Weird auto equalization issue

[dima1stg]

In this particular case "match response to target" leaves area above about 11Khz unaffected, leaving almost 5db gain, despite having 2 unused filters, but there is one generated 15.5khz filter but it has 0 gain. Tried various target level values with no effect. I've seen this behaviour before but very small tweaking of target level fixed it then, but not now. Please see attached.

 

[John Mulcahy]

Hard to say anything meaningful without seeing the measurement. The other filters could have been used initially but ended up being discarded during optimisation.

 

[dima1stg]

Sorry, forgot to attach mdat but it's no longer exists. Will post next time when I see this behaviour.

 

[dima1stg]

Would like to return to this subject.
Attached are mdat and few pictures of equalization results. Images named "...calc target level from response" are equalization based on suggested target level of response. As you could see, in both area around 15.5khz is not corrected even though the gain is 4.5db and there are bunch of available filters that could take care of this. On another hand, having target level just 0.2db higher makes this area corrected.
The issue I started this conversation awhile ago is that undr certain circunstances areas with large gains are not being corrected.

 

[John Mulcahy]

Why did you change the mdat extension to .txt?

Looks like you are using a house curve, please attach that file.

 

[John Mulcahy]

You have Vary max Q above 200 Hz selected, which means:

the Vary max Q above 200 Hz option tells REW to adjust the maximum Q from 10.0 at 200 Hz to 3.0 at or above 10 kHz

A max Q of 3 is likely too broad for that 15 kHz peak.

 

[dima1stg]

Why did you change the mdat extension to .txt?

In a past, I was advised to do this so that file can be attached. Apparently, it's no longer a case.

I unchecked "Vary max Q above 200hz" but it didn't help.
House curve is attached.

On a side note, I'd rename "Vary max Q above 200hz" to "Variably limit max Q above 200hz".

 

[John Mulcahy]

I'm surprised it works at all. The entire response from just above 700 Hz is above the target. You stopped the measurement at 20 kHz, which is also the end of the match range, so at target levels below 85.3 dB the end of the measurement is also above the target. The core of the EQ algorithm deals with the enclosed areas where the measurement crosses the target, the underlying expectation is that the target is within the measurement, not below or above it.

 

[dima1stg]

I was not aware of this, John. My understanding was, and it is actually happening, is that if target crosses the measurement then those areas below the target will get even deeper below with neighborhoud regions "helping" this to happen. My listening area is as bad acoustically as it could get, there are spikes and deeps of high magnitude with high Q, so correcting peaks down bring neighbor deeps, which are already down the curve, even more down. I've seen great many times that, when using suggested "...calc target level from response", areas initially at 1-3db below the target become 4-6db or even more below the target. So what I do is I experiment with target level in such a way so that resulting output becomes no worse than 2-3db below the target, however this comes at the expense of having target lower and lower total SPL. Is this a wrong way of doing things? What methodology would you suggest?

 

[John Mulcahy]

if target crosses the measurement then those areas below the target will get even deeper below with neighborhoud regions "helping" this to happen

That's at least in part because you have set the limit for the individual max boost for filters to just 1 dB, so REW has extremely limited scope to use filters with gain to counter the effect of cuts outside the region they are needed.

when using suggested "...calc target level from response", areas initially at 1-3db below the target become 4-6db or even more below the target

Yes, that's normal, why does it matter? Setting a very low target so the response is all dragged down onto it is, in my view, a very bad approach. However, I also think it's a bad idea to apply filters above a few hundred Hz and that trying to impose a target curve on frequencies above a few hundred Hz is wrong and based on a misunderstanding of Harman's curve. Other people have different opinions.

 

[dima1stg]

If you don't mind, I'd like to ask few things.
1. Why, in your opinion, 4-6 or more db deeps aren't matter.
2. I limit max gain to 1db to not math-overflow DAC. Do you think the limit could be higher, at least in higher frequencies?
3. My understanding of Harman curve is (a) it is experimentally most preferred in blind comparison and (b) it's related to our hearing lost sensitivity in lower frequencies. I listened to various curves and Harmon' is indeed best suits my preference. Though I do agree that above few hundred hz it's much less relevant if relevant at all. In your professional opinion, is my understanding incorrect?

I'm aware of different opinions, however I'm asking about your personal thoughts as very respectful audio specialist.

 

[John Mulcahy]

We are much more sensitive to peaks in the response than to dips. Dips in the response are typically a result of partial cancellations from signals travelling different paths. They are very location-specific, even a small movement of the measuring position can shift them or turn them into peaks. Attempts to fill them would be just as location specific.

Your point (b) is not correct. A bass guitar sounds like a bass guitar. The bass player doesn't have to stand closer to you to allow for your ear's sensitivity. If you alter the frequency balance of that sound you are making it sound like something else.

The overall spectral balance Harman identified as preferable should not be considered in isolation. Another component of that blind comparison preference was for speakers whose direct response is largely flat. The overall balance is a consequence of the direct on-axis sound from the speaker and all the off-axis sound it produces reflecting from the room's surfaces and arriving at the mic position over time. Off-axis sound has less treble energy and room surfaces have higher absorption at higher frequencies, so the overall result has a downward slope. The nature of that slope depends on the polar response of the speaker and the room's surfaces. It is a consequence, not a target. If you alter a speaker's output to try and achieve the shape of that Harman curve you are also altering the direct sound from the speaker, so it no longer has its original sound. You compromise the part your ear can most easily distinguish to try and emulate something which is (a) less important and (b) a consequence of speaker design (off-axis behaviour) and listening environment. That's a mistake.

 

[dima1stg]

John, very much appreciate. Many thanks!