Spacial Average Measurements

Samazar

New Member
Thread Starter
Joined
Aug 21, 2017
Messages
37
I want to know what experienced REW users think about spacial averaging which entails taking 4 individual measurements with REW and averaging them to come to a final end-result. The mic needs to be upright with the 90 degrees cal file and placed within the seating area not in line with each other but offset and mic heights adjusted to be slightly different for all 4 positions. I presume this provides better overall readings and repeatability for future measurements.
 

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Hi John,
I got annoyed by the "Vector Average" function that elaborates an averaged IR from spatial measurements.
It is suggested that this function must be used for amplitude correction. But I guess amplitude correction should not take into account the phase of the different measured IR. This is because a listener will listen the system at one only spatial position at a time. So only amplitudes of the measurements has to be averaged without taking into account the phases.

Can you please comment on this ?

Thx.

Here is a plot showing the huge difference between the "Vector Average" function and the "Average" function in a moderatly reverberating room.

 

jtalden

Senior Member
Joined
May 22, 2017
Messages
887
Location
Arizona, USA
More  
Preamp, Processor or Receiver
Marantz AV7705 Pre/Pro
Main Amp
VTV 6 chnl NC252MP P-amp x 2
Additional Amp
Behringer DCX2496 x 2
Universal / Blu-ray / CD Player
OPPO BDP-103 Universal Player
Front Speakers
DIY SEAS H1456/H1212 Spkr x 5
Subwoofers
DIY JBL 2235H 15" SW x 2
Video Display Device
JVC DLA-X790R
Screen
Da-Lite Da-Snap 39105V - 92"
This result is not consistent with my experience or that of several others:
Did the Impulse overlay chart show that the 4 impulses were properly aligned in time for vector averaging?
Was there an FDW applied to the measurements during vector averaging?
Can you post the mdat of 4 measurements so we can investigate ourselves?
 

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Hi,
Thanks for your help.
Yes impulse are aligned before vector averaging. And no FDW applied.
Here is a link to the 9 measurements you can play with :
www.transfernow.net/226pg8i48ada
 

jtalden

Senior Member
Joined
May 22, 2017
Messages
887
Location
Arizona, USA
More  
Preamp, Processor or Receiver
Marantz AV7705 Pre/Pro
Main Amp
VTV 6 chnl NC252MP P-amp x 2
Additional Amp
Behringer DCX2496 x 2
Universal / Blu-ray / CD Player
OPPO BDP-103 Universal Player
Front Speakers
DIY SEAS H1456/H1212 Spkr x 5
Subwoofers
DIY JBL 2235H 15" SW x 2
Video Display Device
JVC DLA-X790R
Screen
Da-Lite Da-Snap 39105V - 92"
I just reviewed my results and was mistaken. Your findings are consistent with mine for listening position measurements without FDW. Here is an old chart of mine. The red trace is the vector average method at the LP compared to other MMM traces.
I have often used vector averaging at the LP for improved definition of direct sound phase. It really helps to clean up the chart trace. However, since FDW is available and helps achieve the same result. I have not used this method recently.
avgs spl compared.png
 

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Thanks jtalden.
I agree with you, verctor average is better to retreive direct sound phase (as FDW is)
Then the question is why REW recommand to use vector average to perform amplitude correction at the listeniing position whilst it is not consistant with scalar average measurement ?
What we are listening to is not the direct sound energy, but the overal sound energy (at least up to 50 ms delay time).
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
Hi John,
I got annoyed by the "Vector Average" function that elaborates an averaged IR from spatial measurements.
It is suggested that this function must be used for amplitude correction.
It doesn't seem to make sense to get annoyed at the feature, that is like being annoyed at a screwdriver for not being a spanner. I don't recall saying that vector average must be used for any particular purpose, it was added at the request of users who wanted a vector average to allow phase correction to be applied to the result. Whether that is a good or bad thing I couldn't really say, it isn't something I have studied.
 

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Hi John,
Fair enough. Thanks for the reply. It's in line with my thought. This function should be used for phase correction only, not for amplitude correction as I read somewhere.
REW is a so great piece of software. Thank you.
Xavier
 

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
Hi John, Hi All,
Bravo for this feature !!! Vector average work far more better than any other method (FWD, regular average...) even for amplitude correction (I don't use phase correction anyway)
I just tested it using the famous Dirac live 9 points measurements and the result is really amazing.

