{\overset{p}{\Σ}}y(n+i).$

α is convergence factor, and 0 ＜ α ＜ ＜ 1, the value of i is [-p, p]；

Step 2-4, solves the optimum reflection coefficient needed for next iteration:

$K_m(n+1)=K_m\left(n\right)+\frac{\mathrm{\μ}}{d_p\left(n\right)}*{\▿}_{K_m}{P}_{cum}$

Wherein μ is iteration step length, and value is [0,1],For current cost function about current reflectance K_m's Gradient；

Step 2-5, utilizes Burg algorithm to calculate next iteration forward error power, backward error power, utilizes lattice Filter construction carries out recursion:

$\left\{\begin{array}{c}{e}_m^f\left(n\right)=e_{m-1}^f\left(n\right)+K_m{e}_{m-1}^b(n-1)\\ e_m^b\left(n\right)=e_{m-1}^b(n-1)+K_m{e}_{m-1}^f\left(n\right)\end{array}\right.$

Step 2-6, repeats above step 2-2 and completes whole iterative process to step 2-5 process.

Further, step 3 is passed through the tool of the reference unit Parameters of Autoregressive Models structure prewhitening filter estimated Body process is:

Step 3-1, utilizes the reflection coefficient K obtained_mAnd draw based on Third-order cumulants through Levinson recursion criterion The whitening filter coefficients a of autoregression model_p；

Step 3-2, utilizes the whitening filter coefficients a obtained_p, constructing prewhitening filter, the frequency of prewhitening filter is special Property is:

$H\left(w\right)=1+\underset{p=1}{\overset{p}{\Σ}}\widehat{a_p^{*}}{e}^{-jwp}$

Wherein,For a_pConjugate transpose, w is frequency.

Below in conjunction with specific embodiment, the invention will be further described.

Embodiment

In the present embodiment, simulation parameter is set to: carrier frequency is 10GHz, and carrier wave is linear FM signal, and bandwidth 10MHz is adopted Sample frequency 20MHz, wide 80 μ s during pulse, dutycycle is 1:20, and slow time dimension accumulation number of times is 512 times, and target is radially done even Speed linear motion, speed is 0.4m/s, and noise is white noise, and signal to noise ratio is 10dB, and clutter is the land clutter of Rayleigh distributed, Signal to noise ratio is 0dB.

From figure 2 it can be seen that in the case of conventional moving-target detection processes, the clutter energy near zero-frequency is concentrated, Amplitude show with target energy approximation, it is difficult to detect target, false-alarm the most easily occurs.

From figure 3, it can be seen that carry out moving-target detection, the clutter base near zero-frequency after whitening filtering pretreatment again This is suppressed, and whole detection background becomes the situation of similar " white noise ", and target clearly highlights, and can carry out target Detection.

Figure 4, it is seen that after whitening filtering processes, originally have the detection background of a large amount of noise jamming to become The background of similar white noise, target at a slow speed is no longer caused powerful interference by the clutter originally collected around at zero-frequency, and target is examined Record to carry out, illustrate that this whitening pretreatment is effective.

Claims (3)

1. a ground bounce removal method based on prewhitening filter, it is characterised in that comprise the following steps:

Radar return data containing clutter and noise are carried out process of pulse-compression, and reset number according to range gate by step 1 According to；

Step 2, is set to the k-th range gate after pulse compression reference unit, and the radar return data of slow time dimension is built Stand as autoregressive process, estimate reference unit Parameters of Autoregressive Models by Burg algorithm based on Third-order cumulants；K is and works as Pretreatment unit is separated by the range gate number of two range gate；

Step 3, constructs prewhitening filter by reference unit Parameters of Autoregressive Models；

Step 4, utilizes the prewhitening filter constructed that the data of current detection unit are carried out whitening filtering process；

Step 5, carries out moving-target detection process, completes ground bounce removal filtered time-domain signal.

Ground bounce removal method based on prewhitening filter the most according to claim 1, it is characterised in that logical in step 2 The detailed process crossing Burg algorithm based on Third-order cumulants estimation reference distance unit Parameters of Autoregressive Models is:

Step 2-1, initializes the iterative parameter needed for Burg algorithm:

$K_m\left(0\right)=e_{m-1}^b\left(0\right)=e_{m-1}^f\left(0\right)=0$

$e_0^f\left(n\right)=e_0^b\left(n\right)=y\left(n\right)$

Wherein K_mFor reflection coefficient, y (n) is the echo-signal of current time, and n represents current time,It is that m rank forward direction misses Difference power,It is m rank backward error power；

Step 2-2, calculates the cost function based on Third-order cumulants that each Burg algorithm iteration process is corresponding:

$P_{cum}=E\{\underset{i=-p}{\overset{p}{\Σ}}y(n+i)\[|e_m^f\left(n\right){|}^2+|e_m^b\left(n\right){|}^2\]\}$

Wherein, p is the exponent number of prewhitening filter；

Step 2-3, carries out recursion by steepest descent method, definition iterative process intermediate parameters:

$d_p\left(n\right)=(1-\mathrm{\α})d_p(n-1)+\left|S_y\left(n\right)\right|\[|e_m^f\left(n\right){|}^2+|e_m^b\left(n\right){|}^2\]$

$S_y\left(n\right)=\underset{i=-p}{\overset{p}{\Σ}}y(n+i)$

α is convergence factor, and 0 ＜ α ＜ ＜ 1, the value of i is [-p, p]；

Step 2-4, solves the optimum reflection coefficient needed for next iteration:

$K_m(n+1)=K_m\left(n\right)+\frac{\mathrm{\μ}}{d_p\left(n\right)}*{\▿}_{K_m}{P}_{cum}$

Wherein μ is iteration step length, and value is [0,1],For current cost function about current reflectance K_mGradient；

Step 2-5, utilizes Burg algorithm to calculate next iteration forward error power, backward error power, utilizes Lattice filter Device structure carries out recursion:

$\left\{\begin{array}{c}{e}_m^f\left(n\right)=e_{m-1}^f\left(n\right)+K_m{e}_{m-1}^b(n-1)\\ e_m^b\left(n\right)=e_{m-1}^b(n-1)+K_m{e}_{m-1}^f\left(n\right)\end{array}\right.$

Step 2-6, repeats above step 2-2 and completes whole iterative process to step 2-5 process.

Ground bounce removal method based on prewhitening filter the most according to claim 1, it is characterised in that step 3 is concrete For:

Step 3-1, utilizes the reflection coefficient K obtained_mAnd draw based on Third-order cumulants autoregression through Levinson recursion criterion The whitening filter coefficients a of model_p；

Step 3-2, utilizes the whitening filter coefficients a obtained_p, constructing prewhitening filter, the frequency characteristic of prewhitening filter is:

$H\left(w\right)=1+\underset{p=1}{\overset{p}{\Σ}}\widehat{a_p^{*}}{e}^{-jwp}$

Wherein,For a_pConjugate transpose, w is frequency.