Jump to content

Using REW to Find Parametric Equalizer (PEQ) settings


Chris A

Recommended Posts

1 hour ago, rjp said:

It shouldn't matter what the slope of this phase is as long as it's linear.

Yes--true for audio, too.

 

1 hour ago, rjp said:

You mentioned group delay above as a figure of merit. Is this simply the derivative of phase as a function of frequency?  If so then constant group delay is the goal I assume?

Yes--GD is the first derivative of phase.

 

But group delay audibility hasn't been established below 500 Hz.  Here is a table on the most referenced report on the subject of group delay audibility (Blauert & Laws):

 

Blauert and Laws GD Audibility Thresholds.PNG

 

There is some evidence that I've seen that GD audibility extends to bass frequencies (i.e., the audibility of port group delay in bass reflex cabinets), but no one has done the research to establish audibility threshold levels...just like no one has done the work to establish modulation distortion threshold levels vs. SPL and frequency.  We could certainly use the data, but none apparently exists. 

 

Absolute phase audibility has been repeatedly shown to be inaudible (linear phase), but rate of change of phase vs. frequency can be audible (i.e., group delay).

 

Chris

  • Like 1
Link to comment
Share on other sites

1 hour ago, Schu said:

didnt Klipsch, at some point, reverse phase on mid's and uppers (that used a passive network) in order to more seamlessly align time?

I'm not aware of the various passive crossover designs vs. frequency for the different Klipsch Heritage loudspeaker models.  All I've got are the measurements from the loudspeakers models that I own ('79 Cornwalls, '81 Belles, '78 Heresies, '07 Jubilees). 

 

If you change an electrical filter type or slope in an active or passive crossover network, or change the crossover frequencies, then virtually everything in the crossover phasing and phase-alignments can change. 

 

Chris

Link to comment
Share on other sites

2 minutes ago, Chris A said:

Looks like you need an additional 4 ms of delay, but you have to also look at what that's doing to the crossover region--most of this delay may be due to the crossover filters themselves.

 

Chris

So youre looking at the min. GD vs the  1/12 smoothed graph? And "by look at the crossover region" you mean the SPL graph? Should I redo the measure, but without active crossovers? Or do you need the crossovers on to see and calculate the delays? Also, if I'm correct I should apply PEQs before calculating the delays right? So measure with xover and PEQ, calculate the delay, and then do another sweep to see the influence on the crossover region? 

 

Also, on a side note. What is a good end result for a SPL graph? 6db over the whole region? Thats where I'm at right now (with PsychoAcoustic smoothing). I was at 9db before the PEQs. I also have a few PEQs over to make that 6db a bit narrower (there are 2 "big" dips/peaks I think I can effectively correct), but first I want to get the delay right. 

Link to comment
Share on other sites

I still have 2 huge peaks around the crossover point, but the difference has almost disappeared at the crossoverpoint itself. What is the (probable/plausible) cause of these peaks (@ 870hz and 421hz), and how would I take these on? 

 

image.png.77551636bfb5c48ffe28200df296db91.png

Link to comment
Share on other sites

Well just as an update. I now have the miniDSP 2x4 HD and the UMIC-1 and REW. I have to say that as an engineer myself, I am in heaven with all this adjustability and measurement. I have spent about 5 hours tweaking so far and so far haven't found any settings that sound good to me, but it is still a lot of fun.

 

I see each PEQ is actually implemented as a 5 coefficient IIR filter. How neat! And I see there is even an option for a FIR filter with up to 2048 taps. This is an amazing little box! One thing I wish it had is a way to bypass the entire PEQ set in a single click for A/B comparison rather than having to bypass each of the 10 PEQ's one at a time. Is there a way?

 

I have been reading the help documents on the miniDSP site and I am impressed with the technical detail they include. The article made it really clear why I should be interested in the slope of the excess phase (They call this excess group delay) as an indicator os which portions of the spectrum are likely to be "EQ-able" without making a mess of the phase response.

 

Honestly, this device and REW EQ are so technical I wonder how a non electrical engineer could even understand it well enough to use it. It is really easy to mess things up reall bad if you don't know what you are doing. (and I barely know what I am doing at this point) 

 

So at this point, one day into the fun, I would say I love this product and appreciate what it can do, but I can't seem to find a way to use it to improve the sound in my room.

 

I could use some pointers on how to proceed.

 

So far I tried two basic approaches:

 

(1) Measuring just one speaker at 1M and EQing for that and applying to both speakers. I have observed that the options for smoothing, freq range to treat,  and maximum individual gain make a huge difference in the results.

