CN101938317A - Line-spectrum detection method for noise power spectra - Google Patents

Abstract

The invention discloses a line-spectrum detection method for noise power spectra, which comprises: setting a noise signal sequence as s(n), performing the power spectrum estimation of the s(n), and acquiring power spectrum p1(n) and a logarithmic power spectrum p2(n), wherein n is a non-negative integer; acquiring a smooth spectrum by using sliding window orthogonal polynomial related fitting and acquiring a power smooth spectrum ps1(n) of a noise signal and the logarithmic power smooth spectrum ps2(n) of the noise signal by performing the sliding window orthogonal polynomial fitting of the power spectrum p1(n) and the logarithmic power spectrum p2(n); computing the difference spectra of the power spectrum and the logarithmic power spectrum and normalizing the difference spectrum, wherein pd1(n)=[ p1(n)-ps1(n)]/std([ p1(n)-ps1(n)]) and pd2(n)=[ p2(n)-ps1(n)]/std([ p2(n)-ps2(n)]), and the pd1(n) and the pd2(n) represent the normalized difference spectrum of the power spectrum and the normalized difference spectrum of the logarithmic power spectrum; setting an amplitude threshold G1 and a logarithm threshold G2, and extracting line spectrum pl1(n) of the power spectrum and the line spectrum pl2(n) of the logarithmic power spectrum; and acquiring a target line spectrum pl(n) by combining the power spectrum and the logarithmic power spectrum, and displaying the value of the spectrum in a logarithmic from.

Description

Noise power spectrum line spectrum detection method

One, technical field:

The invention belongs to the signal processing technology field, relate to the method that a kind of noise line spectrum detects.

Two, background technology:

Line spectrum in the noise detects has important meaning, traditional line spectrum leaching process still to have shortcomings such as computing complexity, big, the easy error extraction line spectrum of amount of calculation or leakage extraction.Its reason is that the method for trend term extraction is complicated and decision threshold is single.

The present invention is directed to this two shortcomings, propose the method that a kind of power spectrum noise line spectrum extracts.Utilize precalculated orthogonal polynomial sequence to compose the data sliding window and extract trend term, simplified the amount of calculation of trend term extraction and be easy to realization.Employing amplitude thresholding and logarithm thresholding extract the target line spectrum, effectively reduce the error extraction of line spectrum and leak to extract probability.

Three, summary of the invention:

The object of the present invention is to provide a kind of noise power spectrum line spectrum detection method, it can be at balanced background noise, by the line spectrum component in the comprehensive judgement extraction noise of amplitude power spectrum and log power spectrum.

The object of the present invention is achieved like this:

A kind of noise power spectrum line spectrum detection method is characterized in that the power spectrum at noise signal, utilizes quick background equalization methods to eliminate trend term, utilizes difference spectrum and ratio spectrum thresholding comprehensively to extract the power spectrum line spectrum.Comprise following process:

A. establishing the noise signal sequence is s (n), and its power spectrum is estimated that obtain power spectrum p1 (n) and log power spectrum p2 (n), n is a nonnegative integer,

B. utilize the relevant match of sliding window orthogonal polynomial to obtain level and smooth spectrum

If power spectrum data p (n) length N point is got the window that a length is ordered for M, M＜N, step-length m point, m＜M, the initial starting point w=0 that establishes window overlaps with the starting point of power spectrum, and N, M and m are positive integer, the w nonnegative integer, w is the start position of window,

A) calculate 5 groups of standard orthogonal sequences in advance, computational methods:

$x_i\left(n\right)={(\frac{2n}{M-1}-1)}^i,$ I=0 wherein, 1,2,3,4, n=0,1 ..., M-1,

To x _i(n) carry out Schmidt Schimidt orthogonalization, obtain standard orthogonal vectors y _i(n), i=0,1,2,3,4,

B) in power spectrum data p (n), get w, w+1 ..., the w+M-1 point is as pending window power spectrum data s _M(n), that is:

s _M(n)＝p(n+w)，n＝0，1，...，M-1

C) calculation window power spectrum data s _M(n) and standard orthogonal vectors y _i(n) correlation coefficient r _i,

D) calculation window match spectrum

Get window match spectrum s ' _M(n) preceding m point data is inserted among the level and smooth spectrum data ss (n), promptly

ss(n+w)＝s′ _M(n)，n＝0，1，...，m-1

E) when w+m+M＜N-1, make w=w+m, return b), otherwise, enter f),

F) when w+m+M＞=N-1, note w0=w makes w=N-1-M again, is carrying out b successively), c) after, calculate

$s_M^{\′}\left(n\right)=\underset{i=0}{\overset{5}{\mathrm{\Σ}}}{r}_i{y}_i\left(n\right),$ Get ss (w0+m+n)=s ' _M(w0+m-w+n),

N=0 wherein, 1 ..., M-1+w-w0-m,

C. according to the B method, power spectrum p1 (n) and log power spectrum p2 (n) are carried out the sliding window way of fitting, the power that obtains noise signal is smoothly composed the logarithm power of ps1 (n) and noise signal and is smoothly composed ps2 (n),

