Does anyone know any good method to compute the acoustic spectrum accurately? I am doing an aeroacoustic simulation and need to convert a pressure variation signal into acoustic spectrum and compare it with experimental result from others.

I hear that we can do this by doing a Fourier transform or analog filtering. So at first, I tried the FFT in Matlab. It worked well predicting the low frequency but under-estimate the dominant peaks. I also tried some PSD functions and converted them back into dB-Hz spectrum. They generally over-estimated the magnitude. Finally I tried the FFT post-process program in Fluent. It can accurately capture one peak in frequency and SPL, but leaving other range over-estimated. So does any one know any accurate method or software to compute this?

What I have now are about 35000 signal points from steady oscillation stage. I think they are enough for doing a time series analysis.

Thanks,

Bearcat

Check the "numerical recipes" where you can find the theory and the codes you are looking for.

Regarding the use of FFT you have to be sure that the number of signals is a power of 2; if it is not you have to use a DFT.

Regarding the number of samples: 1) you have to process an integer number of periods 2) the minimum spacing in time determines the maximum frequency you can capture.

So it is not a matter of how many samples you have but how many samples per period you have!!

All these things are explained in much more mathematical details in the Numerical Recipes.

Hi

Thank you for your reply.

Some people say DFT is only exact for the signal of infinite extend, and FFT is a way to implement a DFT for a period expressed by 2^n. So I think my FFT computing is only kind of approximate estimation.

What I concern is not the high frequency, but the prediction of the low frequency. My sampling frequency is 1000000, and the dominant frequency should be about 500Hz.

bearcat

hi I often use fft, but not for acoustic application, but this is the general observation i have on sampling time,

-low frequency dominant case, your data collection/sampling time should be big, and a big-time step then will be needed, otherewise u will end with a huge Gbyte of files with insignificant variations amongest sucessive data. I advise u to go for a bigger time step unless divergence of the itteration occurs.

-for high frequency u should sample u'r data at very small time steps, significant variation occurs in the signal at small time interval, even I may suggest u collect the data at every simulation time step advance, u will even get smooth data, and a long time data collection wont be needed.

but always set time step bigger, enogh to get temporal resolution level u are after, to at least get some reasonable result first, at least till u get a good idea of the flow property

regards, taw