Hello,
my problem is not with a specific software, but more with all softwares that allow gating of impulse response.
Let me try to explain it clearly:
-usual wisdom is to use a swept sine
-the minimum swept time is usually high on most if not all soft
-the time of the first reflection limit the resolution in the low part of the spectrum
My current case (nothing fancy here):
my sweepis 16k samples at 44.1k so it's 371ms
i'm measuring a flat panel with a window time of +/-8ms
so when i apply the gate on the signal, i'm effectively removing the high part of the sine sweep, the one that i need to get a meaningful result in the high part of the spectrum.
That's why i said it's a theorical problem, you can't measure frequencies that are not yet excited; at 8ms in my sweep i'm still in the low frequency part of the sweep.
Only way i see to resolve that is to have an excitation signal which size is inferior to the window time, otherwise that make no sense.
Conundrum ?
my problem is not with a specific software, but more with all softwares that allow gating of impulse response.
Let me try to explain it clearly:
-usual wisdom is to use a swept sine
-the minimum swept time is usually high on most if not all soft
-the time of the first reflection limit the resolution in the low part of the spectrum
My current case (nothing fancy here):
my sweepis 16k samples at 44.1k so it's 371ms
i'm measuring a flat panel with a window time of +/-8ms
so when i apply the gate on the signal, i'm effectively removing the high part of the sine sweep, the one that i need to get a meaningful result in the high part of the spectrum.
That's why i said it's a theorical problem, you can't measure frequencies that are not yet excited; at 8ms in my sweep i'm still in the low frequency part of the sweep.
Only way i see to resolve that is to have an excitation signal which size is inferior to the window time, otherwise that make no sense.
Conundrum ?
no, slow sweep and gating don't go together
yes, you can only get useful info from frequencies which periods fit in/are shorter than the gate time
but that sets a lower frequency limit - not an upper limit - gated impulse response is good to however high your system's mic, ADC sample rate allows with the right test signals
yes, you can only get useful info from frequencies which periods fit in/are shorter than the gate time
but that sets a lower frequency limit - not an upper limit - gated impulse response is good to however high your system's mic, ADC sample rate allows with the right test signals
Last edited:
I guess you slightly misunderstood (or i explained it badly)
i clearly understand that the gating set a lower frequency limit, that's kinda obvious, and i clearly understand that part.
What i'm telling , is when i do my gate a 8ms, at that precise time the high frequencies of the sweep still didn't happen (since my sweep was 371ms remember), then i can't have an impulse response from an impulse that doesn't exists yet !
i really don't see how you could measure something that didn't happen yet
i clearly understand that the gating set a lower frequency limit, that's kinda obvious, and i clearly understand that part.
What i'm telling , is when i do my gate a 8ms, at that precise time the high frequencies of the sweep still didn't happen (since my sweep was 371ms remember), then i can't have an impulse response from an impulse that doesn't exists yet !
i really don't see how you could measure something that didn't happen yet
Sweep time =/= impulse gate time. IOW, impulse response is calculated from sweep response and THEN the gating is applied.
So, there is something with the theory i don't understand 😉
need to do some more reading guess, because that doesn't make any sense to me one way or another.
need to do some more reading guess, because that doesn't make any sense to me one way or another.
Collect sweep -> deconvolution -> impulse response -> gate -> FFT.
It may be helpful to look up Farina's papers which handle the problem with rigor, but it really is this simple- the trick is the deconvolution step which gets you the impulse response as if you had taken it directly. From there on, it's exactly like a "true" impulse where you can set the gate before the first echo.
It may be helpful to look up Farina's papers which handle the problem with rigor, but it really is this simple- the trick is the deconvolution step which gets you the impulse response as if you had taken it directly. From there on, it's exactly like a "true" impulse where you can set the gate before the first echo.
Ok i'll read those papers.
Just to be sure if i understood what you said :
the full impulse response contains all excited frequencies, but from the start of the main peak to the time of the first echo, i only get the response from the speaker?
So that would mean the sweep time doesn't matter at all and should actually be the longest possible?
Just to be sure if i understood what you said :
the full impulse response contains all excited frequencies, but from the start of the main peak to the time of the first echo, i only get the response from the speaker?
So that would mean the sweep time doesn't matter at all and should actually be the longest possible?
Sy is correct. You have to remember that you can get an impulse response curve using any signal which has sufficient spectral energy over the entire audio range, a click, noise-like signals, firecracker, a choir of frogs chirping.... You just record the speaker's time domain signal in response to what it is applied and "deconvolve" it with the time signal that is applied to the speaker terminals (or a good analog of it). Different types of original signals have different advantages or disadvantages (a log-swept sine has the advantage that it both rejects and allows you to quantifiy harmonic distortions in the recovered impulse response). The impulse response you get after deconvolving includes all the reflections in the room along with the speaker response. Then you take that time record (the recovered impulse response), window that and FFT it to get the quasi-anechoic frequency response.
Ok i guess then it's maybe the convolution itself i was naive about.
let me restate that another way just to be sure:
i'm necessarily loosing something by gating the impulse
1-the size of the gate determine the lowest frequency i can get , because of the period of the signal that can fit in (that one is kinda straightforward)
2-i'm probably also loosing resolution by choosing a tiny gate time(because of the sampling theorem), but i'm not loosing hi freq , because the hi frequency are kinda stacked at the beginning of the main peak (the part that give my brain headaches)
is it right?
let me restate that another way just to be sure:
i'm necessarily loosing something by gating the impulse
1-the size of the gate determine the lowest frequency i can get , because of the period of the signal that can fit in (that one is kinda straightforward)
2-i'm probably also loosing resolution by choosing a tiny gate time(because of the sampling theorem), but i'm not loosing hi freq , because the hi frequency are kinda stacked at the beginning of the main peak (the part that give my brain headaches)
is it right?
resolution in terms of frequency bin size; longer record, narrower bin
but it really is just the frequency of the longest period sine fitting in the max time window again
the fancy signal spreading/convolution "just" gives a different way of averaging - but the result should be the same as using actual impulses and averaging lots of repetitions of time aligned ADC output files of its response
where "just" involves some crest factor/power/operational trades - but the ultimate "linear" part is just equivalent to averaging lots of impulse responses
but it really is just the frequency of the longest period sine fitting in the max time window again
the fancy signal spreading/convolution "just" gives a different way of averaging - but the result should be the same as using actual impulses and averaging lots of repetitions of time aligned ADC output files of its response
where "just" involves some crest factor/power/operational trades - but the ultimate "linear" part is just equivalent to averaging lots of impulse responses
Ok ok i think i get it, just need to read the theory from now to get the details right.
Thanks for all the explanation 😉
Thanks for all the explanation 😉
- Status
- Not open for further replies.