I am also always suspicious of measures i see. I have seen measures of the same driver with quite different measured responses.Which one do you believe?
dave
I think what happens is a lot of people take measurements in their room without doing a good job of getting away from reflections. They frankly post crap measurements (heck, I'm just as guilty, I've learned some things). This leads to confusion. As someone else mentioned, the measurement to believe is the one that is pseudo-anechoic and the tester knows what that is and what the limits of his/her test are.
And for me, 2nd & 3rd are not enough and CSD has almost no importance. 😉 Nor am I alone in the latter, (Zaph for example doesn't regard it as being of any real value either).
I haven't looked at his site in some time, but I'm not sure that's what Zaph thinks. I believe he doesn't post CSD for other reasons. Although he does post it sometimes.
Ultimately, the time-resolved transient response of a driver *completely* describes its acoustically measurable (via microphone) response. The IR is the inverse FFT of the frequency response and is closer to the truth of a speaker's behavior. We measure FR because it is easier to do than measuring IR with a transient digitizer. The IR tells us if the speaker rings or has sibilance, or is able to provide a sharp and clean transient from percussion instruments (plucked or struck strings, drums, high hats, rim shots, cymbals, etc). FR without the IFFT to IR won't tell you if it rings (although a sharp big peak around cone breakup is a dead giveaway).
I believe you have this backwards. The software measures the IR and then computes the FR based on the IR. Most everything comes from the IR. CSD, HD, FR all come from the IR. I also am not sure how you can predict if a speaker has sibilance or has sharp and clean transient from looking at the IR. Unless you mean, the IR provides us with all those useful graphs to help predict those qualities.
When I think of sharp transients I often think of clean upper midrange that has low CSD. I am in the 'CSD matters' camp. Though I'd suggest it's pretty low on the list. There's much more important things to consider.
I believe you have this backwards. The software measures the IR and then computes the FR based on the IR. Most everything comes from the IR. CSD, HD, FR all come from the IR. I also am not sure how you can predict if a speaker has sibilance or has sharp and clean transient from looking at the IR. Unless you mean, the IR provides us with all those useful graphs to help predict those qualities.
It depends on the software and test equipment. The one that measures impulse many times and averages them requires a higher quality mic and transient recorder. Usually these are pro measurement systems. The consumer stuff like REW and Holmimpulse use a frequency sweep - you can hear it. This doesn't require a high speed mic or transient recorder.
It depends on the software and test equipment. The one that measures impulse many times and averages them requires a higher quality mic and transient recorder. Usually these are pro measurement systems. The consumer stuff like REW and Holmimpulse use a frequency sweep - you can hear it. This doesn't require a high speed mic or transient recorder.
Page 17 of the HolmImpulse user guide: "Time window can be dynamically adjusted using the pointing device (The frequency response is instantly updated)" Why does the FR get updated if that's what is measured. The IR doesn't ever change once it's measured.
How does REW compute CSD from a FR? It can't. This is taken from the IR.
Also see Loudspeaker Measurement by Joseph D'Appolito if you own a copy.
I'm sure there are other software manuals that would agree. Sorry the Holm manual doesn't explicitly address our disagreement and I couldn't find a manual for REW. I only skimmed the Holm manual, maybe have a closer read of it and see if it can clarify some things for you.
EDIT - just because you hear a sweep when you take the measurement doesn't mean it's not measuring the IR. I think you're getting confused with something like an RTA. An RTA does measure the FR, not the IR as far as I know.
Last edited:
REW actually records the wave file of the frequency sweep and performs a sliding window algorithm on the time series data of the frequency sweep to get the CSD. Clever math but there is no direct native IR recorded in REW. Maybe it's different with Holm but I recall when I used it it sounded like a frequency sweep. The packages that do it via stochastic sampling of true impulses don't sound like a sweep. Just a bunch of percussive sounding clicks/clapping sounds.