To be noted that I'd never done my measurements from REW, I always use MLS 0.500 from HOLM and then import to REW and I do manual time alignement for a most accurate result.

Anyway thanks alot for this great feature.

here some images

1.jpg


2.jpg


3.jpg
 
Last edited:

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Hi,
remember that vector average will tend to simulate direct sound measurement (reflexions canceled). That's not what your ears will hear. Human ears and brain average direct sound and early reflexion, and reject far reflexions. 30 to 40 ms seems to be the correct threshold. Then instead of using vector average fonction, I suggest to use time filtering of impulse responses to 30 to 40ms, then perform a scalar average of all measurements. This should be the curve to equalise.
 

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
Hi Niziers,

Following your advice, I did some new measurements at 3 differents points with REW Sine sweep and HOLM MLS 0.500 (I placed the mic a one point did REW record and then HOLM record without moving the mic. same for the 2 others points).
That gave me 6 measurement (3 sine sweeps and 3 MLS).

For the sweeps, I time filtered the 3 at 30 ms and did a scalar average.
For the MLS, I time aligned and then did a vector average.

At the end, I almost had the same curve between 30hz to 6Khz. From that point, frequencies are higher on the sweep until 20 khz
I don't know how to upload screenshots. If not I would have done it.

I corrected both average (regular and vector) and at ears, I found vector more detailed and with better sound spacialization.
So why to try to catch a headaches whit time filtering and trying to determine if a reflection is early or not ?

To be noted that MLS 0.500 using 9 points sofa like "Dirac live" is far more accurate than those 3 points measurements.
 
Last edited:

niziers

New Member
Joined
Jun 10, 2017
Messages
6
More  
Preamp, Processor or Receiver
X-UNI advance acoustic
Front Speakers
DIY LA
Subwoofers
DIY 18"
Other Speakers or Equipment
Mini-DSP Opendrc DA8
Hi 60seconds,
I guess your room is properly insulated so that there is not that much reflexions. that's what your measurments tend to prove as you've no differences until 6kHz. The room might not be that big too so that you have not that much far reflexions. At the end, it's also a question of feeling. Time filtering and vector averaging should provide better fidelity because it take into account pshycoacoustic, but vector averaging can tends to give better sound feeling. Its also a question of balance between low end and high end. Different target curve could help, for example by adding few dBs (3 to 8) bellow 80 Hz and linearly lowering down by 2 to 5 dB highs.
 

AudiocRaver

Senior Reviewer
Staff member
Joined
Nov 21, 2016
Messages
974
Location
North Carolina, USA
More  
Preamp, Processor or Receiver
Onkyo TX-SR705 Receiver
Main Amp
Crown XLS 1502 DriveCore-2 (x2 as monoblock)
Additional Amp
Behringer A500 Reference Power Amplifier
Front Speakers
MartinLogan Electromotion ESL Electrostatic (x2)
Center Channel Speaker
Phantom Center
Surround Speakers
NSM Audio Model 5 2-Way (x2)
Subwoofers
JBL ES150P Powered Subwoofer (x2)
I believe, although I have not yet experimented with it, that vector averaging would be better at helping one focus on the detail that affects soundstage and imaging (SS&I). There is NO substitute, however, for identifying and treating early reflections, which can affect frequency response (FR) only a little while having a big affect on image clarity and soundstage detail.
 

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
True, my room is relatively small and I have carpet on the floor.
I did tried Target curves before but I was’nt really convince.

The thing that was really bothered me was the differences between REW sweep and HOLM MLS impulses. I read somewhere that MLS is less sentive to early reflections. maybe that’s why I have better results using Vector average with it but I’m not sure.

Based on the 6kHz diferences between both impulses I applied a 12dB LP filter that seems to suit perfectly to the MLS curve. I think that I maybe finally, surgically, identified my early reflections. A least, that what my ears are saying. Wayne Shorter's saxophone do too.

vec_avrg.jpg


target.jpg
 
Last edited:

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
Ok, after a trade with one of my acquaintance who work on the sound field about MLS, he confirmed to me that MLS doesn’t record early reflections so the impulse produced is reflections free. He also told me that the fall at 10kHz on my MLS curve (Black) is the mechanical limit of my tweeter.
The real reflections of my room is the space between the black curve and the red one (Sweep) from 6kHz to beyond.
That’s why my 12 dB HP filter at 6kHz worked so well. But when I look the space between the black and the red curve at 12 kHz it seems to be 8 dB/octave. Since REW doesn't allow a linear fall from 6kHz and only for less than 6dB, I' will stay with my HP filter.

