Effect of micro-setting target level when EQing - a sound practice?

[klettermann]

After making some room changes I' m re-EQing my system. Tools are MiniDSP SHD Studio and REW, just focusing on the MLP. I discovered that very small changes in target level can have unexpectedly large impact on FR. As a result I've taken to fine tuning the target level for what seems to be the best result. See below. This series goes from a target level of 77dB (yellow) to 76.7dB (red) to 76.5 (blue). The scale is rather expanded, but even so there seems to be a real sweet spot between about 120-450Hz but with a tiny penalty from ~1.5 - 3KHz. That goal in this case was a pretty flat response with about a -3dB drop from 100-20kHz.

The question: are these changes enough to even bother with? Or audible? Or even real? It's kinda fun tweaking it, I must say. Comments from smarter minds welcome! Thanks and cheers,

 

[thcdru2k]

Are these different filters per target level?

 

[klettermann]

Same filters, the only difference is target level.

 

[thcdru2k]

That is interesting I would not expect such a different shape with such miniscule changes. I guess if it were me I would just try to see if the results are repeatable. If they are I would just pick my actual listening volume and tune for that

 

[Keith_W]

I assume that all 3 were taken with the microphone in the same position, and not moved between measurements?

If you want to see what is going on, click on "Generate measurement from filters". REW will show you exactly what the filters are doing. Very likely, it's just the lo-res nature of PEQ's combined with the fact that the MiniDSP does not have very many of them.

 

[kemmler3D]

This seems pretty odd, I haven't noticed the same phenomenon when I've done measurements. Is this with a UMIK? You definitely didn't move anything in the room (including yourself) between measurements?

Either something has gone wrong with the measurements, maybe a high background noise level, or something funky is going on in your room.

What does it look like unsmoothed, if you zoom in on some of the frequencies that changed a lot?

 

[Titurel]

I'm surprised no one mentioned other potential variables, e.g., changes in humidity, temperature, or where the OP is standing or sitting during each sweep. Unless the OP is running the tests while in another room, I would suspect number 3, the "Heisenberg writ large" answer as the most likely candidate, especially if the OP's position is relatively close to the mic, in effect, making the OP's body an uncontrolled variable element in the test.

Then there's room noise, i.e., noise leakage from outside and bio-generated noise from the OP (breathing).

 

[klettermann]

Yeah, it's a somewhat surprising result. These are all using the same original FR measrement. The ONLY difference is the target level in REW EQ. What's pictured is "Generate measurement from predicted" from the respective levels. I've seen this before. Say, for example, I'll start with "calculate target level from response" and start generating "measurements from predicted" in successive 1dB steps down. Usually there will be a point where there's a significant change, not just the same thing but attenuated 1dB. Then do more measurements interpolating between those. I've usually ended up with a response that's best of both worlds. That's what this is. Seems odd though. And I don't know if it makes a difference. I'm going to measure some corresponding EQd curves later. I will post the outcome.

Same thing zoomed @1/12 smoothing. I went though this several times and looks like it's repeatable. !!??!#?!

 

[klettermann]

And to repeat: these are NOT different measurements. They are the measurements generated by REW for the single original measurement. The only difference was the target level setting for running the EQ. So temperature, sunspots, ozone diffraction and tachyon entanglement aren't at play.

 

[kemmler3D]

And to repeat: these are NOT different measurements. They are the measurements generated by REW for the single original measurement. The only difference was the target level setting for running the EQ. So temperature, sunspots, ozone diffraction and tachyon entanglement aren't at play.

Oh, so these are the predicted responses from generating the EQ filters?

This isn't unexpected, the EQ filter optimizer has somewhat unpredictable outcomes depending on the settings and inputs. You can get similar swings in the curve by tweaking the number of filters, max / min boost, frequency range, flatness target, etc.

I would say you shouldn't think too much about the target level per se... I'm not sure what the best practice actually is for that, but I feel like you should set it to roughly match the smoothed level from 1000hz on up.

What I do after that is just run the EQ filter generator until I get a result with the smallest peaks / dips left over at the end.

It also looks like you're running EQ full-range, but that can be counteproductive, you don't want narrow filters above 300hz or so. Maybe try running it only below that?

 

[klettermann]

Oh, so these are the predicted responses from generating the EQ filters?

Exactly.

This isn't unexpected, the EQ filter optimizer has somewhat unpredictable outcomes depending on the settings and inputs. You can get similar swings in the curve by tweaking the number of filters, max / min boost, frequency range, flatness target, etc.

