CN105911351B - Real-time ZFFT methods based on DSP - Google Patents
Real-time ZFFT methods based on DSP Download PDFInfo
- Publication number
- CN105911351B CN105911351B CN201610217659.1A CN201610217659A CN105911351B CN 105911351 B CN105911351 B CN 105911351B CN 201610217659 A CN201610217659 A CN 201610217659A CN 105911351 B CN105911351 B CN 105911351B
- Authority
- CN
- China
- Prior art keywords
- frequency
- zfft
- real
- data
- dsp
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
- G01R23/16—Spectrum analysis; Fourier analysis
- G01R23/165—Spectrum analysis; Fourier analysis using filters
Landscapes
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Measuring Frequencies, Analyzing Spectra (AREA)
Abstract
The present invention discloses a kind of real-time ZFFT methods based on DSP, including three steps：（1）By application requirement off-line calculation and filter coefficient, the frequency displacement coefficient of temporary ZFFT；（2）The data sampling data choosing pumping for carrying out ZFFT online simultaneously, filtering, shift frequency operation；（3）To step（2）As a result the refinement spectral line that offline FFT can be obtained required bandwidth is carried out.Present invention tool the characteristics of being effectively utilized ZFFT operations and the strong advantage of DSP digital processing capabilities, take out the data choosing of ZFFT and are combined with sampling, and filtering, shift frequency operation, which are distributed to multiple sampling periods, to be executed；Sampling terminates only to need the FFT of a Q point, you can obtains the refinement spectral line of analysis bandwidth range；Compared with existing ZFFT methods：Committed memory space is few, real-time is good, can meet on-line monitoring, real-time signal analysis processing system needs.
Description
Technical field
The invention belongs to frequency spectrum refinement analytical technology, specifically a kind of real time spectrum thinning method based on DSP.
Background technology
ZFFT is also referred to as Zoom-FFT, is a kind of complex electromagnetic parameters method.It is conditional in memory and FFT computational lengths
In the case of, both otherwise frequency analysis range is reduced, and it is contradictory to increase frequency resolution again；For this purpose, the seventies in last century
Occur the various Frequence zooming analysis methods based on different principle in succession afterwards, main purpose is the subtle knot identified on spectrogram
Structure.Such as：Frequency sweep narrow band analyzing method, the Zoom-FFT methods based on multiple modulation directly select pumping method, phase compensation refinement and maximum frequency
The local representation etc. of spectrum.It is most common to have the methods of multiple modulation Zoom-FFT, phase compensation refinement.However computational efficiency,
The all more satisfactory method of precision and flexibility etc. the still Zoom-FFT based on multiple modulation, therefore obtained more answer
With.
The basic thought of ZFFT is to utilize frequency displacement theorem, is the high-resolution Fourier's analysis method based on multiple modulation, it
Enough frequency resolutions can be specified to analyze spectrum structure of a certain bandwidth signal on the frequency axis in any narrowband, including number
Word frequency displacement, digital low-pass filtering, resampling（Choosing is taken out）, FFT transform and weighting processing and etc..It is to be subject to time domain samples
Transformation, makes corresponding frequency spectrum origin move on at the centre frequency of frequency range interested, then resampling is FFT, you can to obtain higher
Frequency resolution.ZFFT is as follows.
（1）Signal discreteization and digital frequency displacement time domain vibration signalx(t) through cutoff frequency bef c≤f s/ 2 low pass is anti-
Mixed filter (LPF), sampling time sequence is converted to by A/Dx 0(n).Analysis on Selecting frequency band based on the actual application requirements
The upper limitf 2, lower limitf 1, in frequency rangef 1~f 2Interior carry out frequency spectrum refinement, the then band center to be observed becomef o=0.5（f 1+f 2）, in order to which band center is becomef oOriginal zero frequency position is moved to, it is right according to the frequency displacement property of DFTx 0(n) with e-i2π(fo/fs)n Progress multiple modulation obtains digital frequency shift signal and is
x (n)=x 0(n) e-i2π(fo/fs)n =x 0(n)[cos(2πf o n/Nf s)-sin(2πf o n/Nf s)]。
（2）The resampling of frequency shift signal is to avoid high frequency aliasing, the frequency shift signal that will be previously obtainedx (n) pass through number
Low-pass filter filters out the radio-frequency component other than observation simulation；Choosing is synchronized with scale factor D to take out, sample frequency is dropped in time domain
As low asf s/ D, scale factor D, which are also known as selected, takes out ratio or refinement multiple.It, must in order to ensure to will not occur the aliasing of frequency spectrum after choosing is taken out
Corresponding limit conditional, the i.e. bandwidth of low-pass filter must be given no more thanf s/2D。
（3）FFT transform and weighting handle the resampling result to frequency shift signalg(m) carry out FFT transform can find out its frequency
Spectrum, after being weighted processing to frequency spectrum, you can obtain analysis frequency rangef 1~f 2Interior zoom FFT.