See Equation 8 in paper below - shows how to use windowing function to get CSD from time-energy spectrum.
http://webistem.com/acoustics2008/acoustics2008/cd1/data/fa2002-sevilla/forumacusticum/archivos/phagen039.pdf
I think B&K's Dirac software uses the impulse method.
http://www.bksv.com/Products/analysis-software/acoustics/building-acoustics/room-acoustic-measurement-dirac
See Equation 8 in paper below - shows how to use windowing function to get CSD from time-energy spectrum.
http://webistem.com/acoustics2008/acoustics2008/cd1/data/fa2002-sevilla/forumacusticum/archivos/phagen039.pdf
I think B&K's Dirac software uses the impulse method.
http://www.bksv.com/Products/analysis-software/acoustics/building-acoustics/room-acoustic-measurement-dirac
Last edited:
I'm not going to get into this with you because historically you're tenacioucly stuborn. I'll just leave you with this quote from equation 8 of the link you provided:
"[CSD] is computed from the system’s impulse response"
This isn't news to me 🙂
"[CSD] is computed from the system’s impulse response"
This isn't news to me 🙂
TrueRTA does a sine sweep - I've sometimes run that at fairly high levels (~20vrms) outdoors and near xmax to swamp noise
I'm not going to get into this with you because historically you're tenacioucly stuborn. I'll just leave you with this quote from equation 8 of the link you provided:
"[CSD] is computed from the system’s impulse response"
This isn't news to me 🙂
Yes, and IR is calculated at different sliding window intervals from the inverse FFT of the frequency sweep data. I won't argue with you either. Just know that when you hear a freq sweep during a measurement it is not using the impulse as the raw data.
Why I'm wasting my time on this, I'm not sure, but I found some REW info.
Impulse Responses
From this link (be sure to differentiate what it's saying about a gun shot click and what the software is actually doing):
"you can work out a system's frequency response by determining the frequency components that make up its impulse response. REW does this by Fourier Transforming the impulse reponse, which in essence breaks it up into its individual frequency components. The plot of the magnitude of each of those frequency components is the system's frequency response."
"When an impulse response is measured by means of a logarithmically swept sine wave"
"After capturing a sweep, FFT processing is carried out to derive the system's impulse response and the corresponding frequency response. There are controls to adjust the position and widths of left and right windows that define the portion used to derive the frequency response"
Impulse Responses
From this link (be sure to differentiate what it's saying about a gun shot click and what the software is actually doing):
"you can work out a system's frequency response by determining the frequency components that make up its impulse response. REW does this by Fourier Transforming the impulse reponse, which in essence breaks it up into its individual frequency components. The plot of the magnitude of each of those frequency components is the system's frequency response."
"When an impulse response is measured by means of a logarithmically swept sine wave"
"After capturing a sweep, FFT processing is carried out to derive the system's impulse response and the corresponding frequency response. There are controls to adjust the position and widths of left and right windows that define the portion used to derive the frequency response"
Just know that when you hear a freq sweep during a measurement it is not using the impulse as the raw data.
By what do you base this on?
Tux,
I admit you are partially right. But REW does not use an impulse as the raw data being measured. It appears that an FFT is used to derive both the IR and the FR from a logarithmically swept sine wave ( data is still a time series of a frequency response).
This was taken from REW website that you linked.
I admit you are partially right. But REW does not use an impulse as the raw data being measured. It appears that an FFT is used to derive both the IR and the FR from a logarithmically swept sine wave ( data is still a time series of a frequency response).
This was taken from REW website that you linked.
When an impulse response is measured by means of a logarithmically swept sine wave, the room's linear response is conveniently separated from its non-linear response.
Impulse Response Windows
After capturing a sweep, FFT processing is carried out to derive the system's impulse response and the corresponding frequency response. There are controls to adjust the position and widths of left and right windows that define the portion used to derive the frequency response, these controls may be accessed by pressing the "IR Windows" button on the toolbar
Last edited:
Cool. Now that we're on common ground let's both try and grow our understanding. This part: "the room's linear response is convenietly separated from its non-linear response". I think what the writer is trying to say here is that we can discern both the direct and indirect (direct and reflected) sound from looking at the IR. Would you agree with that? I'm just not totally sure what is meant by that part of the sentance.
This: "FFT processing is carried out to derive the system's impulse response and corresponding frequency response". This is also a confusing sentance to me, but I think the writer is saying that once the measurement is taken, the software can derive and display the IR and when it has the IR it can compute the FR.
