Question on Wayne's Mimimal EQ post

[mstailey]

Based on Wayne's Minimal EQ and Target levels post I've an example per that doctrine and wanted to get opinions if I'm following it and more importantly getting the expected results.

From Wayne's post:

Optimize REW filters for response smoothing
First, REW calibrate for the standard 75 dB Target in the Settings panel. Take your measurement sweep, then adjust your viewing window for a 20-130 dB setting, using the Graph Limits box (top right of screen), so that your response curve doesn’t look so scary.

Next, re-adjust the Target Level to a good midway point between your response peaks and depressions - or if possible, align the Target to where it most closely tracks the bulk of your response curve.

Under Trace Adjustments, set the Smoothing option to 1/3-octave, which will help induce REW to focus more on the most serious problems instead of the minor ones. Then under Filter Tasks, limit the Find Peaks range to a bit below your crossover point. The response you’re seeing close to the crossover frequency is most likely duplicated by the mains, so it’s to no effect to equalize up there. (Of course, this needs to be verified by comparing graphs for sub only, mains only, and both together.)

The graph showed (sub only, axis and target line set with 1/3 smoothing per Wayne's recomendation) reports no peaks found when I select the find peaks option.

Can this be correct? Would you add some gain filters around where the graph dips a little?

BTW - the sound is not bad except it seems a little bass heavy - a little more than I care for when playing Wayne's recomended Basia CD as a test.


Thanks,
Mike

[brucek]

The graph showed (sub only, axis and target line set with 1/3 smoothing per Wayne's recomendation) reports no peaks found when I select the find peaks option.

Can this be correct?

1- Set the vertical axis to 45dB-105dB.

2- Turn off smoothing. Never use smoothing on low frequency measurements. Filters optimised against a smoothed response will have settings that don't accurately match the room's modes.

3 - Set your target to 75dB.

4- Find Peaks............. you'll be happy.

It's not a big secret that Wayne and I disagree on this subject.

brucek

[mstailey]

brucek wrote:

1- Set the vertical axis to 45dB-105dB.

2- Turn off smoothing. Never use smoothing on low frequency measurements. Filters optimised against a smoothed response will have settings that don't accurately match the room's modes.

3 - Set your target to 75dB.

4- Find Peaks............. you'll be happy.

It's not a big secret that Wayne and I disagree on this subject.

Okay, I did what you suggested. I'd say no kidding that you and Wayne disagree. This is very frustrating

[brucek]

This is very frustrating

It's a complicated subject. People will disagree. No big deal.

Make sure that you remember to run the Gain and Optimization routine on those filters REW finds.

I don't happen to agree with using smoothing before finding filters. John (author of REW) has explained this at length, and I buy into the explaination. I provide some support for my argument here with illustrations. Read point #2 on smoothing. Smoothing shifts the center frequency and Q of the filters and so they will not be as effective in removing room modes.

The vertical axis is important to leave as I indicated so you can get a better look at where the best target level should be set to provide reasonable filters. A wide axis may make you feel better, but it hides the truth too. Once you have the filters, I wouldn't worry too much above 100Hz, since the filters won't have much effect given the crossover reduction to the sub signal. Probably filters 11, 12, 7 could be removed from your chart. Remember to run Gain and Optimize first though. Note there is a button on the filter popup that sorts filters from low to high.....

brucek

[mstailey]

Fair enough brucek. Then let me ask the question - what do you think of the base graph? From your experience do you see anything I might do before turning to electronics or is this a good (graph) jumping off point to start with EQ?

Thanks for taking time to look / comment on this.

Mike.

[Wayne A. Pflughaupt]

Hi Mike,

Twelve filters - as you can see, REW wants to filter your graph to death. Not a good tact, in the experience of many people here.

The graph showed (sub only, axis and target line set with 1/3 smoothing per Wayne's recomendation) reports no peaks found when I select the find peaks option.

Can this be correct? Would you add some gain filters around where the graph dips a little?

Take another look at your first graph - the reason REW didn’t find any peaks is because your response plot isbelow the Target Curve. REW does not look for peaks below the Target Curve.

You’ll need to realign the Target Curve so that it’s a good midway point between your peaks and valleys (75 dB looks too low, based on your second graph). As noted, REW won’t do anything about what’s below the Target, so you’ll need to use some manual boosted filters to address what’s below the Target. In the “A few examples” heading of the Minimal EQ article (right below the one you highlighted in the quote box) I show that you can do some manual tweaking of the filters REW recommends to eliminate and/or re-adjust any notch filters or useless filters.

Double check your smoothing setting – it looks like way more than 1/3-octave. You might need to bump it up to 1/6-octave if indeed it was 1/3-octave before.