More than（1）~（3）The refinement selectable band analysis of step, if all carrying out N point spectrum analysis before and after refinement, the panorama before refinement
Composing has N/2 item independence spectral lines, and reflection 0~f sThe frequency spectrum of/2 frequency ranges；And select band spectrum that there is N items independently to compose after refining
Line, reflectionf 1~f 2Frequency spectrum in frequency range；Obviously it is inconsistent to refine forward and backward spectrum number of lines.Before and after making refinement
Independent spectrum number of lines is consistent, can be used and is by one times of the cutoff frequency constriction of low-pass digital filterf s/ 4D, i.e., every 2D point double samplings
Method.Therefore, there are following limitations for above-mentioned ZFFT analysis methods, cannot be satisfied the requirement of real-time of monitoring system.
1) maximum refinement multiple is limited when the data of N=1024 refine 1000 times, at least needs 1000 × 1024 × 2
The memory headroom of point stores intermediate data.Since it needs the memory headroom for storing intermediate data huge, to limit maximum
Refine multiple.
2) low-pass filter characteristic, which limits precision and the actual low-pass filter of maximum refinement multiple, intermediate zone,
When refinement multiple is bigger, the width of intermediate zone is bigger on filtering accuracy influence, and the analysis precision at selectable band analysis both ends can be made big
To reduce, and generate frequency aliasing phenomenon.When refinement multiple reaches certain value, entire selectable band analysis frequency band can all generate very big
Analytical error, this method actually can not achieve, even more so when especially with software realization.
3) calculation amount is larger in front（1）~（3）Low-pass filtering and double sampling are completed together in step flow, i.e., first
Snap shot is selected in determination, only to selecting snap shot to carry out low pass anti-aliasing filtering；But shift frequency processing must be carried out to all analysis sites, it is elected to and takes out
When multiple is very big, this step calculation amount is sizable.
4) frequency content adjusts the frequency that the more complex frequency content for obtaining FFT and spectrum analysis is adjusted to selected frequency band
The process of ingredient is more complicated, especially for frequency alias caused by the marginal error for avoiding low pass anti alias filter, such as into
400 spectral lines of display centre frequency band, this process are just more complicated when 1024 spectrum analysis of row.
Currently, having scholar for limitation existing for ZFFT, propose that a kind of multiple modulation based on parsing bandpass filtering is thin
Change spectral analysis method；To with sample frequencyf sThe former sequence of samplingx (i), whereini=0,1,2,…,DN+2M；N is fft analysis
Points, D are refinement multiple, M is half exponent number of filter；Three kinds of methods may be used and realize parsing bandpass filtering refinement spectrum point again
Analysis：1. use width forf sThe complex analytic band-pass filter of/D, every D click pumping a bit, shift frequency and the method for making N point spectrum analysis,
Obtain the frequency band indicated with N item independence spectral linesf 1~f 2；2. use width forf sThe complex analytic band-pass filter of/2D, clicks every D
Take out a bit, shift frequency and the method for making N point spectrum analysis, wherein only N/2 spectral line has value reflection frequency bandf 1~f 2Frequency spectrum, it is remaining
The value of N/2 spectral line is 0；3. use width forf sThe complex analytic band-pass filter of/2D, every 2D click pumping a bit, shift frequency and work
The method of N/2 point spectrum analysis obtains the frequency band indicated with N/2 item independence spectral linesf 1~f 2.Method 1.~3. in parsing band logical filter
Wave device selects FIR onrecurrent complex analytic band-pass filters, thinning process to be carried out by following 5 steps.
