REW-Good Looking Freq Response Graph Doesn't Tell The Whole Story - Random Thoughts/Findings & Lessions Learned

[cjf]

So I've been messing around with various Correction Procedure settings within Audiolense to see if I can tighten up the Step Response graph a bit more but there are difficulties trying to do it all within AL only via the simulations. Probably one of few complaints with AL I have is that it doesn't offer an Overlay feature between graphs. My current approach has been pretty caveman like in nature by making a few changes in CPD then generating a filter. I would then move over to the Step simulations and look at them. After Im done I take a screenshot of the plot so I can use it to compare another round of CPD changes. Rinse and repeat.

 

For anyone familiar with that you can imagine that its not very easy to see how the two plots compare with another, especially when viewing Step Response plots and making only .25-1.0 changes in values. The only way to really tell is to do another REW measurement and overlay them in there. Depending on your setup this may be easy enough but in my case I need to unhook all the connections from my Lynx Hlo and my Benchmark DAC3 then reconnect them to my Hapi DAC in order to get a real picture of how it will all look using my full-time DAC.

 

The Hilo and the DAC3 combo are used in order to have enough XLR output channels for my purposes. I use all of the TRS and XLR Analog Outputs on the Hilo then feed two more channels out thru the AES Output to the AES Input on the DAC3. The Hapi has enough channels but I don't have an AD Card in it so use the other two DAC's together for measurements only.

 

These are all just first world problems of course but they add quite a bit of time to the evaluation process.

[Jean Paul D]

32 minutes ago, cjf said:

So I've been messing around with various Correction Procedure settings within Audiolense to see if I can tighten up the Step Response graph a bit more but there are difficulties trying to do it all within AL only via the simulations. Probably one of few complaints with AL I have is that it doesn't offer an Overlay feature between graphs. My current approach has been pretty caveman like in nature by making a few changes in CPD then generating a filter. I would then move over to the Step simulations and look at them. After Im done I take a screenshot of the plot so I can use it to compare another round of CPD changes. Rinse and repeat.

 

For anyone familiar with that you can imagine that its not very easy to see how the two plots compare with another, especially when viewing Step Response plots and making only .25-1.0 changes in values. The only way to really tell is to do another REW measurement and overlay them in there. Depending on your setup this may be easy enough but in my case I need to unhook all the connections from my Lynx Hlo and my Benchmark DAC3 then reconnect them to my Hapi DAC in order to get a real picture of how it will all look using my full-time DAC.

 

The Hilo and the DAC3 combo are used in order to have enough XLR output channels for my purposes. I use all of the TRS and XLR Analog Outputs on the Hilo then feed two more channels out thru the AES Output to the AES Input on the DAC3. The Hapi has enough channels but I don't have an AD Card in it so use the other two DAC's together for measurements only.

 

These are all just first world problems of course but they add quite a bit of time to the evaluation process.

you can x your measurement by your filter in REW and that simulation is quite accurate 

[ecwl]

48 minutes ago, cjf said:

So I've been messing around with various Correction Procedure settings within Audiolense to see if I can tighten up the Step Response graph a bit more but there are difficulties trying to do it all within AL only via the simulations.

Sounds like you’ve pushed your system very far.

 

If you still have suspicion that you can push it further, you can always ask @mitchco or just hire him through his new service to setup the convolution filter for you. You already have all the latest measurements. You already know what your preferred frequency response is. So it’s just a matter of the optimal way to build the filter so I presume the cost would not be exorbitant.

 

That said, sometimes, part of the fun is learning and understanding more about convolution filters. On the other hand, sometimes, it’s just more fun to enjoy all kinds of music without the need to fiddle around with filters and doing repeated A/B testing.

[cjf]

1 hour ago, Jean Paul D said:

you can x your measurement by your filter in REW and that simulation is quite accurate 

Hello,

 

Would you mind expanding on how this is done? I'm not sure I fully understand what you mean.

 

Thanks

[Jean Paul D]

27 minutes ago, cjf said:

Hello,

 

Would you mind expanding on how this is done? I'm not sure I fully understand what you mean.

