Measuring freq. response with a deep notch, cannot measure the correct depth.

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Hi here.

I'am trying to measure an active 998Hz notch, where the notch deep is around -70dB (very sharp).
If i'am measuring it with the RTA / FFT function in REW with my soundcard, I can tell by the numbers, that there are at least -70dB between 998Hz and 8000Hz(way outside the notch)
So I can for sure tell, that the notch is that -70dB deep.

On the picture below showing two notch's, the green is a -30dB notch which I can measure good (Cosmos APU), the yellow one is this very sharp -70dB notch, which only measures around -26dB during frequency sweep.
- This is nomatter what samplerate, and sweep lenght I use the same deept.

This is a matter of frequency resolution i guess, but I cannot figure out howto measure it more precisely.
- Should I use stepped sine ? - And if; Howto do that correctly?
- The measurement from 10Hz to 22000Hz are supposed to be used as a notch compensation file (loaded as a mic. cal file)

This is for sure a learning experience ;)

Thank's in advance; Jesper.

Notch.PNG


AR 998Hz.png


AR 8000Hz.png
 

trobbins

Member
VIP Supporter
Joined
Nov 26, 2017
Messages
121
Location
Australia
Perhaps limit the frequency scan range to just around the notch to diagnose. Perhaps turn off dither. You could step though the notch with single frequency steps to confirm as well.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Perhaps limit the frequency scan range to just around the notch to diagnose. Perhaps turn off dither. You could step though the notch with single frequency steps to confirm as well.
Hi trobbins.

I did try to lower the span to 900Hz ---> 1100Hz, giving the same result.
I tried to step through the notch with RTA / FFT with 1Hz step's, showing me that the notch is -70dB deep, but I don't know howto do that in "measurement" ?

Jesper.
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
In the Analysis preferences choose the option to retain the full IR then measure with a long sweep. The total length of the IR window and the sample rate determine the frequency resolution, you can see the figure in the Info panel, so set a very long window.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
In the Analysis preferences choose the option to retain the full IR then measure with a long sweep. The total length of the IR window and the sample rate determine the frequency resolution, you can see the figure in the Info panel, so set a very long window.
Okay...

That's something I did not know.
I will post back with my findings.

Jesper.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Is it the Decimate IR checkbox.

Sry. I'am not knowing that much about that YET :T

The Decimate IR selection controls whether REW reduces the sample rate of the impulse response to correspond to the range of frequencies in the measurement. Selecting this option greatly reduces the impulse response size for low frequency measurements and speeds up processing of the data.

IR.PNG
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
Change it back afterwards, keeping the full IR will produce very large measurement files and use a lot more memory for each measurement.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
So this is solved now :)

Allway's help John, thanks...
Tried to do a couple of 4M / 48kHz sweep's... Notch is >70dB deep

I will proberly ask some more quistions regarding the calibration file

Jesper.

AR 998Hz -70dB.png
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Hi here...

As I said, proberly need to have some more cleared out.
I created a "mic. calibration" / notch compensation file, as Archimago do. Link ----> Archimago notch compensation
(It's just to allign the measured notch SPL to 0 via. add offset in the SPL & Phase dialog + export measurement as txt, and load this file as a mic. calibration file)

When i measure e.g. the Victor's 998Hz osc. with this notch and without mic. cal. file, I can have the same THD number's by subtracting -70dB (35dB + 70dB = (-)105dB... See picture

When mic. calibration is loaded i can read the number's directly, as supposed.
But the noisefloor is (I THINK) "unrealistic" low, I think this is duo to the very steep -70dB depth of this notch?
(I'am not using manuel fundamental as I do with my passive notch)

The notch is active, and is adding +20dB after the notch.

I don't think I'am doing anything wrong? Should I just accept the "unrealistic" low noisefloor? - Or is there some way to compensate for this?

Must say I'am having fun investigate this :)

Jesper.

AR Notch 997Hz SPL Aligned.png


AR OPA1662.png
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
I don't really follow the logic of setting the bottom of the notch to 0dB, that means everything outside the notch gets 70 dB of adjustment applied to it. I'd normally use 0 dB for the level away from the notch so the bottom of the notch is at -70.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Great...

So you suggest me to try it out adjusted so that the TOP of the notch (away from the notch) is adjusted to 0 SPL instead? - Please see picture.

Or should I do as I did with my passive notch the trace arithmetic (Notch response / No notch response (soundcard loopback)
BTW. Is it correct that this way I compensate for the notch insertation loss also ?? - Or ?

I will try It out tonight.

Thank's John.

AR 997Hz alligned TOP to 0 SPL.png
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
This is really great...

After some hour's with trial and error I finally got it right.
(Notch is dual fliege OPA1662)

This is the way I did it for the result's (see pictures for better explanation) :

1. Adjust Analysis.
Settings Analysis.png


2. Take measurement with notch and without notch. (-25dB - Full input on my hacked Behringer)
Settings measure UMC Loopback.png


3. Do Trace arithmetic.
AR OPA1662 TraceArithmetic.png


4. Measurement without notch and without cal. file
Victors 1007mV to UMC.png


5. Measurement with notch and cal. file + manuel fundamental.
(The fundamental is measured with a good quality DMM when loaded)
Victors_1007mV to ARNotch to UMC.png


I can proberly get much better result's with higher voltage from Victor's 998Hz oscillator, but I keept it around 1V,
to compare directly with full input to my Behringer without clipping.

I think I got it right this time :)
 

Attachments

  • AR Notch - Trace arithmetic.png
    AR Notch - Trace arithmetic.png
    15.7 KB · Views: 47

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
Hi...

A ran some test, to try the two different calibration techniques out.
All test is with Victors 998Hz oscillator @ 2.1V sine.
The Notch is the Cosmos APU adjusted to -41dB / 998Hz.
I carefully calibrated the voltage levels in REW before I started the test's.

I'am in no doubt, that I would allways prefer to use the recommended way by trace arithmetic the notch/nonotch, as this is the way also explained in the REW manuel, but I was just in doubt, as I thought that it was only when using passive notch.

EDIT :: If I can suggest / wish for an new feature for REW, it should be a way to calculate the "correct" -dbfs number's when using the "manuel fundamental", notch over nonotch (explained below)... Donno if it's possible ?

Comment's are very welcome...

First picture to show how the calibration curves look like:
MicCal files.png


Here I measure with the allign to 0 SPL technique.
The result are giving me a very low noisefloor, which i guess is a bit unrealistic. (around -160dB, -170dB)
Also REW is showing 377mV input, which should be 2.1V fundamental (before the notch)
I think the -11dB is correct ?
APU Notch MicCal 0SPL.png


When measuring with calibrationfile created by Notch over Nonotch, as the correct way @John Mulcahy is telling me before in this thread I got approx. the same number's as with the 0 SPL alligned calibration file.
I don't even have to enter the manuel fundamental, when REW is telling me the correct value I measured before the notch... Amazing
The catch here, is that the dbfs number is showing +3.9dbfs, but I know this is duo to the notch calibration vs. the input voltage calculations (@John Mulcahy explained this to me here My other thread with passive notch)
APU Notch MicCal NOTCH_over_NONOTCH.png


Last thing to check in this round :)

When I measure without MicCal. file loaded, I get approx. the same number by subtracting -41dB.
E.g. 2nd. -111.3dB -41dB = -152.3dB... Well not excatly, but very close.
APU +6dB NO MicCal.png


APU Notch -41dB.png

Jesper.
 
Last edited:

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
EDIT :: If I can suggest / wish for an new feature for REW, it should be a way to calculate the "correct" -dbfs number's when using the "manuel fundamental", notch over nonotch (explained below)... Donno if it's possible ?
As explained in the passive notch thread, those are the correct numbers. If 2.1V was actually coming in to the ADC it would be 3.9 dB into clipping. The point of the notch is to extend the range of the ADC by not letting it see the high levels, so you have a system that can accommodate levels above 0 dBFS within the notch span.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
I accept that, also that it's correct.

So anything above my ADC's 0dbfs level (around 1.339v) will be shown as +dbfs
I did know that, and I also realize that it's not making sense what I asked as an option for REW.

Jesper.
 

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
So after some time, I stumbled upon the (A)weighted measurement.
I'm not aware of which kind of numbers I got using my calibration file as pr. instructed here in post #16 to #18.
Or is it possible to change something and get (A) measurements or do I allready get that.
I'm having problems understanding this?
Not sure what the difference is?
(Believe me I searched)
Someone kind to explain this?

Thanks guy's.

Jesper.
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,212
Most of REW's results are unweighted, where the figures are A or C weighted the units show that. For example, the N+D figure is A weighted (hence the "A" shown in the unit field) and the RTA level display includes unweighted, C-weighted and A-weighted figures. The weightings are just responses that the measurement gets multiplied by, A and C weighting responses are attached. You could use a cal file with an inverse A weighted response to force A weighting for all values, though REW wouldn't 'know' they were A weighted, example attached.
 

Attachments

  • a-weight-inverse.txt
    1.7 MB · Views: 6
  • A and C weighting.mdat
    3.1 MB · Views: 9

jesperlykke

Registered
Thread Starter
Joined
May 26, 2020
Messages
20
I see... thanks...

So A is just a calculation i see, makes good sense.

Will dig some more into this next few days :)

Jesper.
 
Top Bottom