（1）It determines centre frequency and refines the centre frequency that multiple determines frequency range to be refined based on the actual application requirementsf o
And refinement multiple D.1. for method, in frequency bandf 1~f 2Frequence zooming analysis is carried out in range, the band center to be observed isf o
=0.5(f 1+f 2).2., 3. for method, in frequency bandf 1~f 2Carry out Frequence zooming analysis in range and frequency band need to be widened forf 1 /~f 2 /。
（2）It constructs complex analytic band-pass filter and designs a complex analytic band-pass filter, 1. for method, filter
Width isf s/ D, 2., 3. the width of filter is methodf s/2D。
（3）Choosing filters wave with designed complex analytic band-pass filter to needing to analyze sample signalx (n) make choosing suction filtration
Filtering can be taken out with choosing using complex analytic band-pass filter and be combined to improve computational efficiency by wave, and real signal is by multiple filter
Wave device just becomes complex analytic signal of the frequency within passband later.1., 2. for method, choosing is taken out than being D, selecting extraction N points,
It is N points to select snap shot number and FFT operations to count,g(m)=y(Dm)；3. for method, choosing is taken out than being 2D, selecting extraction N/2 points,
Sample frequency at this time and analysis frequency aref s/2D；If it is real signal, since sample frequency is not above analysis frequency
It 2 times, will necessarily frequency of occurrences aliasing；But because using complex analytic band-pass filter, negative frequency zero, after choosing is taken out at this time
It only will appear frequency to unroll, i.e., the frequency-portions that amplitude is zero exactly need the other half frequency content analyzed, and are not in
Aliasing avoids the aliasing of signal with the characteristic of complex analytic band-pass filter here.
（4）After multiple modulation shift frequency smokes to choosingg(m) multiple modulation shift frequency is carried out, by the initial frequency of refinementf 1Move on to zero-frequency
Point.Because when D or 2D points take out at 1, sample frequency reduces D or 2D times, 1., 2. has for methodg / (m)= g (m) e-j2π(f1/fs)Dm , 3. then for methodg / (m)= g (m) e-j4π(f1/fs)Dm 。
（5）1. FFT and spectral analysis method are pairg / (m) make point FFT and spectrum analysis, need not be adjusted into line frequency can
To obtain that there is the zoom FFT of N item independence spectral lines；2. method makees N point FFT and spectrum analysis, take positive frequency part, need not be into
Line frequency adjustment can be obtained by the zoom FFT with N/2 item independence spectral lines；3. method makees N/2 point FFT and spectrum analysis, take
Full rate part need not adjust into line frequency and can be obtained by the zoom FFT with N/2 item independence spectral lines.In three of the above
In method, method is 2. small with respect to calculation amount, speed is fast, needs the memory headroom for storing intermediate data small, due to multiple parsing
The amplitude of bandpass filter negative frequency part is zero, so can relax significantly to the required precision of filter, filter order is few,
Intermediate zone width, which can even prolong, rises one times, can still obtain very high filtering accuracy, complicated frequency need not be carried out by having
The advantages that adjustment；But relationship M=8D/ of the exponent number M and refinement multiple D of filter are taken out in choosinga,a=2/3~1 is outside filter
Expand coefficient；When D is smaller, calculating speed is also more satisfactory, and when D is larger, calculation amount increases considerably, it is difficult to meet most of prisons
The requirement of real-time of examining system.For this purpose, thering is scholar to propose to apply two-stage filter algorithm, even D=D1*D2Respectively by two stage filter
Device is realized, although some calculation amounts can be reduced, design comparison is complicated.
In Practical Project signal monitoring, in order to quickly understand the frequency spectrum fine structure of claimed range, signal is just necessarily required
Analysis should have high frequency resolution, have wider frequency range again, and meet system real time.Real-time is wanted
Higher monitoring system is sought, ZFFT methods at this stage are helpless；A kind of real-time ZFFT methods can be invented, at present state
Inside and outside scientific and technological circle never solve, and relevant achievement in research there is no to report.
Invention content
For insufficient existing for current ZFFT methods, the present invention is solves the problems, such as that it is a kind of based on the real-time of DSP that this is provided
ZFFT methods.
The present invention art scheme be：A kind of real-time ZFFT methods based on DSP, it will be based on the polyphony for parsing bandpass filtering
System refinement spectral analysis method operation be split as offline, online two parts, in conjunction with DSP operational capability online partial arithmetic
It decomposes in multiple data sampling periods and carries out to meet the requirement of real-time of Signal Analysis System；Divide following three step：
Step 1, by application requirement off-line calculation and filter coefficient, the frequency displacement coefficient of temporary ZFFT；
Step 2, the data sampling data choosing pumping for carrying out ZFFT online simultaneously, filtering, shift frequency operation；
Step 3 carries out the refinement spectral line that FFT can be obtained required bandwidth to step 2 result.
In the present invention, for analyzing bandwidthf 1~f 2Need refinement time series bex(n), take sample frequencyf s≥2f 2,
Position is taken out in choosingn=M, M+D ..., M+ (Q-1) D, D are refinement multiple, and M is filter order, and Q counts for FFT, filtering
Device coefficienth(k), frequency displacement coefficienty(r) real part, imaginary part be respectively：
In formula,k=0,1,2 ..., M-1,r=0,1,2 ..., Q-1；As long as choosing the work dominant frequency of DSPf z≥(4.5M+
12) f sFiltering, shift frequency on-line operation can be realized.
