Elections, Target curves and measurements

[Le Concombre Masqué]

Measurements like statistics in politics can be manipulated or even ill used with best intentions. In this thread I intend to go under the hood showing the options I chosen along the way while creating convolution filters. So maybe I'll expose my goofs and you're welcome to comment them. Or maybe this will enlighten your own choices and help you take better informed decisions.

[Le Concombre Masqué]

let's start with my speaker's measurements by Cabasse in their world class anechoic chamber.

 

flat response curve, highly controlled directivity... 

 

note that the 20 Hz is 2.5 dB below the 30

[Le Concombre Masqué]

And now what in room measurements do we consider?

 

From 15 sweeps of my not ideally environed Right channel, we can derive drastically different averages.

 

Green and brown couple are FDW (15 cycles) and non FDW scalar averages and they differ up to 7 dB...

 

couple of thick green curves show that non FDW average is consistent with Moving Microphone Method results. Since I'm to apply target curves designed to correct the steady-state, MMM has been my choice for a while. Recently I have chosen the Non FDW scalar average that follows it a lot but that is supposed to be, and probably is, say my ears, more accurate in the Low Frequency range

 

@mitchco @ https://accuratesound.ca/dsp insists upon using FDW, with a degree of sophistication, alright, but it seems we don't start at all to correct Amplitude on the same basis.

 

Please note that we could then present results that would just look alike, alike the target we would aim, but I anticipate the resulting convolution filters to sound vastly different if indeed there's a several dB difference between the measurements we work on to start with. Maybe there's a compensation in his software ? I don't know. But I sure want to correct the steady state with room gain and my ears tell me it's no good when I correct amplitude based on FDW averages in REW.

 

However I have never been able to work the Phase in Rephase with anything but the time & IR aligned FDW Vector Average, logical since we don't want to correct every reflection's influence on phase

 

the thin and thick green couple show that measurement number 2 ( middle of the sweet spot!) would be the one single measurement that would be as correct as an average of 15, as far as Amplitude is concerned. (Phase is not workable, too fuzzy with idiosyncratic reflections)

 

Please note that the 20 Hz is STILL 2.5 dB below the 30 IN ROOM and I had to resist the temptation to electronically null that delta to present beefier bass

 

So... it's complicated and maybe we should buy commercial softwares instead of using REW and RePhase and/or get professional help.

 

Or maybe not : maybe the right choices are implemented in the software but maybe not (see above).

 

there's complexity under the hood and I'd be happy to be proven wrong if I learn and get better SQ but I now have thoroughly tested those options and feel OK with my own choices.

 

 

[Le Concombre Masqué]

before we export VA average to work the time domain in REW, we have to align the impulse on the tweeter

 

that time offset of 0.037 ms  will be set in Rephase too

 

[Le Concombre Masqué]

I won't discuss here Target curves nor present Amplitude results.

 

It's all about election type presentation of measurements vs dirty under the hood.

 

Rest assured you can find my Frequency Response curves look nice and you can see there in other threads.

 

Let's focus on those graphs in REW that we can obtain only if we work with impulses. 

 

So I then need to apply convolution impulses designed in REW plus RePhase on the FDW Vector Average while the Amplitude correction has been applied to another measurement (scalar NON FDW Average) !!! This can't look perfect.

I wish no commercial offer looks to look as good as possible by not correcting the Amplitude on the right basis

 

So what I get follows

 

 

[Le Concombre Masqué]

IN ITALICS : REW's Help

 

Step Response

The step response shows the output which would result if the input signal jumped to a fixed level and stayed there. It is the integral of the windowed impulse response. If there is an offset in the measurement input chain the step response will show an overall rise or fall as time progresses, rather than tending back to zero.

 

Hence it's important to show up to back to zero...

 

Nota Bene : this is the STEP not the Impulse

 

 

[Le Concombre Masqué]

here's the phase, 

[Le Concombre Masqué]

and different views of the impulse

[Le Concombre Masqué]

Group Delay

[marce]

Digesting.🙂

[Le Concombre Masqué]

RT60 Explanation
RT60 is a measure of how long sound takes to decay by 60 dB in a space that has a diffuse soundfield, meaning a room large enough that reflections from the source reach the mic from all directions at the same level. Domestic rooms are usually too small to have anything approaching a diffuse field at low frequencies as their behaviour in that region is dominated by modal resonances. As a result RT60 is typically not meaningful in such rooms below a few hundred Hz. Use the RT60 Decay, waterfall, spectrogram and Decay plots to examine the decay of low frequencies in domestically-sized rooms.

 

I confess I don't pretend RT60 would be meaningful here ; so let's look at RT60 Decay.

 

Interestingly enough, it's cleaner when the convolution is applied to measurement 2 rather than to the VA

 

for all measurements above (phase, step...it's the other way round). As a politician, I showed the best looking

[Le Concombre Masqué]

And now, as a shrewd politician I'll show you a Spectrogram close to ideal

 

An ideal Spectrogram decays very rapidly off the bottom of the scale range. Here is an example of a plot produced from a soundcard loopback measurement in Fourier mode.

 

[Le Concombre Masqué]

The above is obtained when convolving the Vector Average by an Impulse that smoothed it (big bass bump, straight line from 160.

 

I'm far from that good when convolving the VA by an impulse created with the NON FDW scalar average as the measurement to correct for Amplitude.

 

So.... is it too good when a commercial offer is close to ideal, does it mean the Amplitude has not been corrected on the best basis?