Taking actionable measurements

[oohlou]

Sometimes I feel like I know what I'm doing and others I feel lost. Help me understand my measurements.

I have a new MiniDSP SHD with Klipsch Forte speakers and I'm playing around with DIRAC for the first time.
RED: No DIRAC. One position measurement
Green: DIRAC on. One position measurement
Blue: No DIRAC. Moving microphone measurement
Orange: DIRAC on. Moving microphone measurement

The one position measurements were done with REW frequency sweep measurement tool. The moving microphone measurement was done with REW generator pink noise set to the same output level as the previous test and is the average of ~20 measurements with me moving the microphone in concentric circles from my ears.

Questions:
1. Why is the SPL of the moving mic measurements so much lower? It is a good 20dB lower. It doesn't sound that much lower to my ears.
2. Why does the moving mic measurements show so much more high frequency roll off? I played around a bit with one position measurements in the same general area as my moving measurements and the one position measurements are always higher.

This is just a first pass of DIRAC tuning for me but I'm trying to better understand my results before I try to dial in my desired curve.

 

[RayDunzl]

My initial but possibly unsupportable guess:

Pink noise has a 3dB per octave slope to it.

The REW sweep tone does not.

Extra Credit:

Try white noise for the Moving Mic measurement and compare.

Pink Slope Estimate:

 

[DanGuitarMan]

In REW, go to RTA -> Settings -> Appearance Do you have the box labeled "Adjust RTA levels" checked?

 

[Blumlein 88]

The reason the level is lower is because that is correct. In the sweep you at any given moment have the signal level very high, but it is sweeping up in frequency. With pink noise, the total sound level over the entire 20 khz bandwidth is equal to the sweep tone, but when you graph it you are splitting it into narrower bands which have a lower signal level than the total wideband signal. It is what you would expect to see.

So yes to your ears it sounds nearly as loud, but in any given frequency band measured it isn't because the FFT is filtering out most of the pink noise.

And yes, the downward slope is for the reason Ray explained.

 

[RayDunzl]

In REW, go to RTA -> Settings -> Appearance Do you have the box labeled "Adjust RTA levels" checked?

Looking at that here, it just raises the entire RTA trace by 28(?)dB, with no slope change.

 

[bribur]

I'm also new to REW but with pink noise I saw a downward slope when the mode in settings was set to spectrum. 1/48 RTA gives a straight line.

 

[Blumlein 88]

BTW, did you have a target curve for Dirac and what was it? Usually a flat one is too bright and thin sounding. You'd normally want Dirac to shoot for a target curve somewhere between the pink noise one and flat. A 9 or 10 db droop from low to 20 khz.

 

[Blumlein 88]

I'm also new to REW but with pink noise I saw a downward slope when the mode in settings was set to spectrum. 1/48 RTA gives a straight line.

I think this is how it would work.
The reason is in spectrum the frequencies are divided into even sized bands. While in 1/48 or other RTA measures it is divided by portions of an octave. Meaning the higher frequencies cover twice the frequency range of lower frequencies. So the level evens out.

 

[oohlou]

Thanks. I've been reading about pink vs white noise and your responses make sense with respect to pink noise. However, shouldn't white noise then have a similar SPL? Here is the same sweep with white noise. the graphs are virtually the same except for SPL.

 

[RayDunzl]

I'm also new to REW but with pink noise I saw a downward slope when the mode in settings was set to spectrum. 1/48 RTA gives a straight line.

That is correct.

RTA looks at total energy in little bands of frequency, so the frequencies in a band "add up" to make the flat reading with RTA

Spectrum looks at energy at individual frequencies (that's how I explain it to myself). All frequencies must have the same level to show as flat.

If you look at white noise using RTA, it will slope upwards. With spectrum, flat.

If you look at pink noise with RTA, it will be flat. With spectrum, sloping down.

 

[Blumlein 88]

Thanks. I've been reading about pink vs white noise and your responses make sense with respect to pink noise. However, shouldn't white noise then have a similar SPL? Here is the same sweep with white noise. the graphs are virtually the same except for SPL.

View attachment 54921

No it will be flat with white noise like you saw, but still lower in level. Give me a minute and I'll write up a better example if Ray doesn't do it first.

 

[oohlou]

BTW, did you have a target curve for Dirac and what was it? Usually a flat one is too bright and thin sounding. You'd normally want Dirac to shoot for a target curve somewhere between the pink noise one and flat. A 9 or 10 db droop from low to 20 khz.

This was just their default with no adjustments. I know my personal preference would be more bass. I just want to make sure I am measuring correctly and understanding what I am measuring before I start tweak the curve.

 

[oohlou]

That is correct.
If you look at white noise using RTA, it will slope upwards. With spectrum, flat.

If you look at pink noise with RTA, it will be flat. With spectrum, sloping down.

Okay I am starting to understand. Measured results line up with your prediction. I clearly need to do some more reading....

(As an aside, what is funny is I was working with a customer earlier today about spectrum analyser bandwidth and averaging time. I understand the general concepts but this is my first time applying it to audio).

 

[Blumlein 88]

Ok, white noise is every frequency at more or less the same level all at once. So if I put a meter over it I could have 0 db for the entire signal from 0-20 khz. If I filter it into two bands, 0-10khz, and 10khz-20 khz, the filtered meter would show - 3db for each band. Combined they'd be back at 0 db so same signal level without the filtering. If I further filter it into 0-5 khz, 5khz-10khz, 10khz-15khz and 15khz-20khz, the each filtered band would read - 6db on my meter. This is what a spectrum analyzer or FFT does. It divides into many same sized bands, and measures each band separately. So the more FFT bands the lower the level in each narrower frequency band being measured. The total level of all bands with white noise is still 0 db, but it is necessarily lower within each smaller FFT bin (with FFT's each band is called a bin).

Now a swept tone is effectively just one single tone at any given instant. So it can be 0 db at a given instant over all the frequency band because at each given instant there is only one frequency in use. You aren't sub-dividing the energy into various bins. Instead with a sweep you are varying the frequency over time.

This also explains why pink noise slopes down with an FFT or spectrum analyzer. Pink noise has twice the energy in 0-10 khz as it does in 10-20khz. When you are using RTA split into fractions of an octave the filtered bands are of a different size. With pink noise 0-7 khz would have an equal amount of energy as 7-21 khz. The energy goes down with frequency, but your filter expands its bandwidth to even it out. Octave based filters and evenly spaced FFT filters are different in this way.

If this is unclear just ask about it, and hopefully I've not messed it up.

 

[oohlou]

@Blumlein 88 you are a scholar and a gentleman (or gentlewoman). That is an excellent explanation. I understand now. Time domain measurements have always made intrinsic sense to me but I get confused easily in the frequency domain.