Right! But the only difference in this case are tiny changes to the target level

I would say you shouldn't think too much about the target level per se... what I do is just run the EQ filter generator until I get a result with the smallest peaks / dips left over at the end.

It also looks like you're running EQ full-range, but that can be counteproductive, you don't want narrow filters above 300hz or so. Maybe try running it only below that?

I am! I'm aware it's controversial or even ill advised. But I've tried it both ways and never had a worse audible result running full range and generally better. So I just do full range.

 

[kemmler3D]

Right! But the only difference in this case are tiny changes to the target level

I think (don't know) basically you're putting different parts of the response over or under some internal threshold it has for putting a filter at a given frequency. It's a quirk of the software and I don't think it has any further meaning regarding the actual response. Since there are only so many filters it can deploy, it uses an algorithm to make trade-offs, so in this case changing the target by 0.5dB forces it to consider a different part of the response as more important.

 

[janbth]

I am! I'm aware it's controversial or even ill advised. But I've tried it both ways and never had a worse audible result running full range and generally better. So I just do full range.

I would suggest using Var smoothing as input if you’re computing PEQ filters. You need high resolution in the bass in order to compute the correct Q values for the filters.

 

[klettermann]

I think (don't know) basically you're putting different parts of the response over or under some internal threshold it has for putting a filter at a given frequency. It's a quirk of the software and I don't think it has any further meaning regarding the actual response. Since there are only so many filters it can deploy, it uses an algorithm to make trade-offs, so in this case changing the target by 0.5dB forces it to consider a different part of the response as more important.

I think this is true. I tried all the biquads generated from different settings in the MiniDSP and measured those. Bottom line: despite tiny EQ variations they all sounded the same when measured through the system. So I guess, it's something like what you say. I not going to sweat it anymore. Thanks and cheers,

 

[Chris A]

Bottom line: despite tiny EQ variations they all sounded the same when measured through the system.

You might consider reading on psychoacoustic effects such as critical bands and other characteristics of the human hearing system. You have access to different types of smoothing available within REW and other acoustic measurement apps. because of these psychoacoustic characteristics.

Personally, I use psychoacoustic smoothing to more closely follow human hearing acuity. Trying to EQ to finer settings (e.g., 1/12 octave, etc.) will not result in better subjective amplitude response. I find the subjective results are either "no audible difference", or even worse subjectively due to scrambling the phase response of the setup from the use of more higher-Q EQ filters, thus affecting the time-based phase and time delay response of the setup.

______________________________________________________________________________________________________

You're also getting close to the amplitude response performance of the monitor loudspeakers used during mastering. Most mastering monitors and mastering room acoustics will result in greater differences in amplitude response than you might be trying to correct in your listening room. I find this is often one the areas that I run into when doing demastering.

At some point, it becomes more important to consider mastering rooms and monitors used for the exact recordings that you're listening to at the moment, since the mastering environment affects the subjective sound quality...sometimes quite strongly. The effects of using less-than-perfect mastering monitors on the subjective sound quality of produced recordings--such as the Yamaha NS-10M Studio, etc.--systemically affect how these recordings sound.

Chris

 

[thewas]

A good recommendation for smoothing for EQ is variable smoothing which comes from the REW manual itself, as it provides less smoothing and thus usually higher correction filter Q factors in the bass region and the opposite above.
Page 316 https://www.roomeqwizard.com/REWhelp.pdf

 

[Chris A]

Here's what the REW on-line documentation says (without having to search through the entire pdf manual):

Variable smoothing applies 1/48 octave below 100 Hz, 1/3 octave above 10 kHz and varies between 1/48 and 1/3 octave from 100 Hz to 10 kHz, reaching 1/6 octave at 1 kHz. Variable smoothing is recommended for responses that are to be equalised.

Psychoacoustic smoothing uses 1/3 octave below 100Hz, 1/6 octave above 1 kHz and varies from 1/3 octave to 1/6 octave between 100 Hz and 1 kHz. It also applies more weighting to peaks by using a cubic mean (cube root of the average of the cubed values) to produce a plot that more closely corresponds to the perceived frequency response.

ERB smoothing uses a variable smoothing bandwidth that corresponds to the ear's Equivalent Rectangular Bandwidth, which is (107.77f + 24.673) Hz, where f is in kHz. At low frequencies this gives heavy smoothing, about 1 octave at 50Hz, 1/2 octave at 100 Hz, 1/3 octave at 200 Hz then levelling out to approximately 1/6 octave above 1 kHz.

