CN104515983A - Stepped frequency radar signal target extract method based on statistic optimum - Google Patents

Abstract

The invention relates to a stepped frequency radar signal target extract method based on statistic optimum, aims to remove distance mismatching redundancy between a distance range represented by a stepping signal after distance refining IFFT and a distance range represented by a present echo sampling value, and over-sampling redundancy, and solves the problem that a 0th sampling point needs to be calibrated in both a classical algorithm abandoning method and a maximum selecting method, or else a target point cannot be sampled possibly. The key idea of the algorithm is as follows: circulation displacement algorithm is performed on the result of one frame of data after IFFT, rough distance sampling dimension overlapping is performed on aligned data, and a part of data is spliced by taking an overlapped maximum value as a center, so that both points with relatively large amplitudes in identical distance are extracted, and the problem that the 0th sampling point needs to be calibrated in the classical algorithm is also solved.

Description

The stepped frequency radar signal target extract method of Corpus--based Method optimum

Technical field

The invention belongs to radar signal target extract method, be specifically related to a kind of stepped frequency radar signal target extract method of Corpus--based Method optimum, for removing stepped frequency radar IFFT apart from the distance mismatch redundancy between the distance range represented by distance range represented after refinement and current echo samples value, and over-sampling redundancy.

Background technology

Step frequency signal is a kind of important Distance function radar signal, and it adopts the radar pulse launching the linear saltus step of a string carrier frequency, then obtains the effect of synthesis high resolution range by the IFFT process of paired pulses echo.Stepped frequency radar is synthesized wideband signal, and broadband signal is become multiple narrow band signal at frequency domain decomposition by it, is obtained the wide-band-message of synthesis by the transmitting-receiving of multiple narrow band signal.The pulse signal of step frequency signal transmitting and receiving is narrow band signal, and step frequency signal can reduce the requirement to the instant bandwidth of digital signal processor while obtaining high score rate, and system is easy to Project Realization.

In step ped-frequency radar system, target redundancy is a special problem.Therefore need to adopt Target pick-up algorithm to eliminate target redundancy.Distance range after target redundancy refers to IFFT refinement and the distance range represented by current echo samples value are not simple one-to-one relationships, have the information that some are unnecessary.Obtain real range information, just accurately in a certain order, some point must be chosen from the IFFT result of all sampled points, remove redundant information, form complete one-dimensional range profile, the Target pick-up algorithm of Here it is step ped-frequency radar.For a practical step frequency system high resolution range radar system, it similarly is necessary in processor, adopting Target pick-up algorithm thus obtaining complete one-dimensional distance.

Redundancy is divided into two kinds: distance mismatch redundancy and over-sampling redundancy.Its middle distance mismatch redundancy and individual pulse range resolution r _t, single-point is fuzzy distance r not _irelevant, and over-sampling redundancy is primarily of individual pulse range resolution r _t, individual pulse range accuracy r _sdetermine.

1, distance mismatch redundancy:

Transmitted pulse width τ determines monopulse range resolution r _τ=C τ/2, wherein C is the light velocity (same afterwards).And frequency step ladder Δ f (same afterwards) determines single-point not fuzzy distance r _i=C/ (2 Δ f), its ratio is r _τ/ r _i=τ Δ f.Usual step frequency signal demand fulfillment condition: τ Δ f < 1, the distance range after it represents refinement is greater than when the distance range representated by pre-echo, so the one-dimensional range profile after refinement has r _i-r _τzone void (i.e. redundancy), refinement result now contains the information of target complete, but due to distance mismatch, can there is range walk in target, cause dysmetria, must obtain real distance by de-redundancy algorithm.

Fig. 1 is the result schematic diagram in above-mentioned situation after 3 groups of neighbouring sample point IFFT.Wherein white space represents effective circle of good definition, without frame Regional Representative dead space.As seen from Figure 1, the IFFT result often organizing sampled point has r _i-r _τoverlapping area, the actual position of target cannot be judged in aliasing district.But due to r _i> r _τ, the data after folding can not cover circle of good definition, so without aliasing.But, if by the simple continued access of IFFT result in Fig. 1, the actual distance of target still cannot be obtained.Suitable de-redundancy algorithm must being adopted, extracting a part accurately from often organizing IFFT result, extract continued access by 1 in Fig. 1,2,3,4 and obtain real one-dimensional range profile.

2, over-sampling redundancy:

In theory, T _sas long as equal transmitted pulse width, just can obtain omnidistance one-dimensional range profile, in systems in practice, due to echo broadening and disperse, make sampled point not adopt the maximal value of echo, cause amplitude loss.So usually can T be made _s≤ τ/3, thus reduce sampling loss.But improve T _sthe echo of same point target can be made likely to be sampled repeatedly, same impact point is caused repeatedly to appear in the refinement result of different groups, simultaneously, the target echo with multiple scattering center is likely distributed in the middle of plural IFFT result, and this also needs Target pick-up algorithm to process.

