[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Optimal sweep duration for BRIR measurements



Hi all,

I've been asked to clarify the point I was making in the following part of
my last message

> Regarding distortion of the source, this should be an issue for the
processing element (different from burning out you speaker). This is because
one of the strengths of the sweep method (when done correctly) is that any
harmonic distortion components of a higher frequency that the excitation
signal at the time [are] folded back to BEFORE the direct sound after
deconvolution by the excitation signal (as that frequency has yet to be
generated).

Due to a typing error, one word was wrong which was maybe confusing
(corrected above). The principle I can try and explain again, though without
images it is not always easy. So, here goes: 

Let's assume our reference/excitation signal is a linear frequency ascending
sweep. Looking at a spectrogram this is a nice line at 45°. For the recorded
response signal, each frequency excites the room and there is a reverberant
tail for each frequency which can be seen in the resulting spectrogram. When
you perform the deconvolution of the response by the stimulus, the diagonal
of the sweep excitation is the reference and the resulting response is not
an impulse response relative to that excitation. The result is that the
sweep signal diagonal  is in some sense rotated to a vertical impulse
(again, thinking in terms of a spectrogram). With the direct sound for all
frequencies now being time aligned. 

Now, if we look at transducer distortions, these are generated at
frequencies higher than the excitation signal, meaning a speaker with a
signal at 1kHz can produce distortions at 2kHz, 4kHz, etc. When the sweep is
rotated or folded back to the vertical through the deconvolution, these
distortions are also moved. As they occur BEFORE their corresponding
frequency excitation in the stimulus, they will occur BEFORE the direct
sound in the RIR, appearing as pre-RIR. These need to be windowed out in any
processing step when the RIR is to be used. This effectively removes the
speaker distortion from the measurement. 

However, if one is interested, one can also analyse these pre-RIR as they
provide information on the distortion of the speaker (or other devices).
With a correctly designed sweep, this is a powerful tool, as one can now
model a non-linear distorted IR (see the paper I referenced earlier). 

Of course, one should try and minimize the level of distortion in the sweep
excitation when played, as this isn't too friendly or efficient for the
speaker. An advantage of the sweep over other methods is that the speaker is
only generating a single frequency at a time, and is therefore more
efficient in level than for the same input at broadband; speakers like sine
waves, and you get more bang for your buck with sweeps. 

Hope this helped anyone who was interested.

-Brian FG Katz
LIMSI-CNRS


Attachment: smime.p7s
Description: S/MIME cryptographic signature