In the present invention, choosing pumping is carried out within the sampling period primary when DSP samples M data
Multiple parsing filtering, multiple modulation shift frequency operation obtain oneg (r) and the M data are removed D simultaneously, then re-sampling D
The full M data space of data supplement, is so repeated up to obtain Qg (r)。
In the present invention, the multiple parsing filtering operation is that M sampled data is tired out with what parsing filter factor again was multiplied
Adduction is average：
Real part, imaginary part。
In the present invention, the multiple modulation shift frequency operation is multiple parsing filtering operation result and multiple modulation frequency displacement coefficient
Product：
Real part, imaginary part。
In the present invention, step 2 sampling, choosing taken out, filtered, the Q point results of gained carry out FFT after shift frequency, you can obtained
Reflect bandwidthf 1~f 2The Q/2 items of range effectively refine spectral line.
The beneficial effects of the invention are as follows：The characteristics of being effectively utilized ZFFT operations and the strong advantage of DSP digital processing capabilities,
The data choosing pumping of ZFFT is combined with sampling, filtering, shift frequency operation are distributed to multiple sampling periods execution；Sampling terminates only to need
The FFT of one Q point, you can obtain the refinement spectral line of analysis bandwidth range；Compared with existing ZFFT methods：Committed memory space
Less, real-time is good, can meet on-line monitoring, real-time signal analysis processing system needs.
Description of the drawings
Fig. 1 is the operational flowchart of the present invention；
Fig. 2 is the embodiment of the present invention refinement spectrogram.
Specific implementation mode
The solution of the present invention is described in further detail with reference to the accompanying drawings and examples.
Referring to attached drawing 1, a kind of real-time ZFFT methods based on DSP, it refines the multiple modulation based on parsing bandpass filtering
The operation of spectral analysis method is split as offline, online two parts, and online partial arithmetic is decomposed in conjunction with the operational capability of DSP
It is carried out in multiple data sampling periods to meet the requirement of real-time of Signal Analysis System；Point or less three steps carry out.
Step 1, by application requirement off-line calculation and filter coefficient, the frequency displacement coefficient of temporary ZFFT.
The present invention specific practice be：It is required according to Practical Project signal monitoring, determines that signal analyzes bandwidthf 1~f 2, D is
Multiple is refined, sample frequency is takenf s≥2f 2；It is 2. designed according to the multiple modulation refinement spectral analysis method based on parsing bandpass filtering,
The relationship that the exponent number M and refinement multiple D of filter are taken out in choosing is M=8D/a,a=2/3~1 extends out coefficient for filter, according to Q/2
Spectral line shows analysis bandwidthf 1~f 2, then the points of FFT are Q；In this way, being to sampling time sequencex(n) choosing take out position
Forn=M, M+D ..., M+ (Q-1) D.The design philosophy of complex analytic band-pass filter is to be by bandwidthf 2Low-pass filter it is logical
Cross shift frequencyf 1And obtain, impulse Response Function ish(k), it need to usually improve the flatness and stop-band of passband by adding window
Moire effect, impulse Response Function is after adding Hamming windows
In formula,k=0,1,2 ..., M-1,h R(k) it is the real part for parsing bandpass filtering coefficient again,h I(k) it is to parse band logical again
The imaginary part of filter factor.The effect of multiple modulation shift frequency is the initial frequency that will be refinedf 1Move on to zero-frequency point, the π of frequency shift amount ω=2 Df 1/f s,
According to the frequency displacement property of DFT, can obtain multiple modulation frequency displacement coefficient is
In formula,r=0,1,2 ..., Q-1,y R(r) be multiple modulation frequency displacement coefficient real part,y I(r) it is multiple modulation frequency displacement coefficient
Imaginary part.From two formula above：Bandpass filtering coefficient is parsed againh(k), multiple modulation frequency displacement coefficienty(r) calculating only and choosing take out
The points Q of order M, FFT of filter are related, with sample sequencex(n) it is unrelated, can before data sampling off-line calculation and temporary
It deposits.
