CN104502732A - Radiation source screening and positioning method based on STFT time frequency analysis - Google Patents

Abstract

The invention discloses a radiation source screening and positioning method based on STFT time frequency analysis. According to the method, an independent component analysis ICA algorithm is in combination with STFT, and a systematic, comprehensive, standardized radiated EMI analysis and diagnosis method is proposed. Firstly, radiation noise of EUT is separated through an ICA algorithm, and multiple radiation noise sources are screened out; secondly, short-time rapid Fourier transform and time frequency analysis on exceeding-standard frequency ranges of the overall radiation noise are carried out, and signal characteristics causing exceeding-standard radiation are extracted; lastly, the extracted signal characteristics and the multiple radiation noise sources screened out by the ICA are contrasted, positioning diagnosis on the noise sources causing exceeding-standard radiation noise is carried out, and pertinent rectification inhibition measures are adopted. The method can rapidly and effectively inhibit exceeding-standard noise of equipment through searching the corresponding device causing exceeding-standard radiation.

Description

Based on radiation source screening and the localization method of STFT time frequency analysis

Technical field

The present invention relates to the screening of a kind of radiation source based on STFT time frequency analysis and localization method, belong to technical field of electromagnetic compatibility.

Background technology

In recent years, along with a large amount of power electronic equipment or system continue to bring out, radiate EMI noise superposes in spatial field, can affect the normal operation of self and ambient electronics, and according to GB 9254, the test frequency range of radiate EMI noise is 30MHz to 1GHz.Therefore challenge maximum in EMI fault analysis is become to the identification of electromagnetic radiation of electronic equipment noise source and diagnosis.

Radiated noise Study on Problems generally includes Radar recognition, estimates and suppresses three aspects.Radar recognition refers to acquisition to information such as radiation source feature, mechanism, positions and judgement; Radiation source is estimated and is referred to not by standard detection, estimates the issuable greatest irradiation noise of equipment; Radiation source suppresses the measure referred to for reducing radiated noise.Wherein, it is basic goal that radiated noise suppresses, and it is judgment criterion that radiation source is estimated, and Radar recognition is then precondition and theoretical foundation.In addition, as the circuit devcie high frequency noise of primary radiation source, cause radiation field by common mode/differential mode radiation in space, produce source of secondary radiation, mutually will cause serious radiated noise after superposition.Because radiation source quantity is more, mutually superposition, coupling condition are remarkable, and have strong nonlinearity feature, and therefore, radiation source research of tracing to the source seems and is even more important, also comparatively difficulty.

There is many places deficiency in existing Radiative EMI diagnostic method, such as, cannot diagnose the radiation mechanism of circuit-under-test, and actual conditions are difficult to meet test environment requirement, and analytical approach is comparatively complicated, and data volume is large.In view of this, the present invention proposes the screening of a kind of radiation source based on STFT time frequency analysis and localization method, can judge that dipole produces radiation electromagnetic interference noise.

Summary of the invention

Technical matters to be solved by this invention, being the defect overcoming prior art existence, is no longer rectification method by rule of thumb traditionally, but by finding the corresponding device causing radiation to exceed standard, the slight elevated noise of suppression equipment fast and effectively.

The stable signal of electromagnetic radiation noise right and wrong in reality, the signal in electronic equipment is the superposition of radiated noise signals in spatial field that multiple radiation source produces, and the spectral characteristic of this signal changes over time.Short term Fourier transform (STFT) is a kind of most popular method of research non-stationary signal, the basic thought of this conversion carries out windowing with an analysis window slided in time to non-stationary signal to block, non-stationary signal is resolved into a series of approximate stable short signal, then use Fourier transform theoretical analysis each short out time stationary signal frequency spectrum.In the application of electromagnetic radiation noise diagnostics, STFT is applied to the radiated noise measurement result analysis of EUT, in conjunction with independent component analysis ICA algorithm, analyzes to extract and cause radiation to exceed standard the feature of signal.