I'd also like to tease out this topic by asking this question. If the software merely measures the SPL at various frequencies (kinda like an RTA), how will it know which are direct and which are indirect sounds? Consider the first quote while trying to answer this question. Is it possible to measure the FR using a tone generator and SPL meter (old way) and then compute the IR with FFT and then accurately remove the reflection and reverse the FFT to produce a reflection free FR? I would say you cannot do this.
Lastly, if I am partially right, and REW doesn't measure the IR or the FR (you suggest REW derives both IR and FR from the swept sine), then WHAT is it actually measuring??? I think the only answer is that it must be measuring the IR.
This: "FFT processing is carried out to derive the system's impulse response and corresponding frequency response". This is also a confusing sentance to me, but I think the writer is saying that once the measurement is taken, the software can derive and display the IR and when it has the IR it can compute the FR.
I'd also like to tease out this topic by asking this question. If the software merely measures the SPL at various frequencies (kinda like an RTA), how will it know which are direct and which are indirect sounds? Consider the first quote while trying to answer this question. Is it possible to measure the FR using a tone generator and SPL meter (old way) and then compute the IR with FFT and then accurately remove the reflection and reverse the FFT to produce a reflection free FR? I would say you cannot do this.
Lastly, if I am partially right, and REW doesn't measure the IR or the FR (you suggest REW derives both IR and FR from the swept sine), then WHAT is it actually measuring??? I think the only answer is that it must be measuring the IR.
Is it possible to measure the FR using a tone generator and SPL meter (old way) and then compute the IR with FFT and then accurately remove the reflection and reverse the FFT to produce a reflection free FR?
TDS (Time Delay Spectrometry) maybe.Time Delay Spectrometry (TDS)
It uses a swept source and then a windowed input that sweeps along with it so that reflections ar enot measured. They can generate a waterfall, but on a quick look i see no direct reference to the implulse.
dave
IF the system is Linear Time Invariant(LTI) it's Impulse Response and it's Frequency Complex Amplitude Response are intimately linked by the Fourier Transform. This is the basis of nearly ALL good methods of measuring speakers today .. of which Prof. Angelo Farina's method is the best. It uses a log frequency sweep (sounding like a B&K 2307 & 2010) and is capable of deriving a good Impulse Response .This: "FFT processing is carried out to derive the system's impulse response and corresponding frequency response". This is also a confusing sentance to me, but I think the writer is saying that once the measurement is taken, the software can derive and display the IR and when it has the IR it can compute the FR.
I'd also like to tease out this topic by asking this question. If the software merely measures the SPL at various frequencies (kinda like an RTA), how will it know which are direct and which are indirect sounds? Consider the first quote while trying to answer this question. Is it possible to measure the FR using a tone generator and SPL meter (old way) and then compute the IR with FFT and then accurately remove the reflection and reverse the FFT to produce a reflection free FR? I would say you cannot do this.
Speakers are certainly NOT LTI but its convenient for us to assume they are. You just have to watch out that you don't get caught out. eg MLS methods which were once popular are less robust than Farina's log sweep to non-linearity (THD)
Removing the reflection from the Impulse Response is the basis of all modern methods which don't require an anechoic. The frequency limit down to which you can do this is determined by the arrival of the first reflection and is about 200Hz in most domestic rooms.
You can extend this limit further if you adopt Benjamin's method but only by a bit. Extending Quasi-Anechoic Measurements to Low Frequencies which is also an excellent explanation of such quasi anechoic methods.
CDS were first proposed and used by KEF The Application of Digital Techniques to the Measurement of Loudspeakers and their 1977 paper is still the best explanation
Last edited:
REW uses swept sine technique for capturing IR. The response recording is convolved with the swept sine's inverse.
Transfer Function Measurement with Sweeps
Transfer Function Measurement with Sweeps
Holme Impulse allows saving measurement sweep and the inverse sweep. Convolution of the two produces IR of measurement system, a Dirac pulse.
You can use any broadband signal including music but different signals have different pros & cons.
Farina's log sweep and the way he processes it is the theoretical 'best'.
It's likely both REW & Holme Impulse use a crude version. The latest AP attributes their method to Angelo.
A by-product of Angelo's method is it gives you the distortion products too.
Farina's log sweep and the way he processes it is the theoretical 'best'.