With the rapid development of computer technology, digital signal processor (Digital Signal Processor,
DSP) technology will be used wider and wider general, almost have application in entire electronics, information industry.Most widely used is beautiful
TI companies of state production DSP series of products, TMS320C672x series Floating-point DSPs be wherein relatively be suitble to signal analysis and processing and
Cost-effective premium quality product, the chips such as including TMS320C6722, TMS320C6726, TMS320C6727.The DSP core
It is the enhancing version C67x+CPU of the C67xCPU kernels used on C671x, compatible C67xCPU, but it is compact in speed, code
There is notable offer on property and each clock cycle floating-point performance；Under 300MHz clock frequencies, C67x+CPU each clock weeks
Phase can execute 8 instructions parallel（Wherein 6 are floating point instruction）, peak performance reaches 2400MIPS/1800MFLOPS；C67x+
CPU supports 32 fixed points, 32 single-precision floating points, 64 double-precision floating point operations；The each fetchings of CPU be 256 it is advanced very
Long instruction words（VLIW）Instruction packet, the instruction packet are made of elongated execution packet, and it can be 8 work(to execute packet in each clock cycle
Energy unit provides 1 ~ 8 32 bit instruction, and elongated execution packet is the key that save memory, is the significant improvement to C67xCPU；Separately
Outside, CPU includes 2 data channel, and each data channel includes 4 functional units（.D,.M,.S,.L）With a register group,
Each register group includes 32 32 bit registers, shares 64 general registers, quantity is 1 times of C67xCPU, is mitigated significantly
The pressure of register, 4 functional units in each channel can freely use 32 registers in the channel, and each logical
Road is connected to another channel there are one crossedpath, this allows each clock cycle to read an operation from another channel
Number, C67x+CPU allows 2 functional units to use the crossedpath, and C67xCPU only allows 1 functional unit to use the intersection
Path；C67x+CPU can execute all instructions of C67xCPU, in addition to this new floating point instruction is added to again, to improve number
Performance when word signal processing；C67x+CPU also added 2 registers for being responsible for dMAX unit affairs specially, i.e. dMAX events
Trigger register and dMAX event state registers make CPU and dMAX exchange data and just do not have to visit again any memory.It is based on
The above feature of TMS320C672x series DSPs, the preferred TMS320C672x series DSPs of main process task CPU of the embodiment of the present invention, ginseng
See that the present invention is based on TMS320C672x series DSP embodiment assembly language program(me) codes in step 2, from program code：
One is obtained in line computation 1 multiple parsing filtering, multiple modulation shift frequencyg (r) and M data removal D is needed into 0.5* simultaneously
M*9+12 DSP clock cycle；Therefore, in practical applications, as long as choosing the work dominant frequency of DSPf z≥(4.5M+12) f sI.e.
Filtering, shift frequency on-line operation can be achieved.And the work dominant frequency of TMS320C672x series DSPs reaches as high as 300MHz, can meet big
The application requirement of majority signal analysis process system.
By taking the vibration monitoring of certain rotating machinery as an example, the intrinsic frequency of vibration monitoring sensor is 16KHz, and monitoring point
Fault characteristic frequency is not more than 30Hz, and fault characteristic frequency information is typically to be modulated at the transmission of sensor intrinsic frequency signal；Root
According to the frequency spectrum characteristic of modulated signal, vibration monitoring frequency spectrum of concern is only between 16KHz-30Hz to 16KHz+30Hz.Then believe
Number analysis bandwidthf 1=15970Hz,f 2=16030Hz, according to nyquist（Nyquist）Sampling thheorem takes sample frequencyf s≥
2f 2=2.56*16030Hz=41036.8Hz shows analysis bandwidth according to 256 spectral linesf 1~f 2, then needed for frequency spectrum
Resolution axf=0.234375Hz, FFT points Q=512.By sample frequencyf s=41036.8Hz samples 512 point datas through FFT
Obtain the spectral resolution Δ of 256 effective spectral linesf /=80.15Hz, it is clear that can not be observed under spectral resolution interested
Fault characteristic information.For it is observed that the fine structure of bandwidth 15970Hz~16030Hz frequency spectrums is analyzed, usually with the side ZFFT
Method carries out frequency spectrum refinement, refinement multiple D=Δf //Δf=80.15Hz/0.234375Hz≈342.According to based on parsing band logical
2. the multiple modulation refinement spectral analysis method of filtering designs, filter is taken to extend out coefficienta=1, choosing take out the exponent number M of filter with it is thin
The relationship for changing multiple D is M=8D/aThe work dominant frequency of DSP is chosen in=8*342=2736f z≥(4.5M+12) f s=(4.5*2736+
12) filtering of refinement spectrum, shift frequency on-line operation can be realized in the Hz of * 41036.8Hz=112769126.4；According to
The characteristics of TMS320C672x series DSPs, makes work dominant frequency by configuring its internal pll parameterf z>=120MHz can be completed
Refinement spectrum analysis required by the present embodiment.