The present invention proposes the screening of a kind of radiation source based on STFT time frequency analysis and localization method, the steps include:

The first step: carry out near-field test to an equipment under test M magnet field probe, records the near field time domain signal of M group equipment under test by oscillograph, by this M group time-domain signal statistical conversion, by M group observation signal x _it () forms column vector x (t)=[x ₁(t) ..., x _m(t)] ^t, i=1,2 ..., M; The number M that wherein pops one's head in depends on source signal number N, M>=N;

Second step: independent component analysis is carried out to observation signal x (t):

If source signal is by N group independently signal s ₁(t) ..., s _nt () forms, i.e. s (t)=[s ₁(t) ..., s _n(t)] ^t, the pass between observation signal and source signal is x (t)=As (t), and wherein A is hybrid matrix; Independent component analysis namely will under the condition of A, s (t) the unknown, and signal x (t) according to the observation, finds out separation matrix W by the objective function of the criterion of independence weighing variable, thus try to achieve the maximum estimated of s (t); If the separation signal exported is y (t)=[y ₁(t) ..., y _n(t)], then have

y _N(t)＝W _N×Mx _M(t)＝W _N×MA _M×Ns _N(t)＝s _N(t) (1)

Wherein, y (t) is a maximum estimated of source signal s (t), and each separation signal y _iseparate as much as possible between (t); In this model, source signal number M is generally got identical with observation signal number N;

First, carry out whitening pretreatment to observation signal x (t), use principal component method removes the correlativity between each signal source, makes it uncorrelated, namely finds a linear transformation R ₀, make the R after converting ₀x (t) is albefaction vector, R ₀can be obtained by formula (2), make association's method Matrix C of observation signal x (t) _x=E [xx ^t], A=diag [d1...dn] is with C _xeigenwert be the diagonal matrix of diagonal element, U is C _xeigenvectors matrix, then

R ₀＝Λ ^-1/2U ^T(2)

After observation signal x (t) is carried out whitening processing, then iteration is carried out by approximate Newton iteration method, finding a separation matrix W, that y=Wx is had is maximum Gaussian, substitute in y=Wx and obtain the M group independent signal after being separated, be isolated independent radiated noise signals in near-field test, be designated as Z ₁(t), Z ₂(t) ..., Z _m(t);

3rd step: to the time-domain signal Z after separation _m(t) and the observation signal x recorded _it (), carries out short term Fourier transform STFT conversion respectively, by Time-Frequency Analysis Method, extract the signal characteristic of the frequency that exceeds standard;

The continuous Short Time Fourier Transform of signal z (t) is defined as follows:

$\mathrm{STFTz}(t,f)={\∫}_{-\∞}^{\∞}[z\left(t^{\′}\right){\mathrm{\γ}}^{*}(t^{\′}-t)e^{-2\mathrm{\πf}{t}^{\′}}{\mathrm{dt}}^{\′}---\left(3\right)$

In formula γ (t) be a time width very little slidably time window, * represents complex conjugate, and namely signal z (t) is multiplied by the analysis window γ that take t as distribution center ^*(t'-t) Fourier transform done; The frequency spectrum SPEC of short term Fourier transform STFT, i.e. the T/F energy distribution (instantaneous power spectral density) of STFT, be defined as STFT (t, f) modulus value square, that is:

SPEC(t,f)＝|STFT(t,f)| ²(4)

Be not difficult to obtain the digital algorithm of short term Fourier transform STFT and frequency spectrum SPEC thereof by (3), (4) two formulas:

$\mathrm{STFT}(n,k)=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\left[\left(x_1\right){\mathrm{\γ}}^{*}(i-n)\right]\exp (-j\frac{2\mathrm{\πki}}{N})---\left(5\right)$

SPEC(n,k)＝|STFT(n,k)| ²(6)

Adopt Fast Fourier Transform (FFT) FFT to realize the fast algorithm of formula (5), in formula: N represents Fast Fourier Transform (FFT) FFT length, n, k represent discrete time in time frequency plane represented by STFT frequency spectrum and frequency grid " node "; In application, generally realize the fast algorithm of formula (5) with FFT;

After time-domain signal after separation is changed into frequency-region signal, time frequency analysis is carried out to this signal, association frequency domain and time domain data, construct the joint density function of a kind of time and frequency, to disclose the frequency component and evolution properties thereof that comprise in signal, the frequency content rule over time of signal can well be indicated, thus extract the feature of this signal;

4th step: after carrying out time frequency analysis, extracts the signal characteristic of signal and mixed signal after being separated respectively, analyzes the signal characteristic that radiated noise signals is maximum; Contrast with the circuit theory diagrams of equipment under test again, finally determine the radiation source causing radiated noise to exceed standard.