Anyway, thanks alot HOLMImpulse MLS 0.500 for the reflections free impulse, thanks REW for the Sweep impulse full of reflections that helped me determine surgically my real room reflections and for the Vector average killer feature.

Thanks Niziers, you were right about the lowering thing.

AudiocRaver maybe there is now a method for identifying early reflections.
 
Last edited:

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
MLS captures the whole room response, just as sweep measurements do. Reflections after any particular time can be removed by windowing the impulse response, whether it was obtained by MLS, sweep or one of many other techniques. There is a good summary of the various approaches and how they compare to sweep measurements here.
 

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
Hi John,
First of, thanks for the great work you ‘ve done with REW. I’ve learned alot regarding Room correction thanks to you.

Thanks also for the link it was very interesting (except for the mathemathical part where I was totally lost) and you’re right Sweep and MLS capture the same signal even if the 1st one do it better.

That mean that there’s only few explanation left. One of them can be that HOLMImpulse do a kind of post-computation after the recording. If it’s true so the curve that I’ll have using Sweep or MLS from HOLMImpulse will be same if chosen. I decided to take a look and I saw this :

holmimpulse_signal_settings.png


After fews googling, I ran into this document about the use of MLS technique for annoyance minimization. It can be another track.

Here waterfalls and decays for SWEEP and MLS. there is a roll-off from top higher frequencies until about 1kHz on MLS's that is not present ont Sweep’s.

mls0500_waterfall.jpg


sweep_waterfall.jpg


mls0500_decay.jpg


sweep_decay.jpg


Later I’ll do a one point 3 measurements. One Sweep, one MLS with amplitude of 1.00 and another with an amplitude of 0.50.
If what I think is true, Sweep and MLS 1.0 curves must be the same.

I'll post the results.
 

Greg Dunn

Member
Joined
Oct 4, 2017
Messages
107
What I have done with modest success is to set REW's RTA up for long term averaging (not vector) and fed it with a pink PN signal while doing the moving mic around the listening position and staying away from floors and walls. The results are remarkably consistent and reflect any EQ changes quite well. It seems to work much better for me than manually taking several measurements and then averaging them.
 

AudiocRaver

Senior Reviewer
Staff member
Joined
Nov 21, 2016
Messages
974
Location
North Carolina, USA
More  
Preamp, Processor or Receiver
Onkyo TX-SR705 Receiver
Main Amp
Crown XLS 1502 DriveCore-2 (x2 as monoblock)
Additional Amp
Behringer A500 Reference Power Amplifier
Front Speakers
MartinLogan Electromotion ESL Electrostatic (x2)
Center Channel Speaker
Phantom Center
Surround Speakers
NSM Audio Model 5 2-Way (x2)
Subwoofers
JBL ES150P Powered Subwoofer (x2)
What I have done with modest success is to set REW's RTA up for long term averaging (not vector) and fed it with a pink PN signal while doing the moving mic around the listening position and staying away from floors and walls. The results are remarkably consistent and reflect any EQ changes quite well. It seems to work much better for me than manually taking several measurements and then averaging them.

The Moving Mic Method is a good way to get an average over an area, and gives consistent results. Unfortunately, no phase or time info... But great as a final check after room correction.
 

Greg Dunn

Member
Joined
Oct 4, 2017
Messages
107
Yes. And having run sweeps at the listening position, I know the phase characteristics pretty well, so I can make adjustments with confidence - again, small changes at frequencies where room modes are not dominant. Time alignment is fortunately not a concern, as my speakers are planars with all the radiators at a near-identical distance from the mic (there has to be some payoff for having big freaking hinged panels, right?). I have set up time delay for the sub to minimize issues from it being placed at a slightly different distance.

After many MMM measurements and some EQ changes to test their effect, I'm convinced that further tweaking of the response is not warranted; the biggest change so far, after getting rid of the big response irregularities, has been putting drapery-like material over a window and another reflecting surface right where they would give an early bounce toward the listening position.
 

