No, it really isn't. What could a 3 cycle FDW do for phase if it were based on something that smoothed.
It does depend on the implementation of the FDW though, they most certainly don't all act alike.
Last edited:
Me neither, what you said in the above reply is not quite the same as what you said before and I would strongly suspect that the way you have taken it is not how Mitch intended it.
Positioning the speakers carefully relative to the room and then positioning the mic carefully relative to the speakers is important if you hope for the best outcome. If you can't or won't then it does not mean correction will not work or offer an improvement over leaving it as is. If you are always moving the speakers around and never settle on any specific position for long, it makes total sense that the time it takes to make measurements and process them could become onerous.
Everything comes down to trial and error in the end. I'm surprised that someone who has experimented so much in so mays ways is put off from trying.
The question doesn't make sense to me. The basic idea is to progressively allow more room into the measurement where you actually hear it and it forms part of the frequency and time response that you perceive.
You could view it as time smoothing instead of frequency smoothing, they both have some similarities but I think that is missing the forest for the trees.
Take the example of your synergy mids vs running the DCX down lower instead. One sounds much better to you. There can be all sorts of speculation as to why, some of it might even seem extremely plausible. You haven't found a measurement or metric that actually explains it in a rigorously scientific way. But the difference is so obvious to you that you aren't about to take them out. I feel the same way about this, I have heard the difference and am not about to ignore it because I don't have all the answers to explain it.
A question for @fluid and @wesayso - When you are fine tuning the frequency dependent window for in-room measurements below ~ 500 Hz, the size of the window obviously makes a difference as you have said.... Any guidelines on how many cycles should be included?
For talking purposes, say 100 Hz. 4 cycles would be 0.04 seconds. Would this tell me anything about what is happening at 100 Hz, or should I expand the window to capture more cycles... or ?
j.
For talking purposes, say 100 Hz. 4 cycles would be 0.04 seconds. Would this tell me anything about what is happening at 100 Hz, or should I expand the window to capture more cycles... or ?
j.
When you don't know what size the nut is............ 
You mention different window length. On one side that when using the 400 msec window the number of cycles is a function of frequency. At 1k Hz it's 400, at 100 Hz is 40, at 20 Hz is 8. So when comparing the two cases, where the number of cycles is constant, the window length goes like 1/f. Whent he length of the window is constant in time the number of cycles in the window goes like f. Thus when the result for constant cycles is compared to constant time, starting at 20 Hz they both have 8 cycles. At 40 hz one has 8 cycles the other has 16, at 80 Hz, 8 and 32, 160 Hz, 8 and 64, and so on. Yet no difference of significance. So, apparently for there to be a difference the FFt window would seem to have to be shorter than 8 cycles. In fact, I looked at the amplitude at select frequencies 2k, 1k, 500, 100 Hz, with windows that were 1, 2, 4, and 8 cycles long. In all cases there was no significant difference once the window was 4 cycles or longer. Then I went back check to see if this was due to the inclusion of more reflections or inherent in the length of the FFt window. I look at a frequency where I could alter the length of the window from 2 to 8 cycles and still not have any reflections. Even in this quasi anechoic region pretty much any frequency with less than 3 to 4 cycles had incorrect amplitude. Personally I think it important that whatever windowing you are using it better start by being long enough at each frequency to give the correct anechoic response before you worry about limiting reflections. That would appear to be around 4 cycles minimum.
But focusing on room Eq, personally I feel that if you start with a speaker than has good dispersion and flat band pass response (within a couple of dB) then that speaker will launch an accurate wave front into the room and accurate direct sound. If the speaker is linear phase it will also be transient accurate. But as soon as you apply room EQ to currect for room related response irregularities, at least above the modal region, you are destroying the accuracy of the direct sound. The problem is the room, not the speaker and it is the room that should be correct to provide the desired reverberate sound field, not the speaker.
The modal region is different. There is no early or direct sound. What you hear is the excitation of the modes and their sum. The best way to deal with that was discussed in a paper from Harmon (I forget the author) using multiples woofers to excite modes more uniformly and worry about speaker and listener positions.