The window (20-130 dB vs. 45-105 dB) has no effect on REW’s automatic equalizing; that is merely for your benefit when you’re manually equalizing, so that the response doesn’t look as scary and compel you to over-equalize (I think I need to clarify that in that section of the article). As brucek noted, be sure and switch back to the 45-105 dB window before posting graphs here.

Regards,
Wayne

[cixelsid]

IMO the generic assumption that fewer filters is 'always' better misses the point.

High 'Q' filters will raise the Qtc of the system, low 'Q' filters have significantly less impact.

So it's quite possible that a lower number of filters with a high 'Q' will be more deleterious to sound quality than greater number of filters with a low 'Q'.

[Wayne A. Pflughaupt]

And indeed, why notch filters should be avoided is fully discussed in the second post of the article...

From what I’ve seen the two go hand in hand. All things being equal, the use of broad filters will generally require fewer filters to fully equalize a response curve. By comparison, notch filters can only move small “chunks” of graph “real estate,” as it were, so more of them will be required.

Regards,
Wayne

[lcaillo]

Having worked with filtering on a number of complex signal types, audio, video, and biological signals, I can say that I have learned from all of them that less is often more when filtering. You want to do just enough to get what you need and no more. You will chase your tail and create more problems than you solve.

[mstailey]

lcaillo wrote:

Having worked with filtering on a number of complex signal types, audio, video, and biological signals, I can say that I have learned from all of them that less is often more when filtering. You want to do just enough to get what you need and no more. You will chase your tail and create more problems than you solve.

So, if that is the case then putting in the number of filters REW recomends should be just enough. Agreed?

If the program states to put in x number of filters and you ignore the recomendations of the program than what's the point to even using the program?

I would really like JohnM, the author, to weigh in on this. After all he is the person who put the program together (along with all the other many contributions from here as well).

Mike

[stevenkelby]

How do you decide what to aim for when deciding whether to filter or not? Do you draw an imaginary line at say, 3db or something and if it's less than that, leave it, if more, EQ it?

[brucek]

than what's the point to even using the program?

Depending on the target you select (and that can be automatic or set by you), the program will offer more of less filters. You have the option to decide. You may also decide the bandwidth you wish REW to recommend filters for. Hopefully no one wants the option to decide taken away?

brucek

[JohnM]

Currently the REW filter recommendations are aimed at modal resonances. REW examines the response, identifies features that have the characteristics of modal resonances and calculates filters to counter them. The filter recommendations depend on the target level, as only features above the target are considered for filtering. The identification process is not foolproof, REW may miss some features that should be countered or it may apply a filter where it would be more correct to have 2 separate filters or no filter at all. It also does not apply any filters to shape the response to a desired overall trend. I'm working on improvements to the process of choosing and optimising filters, but there needs to be an element of user judgement applied to the REW recommendations. REW also does not currently consider any audibility criteria for the resonances it detects - some research indicates even prolonged resonances at low frequencies have little audibility, see the papers on the Genelec web site for example.

[DrWho]

I think the biggest thing missing here is an actual interpretation of the measurement. And just to strike a contrast - we have lots of people interpreting lines.

I suppose one might try to argue that there is some kind of correlation between the lines and what we hear, but there is so much more going on than just a single line going across the screen. I can think of a few instances where similar looking lines can sound very different...especially when you start introducing smoothing.

I think it makes a lot more sense to become familiar with the actual real world acoustic behavior and then learn how those real behaviors might reveal themselves in the measurements. And with REW, we have more than just one measurement to look at, so interpretations can take all perspectives into account.

So basically, a measurement should be seen as a tool to aid the identification of problems and then provide verification of the solutions.

Instead of discussing "number of filters" and "the shape of the curve" or changing the limits of graphs to "make the curve less scary", why not start asking questions about what is causing what you see? And then from there, we can discuss possible ways of addressing the real world problem - not ways of making the line "prettier".

[Blaser]

The objective is not making the line prettier.

What I understand is there are 2 theories:

1- make the job for the main problem area and don't bother too much about details (apply 1/3rd octave smoothing). Not the very best sound but easier and close enough.

2- OR, Do not apply any smoothing to have better results though more filters will be applied.

I think this is the summary of what has been discussed with an opinion on changing the vertical scale for doing manual tweeks (this won't change anything to the result)

I also believe the reasons for graphs anomalies are generally known, and if they aren't, that won't change a lot as far as the REW ajustment procedures are followed correctly.

And yes, in the end you will have a relatively "better" (or closer to what is initially desired) sound from your sub in a given situation if you have better equalized looking curves (the graphs don't lie).

Of course there could be some variation in sound between using smoothing or not, but when compared smoothing won't be too bad (if discernable) as it makes equalization not address the irrelevant (too small) "anomalies" specially when it is used for movies when only an "acceptable" boom is required.