D. the difference spectrum and the standardization of rated output spectrum and log power spectrum,

pd1(n)＝[p1(n)-ps1(n)]/std{[p1(n)-ps1(n)]}

pd2(n)＝[p2(n)-ps2(n)]/std{[p2(n)-ps2(n)]}

Pd1 (n), pd2 (n) are respectively power spectrum standard difference spectrum and log power spectrum standard difference spectrum,

E. set amplitude thresholding G1 and logarithm thresholding G2, extract power spectrum and log power spectrum line spectrum pl1 (n), pl2 (n):

F. comprehensive power spectrum and log power spectrum obtain target line spectrum pl (n), and the spectrum value provides with logarithmic form:

Compared with prior art, the present invention has following advantage:

1) integrated noise power spectrum and noise log power spectrum extract line spectrum, and it is more reliable that more single power spectrum extracts line spectrum;

2) method of calculating standard orthogonal polynomial is simple, adopts the relevant match of sliding window orthogonal polynomial to obtain level and smooth spectrum, can obtain background trend item preferably;

3) this method has realizability preferably.

Four, description of drawings

Fig. 1 is a FB(flow block) of the present invention, wherein, and 1. noise signal; 2. rated output is composed; 3. the relevant match of sliding window orthogonal polynomial; 4. go trend term, standardization; 5. extract the amplitude line spectrum; 6. calculating log power spectrum; 7. the relevant match of sliding window orthogonal polynomial; 8. go trend term, standardization; 9. extract logarithm amplitude line spectrum; 10. extract comprehensive logarithm line spectrum.

Fig. 2 is in the sliding window, the correlation schematic diagram of N, M, m.

Fig. 3 is 16384 spot noise sequences.

Fig. 4 is noise power spectral sequence (last figure) and noise log power spectrum sequence (figure below).

Fig. 5 is the standard orthogonal sequence.

Fig. 6 is that noise power spectrum smoothing spectrum (last figure) and noise log power spectrum are smoothly composed (figure below).

Fig. 7 is power spectrum standard difference spectrum (last figure) and log power spectrum standard difference spectrum (figure below).

Fig. 8 is the noise power spectrum line spectrum.

Five, embodiment

At the power spectrum of noise signal, utilize quick background equalization methods to eliminate trend term, utilize difference spectrum and ratio spectrum thresholding comprehensively to extract the power spectrum line spectrum.Comprise following process:

Embodiment 1

A kind of noise power spectrum line spectrum detection method,

A. establishing the noise signal sequence is s (n), and its power spectrum is estimated that obtain power spectrum p1 (n) and log power spectrum p2 (n), n is a nonnegative integer,

B. utilize the relevant match of sliding window orthogonal polynomial to obtain level and smooth spectrum

If power spectrum data p (n) length N point is got the window that a length is ordered for M, M＜N, step-length m point, m＜M, the initial starting point w=0 that establishes window overlaps with the starting point of power spectrum, and N, M and m are positive integer, the w nonnegative integer, w is the start position of window, and the relation of N, M, m as shown in Figure 2

A) calculate 5 groups of standard orthogonal sequences in advance, computational methods:

$x_i\left(n\right)={(\frac{2n}{M-1}-1)}^i,$ I=0 wherein, 1,2,3,4, n=0,1 ..., M-1,

To x _i(n) carry out Schmidt Schimidt orthogonalization, obtain standard orthogonal vectors y _i(n), i=0,1,2,3,4,

B) in power spectrum data p (n), get w, w+1 ..., the w+M-1 point is as pending window power spectrum data s _M(n), that is:

s _M(n)＝p(n+w)，n＝0，1，...，M-1

C) calculation window power spectrum data s _M(n) and standard orthogonal vectors y _i(n) correlation coefficient r _i,

D) calculation window match spectrum

Get window match spectrum s ' _M(n) preceding m point data is inserted among the level and smooth spectrum data ss (n), promptly

ss(n+w)＝s′ _M(n)，n＝0，1，...，m-1

E) when w+m+M＜N-1, make w=w+m, return b), otherwise, enter f),

F) when w+m+M＞=N-1, note w0=w makes w=N-1-M again, is carrying out b successively), c) after, calculate

$s_M^{\′}\left(n\right)=\underset{i=0}{\overset{5}{\mathrm{\Σ}}}{r}_i{y}_i\left(n\right),$ Get ss (w0+m+n)=s ' _M(w0+m-w+n),

N=0 wherein, 1 ..., M-1+w-w0-m,

C. according to the B method, power spectrum p1 (n) and log power spectrum p2 (n) are carried out the sliding window way of fitting, the power that obtains noise signal is smoothly composed the logarithm power of ps1 (n) and noise signal and is smoothly composed ps2 (n),

D. the difference spectrum and the standardization of rated output spectrum and log power spectrum,

pd1(n)＝[p1(n)-ps1(n)]/std{[p1(n)-ps1(n)]}

pd2(n)＝[p2(n)-ps2(n)]/std{[p2(n)-ps2(n)]}

Pd1 (n), pd2 (n) are respectively power spectrum standard difference spectrum and log power spectrum standard difference spectrum,

E. set amplitude thresholding G1 and logarithm thresholding G2, extract power spectrum and log power spectrum line spectrum pl1 (n), pl2 (n):

F. comprehensive power spectrum and log power spectrum obtain target line spectrum pl (n), and the spectrum value provides with logarithmic form:

Embodiment 2

At first gathering the target noise burst is s (n), n=0 wherein, and 1 ..., 16383, as shown in Figure 3.Noise signal power spectrum is estimated, obtained power spectrum p1 (n) and log power spectrum p2 (n), as shown in Figure 4, n=0 wherein, 1 .., 8191.Get a long window of ordering for M=41, step-length m=5 point is established the initial starting point w=0 of window.

Promptly utilize the relevant match of sliding window orthogonal polynomial to obtain level and smooth spectrum according to the B method, calculate 5 groups of standard orthogonal sequence y _i(n), i=0,1,2,3,4, as shown in Figure 5.Power spectrum p1 (n) and log power spectrum p2 (n) are carried out the sliding window way of fitting, and the power that obtains noise signal is smoothly composed the logarithm power of ps1 (n) and noise signal and is smoothly composed ps2 (n), as shown in Figure 6.

The difference spectrum and the standardization of rated output spectrum and log power spectrum obtain standard difference spectrum pd1 (n), pd2 (n), as shown in Figure 7.

Setting amplitude thresholding G1=3.0 and logarithm thresholding G2=3.0 extract power spectrum and log power spectrum standard difference spectrum pl1 (n), pl2 (n), and the line spectrum pl (n) that obtains noise after comprehensive is as shown in Figure 8.In Fig. 7, can see, in the power spectrum standard difference spectrum, 200 points, there is the many places range value to surpass 3.0 threshold value between 3000 o'clock to 4000 o'clock, in the standard difference spectrum of log power spectrum, 100 points, 200 points, 3000 have all surpassed 3.0 threshold value, through two results' merging, the value that has obtained 200 and 3000 positions is the line spectrum value, handles the probability that has reduced the error extraction line spectrum than independent power spectrum processing or log power spectrum.

Claims (1)

1. a noise power spectrum line spectrum detection method is characterized in that,

A. establishing the noise signal sequence is s (n), and its power spectrum is estimated that obtain power spectrum p1 (n) and log power spectrum p2 (n), n is a nonnegative integer,

B. utilize the relevant match of sliding window orthogonal polynomial to obtain level and smooth spectrum

If power spectrum data p (n) length N point is got the window that a length is ordered for M, M＜N, step-length m point, m＜M, the initial starting point w=0 that establishes window overlaps with the starting point of power spectrum, and N, M and m are positive integer, the w nonnegative integer, w is the start position of window,

A) calculate 5 groups of standards in advance and only hand over sequence, computational methods:

I=0 wherein, 1,2,3,4, n=0,1 ..., M-1,

To x _i(n) carry out Schmidt Schimidt and end friendshipization, obtain standard and only hand over vectorial y _i(n), i=0,1,2,3,4,

B) in power spectrum data p (n), get w, w+1 ..., the w+M-1 point is as pending window power spectrum data s _M(n), that is:

s _M(n)＝p(n+w)，n＝0，1，...，M-1

C) calculation window power spectrum data s _M(n) and standard only hand over vectorial y _i(n) correlation coefficient r _i,

D) calculation window match spectrum

Get window match spectrum s _M(n) preceding m point data is inserted among the level and smooth spectrum data ss (n), promptly

ss(n+w)＝s _M(n)，n＝0，1，...，m-1

E) when w+m+M＜N-1, make w=w+m, return b), otherwise, enter f),

F) when w+m+M＞=N-1, note w0=w makes w=N-1-M again, is carrying out b successively), c) after, calculate

Get ss (w0+m+n)=s _M(w0+m-w+n),

N=0 wherein, 1 ..., M-1+w-w0-m,

C. according to the B method, power spectrum p1 (n) and log power spectrum p2 (n) are carried out the sliding window way of fitting, the power that obtains noise signal is smoothly composed the logarithm power of ps1 (n) and noise signal and is smoothly composed ps2 (n),

D. the difference spectrum and the standardization of rated output spectrum and log power spectrum,

pd1(n)＝[p1(n)-ps1(n)]/std{[p1(n)-ps1(n)]}

pd2(n)＝[p2(n)-ps2(n)]/std{[p2(n)-ps2(n)]}

Pd1 (n), pd2 (n) are respectively power spectrum standard difference spectrum and log power spectrum standard difference spectrum,

E. set amplitude thresholding G1 and logarithm thresholding G2, extract power spectrum and log power spectrum line spectrum pl1 (n), pl2 (n):

F. comprehensive power spectrum and log power spectrum obtain target line spectrum pl (n), and the spectrum value provides with logarithmic form:

。

Images