At present, Target pick-up algorithm mainly contains following several method:

1, method is given up

Suppose sampled distance resolution r _s(r _s=CT _s/ 2), for each group IFFT result, the length that it comprises ' range information ' is exactly r _s, so if to take out length be r often organizing in IFFT result _s' information ' and continued access get up, just constitute the simplest ' giving up ' Target pick-up algorithm (hereinafter referred to as giving up method).

Give up method principle as shown in Figure 2, if total M sampled point, after IFFT, N number of data can be obtained, thus to obtain a size be the matrix of N × M.For wherein m sampled point, (m=0,1,2 ..., M-1), take out wherein P _mto Q _mw between point _mindividual data (P _m, Q _m=0,1,2 ..., N-1) and (dash area in Fig. 2), as the extraction point mark of current sampling point, then the data that each sampled point is got are spliced, namely obtain the one-dimensional range profile of target.About P _mand Q _mvalue determined by following formula:

1. in the 0th sampled point, if P ₀=0, then W ₀=Trunc (r _s/ Δ r), Q ₀=P ₀+ W ₀-1.

2. in m sampled point (m=1,2 ..., M-1), there is P _m=(Q _m-1+ 1) ModN

③ $\left\{\begin{array}{c}{W}_m=\mathrm{Trunc}\left(\frac{r_s}{\mathrm{\&Delta;r}}\right),m=0\\ W_m=\mathrm{Trunc}\left[\frac{(m+1)r_s}{\mathrm{\&Delta;r}}\right]-\underset{i=0}{\overset{m-1}{\mathrm{\&Sigma;}}}{W}_i,m=\mathrm{1,2},...,M-1\end{array}\right.---\left(1\right)$

Q _m＝(P _m+W _m-1)ModN

Wherein Trunc---block computing; Mod---complementation; The minimum resolution distance unit that Δ r=C/ (2N Δ f) is radar.

Formula (1) describes the strategy giving up method, and computing is simply the great advantage giving up method.Obviously found out by the sequence number in formula (1), giving up method is remained as ' range information ' the most beginning in effective for current I FFT result district, at this moment target just enters current sampling point, usual amplitude is less, reduce the signal to noise ratio (S/N ratio) after extraction, this gives up the topmost shortcoming of method.Suitably adjust P ₀can alleviate this problem, but basic solution adopts other Target pick-up algorithm.

2, large method is selected

Select large method only to take out the length corresponding with pulsewidth τ to each sampled point, and carry out same distance with the result that adjacent sampled point extracts and compare, take out the larger point of amplitude as extraction result.The extraction starting point P of m sampled point IFFT result is determined according to formula (1) _mbut the data amount check extracted is not W _m, but determined by (2) formula

$X_m=\mathrm{Trunc}\left(\frac{\mathrm{C\&tau;}}{2\mathrm{\&Delta;r}}\right)---\left(2\right)$

Obviously, the end part extracting result at each sampled point has the one piece of data of one piece of data and next sampled point extraction result beginning to be overlap, as shown in Figure 3, dash area is the data that each sampled point is chosen, intersection is had, counting as X of coincidence between visible two adjacent sampled points _m-W _m.For the data of intersection, take out the larger conduct of amplitude and export.Select large method under quiet goal condition, target extract result can be obtained well, can maximal value be obtained, and not have pseudo-peak.But calculated amount is comparatively large, and there will be pseudo-peak when there being range walk.

3, summation

For the data pick-up principle of each sampled point with selecting large method, but do not adopt choosing large for the redundant data of neighbouring sample point but carry out that same distance is cumulative to be averaged.Under quiet goal condition, target extract result can be obtained well, and part signal to noise ratio (S/N ratio) can be improved, and not have pseudo-peak.Shortcoming is that calculated amount is comparatively large, and there will be pseudo-peak when there being range walk, and physical significance after being added is not too clear and definite.

Summary of the invention

The technical matters solved

Can find out according to Fig. 2 and Fig. 3, give up method and select large method all to need to make demarcation to " 0 " individual sampled point, otherwise this situation may be there is: i.e. the fine pitch that the thick distance sample at target place extracts is from part, namely the dash area in figure, do not cover the thin distance sample of target, as shown in Figure 4, black five-pointed star represents the position of target, and the probability that this situation occurs is larger, and select the Target pick-up algorithm in large method to be drawn into the probability of impact point compared to giving up method, just the probability being drawn into impact point is doubled, and really cannot solve the problem extracting fall short point.Based on above problem, the present invention proposes the Target pick-up algorithm in a kind of statistical significance, the method first by the result after a frame data IFFT with a certain thick distance sample for benchmark carries out ring shift alignment, again the data after alignment are superposed in thick distance samples dimension, find the maximal value of the rear data of superposition, a part of data of taking out centered by this maximal value are spliced, and namely complete the extraction of target one-dimensional range profile.The method does not need to demarcate " 0 " individual sampled point, and substantially increase extract the probability of distance coverage goal point.