It's likely both REW & Holme Impulse use a crude version. The latest AP attributes their method to Angelo.
A by-product of Angelo's method is it gives you the distortion products too.
Farina generates inverse in time domain. Convolution of sweep with its inverse yields a gigantic sync function. Farina's system isn't based on 2^n samples.
Cleanest IR is obtained with single large sweep.
With sweep construction, HD components end up before t=0 in proportion to the harmonic order and number of samples/octave of the sweep in the convolution result.
Sweeps effectively have a single sharp frequency at any moment along the sweep.
With MLS or other broadband signals harmonic distortion and IM distortion is distributed throughout the recovered linear IR and appears as raised noise floor.
Cleanest IR is obtained with single large sweep.
With sweep construction, HD components end up before t=0 in proportion to the harmonic order and number of samples/octave of the sweep in the convolution result.
Sweeps effectively have a single sharp frequency at any moment along the sweep.
With MLS or other broadband signals harmonic distortion and IM distortion is distributed throughout the recovered linear IR and appears as raised noise floor.
It sounds like REW uses Farina's method as their website says they do a logarithmic sine sweep and HD components are provided automatically for t<0. So the real component of the complex FFT is the frequency domain and the imaginary component is the impulse response in the time domain?
REW uses exponential swept sine with constraint that it is 2^n samples in length. This is so exact solution is generated in frequency domain by FFT, which is 2^N process. In frequency domain complementary spectrum is calculated such that multiplication of the two yields spectrum with equal values in all the frequency bins.
Multiplication in frequency domain is convolution in time domain.
Complementary spectrum is brought into time domain with inverse FFT. Once in time domain direct convolution with exponential sweep returns IR.
I've built Farina inverse sweeps in Cool Edit per his paper(s) on the topic. Exponential sweep has spectrum decreasing 3dB/octave. Sweep is reversed in time and an amplitude envelope is applied that results in spectrum of sweep rising 3dB/octave.
Notation and Format of the Real DFT
Rest assured, REW uses exponential sweep, but it and its inverse are not generated by Farina's method.
Read the Swen Muller paper linked above and study systems of 2^N.
Multiplication in frequency domain is convolution in time domain.
Complementary spectrum is brought into time domain with inverse FFT. Once in time domain direct convolution with exponential sweep returns IR.
I've built Farina inverse sweeps in Cool Edit per his paper(s) on the topic. Exponential sweep has spectrum decreasing 3dB/octave. Sweep is reversed in time and an amplitude envelope is applied that results in spectrum of sweep rising 3dB/octave.
Notation and Format of the Real DFT
The names real part and imaginary part originate from the complex DFT, where they are used to distinguish between real and imaginary numbers. Nothing so complicated is required for the real DFT. Until you get to Chapter 29, simply think that "real part" means the cosine wave amplitudes, while "imaginary part" means the sine wave amplitudes. Don't let these suggestive names mislead you; everything here uses ordinary numbers.
Rest assured, REW uses exponential sweep, but it and its inverse are not generated by Farina's method.
Read the Swen Muller paper linked above and study systems of 2^N.
REW uses exponential swept sine with constraint that it is 2^n samples in length. This is so exact solution is generated in frequency domain by FFT, which is 2^N process. In frequency domain complementary spectrum is calculated such that multiplication of the two yields spectrum with equal values in all the frequency bins.
Multiplication in frequency domain is convolution in time domain.
Complementary spectrum is brought into time domain with inverse FFT. Once in time domain direct convolution with exponential sweep returns IR.
I've built Farina inverse sweeps in Cool Edit per his paper(s) on the topic. Exponential sweep has spectrum decreasing 3dB/octave. Sweep is reversed in time and an amplitude envelope is applied that results in spectrum of sweep rising 3dB/octave.
Notation and Format of the Real DFT
Rest assured, REW uses exponential sweep, but it and its inverse are not generated by Farina's method.
Read the Swen Muller paper linked above and study systems of 2^N.
REW's website says they use logarithmic (inverse of exponential) sweep. See third paragraph in link.
Impulse Responses
- Status
- Not open for further replies.
- Home
- Loudspeakers
- Full Range
- Limitations (or not) of current speaker measurement technology