John Mulcahy has produced an excellent tool in his Room EQ Wizard (REW). That doesn't mean that he's expert over each area that is represented in its functionality. One of the areas that has received some constructive criticism in the past (i.e., when the online help text was written--now ~15 years ago) was from noted mastering engineer Bob Katz. The topic was on which smoothing to use for EQing in-room. Katz disagreed with the advice given, so now we have the three non-fixed smoothing types to choose from.

Bottom line:

If you are going to EQ bass below the room's Schroeder frequency (~150-200 Hz for the majority of listening room sizes), you should understand that this frequency region is dominated by standing wave room modes. Extremely small changes in the position of the loudspeakers relative to the room boundaries and most notably the position of the listener's ears within the room have strong variability of amplitude response that cannot be effectively EQed except for limiting peaks in response.

The problem with showing unsmoothed bass response is that the user is most often induced into trying to boost via EQ the room mode nulls that cannot be corrected (i.e., non-minimum-phase response).

If you realize all this and can resist the urge to apply PEQ filters in this area to try to boost non-minimum-phase response, how the bass response in smoothed really has little effect on what you can achieve in-room.

You really should be aiming for some sort of average which is produced with either psychoacoustic or ERB smoothing--unlike the on-line advice given in REW. The results will be much more useful in-room for any listening position--even the so-called "sweet spot" (head in a vise) position.

Chris

 

[thewas]

The problem with showing unsmoothed bass response is that the user is most often induced into trying to boost via EQ the room mode nulls that cannot be corrected (i.e., non-minimum-phase response).

That would be an user error, with higher resolution bass modes which are usually minimum phase can be better attenuated and of course not narrow dips should be filled as they are usually not.

 

[janbth]

Here's what the REW on-line documentation says (without having to search through the entire pdf manual):



John Mulcahy has produced an excellent tool in his Room EQ Wizard (REW). That doesn't mean that he's expert over each area that is represented in its functionality. One of the areas that has received some constructive criticism in the past (i.e., when the online help text was written--now ~15 years ago) was from noted mastering engineer Bob Katz. The topic was on which smoothing to use for EQing in-room. Katz disagreed with the advice given, so now we have the three non-fixed smoothing types to choose from.

Bottom line:

If you are going to EQ bass below the room's Schroeder frequency (~150-200 Hz for the majority of listening room sizes), you should understand that this frequency region is dominated by standing wave room modes. Extremely small changes in the position of the loudspeakers relative to the room boundaries and most notably the position of the listener's ears within the room have strong variability of amplitude response that cannot be effectively EQed except for limiting peaks in response.

The problem with showing unsmoothed bass response is that the user is most often induced into trying to boost via EQ the room mode nulls that cannot be corrected (i.e., non-minimum-phase response).

If you realize all this and can resist the urge to apply PEQ filters in this area to try to boost non-minimum-phase response, how the bass response in smoothed really has little effect on what you can achieve in-room.

You really should be aiming for some sort of average which is produced with either psychoacoustic or ERB smoothing--unlike the on-line advice given in REW. The results will be much more useful in-room for any listening position--even the so-called "sweet spot" (head in a vise) position.

Chris

No. What we are talking about here is EQing down a peak from a single room mode with a single PEQ filter. If you smooth the frequency response with ERB smoothing you will smear out the peak almost completely. ERB smoothing is very broadband in the bass region. If you think you have heard an improvement, you have probably heard the slight depression of the bass region as a whole, which may subjectively sound slightly better. You have not addressed the underlying problem, which is the physical phenomenon of a resonance - characterized a by a Q value.

Here is a slide from a presentation by Floyd Toole, that was recently posted in another thread. The figure is also shown in Sound Reproduction. (I don't have it here, so I can't provide the number of the figure.) It illustrates the problem nicely.

Do you have a reference to Katz?

 

[Chris A]

Do you have a reference to Katz?

See Feature Request: Frequency Dependent Windowing

No. What we are talking about here is EQing down a peak from a single room mode with a single PEQ filter.

So was I... The problem isn't there, however.

The 1/3 octave smoothing used in psychoacoustic smoothing in REW is just about right to avoid beginners trying to boost room mode nulls below Schroeder frequency.

You can see the non-minimum phase areas of room response by looking at the excess group delay plot. Wherever that curve is non-flat, you've got non-minimum phase problems. (ref: D'Appolito Testing Loudspeakers).

For instance, here's an excess group delay plot from a 1st-gen Jubilee with microphone distance of 1m:

The spike in group delay occurs at the crossover between the HF and bass bin due to their vertical separation distance.

And the same thing for a K-402-MEH in the same room at the same microphone distance:

Any group delay values above ~1ms (above 400 Hz) are audible. Below that group delay audibility threshold appears to be ~1/f (the period of the sound wave).

Chris