Technical scheme

A stepped frequency radar signal target extract method for Corpus--based Method optimum, is characterized in that step is as follows:

Step 1: the data of a CPI being stored as a size is the matrix of T × N, the wherein data that receive of every behavior of matrix each pulse repetition time, matrix be often classified as the different sampled point of pulse repetition time on same thick distance and position;

Step 2: IFFT is done to each row of T × N matrix, after obtaining distance refinement, every frame data matrix A size is M × N, wherein: M is that IFFT counts;

Step 3: calculate the data segment, length W that each thick distance sample needs to extract:

$W=\mathrm{Trunc}\left(\frac{r_s}{\mathrm{\&Delta;r}}\right)$

r _s＝c/2f _s

Δr＝c/(2MΔf)

Wherein: Trunc for blocking computing, r _sfor sampled distance resolution, Δ r is the minimum resolution distance unit of radar, and c is the light velocity, and Δ f is frequency step ladder, f _sfor baseband sampling rate;

Step 4: with " 0 " individual thick distance sample for benchmark, carries out ring shift, i=1:N-1 by i-th thick distance sample, obtains the matrix B that size is M × N, described ring shift computing formula:

B(n,i)＝A(mod(n+(i-1)*W,M),i)

Wherein mod is modulo operation;

Step 5: matrix B superposed by row, obtains the matrix X that size is M × 1;

Step 6: Application comparison method finds out the position X (m, 1) at maximal value mx and maximal value place in the M number in matrix X, intercepts the matrix Y of one section of W × N centered by maximal value in matrix B, intercepts formula as follows:

P＝m-floor(W/2)

Q＝P+W-1

Wherein P is for intercepting matrix initial row, and Q is for intercepting matrix end line, and floor is downward rounding operation; Take out the capable data capable to Q of P in matrix B, obtain the matrix Y that size is W × N;

Step 7: the data in matrix Y are spliced continuously by row head and the tail and obtains the one dimension matrix Z that size is 1 × (W*N), represent the one-dimensional range profile of target.

Beneficial effect

The stepped frequency radar signal target extract method of a kind of Corpus--based Method optimum that the present invention proposes, compared with existing Target pick-up algorithm, the beneficial effect had is:

No matter give up method or select large method, be all based on a hypothesis, namely " the 0th sampled point, if P ₀=0 ", this just needs to demarcate " 0 " individual sampled point in system, and in Practical Project, is inconvenient usually in realization.The data of all thick distance sample are first carried out " alignment " process by the statistics advantest method that the present invention proposes, data after " alignment " are added up along thick distance samples dimension, find the maximal value after superposition, and the data of taking out centered by this maximal value are spliced, and obtain the one-dimensional range profile of target.So neither need to demarcate " 0 " individual sampled point, also greatly improve the mulching measures of target.

Accompanying drawing explanation

The distribution schematic diagram of Fig. 1 one-dimensional range profile

Fig. 2 gives up Target pick-up algorithm schematic diagram

Fig. 3 selects general objective extraction algorithm schematic diagram

Schematic diagram under Fig. 4 gives up method and selects large method invalid situation

Fig. 5 the present invention adds up optimal objective extraction algorithm schematic diagram

Embodiment

Now in conjunction with the embodiments, the invention will be further described for accompanying drawing:

Now in conjunction with the embodiments, the invention will be further described for accompanying drawing:

The scheme of technical solution problem of the present invention is: first identical range unit point in the pulse repetition time different in frame data is carried out IFFT, by the result after IFFT with a certain thick distance sample for benchmark carries out ring shift alignment, for the sake of simplicity, can choose " 0 " individual thick distance sample is benchmark, again the data in follow-up each thick distance sample are carried out ring shift, make same section of thin sampled distance in each thick distance sample be continuous print on target actual distance, in this algorithm, be referred to as " alignment " process.Data are after " alignment " process, again the data after " alignment " are added up along thick distance samples dimension, find out the maximal value of the data after superposition, take out a part of data centered by this maximal value, this part data is carried out head and the tail splicing continuously, namely complete the Target pick-up algorithm of Corpus--based Method optimum.