 

(2) MEasuring the "room" by taking 6 to 10 mic position measurements and averaging them, smoothing, and EQing.

 

(3) Just manually playing around with a single PEQ to find annoying frequencies and cut them out. This has been remarkably effective actually. The method I use is to generate a single PEQ with high gain and narrow BW and move it around in frequency until I find the spot that sounds the absolute worst, then change the gain to negative and Bam! it is gone. I find it is much easier to find what sounds bad than what sounds good. And I'm fairly used to this technique from working sound for bands back in the day so I guess it comes easier.

 

But I want to get the measurement method to work better. Any tips appreciated. Thanks!

 

 

 

 

 

 

 

Link to comment
Share on other sites

Got some pretty good sounding EQ settings last night. I decided to explore the "generalized room correction" approach a bit more.  I made a collection of measurements at 8 listening positions and averaged them together and then filtered at 1/12 octave. I then constructed 10 PEQs over full range to match this. I downloaded the 90 degree mic cal file for this and used the mic pointing at the ceiling. The results sound pretty good. Fixed some bass resonances I didn't even realize were there. I'd still like to soften up the mids a bit more.

 

Overall I found that I tend to get better results (to my ears)  with the PEQ's cutting rather than boosting. If I allow the PEQs to have unlimited positive gain and position the target response in the middle of the recorded data the predicted response is nearly perfectly flat, but to my ears it sounds a little better if I instead set the target response a little lower and restrict the PEQ gains to no more than +3dB (or even 0dB). I believe this forces the algorithm to do the best it can with cuts only. 

  • Like 2
Link to comment
Share on other sites

I do something similar with REW and a UMIK-1.  90-degree calibration.  Use REW's generator to produce a full-range Pink-PN while using REW's RTA and the UMIK-1 to analyze the Pink-PN signal while I do about 10-15 seconds of figure-8 patterns around my listening area.  After saving it down, I use REW's EQ (variable smoothing), feeding it general parameters to automatically get my SPL corrections in the right range.  But then I adjust from there - usually dialing the automatic corrections back and avoiding correction gains.  Basically, less is more and I'm just trying to correct some obvious room gains and very lightly apply a 'house sound.'  In my case, I then export the EQ as wav files which I feed into Roon's convolution filter.

  • Like 1
Link to comment
Share on other sites

  • 1 month later...

Just redid all the PEQ settings I had, and tried something different to measure the delays. I just used the built in "acoustical delay" function in the REW measurement tool. I redid the in/output settings in the processor  so the High frequency driver got the left signal, and the Low frequency driver the right. That way it calculated the delays as it would between 2 speakers. Only took me a minute or so to do, without having to change any cables. It also extremely simple to do, and I dont think you can make any mistake interpreting it as it just says "you need to apply .... ms delay". When I applied the new delays I got a totally new speaker! Damn, the voices and intelligibility improved so much! 

  • Like 1
Link to comment
Share on other sites

17 hours ago, Droogne said:

Just redid all the PEQ settings I had, and tried something different to measure the delays. I just used the built in "acoustical delay" function in the REW measurement tool. I redid the in/output settings in the processor  so the High frequency driver got the left signal, and the Low frequency driver the right. That way it calculated the delays as it would between 2 speakers. Only took me a minute or so to do, without having to change any cables. It also extremely simple to do, and I dont think you can make any mistake interpreting it as it just says "you need to apply .... ms delay". When I applied the new delays I got a totally new speaker! Damn, the voices and intelligibility improved so much! 

Dontcha love it when a plan comes together. It takes time to time align and all that other nasty stuff to make good noise from any speaker! LOL.

Link to comment
Share on other sites

  • 4 weeks later...
On 7/10/2018 at 5:10 PM, Schu said:

Don't you need to double blind in order you're not experiencing a bias.

 

😛 

 

That being said....a smeared image is very ugly.

Probably, but it's been proven that it can't hurt the sound, only improve it. There are other things to listen for that are more important to some ears.

Link to comment
Share on other sites

  • 2 months later...
On 5/22/2018 at 11:02 AM, rjp said:

Overall I found that I tend to get better results (to my ears)  with the PEQ's cutting rather than boosting. If I allow the PEQs to have unlimited positive gain and position the target response in the middle of the recorded data the predicted response is nearly perfectly flat, but to my ears it sounds a little better if I instead set the target response a little lower and restrict the PEQ gains to no more than +3dB (or even 0dB).

That's what I typically aim for, but there are many instances where this reduces the overall sensitivity too far for the preamp, etc. So I try to use attenuating PEQs as much as is reasonable and use higher channel gains to compensate.

  • Like 1
Link to comment
Share on other sites

On 5/22/2018 at 11:02 AM, rjp said:

Fixed some bass resonances I didn't even realize were there. I'd still like to soften up the mids a bit more.

There are different ways to see whether those resonances are minimum phase or excess phase by looking at the excess phase and group delay curves, and also by looking at the SPL plot at higher resolution to look for higher-Q resonances and nulls.  Non-minimum-phase resonances/nulls are usually found below the Schroeder frequency of the room.  (My room's Schroeder frequency is about 100 Hz.) 

 

Paying close attention to the absolute maximum deviations from average SPL and local deviations for the peaks, especially in the midrange to low-treble region, is really important to achieving the best sound quality from the setup.  This observation is also supported by Sean Olive's subjective loudspeaker assessment model quite strongly.  It's important to flatten out the frequency response as much as possible.  I've also used REW's capabilities to use a stepped-sine measurements over a much longer measurement period (~16 minutes in my latest measurements).  I've used this capability to verify the on-axis frequency response and have found a little fine tuning adjustment is possible using this method.

 

Chris

  • Like 1
Link to comment
Share on other sites

On 5/21/2018 at 4:34 PM, rjp said:

Honestly, this device and REW EQ are so technical I wonder how a non electrical engineer could even understand it well enough to use it. It is really easy to mess things up really bad if you don't know what you are doing. (and I barely know what I am doing at this point)

If this were true, then we'd also be walking and riding horses everywhere instead of riding in "complex machines".  One of my main skills has been taking complex problems and technical systems and reducing them down to much simpler problems and explanations so that many more people can understand how to use them or get the information they need.  That's the reason for this thread.  On being taken apart and examined in its various pieces, and a process described and used to employ REW's EQ facility, I find that it becomes very easy and intuitive--much like riding a bicycle. 

 

I strongly recommend taking measurements with the microphone at one metre from each loudspeaker under test, with the microphone height adjusted to approximately mid-height of the loudspeaker and the microphone itself pointed upward at 45 degrees from the horizontal.  Then using psychoacoustic smoothing of the data, you can set the average level of the target SPL and the desired flatness of response in dB, then use REW's EQ optimization routine so that it can iteratively find the minimum number of optimized PEQs to achieve your goal.  When the optimizer finishes it's calculations, you can observe the results and update any constraints to re-run the optimizer, until something reasonable is achieved.  The proposed PEQs can be turned on and off within the optimizer to see the results of not using certain PEQs, or you can manually change those PEQ settings to see the results on the overall frequency response. 

 

Chris

  • Like 2
  • Thanks 1
Link to comment
Share on other sites

  • 1 month later...
  • 10 months later...
On ‎11‎/‎28‎/‎2016 at 10:18 PM, Chris A said:

I recommend using "Psychoacoustic" smoothing before running the EQ filter optimizer, so that you're not trying to create a lot of high-Q (i.e.,very narrow) filters that are probably trying to correct for room modes. 

 

I see by entering numbers into my Xilica the smaller the number (BW Oct) the REW EQ wizard gives me, the larger the corresponding Q number Xilica puts in for that value. For example if REW gives you .15 the Xilica Q=9.613, likewise if you get a larger number form REW like .81 you get a Q=1.757.

 

Can you provide some general guidance as where to throw out a PEQ that the REW Wizard provides? I see the default Xilica value is .33Oct which equals a Q of 4.362.

 

Meaning how small is too small from REW? .25? .15? etc?

Link to comment
Share on other sites

Just now, rplace said:

Can you provide some general guidance as where to throw out a PEQ that the REW Wizard provides?...Meaning how small is too small from REW? .25? .15? etc?

In general, you can turn off the individual PEQs within the REW EQ facility to see the effect that each filter has on the overall frequency response (i.e., SPL flatness).  I usually go through and turn off all the individual filters, one at a time, while leaving all the other PEQs on, to see which PEQs are good, and which ones that I can omit or change manually. It gets a lot more complicated in the crossover interference bands, but in general you're using input channel PEQs (instead of output channel PEQs), so you see the combined effects of both ways...sets of drivers/horns...at the same time.  REW doesn't predict a perfect response in the interference bands even using only input channel PEQs, but it gets you close enough to iterate the PEQs on input channel and output channels to flatten the response successfully.

 

PEQs that are generated using very small values of bandwidth (BW) (conversely said, high values of "Q") are the first filters that I look at omitting.

 

Chris

  • Thanks 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...