HOLMImpulse Measurements, Need step by step dummy guide.

[Dennis H]

Doug, are you using the BMS or the Celestion? The Celestion doesn't like to cross as low.

FWIW, ARTA couldn't import the .wav files, said it can only import PCM wavs whatever that means.

[penngray]

Doug, are you using the BMS or the Celestion? The Celestion doesn't like to cross as low.

FWIW, ARTA couldn't import the .wav files, said it can only import PCM wavs whatever that means.

One speaker is the Celestion and the other speaker is the BMS. Right now we are measuring the Celestion.

Why do you say it doesnt like to be crossed that low? Even Augerpro's measurements show its okay down to 1500Hz.

[Saurav]

FWIW, ARTA couldn't import the .wav files, said it can only import PCM wavs whatever that means.

I think I know. I asked penngray to export the WAV as 32-bit floats, figuring that would have good resolution. There are a couple of PCM options in the 'impulse as WAV' export dialog (don't remember the details, don't have it on front of me right now). I'll bet that if he used one of those, ARTA would be able to import it. AFAIK HOLM saves the last set of measurements automatically, so he should be able to just start up the application and re-export the impulse as a WAV using PCM. I think the options were 16 and 32 bit, but I'm not sure.

You can easily run the crossover down to 1200Hz if you choose too.

I vaguely remembered something about Geddes using CDs down pretty low, but I didn't remember the details. I think the problem with going really low is that there isn't much left for the crossover to work with, so to speak. That's another area I don't fully understand, how the electrical attenuation and acoustic attenuation are related. I remember reading something about how the electrical attenuation affects driver excursion, but again, don't remember the details. For example, once I flatten the drooping treble from the waveguide, the response I'm left with looks fairly close to LR4, so in theory I could run with no filter on the tweeter or a minimal filter - just use the tweeter's natural rolloff, and cross the woofer at that frequency. But I think that would have other consequences on the tweeter.

I wouldnt run the woofers up to 2KHz at all. I do not like their sound even past 2000Hz.

Can you describe how you tried this? Was this a full XO, woofer and tweeter, or were you listening to just the woofer with a low-pass, or something else?

[Dennis H]

One speaker is the Celestion and the other speaker is the BMS. Right now we are measuring the Celestion.

Why do you say it doesnt like to be crossed that low? Even Augerpro's measurements show its okay down to 1500Hz.

Yeah, you're right, the distortion looks okay down to 1K. I was thinking the resonance in the impedance curve was higher but it's not all that much looking at Brandon's curves and the peaks are much more damped with the Celestion. What the heck, go for it.

[Saurav]

OK, show of hands, what XO frequency should I try to design tonight? I'll start with LR4 targets, if only because it's easier to ensure correct phase matching with this 'remote design' process.

[penngray]

OK, show of hands, what XO frequency should I try to design tonight? I'll start with LR4 targets, if only because it's easier to ensure correct phase matching with this 'remote design' process.

1500Hz please

[penngray]

BTw, could you create the crossover assuming there is a -9dB Lpad included on the tweeter? Im creating one soon or maybe I should wait?

[Dennis H]

I think I know. I asked penngray to export the WAV as 32-bit floats, figuring that would have good resolution. There are a couple of PCM options in the 'impulse as WAV' export dialog (don't remember the details, don't have it on front of me right now). I'll bet that if he used one of those, ARTA would be able to import it. AFAIK HOLM saves the last set of measurements automatically, so he should be able to just start up the application and re-export the impulse as a WAV using PCM. I think the options were 16 and 32 bit, but I'm not sure.

From one of Doug's graphs above, it looks like he did the measurements at 44.1/16 so a 16 bit wav would probably be good enough. It takes a pretty good sound card to show any better noise floor going up to 24 or 32 bits. Doug, if you can do a 16 bit PCM export (maybe load the wavs you have and export them again?) I can take a look at the phase and acoustic center stuff in ARTA.

[penngray]

Im trying out acourate software (trial version is very limited) but I have this right now. I imported the wave files from HOLM into it.

Image not available

Second one is the phase.

Dennis, I will have to get those measurements for you.

I see in/out stream samplerate = 44.1Khz but where do I find if its 16bit?