Step 2, the data sampling data choosing pumping for carrying out ZFFT online simultaneously, filtering, shift frequency operation.
In the present invention, it is synchronous progress that data choosing, which is smoked with data sampling, i.e., when DSP samples M data, at this
Primary multiple parsing filtering is carried out in sampling period, multiple modulation shift frequency operation obtains oneg (r) and simultaneously remove the M data
D, then D full M data space of data supplement of re-sampling, is so repeated up to obtain Qg (r).In conjunction with attached drawing 1
With based on TMS320C672x series DSP embodiment assembly language program(me) codes, specific practice of the invention is：（1）Given step
Calculated filter coefficient in oneh (k), frequency displacement coefficienty (r) first address, arrange M sampled data buffer unit, Q a
As a result its counting pointer of storage unit juxtapositionj=0、r=0；（2）Circulating sampling data are stored to M sampled data buffer unit,
Often storing a sampled data, it counts pointerjAdd 1, untiljCycle juxtaposition is exited when=M, and it counts pointerk=0、j=0；（3）
Read sampled datax (k) and filter factorh (k), it often reads primary its and counts pointerkAdd 1, ifk>=D, then by datax
(k) be saved injSpecified sampled data buffer unitx (j), often preserving a data, it counts pointerjAdd 1；（4）Calculate M
Sampled data is multiplied cumulative and average with parsing filter factor again：
Real part, imaginary part, sentence after primary per product accumulation
It is disconnected to count pointerkIfk<M returns to the（3）Step, which recycles, reads data, otherwise exits cycle and ask cumulative and mean deviation reading frequency displacement system
Numbery (r)；（5）Choosing is taken out, filtered result progress multiple modulation shift frequency, it will parsing filtering operation result and multiple modulation frequency displacement again
Multiplication：Real part, imaginary part, preserve result of productg (r), often protect
Deposit oneg (r) it counts pointerrAdd 1, ifr<Q returns to the（2）Step D data of circulating sampling add to full M data cell,
Otherwise it exits cycle and FFT is done to multiple modulation frequency displacement result.For the overall machine vibration monitoring embodiment in step 1, use
TMS320C672x series DSPs pass through phase-locked-loop configuration CPU work dominant frequency as master cpuf z=120MHz, in conjunction with C67x+
CPU can write that the present invention is based on TMS320C672x series DSP embodiment assembler languages in terms of Digital Signal Processing the advantages of
Program code is as follows：Wherein, data first address B13, filter factor first address A13, frequency displacement coefficient first address A14, result data
First address B4
Step 3 carries out the refinement spectral line that FFT can be obtained required bandwidth to step 2 result.
Step 2 sampling, choosing are taken out, filtered, the Q point results of gained carry out FFT after shift frequency, you can obtains reflection bandwidthf 1
~f 2The Q/2 items of range effectively refine spectral line.For the overall machine vibration monitoring embodiment in step 1, Fig. 2 is its refinement
Spectrogram, the first half is time-domain signal in figure, lower half is zoom FFT, and it is that monitoring passes that intermediate highest spectral line, which is known as dominant frequency spectral line,
The intrinsic vibration of sensor, it is monitoring point rotation to be distributed in dominant frequency spectral line both sides and be known as side frequency spectral line about the symmetrical spectral line of dominant frequency
The fault characteristic information of component of machine；Signal analyzes bandwidthf 1=15970Hz,f 2=16030Hz, sample frequencyf s=
41036.8Hz, refinement multiple D=342, analytic band-pass filter exponent number M=2736, FFT points Q=512, sampling required for refining
Data amount check be N=DQ=342 × 512=175104, the spectral resolution Δ after refinementf=0.234375Hz.No matter using
Which kind of frequency spectrum analysis methodRealize spectral resolution Δf=0.234375Hz, then its sampling time T=1/ Δsf≈4.267
Second, sampled data number N within the sampling time=f s/Δf≈ 175090, therefore, the present invention in terms of data collection capacity with
Other methods are consistent, but other methods at least need DQ=175104 data buffer storage unit, and the method for the present invention only needs
8D=2736 data buffer storage unit is wanted, is only the 1/64 of other methods.The method of the present invention or other ZFFT methods can use
Offline mode calculates filter coefficienth (k), frequency displacement coefficienty (r), other ZFFT methods be completed in data sampling it is laggard
Row filtering, the complex multiplication of frequency displacement, complex addition operations, and the method for the present invention is that filtering, the complex multiplication of frequency displacement, plural number are added
Method operation is split in multiple periods of data sampling and carries out, and data sampling terminates a FFT for a Q point is only needed to can be obtained required
Zoom FFT；For FFT calculating, the libraries FFT that TI companies provide, the storage sky that Q31,32-bit Real FFT, 0 wait for
Between the speed of operation be 128 points to fast：Need 6509 clock cycle, 256 points：Need 14756 clock cycle, 512
Point：Need 33081 clock cycle, 1024 points：Need 73422 clock cycle；For the rotating machinery vibrating in step 1
Embodiment is monitored, only 33081 clock cycle is needed to can be obtained zoom FFT after sampling terminates using the method for the present invention, and
Q (4.5M+12)+33081=6309888+33081=6342969 clock cycle is at least needed using other ZFFT methods, this
Inventive method required operation time after sampling terminates only is the 1/192 of other ZFFT methods.