In order to more clearly set forth embodiment, the data of a CPI i.e. frame data are stored as a matrix by the present invention, the wherein data of each pulse repetition time reception of every behavior of matrix, matrix be often classified as the different sampled point of pulse repetition time on same thick distance and position, hypothesis matrix size is T × N.Before doing Target pick-up algorithm, the data of a CPI need advanced row distance refinement, method is that the sampled point of the different pulse repetition times in same CPI on same thick distance and position is done IFFT process, namely IFFT is done to each row of matrix, suppose that IFFT counts as M, after so doing distance refinement, every frame data matrix size is M × N.Usual M=T; If (because M<T, only use the partial information in a front M PRF, do not use full detail, minor increment resolution element precision can be reduced; If M>T, then only operand can be increased, to raising minor increment resolution element precision without in all senses; So usual M=T.)

Fig. 5 is the Target pick-up algorithm schematic diagram of Corpus--based Method advantest method.The algorithm steps of statistics advantest method divides the following steps:

(1) draw all sampled point IFFT results, be set to matrix A, size is M × N;

(2) determine each thick distance sample value length W according to systematic parameter, namely each thick sampled point is refined as several thin distance sample (being converted into the data segment, length needing in each row of matrix to extract), and formula is as follows:

$W=\mathrm{Trunc}\left(\frac{r_s}{\mathrm{\&Delta;r}}\right)$

Wherein: Trunc for blocking computing, r _sfor sampled distance resolution, Δ r is the minimum resolution distance unit of radar, and has: r _s=c/2f _s, Δ r=c/ (2M Δ f), c is the light velocity, and Δ f is frequency step ladder, f _sfor baseband sampling rate.

(3) with " 0 " individual thick sampled point for benchmark, thick to i-th (i=1:N-1) distance sample is carried out ring shift, as shown in Figure 5, data by dash area in figure are displaced to figure bend part in the direction of arrows, namely to N row, often arrange M data to the 2nd row of matrix and carry out ring shift, ring shift is carried out according to following formula:

B(n,i)＝A(mod(n+(i-1)*W,M),i)；

Wherein mod is modulo operation, and after process, every frame data size is still M × N.

(4) data superposed along thick distance samples dimension, superpose by row by matrix B, obtain matrix X, size is M × 1.

(5) in the M number in matrix X, Application comparison method finds out the maximal value mx in matrix X, and the position X at its place (m, 1), and in matrix B, intercepts the matrix Y of one section of W × N centered by this maximal value, intercepts and is undertaken by following formula:

P＝m-floor(W/2)；

Q＝P+W-1；

Wherein P is for intercepting matrix initial row, and Q is for intercepting matrix end line, and namely take out the capable data capable to Q of P in matrix B, obtain matrix Y, size is W × N, and floor is downward rounding operation.

(6) data in matrix Y are obtained one dimension matrix Z by the continuous splicing of row head and the tail, the size of Z is 1 × (W*N).Thus obtain the one-dimensional range profile of target, complete Target pick-up algorithm.

Claims (1)

1. a stepped frequency radar signal target extract method for Corpus--based Method optimum, is characterized in that step is as follows:

Step 1: the data of a CPI being stored as a size is the matrix of T × N, the wherein data that receive of every behavior of matrix each pulse repetition time, matrix be often classified as the different sampled point of pulse repetition time on same thick distance and position;

Step 2: IFFT is done to each row of T × N matrix, after obtaining distance refinement, every frame data matrix A size is M × N, wherein: M is that IFFT counts;

Step 3: calculate the data segment, length W that each thick distance sample needs to extract:

$W=\mathrm{Trunc}\left(\frac{r_s}{\mathrm{\&Delta;r}}\right)$

r _s＝c/2f _s

Δr＝c/(2MΔf)

Wherein: Trunc for blocking computing, r _sfor sampled distance resolution, Δ r is the minimum resolution distance unit of radar, and c is the light velocity, and Δ f is frequency step ladder, f _sfor baseband sampling rate;

Step 4: with " 0 " individual thick distance sample for benchmark, carries out ring shift, i=1:N-1 by i-th thick distance sample, obtains the matrix B that size is M × N, described ring shift computing formula:

B(n,i)＝A(mod(n+(i-1)*W,M),i)

Wherein mod is modulo operation;

Step 5: matrix B superposed by row, obtains the matrix X that size is M × 1;

Step 6: Application comparison method finds out the position X (m, 1) at maximal value mx and maximal value place in the M number in matrix X, intercepts the matrix Y of one section of W × N centered by maximal value in matrix B, intercepts formula as follows:

P＝m-floor(W/2)

Q＝P+W-1

Wherein P is for intercepting matrix initial row, and Q is for intercepting matrix end line, and floor is downward rounding operation; Take out the capable data capable to Q of P in matrix B, obtain the matrix Y that size is W × N;

Step 7: the data in matrix Y are spliced continuously by row head and the tail and obtains the one dimension matrix Z that size is 1 × (W*N), represent the one-dimensional range profile of target.