Home Theater Forum and Systems banner

1 - 6 of 6 Posts

·
Registered
Joined
·
8 Posts
Discussion Starter #1
Hello all,
my name is Gaetano, from Italy. I'm an electronic engineer and I've a "medium" recording studio in Naples.
A few days ago, I decided to re-perform some measurements of my speakers, Yamaha MSP7 studio, in my control room, taking also the opportunity for comparing the MSP7s with the smalls newcomers Iloud Micro Monitors from IK multimedia :)yikes:)
I've already used REW in the past for measuring this control room, using an ECM8000, and an RME Fireface 800; but this is the first time I do it using an Imac, OSx El Capitan (still using ECM8000 and RME Fireface 800). Here you can see some pictures of the control room:

https://www.facebook.com/Bluwavestudio/photos/a.734245210035997.1073741828.734242673369584/734244903369361/?type=3&theater

First time I've performed measurements using REW version 5.18 (the one with JRE integrated in it) and I've noticed something really strange in all the impulse responses: they have multiple big peaks, sometimes after the zero time (meaningless...) , sometimes before (indicating absurd distortions levels??). Moreover, obviously, Waterfall plots appears totally wrong for the impulse responses that have multiple peaks after the zero...You can find at the following link the first mdat file containing the measurements (named starting with 5.18):

https://drive.google.com/file/d/0B7SRBJ1_E7z9cUJ4Z1B4X0pHT2s/view?usp=sharing

So I decided to re-perform the measurements using REW version 5.19 Beta6 (the one with JRE integrated in it) and, this time, I tried also to perform a sound card calibration in REW, without success (fuzzy curves came out, probably because the Java drivers behaviour with this Asio sound card is a mess) ; still, this step is not really necessary with an RME Fireface 800, that is amazingly linear and flat, so I gave up on this point and continued with the measurements: following you'll find also this new set of data (file mdat is named starting with 5.19 Beta6):

https://drive.google.com/file/d/0B7SRBJ1_E7z9VGVEcnYxVlFCSUE/view?usp=sharing



As you can see, the two measurements set, for what concerns SPL curves, matches pretty well; nevertheless, they have really different Waterfall plots, RT60, clarity, Distortions etc ect. To me, it seems obvious that the second set 5.19 Beta6, is the "correct one".
Still, two different results obtained in the same conditions, generated in me apprehension about the uniqueness/repeatibility of the results that one has to espect from REW, hence the reliability of the generated filters. In particular, try for example using Frequency Dependent Window, let's say 15 cycles, on the two measurements sets: 5.18 set gives totally absurd results, 5.19 Beta6 set gives "reasonable" results, a part from some really severe dips; fact is that, at this point, I no more know even if this deeps are a normal behaviour or some other sort of artefact/bug REW is introducing.

Obviously still plenty of gratitude for this great piece of code John wrote! Just a little bit afraid about the reliability of the results/filters. Any helps/toughts?

Thank you!

Gaetano
 

·
Premium Member
Joined
·
6,711 Posts
Hi Gaetano,

For your first measurement the soundcard was not maintaining synchronisation between the 4 sweeps you were performing for each measurement, they are offset from each other by about 140 ms. That might have been because the soundcard was configured in Audio Midi Settings to run at a different sample rate than the 44.1 kHz selected in REW so the OS was resampling the data. In any case, use single sweeps for measurement and that problem should be cured.
 

·
Registered
Joined
·
8 Posts
Discussion Starter #3
Hi John, thanks for the feedback. I will avoid multiple sweeps, got it. Another question : What about the deep dips that come out using frequency dependent window? Are they a collateral effect of this technique? For creating filters do you suggest using it? If yes, do you suggest to use some smoothing after it and before creating the filters? Moreover, do you suggest to aim for a no-flat target curve even when using frequency dependent window?
 

·
Premium Member
Joined
·
6,711 Posts
Dips that go below a conventionally smoothed response are a characteristic of using FDW - as frequency rises the window gets narrower, so the reflections close to the peak come to dominate the result and dips they cause are no longer filled in by later arriving reflections. I suggest using the variable smoothing for inputs to EQ, but even if using FDW dips shouldn't matter very much as they shouldn't be targets for EQ anyway. For listening position measurements I would suggest a target curve that slopes down gently as frequency increases, whether using FDW or not. I wouldn't suggest using EQ above the bass range generally though. If there is some known deficiency in the anechoic speaker response that EQ is being applied to correct then anechoic or pseudo-anechoic measurements (windowed to remove reflections from surfaces) should be used.
 

·
Registered
Joined
·
8 Posts
Discussion Starter #5
John, thanks for the patience :)
Just one more question: shouldn't FDW, with a short cycle number, be more representative of an anechoic-like measurement of a speaker, at least in the medium-high freq ? Wasn't it in origin created just to mitigate the effect of the reflections in the medium-high regions hence to give much more importance to direct (hence anechoic-like like) sound? If so, why a speaker that measures flat in an anechoic-like room should have deep dips when using FDW ? Moreover, considering this anechoic-like behavior of FDW, why one should still aim to a soft slope for the target curve instead of a flat one?
 

·
Premium Member
Joined
·
6,711 Posts
FDW is in some ways more representative of hearing in the way it has a time window that reduces with frequency, similar to the ear's varying time window within which it treats sound as coming from the source rather than being an echo. At very high frequencies, depending on the window setting, the window gets narrow enough to get closer to the anechoic response, but at low frequencies and through the midrange the window is still quite wide so typically includes some reflections. To get closer to an anechoic measurement you would need to remove the causes of the reflections which lie within the varying window.

The attenuation of sound in air is frequency dependent, by 20 kHz it has increased to about 0.5 dB/m though the figure depends on temperature and humidity.
 
1 - 6 of 6 Posts
Top