The beneficial effects of the invention are as follows：The characteristics of being effectively utilized ZFFT operations and the strong advantage of DSP digital processing capabilities,
The data choosing pumping of ZFFT is combined with sampling, filtering, shift frequency operation are distributed to multiple sampling periods execution；Sampling terminates only to need
The FFT of one Q point, you can obtain the refinement spectral line of analysis bandwidth range；Compared with existing ZFFT methods：Committed memory space
Less, real-time is good, can meet on-line monitoring, real-time signal analysis processing system needs.
Claims (3)
1. a kind of real-time ZFFT methods based on DSP, it is characterised in that：The operation of ZFFT is split as offline, online two parts,
In conjunction with DSP operational capability online partial arithmetic decompose in multiple data sampling periods carry out with meet signal analysis system
The requirement of real-time of system；Point or less three steps：
Step 1, by application requirement off-line calculation and filter coefficient, the frequency displacement coefficient of temporary ZFFT；
For analyzing bandwidthf 1~f 2Need refinement time series bex(n), take sample frequencyf s≥2f 2, choosing pumping position isn=M,
M+D ..., M+ (Q-1) D, D are refinement multiple, and M is filter order, and Q counts for FFT, designs its complex analytic band-pass filter
Coefficienth(k), multiple modulation frequency displacement coefficienty(r) real part, imaginary part be respectively：
In formula,k=0,1,2 ..., M-1,r=0,1,2 ..., Q-1, as long as choosing work dominant frequencyf z≥(4.5M+12) f s It is floating
Filtering, shift frequency on-line operation can be realized in point DSP；
Step 2, the data sampling data choosing pumping for carrying out ZFFT online simultaneously, filtering, shift frequency operation；
When DSP samples M data, primary multiple parsing filtering is carried out within the sampling period, multiple modulation shift frequency operation obtains
To oneg (r) and the M data are removed D simultaneously, then D data of re-sampling supplement full M data space, so weight
It is multiple to carry out until obtaining Qg (r)；
Step 3 carries out the refinement spectral line that FFT can be obtained required bandwidth to step 2 result；
Step 2 sampling, choosing are taken out, filtered, the Q point results of gained carry out FFT after shift frequency, you can obtains reflection bandwidthf 1~f 2Model
The Q/2 items enclosed effectively refine spectral line.