Other than that, use what every approach you like and make things sound the way that pleases you.
You mention different window length. On one side that when using the 400 msec window the number of cycles is a function of frequency. At 1k Hz it's 400, at 100 Hz is 40, at 20 Hz is 8. So when comparing the two cases, where the number of cycles is constant, the window length goes like 1/f. Whent he length of the window is constant in time the number of cycles in the window goes like f. Thus when the result for constant cycles is compared to constant time, starting at 20 Hz they both have 8 cycles. At 40 hz one has 8 cycles the other has 16, at 80 Hz, 8 and 32, 160 Hz, 8 and 64, and so on. Yet no difference of significance. So, apparently for there to be a difference the FFt window would seem to have to be shorter than 8 cycles. In fact, I looked at the amplitude at select frequencies 2k, 1k, 500, 100 Hz, with windows that were 1, 2, 4, and 8 cycles long. In all cases there was no significant difference once the window was 4 cycles or longer. Then I went back check to see if this was due to the inclusion of more reflections or inherent in the length of the FFt window. I look at a frequency where I could alter the length of the window from 2 to 8 cycles and still not have any reflections. Even in this quasi anechoic region pretty much any frequency with less than 3 to 4 cycles had incorrect amplitude. Personally I think it important that whatever windowing you are using it better start by being long enough at each frequency to give the correct anechoic response before you worry about limiting reflections. That would appear to be around 4 cycles minimum.
But focusing on room Eq, personally I feel that if you start with a speaker than has good dispersion and flat band pass response (within a couple of dB) then that speaker will launch an accurate wave front into the room and accurate direct sound. If the speaker is linear phase it will also be transient accurate. But as soon as you apply room EQ to currect for room related response irregularities, at least above the modal region, you are destroying the accuracy of the direct sound. The problem is the room, not the speaker and it is the room that should be correct to provide the desired reverberate sound field, not the speaker.
The modal region is different. There is no early or direct sound. What you hear is the excitation of the modes and their sum. The best way to deal with that was discussed in a paper from Harmon (I forget the author) using multiples woofers to excite modes more uniformly and worry about speaker and listener positions.
Other than that, use what every approach you like and make things sound the way that pleases you.
As was said, not really any difference, but why not just use a single FIR to d both? On the othe rnad, the idea of ringing. Assume a speaker with a 30 Hz hogh pass response and an LR4 acoustic crossover at 2k Hz. Now, make it linear phase. The problem is not ringing because of the low frequency cut off if these is not input content below 30 Hz. However, the LR4 HP and LP section with individually become linear phase as well and these can introduce ringing in the crossover region. On the design axis that will cancel. Off axis it may or may not. Opinions vary of audibility.
Regarding window length..
https://www.diyaudio.com/community/threads/why-crossover-in-the-1-4khz-range.258487/post-3984954
https://www.diyaudio.com/community/threads/moving-mic-measurement.262246/post-4112321
https://www.diyaudio.com/community/threads/why-crossover-in-the-1-4khz-range.258487/post-3984954
https://www.diyaudio.com/community/threads/moving-mic-measurement.262246/post-4112321
Your point is valid. I'd like to try to clarify the context by saying that designing a speaker to work with a room can also mean the same thing.
3 to 6 cycles is usually the range that works best for me but lower frequencies can be longer.
In terms of DRC-FIR, a recommendation is to start with the minimal template and then go up and down from there to assess the difference and if any artefacts start to become obvious. The minimal template has this configuration
With a sampling frequency of 44,100 the samples to time range have been included after the # 250ms at 20Hz, 0.25ms at 20KHz
With an exponent of 1.0, this gives a straight line from 1/2 cycle at 20K to 50 cycles at 20Hz.
Just remember that being able to vary the number of cycles on a sliding scale like DRC-FIR can is more flexible and can produce a better correction more easily than trying to replicate the process using multiple fixed window sizes. Perhaps no-one has asked @JohnPM for the ability to vary the window in this way? It seems like REW could provide an easier entryway if it did as more people are comfortable with it.