[Saurav]

And maybe the PCM files will import more easily into HOLM. I swear it crashed a dozen times this morning as I was working with the WAV files.

maybe load the wavs you have and export them again?

I would advise against trying to import those WAVs, you might get crashes too If you just open up HOLM, it should still have the last set of measurements intact. Just go through the impulse export process again, but in the drop-down menu, select one of the PCM options, I think there's one for 16 bit.

BTw, could you create the crossover assuming there is a -9dB Lpad included on the tweeter? Im creating one soon or maybe I should wait?

Hmm. I'm not sure if an L-pad will give a textbook-accurate drop in SPL without changing the frequency response, since the driver's impedance isn't purely resistive. I think I can have something for you to try tonight, so unless you think you'll have the L-pad done by then, I'll probably go ahead and run the model without it.

Also, if the L-pad works perfectly, all you have to do is go into the DCX and increase the tweeter level by 9dB from whatever I come up with tonight, and in theory you should be back to where you were (but with less hiss - I understand why you're trying the L-pad). In reality, I suspect it will change the FR a little bit, so then we'd need to measure it again and run it through LspCAD again.

BTW, how are you planning to evaluate these? Listen to the XO in mono, or are you going to put the Celestion into the other speaker as well?

[penngray]

And maybe the PCM files will import more easily into HOLM. I swear it crashed a dozen times this morning as I was working with the WAV files.


I would advise against trying to import those WAVs, you might get crashes too If you just open up HOLM, it should still have the last set of measurements intact. Just go through the impulse export process again, but in the drop-down menu, select one of the PCM options, I think there's one for 16 bit.


Hmm. I'm not sure if an L-pad will give a textbook-accurate drop in SPL without changing the frequency response, since the driver's impedance isn't purely resistive. I think I can have something for you to try tonight, so unless you think you'll have the L-pad done by then, I'll probably go ahead and run the model without it.

Also, if the L-pad works perfectly, all you have to do is go into the DCX and increase the tweeter level by 9dB from whatever I come up with tonight, and in theory you should be back to where you were (but with less hiss - I understand why you're trying the L-pad). In reality, I suspect it will change the FR a little bit, so then we'd need to measure it again and run it through LspCAD again.

BTW, how are you planning to evaluate these? Listen to the XO in mono, or are you going to put the Celestion into the other speaker as well?

I will not have the Lpads done that soon. Yes, my only goal with the Lpad is to reduce that hiss.

I wanted to listen in Mono but I can put the XO in the other speaker if you think Stereo will matter.

btw, I do not have those meaurements in tact because I "clicked" on lots of buttons I clicked on "Time Zero, auto detect" in the options button for the tweeter measurement to see what happens, it did change the response.

I also clicked on Raw, Gated...etc. You posted above I should make sure the gating between the woofer and tweeter was indentical, they were not they were slightly different.

Btw, thanks guys...most of this is still extremely confusing but its fun. I just wish there was software out there that could take the wave files and build inital crossovers for me (schematics and all)!! There should be something that would general maybe 3 or more different crossovers samples for me to try.

I will import the Wave files back in to see what happens.

[penngray]

Yep, HOLM crashes when I import the wave file back in

[Saurav]

I wanted to listen in Mono but I can put the XO in the other speaker if you think Stereo will matter.

Nah, it's up to you. You'd have to swap out the tweeter anyway, the XO won't work with the DE250. I like to listen in stereo, but I also have the problem of stopping when things start sounding good enough, because at that point I'd rather just listen to music than work any more So there are definitely advantages to listening in mono until you have everything ironed out.

You posted above I should make sure the gating between the woofer and tweeter was indentical, they were not they were slightly different.

That's OK. That affects how HOLM calculates frequency response from the impulse response. Since you exported the impulse response, none of your experiments there should have affected what you exported. When I imported your data I fixed the gating to make sure it's the same.

Yep, HOLM crashes when I import the wave file back in

That makes me feel better

[penngray]

Nah, it's up to you. You'd have to swap out the tweeter anyway, the XO won't work with the DE250. I like to listen in stereo, but I also have the problem of stopping when things start sounding good enough, because at that point I'd rather just listen to music than work any more So there are definitely advantages to listening in mono until you have everything ironed out.

hehe, that would be me too....honestly your crossover values plugged into my DCX and both speaker running would probably have me not caring about anything more for a long time

I enjoy the event (music or movies) more then I do hearing the best crossover possible

I care a great deal about doing it right though but its a struggle to keep working on it.

[Dennis H]

Duplicate post -- ISP problems.

[Dennis H]

Duplicate post -- ISP problems.

[Dennis H]

Im trying out acourate software (trial version is very limited) but I have this right now. I imported the wave files from HOLM into it.

Now those impulse responses look more like what I'd expect. I think HOLM is doing too much automatic "stuff" to the impulse for my taste. At a glance, you can see that the woofer signal arrives first as you'd expect with a horn. You can also see that the tweeter is wired with reverse polarity (impulse goes negative first rather than positive). That's confirmed by the phase which curves upward rather than downward. In ARTA, I'd manually invert the tweeter impulse to get it going positive first but at least I'd know I had done it when it came time to design the crossover and wire up the drivers.

[Saurav]

I think HOLM is doing too much automatic "stuff" to the impulse for my taste.

Only if you set it up that way, I think, although I may not be seeing everything you're seeing. The woofer impulse arrives before the tweeter, and I set the time-0 reference to slightly before the start of the woofer impulse. The effect shows in the tweeter phase, which is wrapping faster than the woofer phase. I didn't get to the point of extracting minimum phase / figuring out the AC offset.

The inverted impulse from the tweeter is puzzling, I didn't see that when I imported the measurements, and I don't see it in Doug's measurements either. I'll have to check and see if there's an auto-invert option in the import that I missed.

Here's the impulses from HOLM again, after the import. Other than the tweeter polarity, I'm not really sure what's different:

[penngray]

Thanks,Acourate seems to have some cool features. I could do measurements using it along with Audacity for wave recording and re-importing back into it.

[penngray]

The inverted impulse from the tweeter is puzzling, I didn't see that when I imported the measurements, and I don't see it in Doug's measurements either. I'll have to check and see if there's an auto-invert option in the import that I missed.

I didnt reverse the polarity on my tweeters (atleast I didnt think I did).

Let me import the wave back into acourate because I was definitely playing around with it and I might have reversed the impulse on the tweeter.


EDIT: just checked and it was me "Clicking" on things in Acourate.

Sorry for any confusion. I like Acourate's ease of use...$300 Euro for a PC based product...replacing my DCX ...hmmm.

[Saurav]

The thing I don't get is the apparent lack of any reflections in the impulse responses. Either your measurements setup is really clean, or the data exported by HOLM already went through some gating process, which would then probably render it useless.

Edit: When you exported the impulse, I think there's an option that lets you select the range to export. Do you remember seeing something like -10,000 to 10,000? I don't have it in front of me, I can give you the exact field later tonight.

I think you did it right, because when I imported the data, the impulses were right around the 10,000 sample range (9970 is what I used for 0), so that would be consistent with measuring the tweeter, setting that as the 0 reference, measuring the woofer with 0 locked, and exporting -10,000 to 10,000 on that scale.

Eh, whatever... I'll design a crossover, post what LspCAD thinks the output should be, you dial that into the DCX and post the measured results, and we see how far off we are

[Dennis H]

Ah, okay, I see the woofer impulse in HOLM now. It is before the tweeter.

The thing I don't get is the apparent lack of any reflections in the impulse responses. Either your measurements setup is really clean, or the data exported by HOLM already went through some gating process, which would then probably render it useless.

Yeah, I've got a feeling it's exporting the gated impulse. He should be recording the raw impulse, however you do that in HOLM. All that other stuff comes later for converting to frequency response.

[Saurav]

As far as I know, the directions I provided should have resulted in the full impulse being recorded. I'll check it tonight, zoom out and see if any noise shows up further down. Actually, I know some noise is present, because if I move the gate out, the FR curves start to get fuzzy. I didn't check the fully ungated response, which is usually pretty noisy for me.

Doug, what's your measurement setup like? Distance between speaker and mic, height above the floor, height to the ceiling, distance to nearby reflecting surfaces, do you have any blankets / pillows / foam on the floor, stuff like that?

[penngray]

The thing I don't get is the apparent lack of any reflections in the impulse responses. Either your measurements setup is really clean, or the data exported by HOLM already went through some gating process, which would then probably render it useless.

Edit: When you exported the impulse, I think there's an option that lets you select the range to export. Do you remember seeing something like -10,000 to 10,000? I don't have it in front of me, I can give you the exact field later tonight.

I think you did it right, because when I imported the data, the impulses were right around the 10,000 sample range (9970 is what I used for 0), so that would be consistent with measuring the tweeter, setting that as the 0 reference, measuring the woofer with 0 locked, and exporting -10,000 to 10,000 on that scale.

Eh, whatever... I'll design a crossover, post what LspCAD thinks the output should be, you dial that into the DCX and post the measured results, and we see how far off we are

It could be that my room is okay for measurements? The room is full of treatements, the walls are 12 feet away, left and right from the speaker, Behind the speaker I have two big 4" thick DIY portable panels, on the ground in front of the speaker I have another 4" thick DIY portable panel and on the back wall I have 6" thick sound absorption. The speaker is on a stand 3 feet of the floor, the mic is 1 meter from the speaker on a mic stand (or was it 2 meters) The speaker is 3 feet away from the front wall and 12 feet from the back wall.



I was posting about gating vs non-gated in other forum and I was wondering about the differences. The raw measurements in HOLM are not much different then the gated measurements for me. Either Im screwing or I did set some value properly, Most of its default settings outside of your recommendations.

The export impulse sample was from -10000 to 10000

[penngray]

Ah, okay, I see the woofer impulse in HOLM now. It is before the tweeter.

Yeah, I've got a feeling it's exporting the gated impulse. He should be recording the raw impulse, however you do that in HOLM. All that other stuff comes later for converting to frequency response.

I will do the measurements again tonight for you guys. Thanks again for taking the time to help me !!!

[penngray]

Ah, okay, I see the woofer impulse in HOLM now. It is before the tweeter.

Yeah, I've got a feeling it's exporting the gated impulse. He should be recording the raw impulse, however you do that in HOLM. All that other stuff comes later for converting to frequency response.

yes, I believe that is true.

[Saurav]

When I get home tonight I can post some of my impulse responses. Your room is *much* better for measurements than mine - my impulse response graphs show clear reflections, and much sooner than anything I've seen in yours. If I include those (say with an un-gated response), the frequency response curve turns to noise, you can't see anything in there.

I'd also suggest trying the ARTA demo, even if just to post pictures of the impulse response. That won't need any calibration other than loading in the mic cal file. IMO it's always helpful to check with multiple programs, that way a silly mistake in configuration won't pollute all your measurements.

[penngray]

Sorry, I didnt get the measurements done tonight. Kid duties right now (sick baby).

I have a question about the crossover design. How does the impedance measurements work into this? Do I assume we can just use what Augerpro has or should I do that measurement too? I have the WT3 software.

[Saurav]

How does the impedance measurements work into this?

That's one of the advantages of going with an active crossover - it doesn't. For the most part (unless you're talking about specialized tube and some SS amps), the amp's output impedance is much lower than the driver impedance, so the variations in the driver impedance don't affect the response. With a passive crossover, the impedance of the crossover components is similar to the impedance of the drivers, so you need to factor in driver impedance as well, because the two interact.

Hope that made sense

[Saurav]

I think Doug got the raw impulse, and his room is just so clean that there are hardly any reflections. Here's the woofer impulse zoomed out, and the FR with a ~5ms gate:



10ms gate:



20ms gate:



Ungated, unsmoothed (aka 'raw response' in HOLM lingo):



The tweeter shows a similar increase in noise and LF wiggles as the gate time is increased. And more importantly, the reflections are visible:

[Saurav]

Another neat HOLM feature that I haven't really played with - you can plot the impulse response on a dB scale, and then scale up and down. As far as I can tell, it takes the first/largest peak and uses that as 0dB. This makes for an interesting way to see the reflection peaks.

On a 30dB scale:



Click the down arrow once to go to a 40dB scale, and now the first reflection is visible quite clearly:

[Saurav]

OK, crossover. The woofer wasn't too hard, at least for a first cut I went with one notch and an LR4 low-pass. I need some advice on how to deal with the tweeter. Here's one attempt:



As you can see, that's a 1st-order filter, and most of the XO region ends up getting handled by the EQ. I ran the optimizer with a lower frequency of 750Hz, then extended the range to 350 to show how the response tracks.

So the FR looks good, but I'm not sure the transfer function is right:



Is that enough attenuation below the XO frequency? Something tells me it might not be.

I've tried 2nd, 3rd and 4th order filters, but those end up making the response curve too sharp. So then the optimizer ends up moving the XO down to the 500Hz range, where all it seems to do is help the xfer function below the XO frequency. I think that might be a better approach, but I figured I should ask. I had the same problem when I had a CD tweeter in my speakers - when I try to run it down close to the natural cutoff frequency, I don't really know the best way to handle it in the XO.

[Saurav]

OK, this is a little better, I think. Here's the first XO to try:



Notes:

* For the crossovers, note that one is Butterworth-18, and the other is LR-24.

* The parametric EQs should be easy to enter, the DCX interface pretty much matches the LspCAD output exactly. Ditto for the gain and delay pages.

* Note that the delay and EQ pages both have a global on/off setting, so remember to turn them on. That's tripped me up a couple of times, where I forgot to turn the delay on, then spent several minutes wondering why changing the delay wasn't changing the response.

* For the shelving filter on the tweeter - make sure you use LP, and 6dB/octave (that's the last option on that page, and defaults to 12dB/octave). Also, LspCAD handles shelving filters differently from the DCX, so it needs a 2nd calculation to convert from one to the other. That's the '1394' you see on the side - LspCAD's calculated value of 2920Hz / -12.8dB translates to 1394Hz when you enter it into the DCX. The exact value '1394Hz' probably won't be available, so use whatever is closest.

LspCAD's predicted response:





Transfer functions:



You need to check the phase alignment, it probably won't be right (based on my experience so far). Here's how I do that:

* Reverse the tweeter. You can do this with the physical cabling, or invert the response in DCX, doesn't matter.

* Measure the response with both drivers, and the full XO/EQ/delay/etc. active. You should see something like the reverse-null image posted above, but your null will probably be shallower.

* Go to the woofer's settings, and go to the delay page. Adjust the delay until you get the deepest / sharpest null. I generally try 2 clicks up, measure, 2 clicks down, measure. That will tell you which direction to go in. Then keep adjusting the delay by 2 clicks, and measuring each time. You'll see the null get deeper, then start to get shallower again. At that point, adjust by 1 click around that area until you find the setting with the deepest notch. Post the delay value you end up with.

Also, I haven't done anything about the bass response. I'm not sure exactly what to do about BSC, so the speaker will probably sound bass-heavy or bass-shy or something (you can tell I haven't done any box designs ). But that shouldn't change these settings, so you can work on getting this entered into the DCX and seeing how that works.I'll ask for advice on BSC tomorrow

Also, if you get round to it, measure a few off-axis curves (both drivers, XO/EQ in place), and post those. You can leave the time-zero at auto-detect, and adjust the gate manually if HOLM picks a really short gate like it did before. I just want to see what the FR looks like, I don't need those for the simulations... yet Also, tomorrow I'll try a better circuit on the woofer - notch out the peaks first, then apply the filter.

I wouldn't be surprised if your room is more dead / clean than HOLM expects, and that's why the auto-gating got confused and gave you the bizarre graphs you probably saw in the last few days.

[penngray]

That's one of the advantages of going with an active crossover - it doesn't. For the most part (unless you're talking about specialized tube and some SS amps), the amp's output impedance is much lower than the driver impedance, so the variations in the driver impedance don't affect the response. With a passive crossover, the impedance of the crossover components is similar to the impedance of the drivers, so you need to factor in driver impedance as well, because the two interact.

Hope that made sense

Thanks, the reason I love Active is because it removes impedance issues and even acoustical center stuff can be managed through easy delay settings but I definitely want to understand the passive design a little better.

[penngray]

Suaray, thank you so much for the rest....incredible education here for me!!!

I will digest it all today, reply back later and get something going tonight.

[Saurav]

Forgot to mention one more thing - in the DCX interface, the parametric EQ's are called 'BP'. Thought I'd mention that in case you haven't played with the EQ settings much. Also, you've probably got the global settings done already - make sure the inputs are set correctly to analog or AES/EBU, link inputs A and B together, link the outputs correctly (I use LMH LMH, but you might have wired it some other way), and so on.

I definitely want to understand the passive design a little better.

Can't help you there. Passive XO design is a black art and my knowledge there is pretty much zero

[penngray]

Forgot to mention one more thing - in the DCX interface, the parametric EQ's are called 'BP'. Thought I'd mention that in case you haven't played with the EQ settings much. Also, you've probably got the global settings done already - make sure the inputs are set correctly to analog or AES/EBU, link inputs A and B together, link the outputs correctly (I use LMH LMH, but you might have wired it some other way), and so on.

Thanks, I have 2 DCX setups running already.

Can't help you there. Passive XO design is a black art and my knowledge there is pretty much zero

You and me both...

[Saurav]

I have 2 DCX setups running already.

OK, then you're ahead of me there

I also realized that I can re-export your impulse response measurements as PCM WAV files for Dennis to play with. Don't know why I didn't think of that earlier.

http://home.comcast.net/~saurav/penngray/Celestion1425_HF-PCM.wav

http://home.comcast.net/~saurav/penngray/TD12M_LF-PCM.wav

Dennis, I thought you said you wanted to look into the AC offset. I get 0.3ms by eyeballing in HOLM (impulse peaks are at 0.22 and 0.52ms from my current 0 reference), and LspCAD ended up with the deepest notch with a delay of 0.21ms. It would be helpful to see what you come up with using the minimum / excess phase method.

[penngray]

Here are the ARTA measurements

Image not available

Image not available

Comparison of ARTA vs HOLM

Image not available

It looks like Im doing it the right way.

btw, Sauray, could you post your impulse so I can understand what you are saying about the room impact?

[Dennis H]

I got .35 ms difference between the two. I'm not too confident about the tweeter's excess phase plot though. I couldn't get it quite flat and near zero at 1500, possibly because that's near resonance. The gate starts at sample 10000 and is a dual gate, 6 and 100 msec.

[penngray]

Okay, time to set up the DCX and run more measurements.

[Saurav]

btw, Sauray, could you post your impulse so I can understand what you are saying about the room impact?

It's Saurav

I'll try and post one tonight, I don't think I have any images saved. If you look at your ARTA impulse response, you can see that the impulse starts at around 5.something ms, and the first real wiggle further out is at 12.something ms. So you have about a 6-7ms window that's clear of reflections. If I did my math right, that translates to a path difference of about 7-8'. Path difference means the "speaker -> reflecting-surface -> mic" path was 7-8' longer than "speaker -> mic" direct path. Given what you described of your room setup, that's probably the ceiling reflection?

My room isn't treated (it's my living room, so I get by with what I can), and it's smaller, so the first reflection is stronger, and it occurs sooner. So comparing with your ARTA impulse response, in my room the first wiggle would show up somewhere around the 9ms range, and it would be taller.

Regardless, the principle is the same - you want to remove that reflection from the measurements, so you set the gate such that it ends just before that wiggle. And this is why people measure outdoors / in a garage / etc. - you want to get as long of a reflection-free window as possible. As the window gets shorter, the calculated low-frequency response becomes less accurate (loosely speaking). HOLM shows this in the FR plot (if you're looking at gated response) with a little marker that says 'gating', which means that around / below that frequency, the plot isn't accurate.

[penngray]

It's Saurav

Crap, sorry about that...It looked like a "Y"

Measurements with settings above.

Blue with tweeter polarity inverted on DCX, Red with polarity normal.

Image not available

Measured 45 deg.

Image not available

dotted line is phase, I just want to understand all this phase stuff since it seems phase is the SQ killer.

[Saurav]

Something's wrong. I would suggest double-checking that you entered the delay number into the woofer's settings, and that you changed the on/off setting at the top of the woofer delay page to 'on'. Your measurements look like what my sims looked like before I dialed in the delay, so my first suspicion would be that the delay isn't set right.

[penngray]

I will check that delay, I might have it off now that you post that!!

Give me 5 minutes.