 

Thanks

 

[cjf]

1 hour ago, ecwl said:

Sounds like you’ve pushed your system very far.

 

If you still have suspicion that you can push it further, you can always ask @mitchco or just hire him through his new service to setup the convolution filter for you. You already have all the latest measurements. You already know what your preferred frequency response is. So it’s just a matter of the optimal way to build the filter so I presume the cost would not be exorbitant.

 

That said, sometimes, part of the fun is learning and understanding more about convolution filters. On the other hand, sometimes, it’s just more fun to enjoy all kinds of music without the need to fiddle around with filters and doing repeated A/B testing.

Hello,

 

Yes I've enjoyed reading about the information Mitchco has been generous enough to share around the web and agree, at least in my case, part of the fun is forcing myself to understand how all this works. At some point I may consider contacting him to see how much further I can go once a hack like me is done playing with it all.

 

Thus far though I've been quite pleased with the new Sub layout and filters I've managed to cobble together with this last round of changes I will be posting up. I'm sure they won't check all the boxes for what some might want to see in an ideal set of measurements but for my listening habits and music preferences it seems to work quite well IMO.

 

It seems to come together well at higher SPL levels and the Subs are blending in as if they aren't there and it sounds like you are just listening to much larger pair of main speakers with a bunch more power behind them. When the bass is there in the source material the Subs make themselves known but otherwise despite having a bunch more additional cone space pushing air around the room they are basically invisible and blend in nicely with the mains which was my ultimate goal.

 

When doing demos for buddies I know they are probably expecting to be vibrated out of the listening chair based on the number and size of the subs involved but it just doesn't sound that way at all unless the source has such low Freq content in it...which almost none of it does. What is quite obvious though is the scale and size of the sound of the venue which is hard to explain to those who are not exactly Audiophiles and don't know what to listen for. They know know it sounds good but don't necessarily understand why which is fine by me as long as it works for me, the primary listener 🙂

[cjf]

7 minutes ago, Jean Paul D said:

 

@Jean Paul D

 

Thanks for the info on that. How would you describe the level of detail in the documentation for REW for how to use this feature? I've never tried to use it or noticed it before so do you think the documentation is sufficiently detailed in describing how one can use this feature?

 

Thanks again

[cjf]

Hello,

 

This post is a follow-up to show how things are now measuring after moving the Subs around one last time and after dialing a bit more some of the Audiolense Filters/Corrections. Will call this last set of measurements "Config #3" and it will be compared to the previous "Config #2" which as shown before did not have as nice a Freq Response graph as the original "Config #1" but despite this, Config #2 did beat Config #1 in all other measured categories minus the Freq Response chart...IMO.

 

I'll dive right in with the comparisons but first I will show where the new Sub placements for Config #3 ended up and why I decided on this final placement.

 

S1 = Rythmik F18-SE

S2 & S3 = Rythmik E15HP2-SE's

 

 

Its not shown in the diagram above but there is a half wall that runs in between the placement of S2 & S3. This half wall is basically just providing a separation of space between the main hallway and the listening room. I've suspected for quite awhile that this wall was causing me all sorts of headaches in terms of room acoustics (in addition to the 12ft peaked ceiling above it). This time around I decided to place one of the Subs on each side of that half wall which I hadn't tried to do before. I hadn't tried it mostly because I felt it would have a negative effect on aesthetics and the livability of S2's location which is near the front door. But as it turns out, its not as bad as I expected it to be. I will need to relocate an AC Outlet because of S2's new location but that's not too horrible of a compromise IMO and I've already found a nice table that covers S2's location pretty nicely.

 

For those interested in a pretty cool looking table that can cover up a Sub the size of a Rythmik E15HP2-SE (and smaller of course) below is a link to the one I bought. I did perform before and after measurements with the table in place or not and it measured almost exactly the same except for an inconsequential difference way up in the Freq Range of the Sub long after its usefulness had already started to die off (250hz ++).

 

https://www.wayfair.com/furniture/pdp/gracie-oaks-tannenbaum-end-table-w000573359.html

 

 

Ok now onto the measurements:

 

Frequency Response - ALL CH Combined

 

Config # 2 = Blue Line (1/12th smoothing in effect)

Config # 3 = Pink Line (1/12th smoothing in effect)

 

 

Frequency Response - Left/Right Channels (ie..Ch Matching)

 

Config # 2 = Orange/Green Lines (1/12th smoothing in effect, Line values 1&2)

Config # 3 = Red/Purple Lines (1/12th smoothing in effect, Line values 4&5)

 

 

So comparing these first two charts the main difference is seen from the X-Over Point (60hz) and below where I was able to get Config #3 to measure a decent bit smoother than Config #2. Channel matching is pretty much a wash and both look OK to these eyes.

 

 

Distortion - ALL CH Combined (Cross-hair at largest peak @70hz)

 

Config # 2 = Blue Line (8.4% THD)

Config # 3 = Pink Line (1.4% THD) ..picked up a few percentage points reduction here with the new configuration and is pretty close everywhere else across the board to Config #2

 

 

 

Impulse Response - Left/Right CH

 

Config # 2 = Bottom Red/Green Lines (Line values 1&2)

Config # 3 = Top Orange/Green Lines (Line values 4&5)

 

 

For this comparison it appears both Config #2 & Config #3 are in Time Alignment with their L/R Channels. I'm not sure if one could be considered better than the other or not. The obvious signs of Pre-Ringing as seen in Config #1 are not seen here but there are a few more wiggles before the initial spike with Config #3 than with Config #2. My graph interpretation knowledge of Impulse Response measurements is a bit lacking here to say which one could be considered better.

 

Step Response - Left/Right CH's - Config #2

 

 

Step Response - Left/Right CH's - Config #3

 

 

For the Step Response comparison I was able to tighten up the Config #3 chart compared to Config #2 across the board. The channels appear to be better matched with Config #3 and as you can see the lines converge with each other at the far end of the graph unlike in Config #2 where they follow their own path separate from each other after the mid point

 

Group Delay - Left/Right & Combined CH's - Config #2

 

 

Group Delay - Left/Right & Combined CH's - Config #3

 

 

For the Group Delay comparison Config #3 shows a smoother series of lines compared to Config #2, especially from the X-Over point (60hz) and below.

 

So for these last few sets of measurement comparisons (Impulse/Step/Group Delay) the changes were dialed in by tweaking the Correction Procedure Designer settings in Audiolense in addition of course due to the new Sub placements.

 

The only difference between Config #2 & Config #3 CPD settings are that in Config #3 I have Disabled "Prevent Treble Boost" and increased the TTD Window at 10hz from "5.00" to "5.20".

 

 

Waterfall - All CH Combined - Config #2

 

 

Waterfall - All CH Combined - Config #3

 

 

Spectrogram- All CH Combined - Config #2

 

 

Spectrogram- All CH Combined - Config #3

 

 

So comparing the smoothness, coverage and discontinuities of these last series of graphs between Config #2 & Config #3 we can see a pretty big difference in the Waterfalls plots between the two. We can see far fewer peaks/valleys in the lower Freqs and a less overhang as well in those same lower Freqs.

 

The Spectrogram plots show another way to see the same smoothing effect between Config #2 & Config #3 but the Waterfall plot makes the differences much more obvious to see.

 

To bring a close to this post I want to circle back to where it all started where I made the statement that the Freq Response graph alone doesn't really tell the whole story. When looking at all the comparisons thus far Config #2 measured better than Config #1 in all categories besides Freq Response. And now, at least as far as my interpretations go of these last set of graphs between Config #2 & Config #3 it appears that the new Config #3 now betters Config #2 in almost all of those same measured areas.....but yet again, it does not appear to better Config #1 in terms of Freq Response as shown below:

 

Frequency Response - Left/Right Channels & Combined

 

Config # 1 = Top Brown/Teal/Purple Lines (1/12th smoothing in effect, Line values 4,5&6)

Config # 3 = Bottom Orange/Green/Blue Lines (1/12th smoothing in effect, Line values 1,2&3)

 

 

I guess its debatable depending on who you ask which Freq Response graph looks better but at first glance I suspect most people would choose Config #1 over Config #2 & Config #3 if they didn't look at any of the other measurements involved but the way I see it now, after going thru this whole exercise and analyzing all the data, a better way to chose between two or more sets of Convolution filters/corrections would be to pick the one that measures better in the most number of areas being analyzed with REW.