60secondsassassin

New Member
Joined
Feb 13, 2018
Messages
7
Location
NORMANDY (FRANCE )
Mea culpa.

I was still bothered by the difference I had between my sweeps and MLS measurements so I reviewed all my measurement chain and dig-up a little bit more. This post on gearslutz forum putted me on the right track.

For all my measurement my mic was pointing at the ceiling (vertically) at 90 degree instead of pointing horizentaly at 0 degree. According to this other post on minidsp forum difference can be relatively subtle at high frequency (~ 2dB).
I've re-done 4 measurements. 2 with mic at 90 degree using both signals and 2 at 0 degree again with both signals. the aproximatively 2dB difference was true with sweep signal but far more with MLS.
Turns out MLS signal is more sentive to mic orientation than Sweep. MLS and sweep signal at 0 degree matched perfectly except a subtle difference at lower frequencies around 40Hz.

I just lost my PHD of surgery :-( but now I know tha my mic (an Apex 220, I don't know if mentioned it before) is a "Free field mic".

For me, trying to correct listenning area with only one point measurement is like trying to analyse annual sales records by looking only the sales of July. So I've re-done the "9 points dirac live sofa measurements" (no shorter name yet) using sweep with mic at 0 degree and time aligned impulses at the higher positive peak.

Sorry but now that I finally resolve my high frequencies issue (it took me more than 60 seconds this time) I maintain my statement, Vector average is far more better for frequencies correction. Never done phase correction so far.

One question. With wich filter do you sirs deal with room mode ? For now, I use lowshelf filter at 100Hz to minimize it but maybe you now better solution that don't include using wall bass trap.
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
Room modes are typically best targeted with parametric EQ filters matched to the mode's centre frequency and bandwidth.
 

Matthew J Poes

AV Addict
Joined
Oct 18, 2017
Messages
1,903
Mea culpa.

I was still bothered by the difference I had between my sweeps and MLS measurements so I reviewed all my measurement chain and dig-up a little bit more. This post on gearslutz forum putted me on the right track.

For all my measurement my mic was pointing at the ceiling (vertically) at 90 degree instead of pointing horizentaly at 0 degree. According to this other post on minidsp forum difference can be relatively subtle at high frequency (~ 2dB).
I've re-done 4 measurements. 2 with mic at 90 degree using both signals and 2 at 0 degree again with both signals. the aproximatively 2dB difference was true with sweep signal but far more with MLS.
Turns out MLS signal is more sentive to mic orientation than Sweep. MLS and sweep signal at 0 degree matched perfectly except a subtle difference at lower frequencies around 40Hz.

I just lost my PHD of surgery :-( but now I know tha my mic (an Apex 220, I don't know if mentioned it before) is a "Free field mic".

For me, trying to correct listenning area with only one point measurement is like trying to analyse annual sales records by looking only the sales of July. So I've re-done the "9 points dirac live sofa measurements" (no shorter name yet) using sweep with mic at 0 degree and time aligned impulses at the higher positive peak.

Sorry but now that I finally resolve my high frequencies issue (it took me more than 60 seconds this time) I maintain my statement, Vector average is far more better for frequencies correction. Never done phase correction so far.

One question. With wich filter do you sirs deal with room mode ? For now, I use lowshelf filter at 100Hz to minimize it but maybe you now better solution that don't include using wall bass trap.

As John said you want PeQ. A shelf filter just changes the level but doesn’t actually address the mode itself. You want to knock down the peaks with an inverse preconditioned signal, I.e. the peq filter.

Mic orientation seems to be a complex topic with little solid information. I’m actually not sure what is on the links you posted is totally correct either (I’ve read and even shared those same links myself). The argument for pointing the mic at the ceiling is two fold: 1) the mics natural response is corrected for use in a room by compensating for the high frequency rise, 2) the mic captures more room information by more evenly measuring the mic of direct and reflected sound. However the response shape is not accurate in either position for inexpensive mics in a room so a correction is needed. It seems that for most cheap mics pointing it at the speaker may be better (they are also more directional at high frequencies) if you don’t have an accurate true 90 degree correction. I myself still point the mic at either 45 or 90 degrees and use an appropriate correction file for the reason I gave in number 2 above. Number 1 is not a concern since I use accurate correction files.
 
Top Bottom