I agree with you much more than I disagree with you. I said it before but I view the problem much like diffraction. Diffraction makes it difficult to EQ a speaker as the on and off axis become too different from each other. Nothing you do with EQ fixes the problem but there is usually a compromise to be found where things sound subjectively better. Experimentally I find the same here. Fixing the problem at the source would be better but for most practical scenarios that still leaves a lot to be desired.
Perhaps you mean this paper from Welti and Devantier
https://audioroundtable.com/misc/Welti_Multisub.pdf
This is really about reducing spatial variability, there is a video interview with Welti where he states clearly that for a single listening position putting the sub in the corner for maximum excitation and then EQing the response at the listening position is a valid approach.
If anyone wants to discuss any of this with me any further send me a PM, as this kind of back and forth is not something I want to continue.
There's still as much assumptions being made as quite a few posts back, we've made no progress at all.
There's no easier way to explain than to say: worry not only about what to correct but when to correct it. It really does make a difference.
Simply run a correction using DRC-FIR with ~4 cycles and one with 8 cycles and listen to each result. You can even use the basic tools that are pré-constructed by gmad in this thread. If you don't hear a difference between those two different windows, you probably have one heck of a room. And this window recipe mentioned here is only for magnitude and demo purposes. Correcting phase is done with a separate (even shorter) FD window. Or not, you can choose yourself using these tools. Don't expect great results though, as that would depend on a lot more factors than simply using these so called "room correction software" packages. As said earlier, it's just a tool.
I use a variable FD window that is based on my speakers + room. Speakers that I don't have an "anechoic flat on axis" curve for, as they actually need the room.
So I have to "draw" my own room curve, you know, the procedure that Toole advises against. The only thing that my speakers have going for them is a very cool DI result, when used inside the room. See my thread for examples, searching for DI might help.
Viewing it as a pure: 'correcting the room' is kind of a flaw i.m.h.o. An 'easy to get' idea, seeing as these are always called "Room correction software". Want to correct the room? bring on the absorbers and what not (I know I did). I use this tool as a needed (for me) speaker correction to battle the downwards FR shift of an array inside a room. And I use it (DRC-FIR) outside of it's suggested limits to have it do what I want done. Well, I used to, I still use the tool, but manually do the last parts.
Which has more to do with the way DRC-FIR acts if I end the phase window prematurely, say at 200 Hz. I get cleaner results using DRC-FIR in its minimum phase mode and use the DRC-FIR generated frequency dependent window separately with Rephase to do any manipulation of phase.
Luckily I don't need the confirmation of any other than myself to do this type of stuff. I've made this a study born out of interest and have offered my help to anyone with an interest in it, trying to explain the why (I do it) etc. to people with an actual or genuine interest. I don't see that interest here, so time to move on.
There's no easier way to explain than to say: worry not only about what to correct but when to correct it. It really does make a difference.
Simply run a correction using DRC-FIR with ~4 cycles and one with 8 cycles and listen to each result. You can even use the basic tools that are pré-constructed by gmad in this thread. If you don't hear a difference between those two different windows, you probably have one heck of a room. And this window recipe mentioned here is only for magnitude and demo purposes. Correcting phase is done with a separate (even shorter) FD window. Or not, you can choose yourself using these tools. Don't expect great results though, as that would depend on a lot more factors than simply using these so called "room correction software" packages. As said earlier, it's just a tool.
I use a variable FD window that is based on my speakers + room. Speakers that I don't have an "anechoic flat on axis" curve for, as they actually need the room.
So I have to "draw" my own room curve, you know, the procedure that Toole advises against. The only thing that my speakers have going for them is a very cool DI result, when used inside the room. See my thread for examples, searching for DI might help.
Viewing it as a pure: 'correcting the room' is kind of a flaw i.m.h.o. An 'easy to get' idea, seeing as these are always called "Room correction software". Want to correct the room? bring on the absorbers and what not (I know I did). I use this tool as a needed (for me) speaker correction to battle the downwards FR shift of an array inside a room. And I use it (DRC-FIR) outside of it's suggested limits to have it do what I want done. Well, I used to, I still use the tool, but manually do the last parts.
Which has more to do with the way DRC-FIR acts if I end the phase window prematurely, say at 200 Hz. I get cleaner results using DRC-FIR in its minimum phase mode and use the DRC-FIR generated frequency dependent window separately with Rephase to do any manipulation of phase.
Luckily I don't need the confirmation of any other than myself to do this type of stuff. I've made this a study born out of interest and have offered my help to anyone with an interest in it, trying to explain the why (I do it) etc. to people with an actual or genuine interest. I don't see that interest here, so time to move on.
I decided to diplomatically sidestep how i take it to make sure i cause no harm....I have great respect for Mitch, and don't want my way of thinking to cause any difficulty.
yes, it just comes down to trial and error we think is worth pursuing
That's what I'm not seeing....because they seem more similar than not.
I see Impulse and transfer function as an identity. Smoothing in the time domain will show up as smoothing in the frequency domain and vice versa.
Good example, and very fair.
I do think the small mids can be explained by some form of modulation reduction, but you are quite correct I have no rigorous proof yet to back it up.
But the idea of reducing driver bandwidth spans makes logical sense to me, so I'm willing to pursue it despite the effort.
Honestly, If FDW made logical sense to me, above and beyond the frequency dependent FIR generation I can already easily do , I'd go to the effort.
I just need to see some compelling objective evidence to move me.
Just like folks no doubt need some compelling objective evidence, to move them to work on reducing driver bandwidth spans, huh?
Hi Mark.
I agree that compelling evidences are needed but at same time i don't think dismiss of practical evidence are a good way to approach things either.
After all those are 'new' tools and we are in the enjoyable situation were the experimenters about it are open to share their findings and encourage anyone to try by themself, even helping anyone willing to try.
I totally get your pov and i think it is shared by most in here: (as much) blameless loudspeakers as possible first.
From there room treatments at minima for ER management and this is usually were reluctance starts to happen ( for no good reason and with misconception of what it is and how it can be implemented imho).
Then digital tools to help. Whatever their name and what they correct ( being it the loudspeakers (mainly in my view -but ymmv) or the room- Mitchba's example linked previously correct the room behavior to me and is documented).
Innovative and tbh where the amateur experimenters are is not far away from what Trinnov achieve in my view.
For once there is more acceptance (and sooner) in proworld about it.
'Une première' in a very conservative world.
I must conceed my own pre conception were the same as yours Mark when you started this thread. I think there have been enough evidence and explanation prooving it was misconception and not this well understood principle of application.
Wesayso, Fluid, Mitchba you are onto something imho. And i find amazing it happened with 'free' tools and from 'amateurs' ( given it's a hobby... you are more 'experts' imv).
Thank you all for made me reconsider all this and change POV.
I agree that compelling evidences are needed but at same time i don't think dismiss of practical evidence are a good way to approach things either.
After all those are 'new' tools and we are in the enjoyable situation were the experimenters about it are open to share their findings and encourage anyone to try by themself, even helping anyone willing to try.
I totally get your pov and i think it is shared by most in here: (as much) blameless loudspeakers as possible first.
From there room treatments at minima for ER management and this is usually were reluctance starts to happen ( for no good reason and with misconception of what it is and how it can be implemented imho).
Then digital tools to help. Whatever their name and what they correct ( being it the loudspeakers (mainly in my view -but ymmv) or the room- Mitchba's example linked previously correct the room behavior to me and is documented).
Innovative and tbh where the amateur experimenters are is not far away from what Trinnov achieve in my view.
For once there is more acceptance (and sooner) in proworld about it.
'Une première' in a very conservative world.
I must conceed my own pre conception were the same as yours Mark when you started this thread. I think there have been enough evidence and explanation prooving it was misconception and not this well understood principle of application.
Wesayso, Fluid, Mitchba you are onto something imho. And i find amazing it happened with 'free' tools and from 'amateurs' ( given it's a hobby... you are more 'experts' imv).
Thank you all for made me reconsider all this and change POV.
I was thinking about these FDWs and trying to get a better understanding of what they represent. I don't know if this has been discussed before, so here are my thoughts. I thought of RT60 and CSD plots (water falls) and how the decay time vary with frequency, . Typically the higher the frequency the shorter the decay time is. You can generate a 3-D plot (CDS) of the decay of a signal, Amplitude vs time as a function of frequency, as pictured crudely below. High frequency to low, front to back. Then if you cut a horizontal plane through that curve at a given amplitude you get a time vs frequency for the the signal to decay to that level. In a sense, would not each of these horizontal curves represents a FDW where you would be windowing out any contribution to the impulse where the level of the contribution at that frequency dropped below a certain dB level? Thus, in essence, using a FDW seems like imposing a decay curve for the room. So, when you choose a FDW for a measurement, the resulting response is perhaps similar to that of an un-windowed response of a room with a similar decay vs frequency, at the measurement position.
Do you suppose one could hear the difference between this:
And this:
As measured, a stereo pair of speakers at the listening spot? The program used to view this "wavelet" is APL_TDA which was available as a Demo.
I used that program to make these two comparisons, one without and one with my correction in place. I did this just for fun as a trial measurement on the thread: https://www.diyaudio.com/community/...can-we-do-with-dsp-power-now-availabl.284916/
(the correction was made using DRC-FIR, REW, JRiver and a lot of leg-work around the room to be sure it wasn't a single point wonder)
Comparing input to output, a view of my DAC output, measured the same way (running the output of DAC to the Mic in):
And the Stereo result of my speakers at the listening spot, as measured with a real microphone:
It won't act as nice as a real nice room with lots of absorption and diffusion used, as member jim1961 demonstrated in his "Studio like" room:
(that blue ridge in the "deep blue sea of silence" is a Haas kicker, strategically placed panels that reflect/diffract back to the listening spot)
No DSP was used for this above in-room measurement. I asked Jim to run the APL_TDA demo as I was deeply impressed with the work he had done in his room. This is a Troels second order speaker pair in a heavily treated room I think his subs were off, I can't recall. As said: no DSP or EQ used. That room looks nothing like a regular living room.
You can see the work involved here: https://gearspace.com/board/studio-building-acoustics/817205-my-listening-room.html
And this:
As measured, a stereo pair of speakers at the listening spot? The program used to view this "wavelet" is APL_TDA which was available as a Demo.
I used that program to make these two comparisons, one without and one with my correction in place. I did this just for fun as a trial measurement on the thread: https://www.diyaudio.com/community/...can-we-do-with-dsp-power-now-availabl.284916/
(the correction was made using DRC-FIR, REW, JRiver and a lot of leg-work around the room to be sure it wasn't a single point wonder)
Comparing input to output, a view of my DAC output, measured the same way (running the output of DAC to the Mic in):
And the Stereo result of my speakers at the listening spot, as measured with a real microphone:
It won't act as nice as a real nice room with lots of absorption and diffusion used, as member jim1961 demonstrated in his "Studio like" room:
(that blue ridge in the "deep blue sea of silence" is a Haas kicker, strategically placed panels that reflect/diffract back to the listening spot)
No DSP was used for this above in-room measurement. I asked Jim to run the APL_TDA demo as I was deeply impressed with the work he had done in his room. This is a Troels second order speaker pair in a heavily treated room I think his subs were off, I can't recall. As said: no DSP or EQ used. That room looks nothing like a regular living room.
You can see the work involved here: https://gearspace.com/board/studio-building-acoustics/817205-my-listening-room.html
Indeed, in my professional life proving a negative is really hard, so absence of proof doesn't deter me from seeking it. Most of the important realisations I have made only came later, not at the time. So experimenting is good, even if it fails, it may prove crucial later. Sound follows it's own rules, they are always predictable but not often logical.I just need to see some compelling objective evidence to move me.
Just like folks no doubt need some compelling objective evidence, to move them to work on reducing driver bandwidth spans, huh?
Do you suppose one could hear the difference between this:
View attachment 1162696
And this:
View attachment 1162697
As measured, a stereo pair of speakers at the listening spot? The program used to view this "wavelet" is APL_TDA which was available as a Demo.
I used that program to make these two comparisons, one without and one with my correction in place. I did this just for fun as a trial measurement on the thread: https://www.diyaudio.com/community/...can-we-do-with-dsp-power-now-availabl.284916/
(the correction was made using DRC-FIR, REW, JRiver and a lot of leg-work around the room to be sure it wasn't a single point wonder)
Comparing input to output, a view of my DAC output, measured the same way (running the output of DAC to the Mic in):
View attachment 1162698
And the Stereo result of my speakers at the listening spot, as measured with a real microphone:
View attachment 1162699
It won't act as nice as a real nice room with lots of absorption and diffusion used, as member jim1961 demonstrated in his "Studio like" room:
(that blue ridge in the "deep blue sea of silence" is a Haas kicker, strategically placed panels that reflect/diffract back to the listening spot)
No DSP was used for this above in-room measurement. I asked Jim to run the APL_TDA demo as I was deeply impressed with the work he had done in his room. This is a Troels second order speaker pair in a heavily treated room I think his subs were off, I can't recall. As said: no DSP or EQ used. That room looks nothing like a regular living room.
You can see the work involved here: https://gearspace.com/board/studio-building-acoustics/817205-my-listening-room.html
Looking at the top two plots, I would imagine that there would be an audible difference. WIth the exception of the modal region I would imagine the response that generated the lower plot would sound deader, with more precise imaging. But that becomes a matter of degree and personal preference, and obviously I couldn't judge w/o actually listening. .Some like pin point imaging. Some like a more open, natural sound.
Happy Easter to all.
So I have a question on system high pass
Wesayso I believe about a year ago helped me and looked at my sub response. He saw I had about a 5-10db boost between DC and 20hz
He recommended a BW6 at 18hz , so I did that and it did make bass more punchy and transparent and seemed to also make it have much better transition to the mains
So …… let’s say there is a system with no natural HP on the sub, in that instance would you make the phase flat all the way to dc? Then add that minimum phase high pass filter and leave the phase alone? Furthermore would you add a cascade filter to the mains 80hz , that also adds that BW6 at 18hz on the mains?
Right now I have it follow the minimum phase and fluid I believe is onto something because the bass sounds more natural like this… like a live instrument would.
But I want to make sure I’m understanding what was being said.. because I have a tilt on my sub to mains that’s about 10db difference
So obviously the phase is tilting down instead of up…..
If I make it flat again, to try this, and there is no sub high pass….. yeah should I go to zero all the way through?
Thanks
Wesayso I believe about a year ago helped me and looked at my sub response. He saw I had about a 5-10db boost between DC and 20hz
He recommended a BW6 at 18hz , so I did that and it did make bass more punchy and transparent and seemed to also make it have much better transition to the mains
So …… let’s say there is a system with no natural HP on the sub, in that instance would you make the phase flat all the way to dc? Then add that minimum phase high pass filter and leave the phase alone? Furthermore would you add a cascade filter to the mains 80hz , that also adds that BW6 at 18hz on the mains?
Right now I have it follow the minimum phase and fluid I believe is onto something because the bass sounds more natural like this… like a live instrument would.
But I want to make sure I’m understanding what was being said.. because I have a tilt on my sub to mains that’s about 10db difference
So obviously the phase is tilting down instead of up…..
If I make it flat again, to try this, and there is no sub high pass….. yeah should I go to zero all the way through?
Thanks
Have to say: I have no recollection of that...
Usually I remember just about every conversation, sorry.My advise, hopefully understanding what you want, make phase follow the FR curve you have. Don't correct the phase at the natural roll off of the subs.
You can see the turn in the wavelets I posted, I never correct the lower end roll off. I'm with @fluid on that. Should phase be "EQ-ed flat to DC" and you add a high pass -> indeed leave phase alone for that high-pass under these circumstances. All of the phase need not be flat throughout the audio bandwidth, simply make phase follow the (overall) frequency response curve.
If you were to draw the frequency response you desire in a tool like like RePhase without the crossovers etc., with the bass bump and everything using minimum phase tools, that (minimum) phase that follows the magnitude curve you created would be the goal I'd suggest to aim for. As if You'd have one ideal driver that does it all with the frequency shape you like best (like a boost at lower frequencies etc).