Independent component analysis ICA algorithm combines with STFT by the present invention, proposes set of system, comprehensive, standardized Radiative EMI analysis and diagnosis method.First be separated by the radiated noise of ICA algorithm by EUT, filter out several radiated noise sources.Secondly, the frequency range that exceeds standard of radiation overall noise is carried out short term Fourier transform and time frequency analysis, extracts the signal characteristic causing radiation to exceed standard.Several radiated noise source signal characteristics comparisons signal characteristic extracted and ICA filtered out, the noise source exceeded standard to causing radiated noise positions diagnosis, takes to rectify and improve braking measure targetedly.If diagnose the noise source drawn to be Commonmode model, then take to reduce common mode current or reduce the measures such as the short straight antenna length of equivalence, if be diagnosed as differential mode noise, then take to reduce the measures such as loop area, radiated noise is suppressed accurately and effectively.

Accompanying drawing explanation

Fig. 1 is experimental arrangement figure of the present invention;

Fig. 2 is the time domain waveform of signal after ICA is separated;

Fig. 3 is the signal z after being separated ₁the time frequency analysis result of (t);

Fig. 4 is the signal z after being separated ₂the time frequency analysis result of (t);

Fig. 5 is the time frequency analysis result of mixed signal.

Embodiment

Use PCB example below, be described in further detail the present invention, this PCB is made up of two of 20MHz and 6MHz crystal oscillators.Pcb board, oscillograph, two magnet field probes are positioned on experiment table after being connected according to Fig. 1 with wire.

The first step: first carry out waving map to pcb board, records two groups of time-domain signals with two magnet field probes, draws two groups of observation signal x ₁(t), x ₂(t).

Second step: for the mixed time domain signal recorded, first these two groups of mixed signals are formed the matrix x of a 2*10000, carry out independent component analysis to observation signal x, namely finding separation matrix W, that y=Wx is had is maximum Gaussian, separates the signal into two independently signal z ₁(t), z ₂(t), as shown in Figure 2.

3rd step: to the signal z after separation ₁t () carries out STFT time frequency analysis, result as shown in Figure 3, extracts the some groups of data that in this figure, energy is the strongest, as shown in table 1.

The most intensity values of the energy that table 1 extracts

X/*10e -6	0.891	1.921	3.584	5.366	7.703	13.68	15.43	16.53	19.5
										Y/*10e 8	1.319	1.321	1.324	1.315	1.323	1.317	1.317	1.305	1.315
index	2.191	2.367	2.155	3.769	2.195	2.752	2.592	2.759	2.77

As can be seen from the figure this noise signal is maximum in 130M to 135M left-right signal intensity, the components and parts in comparison pcb board, finds that the frequency multiplication of 6M crystal oscillator is 132M.Next, take 132M as the variance extracting data in benchmark table 1, try to achieve standard deviation S=0.007 and level off to 0, signal z can be drawn ₁t () is that 132M place signal intensity is maximum in frequency, be the radiated noise signals that the 6M crystal oscillator in pcb board is produced by common mode/differential mode radiation.

To the signal z after separation ₂t () carries out STFT time frequency analysis, as shown in Figure 4, extracting this time frequency analysis figure, to obtain data as shown in table 2 for result.

The most intensity values of the energy that table 2 extracts

X/*10e -6	0.814	1.049	2.713	4.891	7.703	11.9	13.37	14.87	19.7
										Y/*10e 7	1.965	1.99	2.017	2.019	1.971	1.995	1.946	2.01	2.01
index	1.313	1.21	1.215	1.026	1.168	1.151	1.158	1.166	1.295

As can be seen from the figure this noise signal is maximum in 19 to 21M left-right signal intensity, the components and parts in comparison pcb board, be assumed to be 20M crystal oscillator produce.Next, take 20M as the variance extracting data in benchmark table 2, try to achieve standard deviation S=0.024 and level off to 0, signal z can be drawn ₂t () is that 20M place signal intensity is maximum in frequency, be the radiated noise signals that the 20M crystal oscillator in pcb board is produced by common mode/differential mode radiation.

Carry out time frequency analysis to mixed time domain signal, after mixed signal carries out STFT time frequency analysis, result as shown in Figure 5, and extracting this time frequency analysis figure, to obtain data as shown in table 3.

The most intensity values of the energy that table 3 extracts

X/*10e -6	0.732	1.762	3.91	6.633	8.376	12.38	14.04	15.9	17.01	18.67
											Y/*10e 8	1.318	1.303	1.31	1.325	1.32	1.325	1.315	1.315	1.313	1.322
Index/*10e -4	13.18	10.28	10.25	10.65	12.77	12.62	11.14	11.96	13.91	10.89

As can be seen from the figure this noise signal is maximum in 125 to 135M left-right signal intensity, the components and parts in comparison pcb board, and 132M is the frequency multiplication of 6M crystal oscillator, supposes that this signal is that 6M crystal oscillator major effect produces.Next, take 132M as the variance extracting data in benchmark table 2, try to achieve standard deviation S=0.007 and level off to 0, signal z can be drawn ₂t () is that 132M place signal intensity is maximum in frequency, be the radiated noise signals that the 6M crystal oscillator in pcb board is produced by common mode/differential mode radiation.

4th step: sum up above-mentioned data conclusion, can show that the radiated noise that pcb board produces is caused by 6M and 20M crystal oscillator, and 6M crystal oscillator is main radiation source, rectifies and improves this device.

Claims (2)

1., based on radiation source screening and the localization method of STFT time frequency analysis, the steps include:

The first step: carry out near-field test to an equipment under test M magnet field probe, records the near field time domain signal of M group equipment under test by oscillograph, by this M group time-domain signal statistical conversion; By M group observation signal x _it () forms column vector x (t)=[x ₁(t) ..., x _m(t)] ^t, i=1,2 ..., M; The number M that wherein pops one's head in depends on source signal number N, M>=N;

Second step: independent component analysis is carried out to observation signal x (t):

If source signal is by N group independently signal s ₁(t) ..., s _nt () forms, i.e. s (t)=[s ₁(t) ..., s _n(t)] ^t, the pass between observation signal and source signal is x (t)=As (t), and wherein A is hybrid matrix; If the separation signal exported is y (t)=[y ₁(t) ..., y _n(t)], then have

y _N(t)＝W _N×Mx _M(t)＝W _N×MA _M×Ns _N(t)＝s _N(t) (1)

Wherein, y (t) is a maximum estimated of source signal s (t), and each separation signal y _iseparate as much as possible between (t);

First, carry out whitening pretreatment to observation signal x (t), use principal component method removes the correlativity between each signal source, makes it uncorrelated, namely finds a linear transformation R ₀, make the R after converting ₀x (t) is albefaction vector, R ₀can be obtained by formula (2), make association's method Matrix C of observation signal x (t) _x=E [xx ^t], A=diag [d1...dn] is with C _xeigenwert be the diagonal matrix of diagonal element, U is C _xeigenvectors matrix, then

R ₀＝Λ ^-1/2U ^T(2)

After observation signal x (t) is carried out whitening processing, then iteration is carried out by approximate Newton iteration method, finding a separation matrix W, that y=Wx is had is maximum Gaussian, substitute in y=Wx and obtain the M group independent signal after being separated, be isolated independent radiated noise signals in near-field test, be designated as Z ₁(t), Z ₂(t) ..., Z _m(t);

3rd step: to the time-domain signal Z after separation _m(t) and the observation signal x recorded _it (), carries out short term Fourier transform STFT conversion respectively, by Time-Frequency Analysis Method, extract the signal characteristic of the frequency that exceeds standard;

The continuous Short Time Fourier Transform of signal z (t) is defined as follows:

In formula γ (t) be a time width very little slidably time window, * represents complex conjugate, and namely signal z (t) is multiplied by the analysis window γ that take t as distribution center ^*(t'-t) Fourier transform done; The frequency spectrum SPEC of short term Fourier transform STFT, i.e. the T/F energy distribution of STFT, be defined as STFT (t, f) modulus value square, that is:

SPEC(t,f)＝|STFT(t,f)| ²(4)

Be not difficult to obtain the digital algorithm of short term Fourier transform STFT and frequency spectrum SPEC thereof by (3), (4) two formulas:

SPEC(n,k)＝|STFT(n,k)| ²(6)

Adopt Fast Fourier Transform (FFT) FFT to realize the fast algorithm of formula (5), in formula: N represents Fast Fourier Transform (FFT) FFT length, n, k represent discrete time in time frequency plane represented by STFT frequency spectrum and frequency grid node;

After time-domain signal after separation is changed into frequency-region signal, time frequency analysis is carried out to this signal, association frequency domain and time domain data, construct the joint density function of a kind of time and frequency, to disclose the frequency component and evolution properties thereof that comprise in signal, the frequency content rule over time of signal can well be indicated, thus extract the feature of this signal;

4th step: after carrying out time frequency analysis, extracts the signal characteristic of signal and mixed signal after being separated respectively, analyzes the signal characteristic that radiated noise signals is maximum; Contrast with the circuit theory diagrams of equipment under test again, finally determine the radiation source causing radiated noise to exceed standard.

2. the screening of the radiation source based on STFT time frequency analysis according to claim 1 and localization method, is characterized in that, in the formula (1) of described step 3, get M=N.