 

In this case we looked at 7 different sets of measurements. Config #1 wins in one out of seven categories while Config #3 wins in six out of seven categories. Based on my listening tests the configuration that has more "Wins" (Config #3) sounds the best in a very noticeable way when compared to the configuration that looks best in just one area.

 

I want to say Thanks again to all those who have contributed to this thread and I appreciated hearing the different insights and opinions on the information I have posted here.

[pompon]

Scaling bad ...

 

try 60 to 110dB unless your graph is not very precise.

[cjf]

4 hours ago, pompon said:

Scaling bad ...

 

try 60 to 110dB unless your graph is not very precise.

I can post a forensic view of the Freq Response plots tomorrow if that's what you mean?

 

I personally don't find them useful once they start looking like an upside down "U" just to squeeze the whole range into the zoomed view but understand others may feel differently.

[Bill Brown]

Nice progress, per the spectrogram you have really helped the 32 Hz mode, but perhaps worsened 82 and 130 hz?

 

Agree re. the scaling of your frequency response graphs.  I think you really need to look in closer.  A 230 db scale (-9db to 140db) is astoundingly high.  In the positive, it will help you identify problem areas, while the negative is that your frequency responses won't look nearly as good to the eye.

 

Bill

[cjf]

OK below are the forensic views of the Freq Resp graphs for Configs #1,2 & 3 individually as well as comparisons between Config1, 2 & 3 on a per channel and combined All Channel basis.

 

Individual Graphs:::

 

FreqResponse - Original Config #1 - RLCHs_PlusCombined

 

 

FreqResponse - Config #2 - RLCHs_PlusCombined

 

 

FreqResponse - Config #3 - RLCHs_PlusCombined

 

 

Comparison Graphs::::

 

FreqResponse - ALL Channels Combined

 

Config #1 = Blue Line

Config #2 = Purple Line

Config #3 = Pink Line

 

 

FreqResponse - Left Channel Only

 

Config #1 = Orange Line

Config #2 = Grey'ish/Green  Line

Config #3 = Aqua Line

 

 

FreqResponse - Right Channel Only

 

Config #1 = Lime Green Line

Config #2 = Brown  Line

Config #3 = Red Line

 

[cjf]

7 hours ago, Bill Brown said:

Nice progress, per the spectrogram you have really helped the 32 Hz mode, but perhaps worsened 82 and 130 hz?

 

Agree re. the scaling of your frequency response graphs.  I think you really need to look in closer.  A 230 db scale (-9db to 140db) is astoundingly high.  In the positive, it will help you identify problem areas, while the negative is that your frequency responses won't look nearly as good to the eye.

 

Bill

Thanks,

 

Yeah its pretty much like playing a game of whack-a-mole. A few wins here and there and a few losses elsewhere. I think at this point I will be living with Config #3 and whatever remaining warts that come with it 🙂

[Bill Brown]

Wow!  They still look very good with the 60 db scale.  Great.  Sometimes I have zoomed in even closer (on the FR and CSD graphs) to see the exact modes and their width, to design my PEQs.

 

But yes, maybe it is time to call it good :)

 

Bill

[cjf]

7 hours ago, Bill Brown said:

Wow!  They still look very good with the 60 db scale.  Great.  Sometimes I have zoomed in even closer (on the FR and CSD graphs) to see the exact modes and their width, to design my PEQs.

 

But yes, maybe it is time to call it good :)

 

Bill

Thanks Bill, yeah I'm throwing in the towel and getting back to the music...for now at least : )