2. the real-time ZFFT methods according to claim 1 based on DSP, it is characterised in that：The multiple parsing filtering fortune
It is that M sampled data adding up and being averaged with what parsing filter factor again was multiplied：
Real part, imaginary part。
3. the real-time ZFFT methods according to claim 1 based on DSP, it is characterised in that：The multiple modulation shift frequency fortune
Parse the product of filtering operation result and multiple modulation frequency displacement coefficient again at last：
Real part, imaginary part。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610217659.1A CN105911351B (en) | 2016-04-11 | 2016-04-11 | Real-time ZFFT methods based on DSP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610217659.1A CN105911351B (en) | 2016-04-11 | 2016-04-11 | Real-time ZFFT methods based on DSP |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105911351A CN105911351A (en) | 2016-08-31 |
| CN105911351B true CN105911351B (en) | 2018-10-23 |
Family
ID=56744842
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610217659.1A Active CN105911351B (en) | 2016-04-11 | 2016-04-11 | Real-time ZFFT methods based on DSP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105911351B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106199631B (en) * | 2016-09-28 | 2020-04-10 | 成都凯天电子股份有限公司 | Coherent wind finding radar wind speed measuring method |
| CN106383512B (en) * | 2016-10-10 | 2019-02-01 | 中国科学技术大学 | A kind of electromagnetic protection reinforcement means of DSP control system |
| CN110346594B (en) * | 2019-05-29 | 2022-04-08 | 浙江工业大学 | Portable totally-enclosed compressor high-precision rotation speed measuring method and device |
| CN111257815B (en) * | 2020-03-06 | 2022-04-05 | 云南电网有限责任公司电力科学研究院 | High-precision frequency spectrum correction method |
| CN111427018A (en) * | 2020-04-22 | 2020-07-17 | 安徽华可智能科技有限公司 | Radar interference equipment interference effect evaluation method |
| CN111474524A (en) * | 2020-04-22 | 2020-07-31 | 安徽华可智能科技有限公司 | Radar interference equipment interference effect monitoring and decision support system |
| CN112213560A (en) * | 2020-09-01 | 2021-01-12 | 东南大学 | High-precision power grid broadband signal measuring method based on Z-ADALINE |
| CN113433521A (en) * | 2021-05-29 | 2021-09-24 | 湖南科技大学 | Frequency spectrum refining method for synchronous detection of lightning field target |
| CN113419288B (en) * | 2021-05-29 | 2023-06-06 | 湖南科技大学 | Data detection and pretreatment method for inversion of underground shelter |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3664671B2 (en) * | 2001-08-02 | 2005-06-29 | 株式会社日立製作所 | Millimeter wave radar equipment |
| CN101605001A (en) * | 2009-07-13 | 2009-12-16 | 中国船舶重工集团公司第七一五研究所 | A kind of Doppler measurement of MFSK underwater sound communication and modification method |
-
2016
- 2016-04-11 CN CN201610217659.1A patent/CN105911351B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3664671B2 (en) * | 2001-08-02 | 2005-06-29 | 株式会社日立製作所 | Millimeter wave radar equipment |
| CN101605001A (en) * | 2009-07-13 | 2009-12-16 | 中国船舶重工集团公司第七一五研究所 | A kind of Doppler measurement of MFSK underwater sound communication and modification method |
Non-Patent Citations (2)
| Title |
|---|
| 基于DSP的移频参数检测***设计;薛凯凯;《CNKI优秀硕士学位论文全文库 工程科技II辑》;20081231;第23-26页及图5-2 * |
| 基于密集频谱校正的电流信号高准确度分离技术;雷小亚 等;《中国测试》;20151231;第41卷(第12期);第112页第1.1节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105911351A (en) | 2016-08-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105911351B (en) | Real-time ZFFT methods based on DSP | |
| Olhede et al. | A generalized demodulation approach to time-frequency projections for multicomponent signals | |
| van Straten et al. | DSPSR: digital signal processing software for pulsar astronomy | |
| Cheng et al. | An order tracking technique for the gear fault diagnosis using local mean decomposition method | |
| CN108020282B (en) | Coriolis mass flowmeter signal processing method based on complex coefficient filtering | |
| CN105117196B (en) | A kind of parallel organization Sinc interpolation methods based on FPGA | |
| CN105865616B (en) | Modulation spectrum rapid refinement method based on FFT | |
| CN104006962A (en) | Gear fault feature extraction method and system | |
| CN104136896B (en) | For the system for the vibration signal for obtaining rotation motor | |
| CN109271133A (en) | A kind of data processing method and system | |
| CN109194307A (en) | Data processing method and system | |
| CN107407703A (en) | The spectrum analyzer of real-time bandwidth with raising | |
| CN103155476B (en) | Fixed frequency analog digital conversion is used to quantify the input sampled by interpolation | |
| CN107121581B (en) | A kind of data processing method of data collection system | |
| CN103969508A (en) | Real-time high-precision power harmonic analysis method and device | |
| JPH0363875A (en) | Computing system of discrete fourier transform using cyclic technique | |
| He et al. | Downsampling-based synchrosqueezing transform and its applications on large-scale vibration data | |
| Javorskyj et al. | Detection of distributed and localized faults in rotating machines using periodically non-stationary covariance analysis of vibrations | |
| CN109991458A (en) | A kind of waveform longitudinal direction average system based on FPGA | |
| CN106643944A (en) | Method for tracking and estimating high-precision frequency of coriolis mass flowmeter | |
| US20150309097A1 (en) | Systems and methods for joint angle-frequency determination | |
| Zhu et al. | Generalized ridge reconstruction approaches toward more accurate signal estimate | |
| CN106649205B (en) | The resampling device of statistical information | |
| CN108776259A (en) | Power analyzer based on digital bandwidth restriction technologies | |
| CN110017957A (en) | A kind of synchronized analyzing method, apparatus and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |