The window needs to be frequency dependant and 'early' in this context is relative to frequency, short window at high frequency, longer at low frequency. I was trying to make the distinction between the common 500ms in room measurement window.
Mark,
You know i'm in agreement with you about the complimentary filter and i share(d?) your pov about global phase linearisation.
That said how would you classified what D.Gunness do on Fulcrum's coax? It's not complementary and despite the loudspeakers i've heard using the licenced tech ( Presonus's Scepter) sounded very good to me.
As well Mitchba used non complementary treatment to correct a low freq early reflection in his own home setup to great success it seems ( it's documented in his ebook about aAcourate but you can find the same in his online review of Audiolense).
I have still to try it at home though ( i face more or less same issue as his but less severe -room assymetry). But it seems to me it could help solve real potential issue here ( ok it's room correction territory rather than loudspeaker but still).
Start with 'fine tuning tips and tricks':
https://audiophilestyle.com/ca/ca-a...nd-room-correction-software-walkthrough-r682/
You know i'm in agreement with you about the complimentary filter and i share(d?) your pov about global phase linearisation.
That said how would you classified what D.Gunness do on Fulcrum's coax? It's not complementary and despite the loudspeakers i've heard using the licenced tech ( Presonus's Scepter) sounded very good to me.
As well Mitchba used non complementary treatment to correct a low freq early reflection in his own home setup to great success it seems ( it's documented in his ebook about aAcourate but you can find the same in his online review of Audiolense).
I have still to try it at home though ( i face more or less same issue as his but less severe -room assymetry). But it seems to me it could help solve real potential issue here ( ok it's room correction territory rather than loudspeaker but still).
Start with 'fine tuning tips and tricks':
https://audiophilestyle.com/ca/ca-a...nd-room-correction-software-walkthrough-r682/
Last edited:
Hi krivium,
You raise two good examples that contrast what I've learned to do, and give me constant thoughts/explorations.
I guess, first i need to emphasize, everything I think I've learned (and advocate) is about quasi-anechoic speaker tuning...no room at all.
And particularly about how to do that easily.......and as technically excellent as possible.
With no delay (latency) constraints, so far I haven't found anything easier, or giving better measurement results both on and off-axis, than steep complementary linear-phase xovers.
Fulcrum however, doesn't have the luxury of ignoring latency.
The FIR filters they employ are only 384 taps @ 48kHz. (that's what they are in the Q-sys presets they provide, I assume 384 taps is what they use for all the processing platforms they support)
If used linear-phase, a 384 tap filter gives only 250 Hz resolution. I have noticed, looking at the FIR filters in Fulcrums Q-Sys containers (which also include IIR xovers and parametric EQs) that impulse is typically moved towards start to gain frequency resolution, but still, even moved all the way to start gives only 125Hz rez.
So really, with such low frequency rez, the FIR filter is mainly for CD correction, imo....ala CD temporal EQ.
I think when time domain correction is done within a driver's passband; when there is little natural magnitude rolloff or IIR xover phase rotation to deal with, pre-ring potential may not even exist. (How I wish sometimes, Dave G would pop up and explain 😀)
Yep, I've got his book, and study the Acourate and Audiolense forums. Those guys are so way ahead of me, in terms of what can and can't be done in a room.
(i'm still just trying to perfect my knowledge anechoic lol)
That said, I do know from trying, how well room corrections can be made to any specific location. But specific location is the rub for me....
I just continue to feel I'm putting the cart ahead of the horse, if I make room corrections before I totally understand what constitutes valid time domain corrections.
Going back to Fulcrum, for example...their using TEQ within a CD's passband. Whatever time domain corrections are being made for the CD/horn, they are affecting all radiated angles of output.
Seems to me, the key to how valid the corrections are is how uniformly they hold up over how large an area. And I know how to do that quasi-anechoically, for the CD's in my builds. (or any other driver section)
I need to find a mental-picture equivalent of how to do that indoors with reflections, that has separated correctable from uncorrectable, in both freq and time domains. (I'm just one of those types, who when taught the Pythagorean theorem, who can't help but stop and figure out how he came up with it.)
That's the one I posted the snip from...the current/existing model. pg 119.
My guess is that generation of products has about 5000 taps @ 96kHz capability. That's all really...
The 48dB to 300dB per octave lin-phase xovers specification is true, but quite misleading imo....
Looking forward to the new model specs....
Hi Mark.
Yes both examples challenge the use i have of FIR too.
I don't think Fulcrum does correction to the Compression Driver but too the woofer: being Altec style coax ( with protubering horn) it will give some reflections for the higher part of the woofer bandpass.
Altec/Urei used some absorbing material on the outside of their horn to attenuate the effect on the 'time aligned' monitor i've seen( the blue 'mantaray' one).
I can't remember if Scepter had it too but it won't surprise me if it is present too.
Anyway i think the Teq is there for that reason: taking care of the last remaining bit.
I think it is the reason behind Fulcrum range of coax is claimed to sound the same whatever the size of driver.
Regarding Mitchba example i think it works on a relatively large area ( couch size) because it is past the minimum distance for the whole system to sum as one wavefront.
We discussed about it on Camplo's thread iirc, when i talked about Dunlavy's loudspeaker...
Yes latency... i've got same limitation in Lake that with Deqx: 25ms latency max define the max steepness of slope (there is no choice in 4way mode, the algo make choice vor you, steepness is imposed : 40db/octave at 100hz, past that it's 80db range until 4khz where 100 and up are availlable).
I think this is why the unit was so praised: engineers limited to less audible artefact possible imho.
Yes both examples challenge the use i have of FIR too.
I don't think Fulcrum does correction to the Compression Driver but too the woofer: being Altec style coax ( with protubering horn) it will give some reflections for the higher part of the woofer bandpass.
Altec/Urei used some absorbing material on the outside of their horn to attenuate the effect on the 'time aligned' monitor i've seen( the blue 'mantaray' one).
I can't remember if Scepter had it too but it won't surprise me if it is present too.
Anyway i think the Teq is there for that reason: taking care of the last remaining bit.
I think it is the reason behind Fulcrum range of coax is claimed to sound the same whatever the size of driver.
Regarding Mitchba example i think it works on a relatively large area ( couch size) because it is past the minimum distance for the whole system to sum as one wavefront.
We discussed about it on Camplo's thread iirc, when i talked about Dunlavy's loudspeaker...
Yes latency... i've got same limitation in Lake that with Deqx: 25ms latency max define the max steepness of slope (there is no choice in 4way mode, the algo make choice vor you, steepness is imposed : 40db/octave at 100hz, past that it's 80db range until 4khz where 100 and up are availlable).
I think this is why the unit was so praised: engineers limited to less audible artefact possible imho.
Yep, some Fulcrum q-sys presets show FIR filters on both CD and woofer coax section, but many don't.
Here's the FIR files used on the RM28, which does have both.
Top is high section; bottom low.
You can see the xover implementation is IIR replication, not linear-phase (which is of course isn't possible with a very low latency design)
Both of these files have the impulse peak at around 1-1.1ms, or about 100 taps in.
Leaving 284 after impulse peak, for a frequency rez of about 170Hz. (Greg might laugh at my simplified take on freq rez, but it works for me!)
Anyway, I just can't see that low a frequency rez can have a huge effect on the woofer's response, other than some mild shaping and providing a smooth low pass. Quite gentle !
That manual doesn’t have the new product model on it , that’s why I sent both links
It’s the old manual , look at specs on new one tho
4096 taps at 96k is the adsp2148x (I have that sharc in my minidsp) this is the next sharc
The adsp2159x has much more ability
And has gpu. So it should do way more
Yes, but how do you get "early" at low frequency. Or, perhaps I should ask "what do you mean by early at low frequency"?
In the modal region a room is basically a lot of oscillators and what you hear vs time is the turn on and rise to steady state of those oscillators followed by their decay in response to the driving function (source). If you assume that it takes X cycles for a mode to be fully excited, then if T is the length of the window, any mode with frequency higher than X/T will be fully excited in the measurement. In my room X is about 7. Thus if I want to measure the response down to 10 Hz, any mode over 70 Hz will be fully excited. Any mode below that will developing. You can see this but taking low frequency measurement with increasing window length, as below. Notice in particular how the violet curve (44 msec) agrees well with the yellow (88msec) and white (166 msec) curves above about 170 Hz but deviates significantly below that. Additionally, X, itself, will vary with frequency due to how the room damps vs frequency.
(Sorry for the hard to read figure. My measurment system is so old I have no way to transfer the images to my laptop other than take a photo.)
Last edited:
But is obsolete, just like all the DEQX's appear to be at the moment. They have some nice renders of the new versions but using CAD pictures suggests it is currently vapourware.
Apologies for the vagueness but there is no defined answer I can give, this is the element of alchemy and experimentation needed, each speaker and room wants something different. I am really just trying to make it clear that the time period needs to change with frequency so that at mid to high frequencies the window is short so that there is very little room in the measurement and as frequency goes down the window should get bigger. Exactly how much varies in practice.
The software I use separates the minimum phase and excess phase components and they can be windowed differently to each other. There is a window size at low frequency one at high frequency and an exponent that determines how the curve moves between the two. An exponent just below 1.0 seems to work well, but the difference between 2 decimal places on the value can be heard so getting it just right takes a bit of time and testing. I use different exponent and window lengths for the MP and EP options.
This image might be helpful
Minus the math , that’s really good and also the way I understand it…. (As much As i can understand at least)
Yes LF is spread out over time too much to have a “early” …..which is also why centering works later in time….
Thank you that is very good info. I definitely learned by this post.
For me, the following worked, your miles may vary:
Frequency dependant windows use a number of cycles per frequency that can be user adjustible.
It took me quite a bit of time to find a good balance of using the least amount of cycles that still had a desirable effect. (About 3 cycles at bass frequencies was enough for phase manipulation). After finding that particular window, there's not much change in phase over time within the room accross a pretty wide listening area, lots of controll measurements. Not searching for a one spot trick.
These corrections are at frequencies of 200 Hz and down. Above it, minimum phase is all you'd need. I did make sure to have a ~ 20 ms window with low levels of early reflections (room treatment at first reflection area's). That is for the benefit of the 200 Hz + frequencies. I showed the resulting wavelet earlier. Consisting of a Stereo measurement with mains + subs.
I use the same package @fluid mentions and I did not get it right in one go. These days I even do it a little different. Even more manual labour as I use minimum phase correction and manually correct phase using RePhase. Still using the frequency dependant windows, with a sliding window as the tool, but separating that job from the automated functions. It simply gives me more control, yet it does take time (and practise). The difference with not doing this isn't small or hardly noticable.
It does help to view the speaker + room as a system that needs to be able to work together.
Frequency dependant windows use a number of cycles per frequency that can be user adjustible.
It took me quite a bit of time to find a good balance of using the least amount of cycles that still had a desirable effect. (About 3 cycles at bass frequencies was enough for phase manipulation). After finding that particular window, there's not much change in phase over time within the room accross a pretty wide listening area, lots of controll measurements. Not searching for a one spot trick.
These corrections are at frequencies of 200 Hz and down. Above it, minimum phase is all you'd need. I did make sure to have a ~ 20 ms window with low levels of early reflections (room treatment at first reflection area's). That is for the benefit of the 200 Hz + frequencies. I showed the resulting wavelet earlier. Consisting of a Stereo measurement with mains + subs.
I use the same package @fluid mentions and I did not get it right in one go. These days I even do it a little different. Even more manual labour as I use minimum phase correction and manually correct phase using RePhase. Still using the frequency dependant windows, with a sliding window as the tool, but separating that job from the automated functions. It simply gives me more control, yet it does take time (and practise). The difference with not doing this isn't small or hardly noticable.
It does help to view the speaker + room as a system that needs to be able to work together.
The more specific the location the stronger the correction can be without suffering artefacts from that correction. The more even the directivity of the speaker in the room the stronger the correction can be without artefact. There are small corrections that can be made that are valid everywhere, there is no need to view it as an either or choice, there is a continuum.
There should not be seen as any competition between using quasi anechoic and room based measurements. They should complement each other, almost all speakers are better designed with anechoic or quasi equivalent measurements. But once the speaker has been designed, it, in most cases is going to go into a room where the two interact. Now a new strategy needs to be used to optimize the system. You don't hear what a microphone does, but instead of accepting that status quo you can use multiple measurements, processing and programs to gain a a better understanding of what is going on when the two combine to find ways to a better overall result.
This all seems overly complicated. The bottom line is that a reference measurement of the response at low frequency is made and from that IIR EQ can be applied, within reason, to move the response towards the desired target. At that point global FIR can be applied to modify the phase response as desired, again, within reason.
Which will change the outcome of the response curve again. So it might get more complicated than you'd imagine. Especially with a transition between mains and subs. However, it is well worth it (to me).
Funny what different people find complicated. I find it frustrating to be "bottom lined" or soap boxed by anyone who has not tried it out for themselves. I did not really expect the results I got either. I have spent literally years of my own time experimenting and I have tried every strategy I can think of in processing to see/hear what the results were. From simple, to things that make the above complication look like a kids cartoon. All of the most successful strategies have frequency dependent windowing at their core and the difference in getting the window right is significant. How the correction is ultimately made either by an automated process or by manual use of IIR and FIR filters does not really matter if it does the same same thing, but it cannot do the same thing if the measurement that it was based on was not the best choice.
I post the information to try and be helpful and point out things that have worked for me. Hopefully it provokes someone into thinking and trying for themselves.
So ppl been living with minimum phase crossover shifts for years and years. And ppl had flaring arguments on the most correct way to have an abomination of time shift and then utterly convinced that it’s the proper way to listen to music
So , doesn’t anyone else see a little bit of this mixed with a little why?
Like , example. What a minimum phase crossover does to the timing.
If we’re “fixing” all that , and arguing which was the most correct… aren’t we doing that in a small way…
If the minimum phase ppl swear you can’t hear the time shift, and we’re changing the time domain, weather we’re right or wrong, as long as all the changes we make are made in the same manner it will be like that minimum phase crossover, or all pass filter , and not hear it…. The phase shift right
Altho I would believe we are way closer then that to correct, but the difference between a pure minimum phase behavior and a zeroed phase using whatever windowing and whatever averaging….
If we find a method that works for us and we get good measurements on our instruments with our methods , I tend to think we’re still way better off then the minimum phase ppl, and we’re all wayyyy closer to a faithful representation then ever before….
So I see a lot of folks all being right , in there own respect. No one wrong on this that I can tell….
And I’m no expert lol (it would have to be an April fools joke to say I was) but I am a thinker and reader and I read and listen carefully and think about it….
Anyways…. The minimum phase thing is working for me , but I’m also using smaart measurements and then sending to REW
The only thing I see as not working at all is the vector average in REW…. We need a new way to get coherent measurements in REW so we can all be more unified in our experience….
A friend wrote an analogy that I found funny so I’ll do my best (and he’s a very respectable person that used to work for JBL)
“If wind blows to the east at 100mph and then blows to the west at 100mph , the vector average would say it blows at 50mph….. no dummy it blows at 100mph”
I like that a lot and it resonates with my quality measurements using vector averages windowed or not and I’ve tried them all, the vector average is just not doing it ok….
So , doesn’t anyone else see a little bit of this mixed with a little why?
Like , example. What a minimum phase crossover does to the timing.
If we’re “fixing” all that , and arguing which was the most correct… aren’t we doing that in a small way…
If the minimum phase ppl swear you can’t hear the time shift, and we’re changing the time domain, weather we’re right or wrong, as long as all the changes we make are made in the same manner it will be like that minimum phase crossover, or all pass filter , and not hear it…. The phase shift right
Altho I would believe we are way closer then that to correct, but the difference between a pure minimum phase behavior and a zeroed phase using whatever windowing and whatever averaging….
If we find a method that works for us and we get good measurements on our instruments with our methods , I tend to think we’re still way better off then the minimum phase ppl, and we’re all wayyyy closer to a faithful representation then ever before….
So I see a lot of folks all being right , in there own respect. No one wrong on this that I can tell….
And I’m no expert lol (it would have to be an April fools joke to say I was) but I am a thinker and reader and I read and listen carefully and think about it….
Anyways…. The minimum phase thing is working for me , but I’m also using smaart measurements and then sending to REW
The only thing I see as not working at all is the vector average in REW…. We need a new way to get coherent measurements in REW so we can all be more unified in our experience….
A friend wrote an analogy that I found funny so I’ll do my best (and he’s a very respectable person that used to work for JBL)
“If wind blows to the east at 100mph and then blows to the west at 100mph , the vector average would say it blows at 50mph….. no dummy it blows at 100mph”
I like that a lot and it resonates with my quality measurements using vector averages windowed or not and I’ve tried them all, the vector average is just not doing it ok….
Last edited:
Not complicated at all. No different that aligning phase of any other crossover.
I'm trying to find out exactly what you mean by "early" at low frequency. The chart you posted is unreadable at low frequency. So, for example, what is the length of the window at 100 Hz, 50Hz, 20 Hz? You say you used "we" about 1 but small changes can be heard and getting it right takes testing. What is "right". This sounds very subjective to me. But that's fine, music is personal.
On the other hand, I can say at low frequency I want my system to have a target response of a LR2 HP response with Fc = 25 Hz at my listening position and construct a filter to come as close to that as is reasonable. I don't have to worry about early or late, transient or steady state, because they are proprieties of the targeted response. If doesn't sound the way I would like it to maybe I'd change the target to a B2 or a Bessel HP, or change Fc.
Hi John K...,
Have you read the link i gave in post #622 ?
When i first saw this low end 'early reflection' thing it challenged my own understanding of what happen below Schroeder freq in a room, but Micthba's results ( measurements) talk by themself imho.
Maybe it's a misnomer, the correct term should be different than the one used ( low freq ER) but in practice it seems to work.
I agree the graph Fluid shared is difficult to read and maybe it would need a better way to display the info.
Iirc it's from DRC-Fir manual and as it is/was a personal effort development it could explain why limited ( Fluid or Wesayso will correct me if i'm wrong i'm sure).
And as Mitchba joined the fun i'm sure he would explain too.
Oabeieo,
Yes measurements are the core issue in all this.
I would be careful in saying vector averaging do not work: it could be an issue in how you implemented it's use or that you are trying to use it for something that it's not designed for.
Intrerpretation of this techniques are not easy ime.
Have you heard about Jean-Luc Ohl's 'MMM' technique? My experiment with it gave interesting results. Search about it, there is a thread Jean-Luc started where there is link to documents describing it ( or directly search for Jean-Luc own site).
Have you read the link i gave in post #622 ?
When i first saw this low end 'early reflection' thing it challenged my own understanding of what happen below Schroeder freq in a room, but Micthba's results ( measurements) talk by themself imho.
Maybe it's a misnomer, the correct term should be different than the one used ( low freq ER) but in practice it seems to work.
I agree the graph Fluid shared is difficult to read and maybe it would need a better way to display the info.
Iirc it's from DRC-Fir manual and as it is/was a personal effort development it could explain why limited ( Fluid or Wesayso will correct me if i'm wrong i'm sure).
And as Mitchba joined the fun i'm sure he would explain too.
Oabeieo,
Yes measurements are the core issue in all this.
I would be careful in saying vector averaging do not work: it could be an issue in how you implemented it's use or that you are trying to use it for something that it's not designed for.
Intrerpretation of this techniques are not easy ime.
Have you heard about Jean-Luc Ohl's 'MMM' technique? My experiment with it gave interesting results. Search about it, there is a thread Jean-Luc started where there is link to documents describing it ( or directly search for Jean-Luc own site).
Last edited: