How to set lpads when using DSP?

[DingDing]

Hi,

I reason that with DSP the Lpads on 4343 are as useful as a big toe on my forehead. Wondering how it would be best to set the lpads such that they do not interfere (don't want to disconnect them unless there will be substantial gain)?

Max, 0? It's hard to set them exactly at 0, but I guess I could measure the resistance?

[pos]

Are you using your DSP as an active crossover feeding each driver independently, or as an EQ in front of the existing passive crossover?

[DingDing]

Are you using your DSP as an active crossover feeding each driver independently, or as an EQ in front of the existing passive crossover?

Both. Using miniDSP 4x10 HD to drive 4343 half active, half passive. The bass drivers are actively crossed over from the rest which are passive. I've also done some close range measurements (50 mm) of the bass and mid driver to correct small issues with PEQ.

So for everything but the bass, there is EQ infront of the passively divided mid, hf and uhf (which is where the lpads are, as the bass seems to be connected directly to the binding posts [passive xover by Guido]).

My idea was to set them optimally so that I can retire the lpads and rather do it digitally with EQ before the passive xover of the mid, hf and uhf. Problem is I don't know what would be the optimal setting.

[pos]

Ok then you should definitely use the lpad, but instead of trying to obtain a flat overall response you should try to get the smoothest transition between drivers at their crossover frequencies, and then EQ the global response back to flat (or whatever target you like).
Global EQ will let you correct response variations, but proper summation at crossover frequencies must be done within the crossover.
Lpad will let you compensate for drivers and passive components variation (including the lpad themselves).

Of course you should first set the lpad of the lowest sensitivity channel at its maximum (the 2121, most probably), and work your way from that., with measurements.
You will most probably end up with different settings for L and R, and thus different EQ needs, but that is what it takes.

On a time-aligned system you could reverse the polarity of the middle driver during measurement (the compression driver here) and look for the deepest null at the crossover point, but I am not sure you would get anything useful with this method here...

[DingDing]

Thanks, pos.

So I guess those lpads are usefull after all then :)

Do have a question though. To optimize the (passive) xover points should I measure close range, 1m or at different locations (with small variations) in the sweetspot? If close range, where should the microphone be placed in relation to the two drivers which are crossed over? Have not done this before, and are currently in an experimental phase.

I've been contemplating getting miniDSP's openDRC and time/phase align the drivers by going active all the way. I've heard rumors you're doing just that (http://www.audioheritage.org/vbulletin/showthread.php?35984-2235h-subwoofer-build-for-JBL-4343B&p=364718&viewfull=1#post364718). How is that working out for you? Do you have a project thread or something like that going? If so, please share.

Have to work with IIR and gain some experience before I take the big jump and go fully active with FIR.

[pos]

Do have a question though. To optimize the (passive) xover points should I measure close range, 1m or at different locations (with small variations) in the sweetspot? If close range, where should the microphone be placed in relation to the two drivers which are crossed over? Have not done this before, and are currently in an experimental phase.
Ideally you should measure at the listening position, but it is often impossible to get a usable measurement far enough from the speaker in a home environment, because of the short gating time needed to remove reflections.
A good start is probably 1m away, between each pair of drivers you want to align.
The compression/slot crossover will be much easier to measure as even a short gating will give you enough precision.

You should do several measurements each time (some measurement software will let you average them) at slightly different positions to detect and ignore diffraction and other position-specific artifacts.
(by the way you should leave the lens in place as they will greatly influence the response in the upper range of the compression)

For this kind of measruement spacial averaging (ie several measurement at different positions) is much more efficient than frequencial averaging (smoothing), because you need precision.

One thing you can also try is to use a very long MLS signal (in HOLM for example you can specify the length, up to 90sec at 48k I think), and slowly move the mic around (with a figure of 8 for example). This should give usable steady-state measurements even at the listening position. Phase information will be lost in the process, but that is not a problem.


I've been contemplating getting miniDSP's openDRC and time/phase align the drivers by going active all the way. I've heard rumors you're doing just that (http://www.audioheritage.org/vbulletin/showthread.php?35984-2235h-subwoofer-build-for-JBL-4343B&p=364718&viewfull=1#post364718). How is that working out for you? Do you have a project thread or something like that going? If so, please share.

Have to work with IIR and gain some experience before I take the big jump and go fully active with FIR.

I believe the 4344 would greatly benefit from a fully active time-aligned crossover, as the deep horn is probably is far from geometrically aligned with the 10" and slot.
This is doable with any good IIR DSP crossover.
What FIR will let you do will be to linearize the phase. This can be done on the stereo signal in front of a time-aligned crossover (which the current passive version is not, obviously), with an openDRC or any software convolution engine if you use a computer as a source.
Phase linearization give a subtle improvement (you should be able to hear its effect on the 300Hz crossover), but the main improvement will be get from a properly time aligned crossover...

[DingDing]

Ideally you should measure at the listening position, but it is often impossible to get a usable measurement far enough from the speaker in a home environment, because of the short gating time needed to remove reflections.
A good start is probably 1m away, between each pair of drivers you want to align.
The compression/slot crossover will be much easier to measure as even a short gating will give you enough precision.

You should do several measurements each time (some measurement software will let you average them) at slightly different positions to detect and ignore diffraction and other position-specific artifacts.
(by the way you should leave the lens in place as they will greatly influence the response in the upper range of the compression)

For this kind of measruement spacial averaging (ie several measurement at different positions) is much more efficient than frequencial averaging (smoothing), because you need precision.

One thing you can also try is to use a very long MLS signal (in HOLM for example you can specify the length, up to 90sec at 48k I think), and slowly move the mic around (with a figure of 8 for example). This should give usable steady-state measurements even at the listening position. Phase information will be lost in the process, but that is not a problem.

Thank you for such great advice. :) For measurements I'm using UMIK-1 with open source REW (Room EQ Wizard). For DSP I use a combination of MiniDSP 4x10 HD and also open source Equalizer APO (more flexibility and only compurter as source, also good integration with REW as it interprets the filter txt files that REW generates).

Followed your advice and averaged different measurements as well as playing pinknoise through each speaker's mid, hf and uhf individually with the microphone at relatively close proximity (about 1m on axsis) while looking at the RTA as I slowly adjusted the lpad. I then EQ'ed the response so they matched closly in level from the sweetspot. Probably as good as it gets by my hands with the passive network now, so thanks a bunch.


I believe the 4344 would greatly benefit from a fully active time-aligned crossover, as the deep horn is probably is far from geometrically aligned with the 10" and slot.
This is doable with any good IIR DSP crossover.
What FIR will let you do will be to linearize the phase. This can be done on the stereo signal in front of a time-aligned crossover (which the current passive version is not, obviously), with an openDRC or any software convolution engine if you use a computer as a source.
Phase linearization give a subtle improvement (you should be able to hear its effect on the 300Hz crossover), but the main improvement will be get from a properly time aligned crossover...

Yes, you must be right about the geometry of the horn. Have thought about that since I recently learned why some manufacturers use stepped baffles. :D This will be quite the undertaking for a noob, but I figure it will probably be the biggest improvement that can be done to this system. This is all very exciting because with DSP you get instant gratification, haha.

I recently time aligned the parts of the system that was possible (two DIY subs xover @ 80 Hz and 300 Hz on the speakers) with 4x10 HD. The improvement was huge. I thought the system was great before time aligning, but the difference when switching between the time aligned and not time aligned settings makes the old setup sound smeared and blurry. Makes me anxious to hear what a completely time aligned 4343B would sound like. The big question is what kind of power to use without breaking the bank but still keep good sound quality.

[martin2395]

I think that you also have to know where the limit of such a speaker is and when it's time to stop pumping money in it and upgrade to newer model :)

BTW I also wonder what a 4-way active 4343/4344 can, compared to the original. Only the 4 amps will cost a fortune.

[DingDing]

I think that you also have to know where the limit of such a speaker is and when it's time to stop pumping money in it and upgrade to newer model :)

BTW I also wonder what a 4-way active 4343/4344 can, compared to the original. Only the 4 amps will cost a fortune.

Haha, true that, martin. However, I can't think of a better looking speaker than 4343B and mine are still in pretty terrible condition as I've not yet come around to restore them. Good sound quality makes for utter laziness.

I'd love to get K2 S9900 at one point, those are the best speakers I've ever heard, but unfortunately they're way out of my league. Also, I've got to admit that the journey and joy succeeding in making something better is a driving force for me within this hobby. If I really wanted K2, I could spend all the time I dabble in this hobby working extra hours, saving the money and probably afford them and semi decent amps to drive them in a few years. At the end the sound quality would be better than my current path but it would not be as fun getting there. :)

[pos]

I recently time aligned the parts of the system that was possible (two DIY subs xover @ 80 Hz and 300 Hz on the speakers) with 4x10 HD. The improvement was huge. I thought the system was great before time aligning, but the difference when switching between the time aligned and not time aligned settings makes the old setup sound smeared and blurry. Makes me anxious to hear what a completely time aligned 4343B would sound like. The big question is what kind of power to use without breaking the bank but still keep good sound quality.

Building a 5-way active system is no easy task!
You have to take extra care about phase coherency and the effect of one crossover point on the others: http://www.linkwitzlab.com/frontiers_5.htm#V

So for example in your case I suppose you are using different channels of your 4x10 for the sub, woofer and mid+cd+tweet.
So that means that if you are using 24dB/oct filters for the sub/woofer crossover then you should add a 80Hz 2nd order allpass filter to the mid+cd+tweet section.
If you go fully active you will have to add allpasses on all channels but, reflecting crossovers from the lower channels.

[pos]

By the way, how did you time-align the 80Hz and 300Hz crossovers?

[DingDing]

Will reply to the other post a little later. Got to read and understand the resource you provide and all-pass filters.


By the way, how did you time-align the 80Hz and 300Hz crossovers?

By compromise. As I was measuring the IR I saw that the frequency at which I started the measurement had significant impact on the timing at which the signal arrived at the listening position.

Since the subs have a LPF at 80, the bass driver a HPF at 80 and LPF at 300 and (mid+hf+uhf) a HPF @ 300, the problem was time aligning the bass driver. If I aligned it at 80 it would be aligned at the xover with the sub, if at 300 the top drivers.

So reasoned that bass at 80 is more important for tactile feel (which is a preference of mine), hence I aligned it @ 80. Would you say doing it at [(300-80)/2 + 80] = 190 Hz would be a better compromize? Should probably have tried several points to compare, but I was so happy with the first result I just sat down listening. The fun thing is that I don't really know what I'm doing, so it would be hilarious if I did it wrong and the sound actually sucks, haha.

Here's a rundown of how I measured:
(1) Calibrated USB sound card *
(2) Calibrated dB-meter
(3) One channel for the meter, another for loopback to get a timing reference.
(4) I disabled all HPF and LPF on all drivers, then started the measurements for each component of the system at the frequency at which I wanted to time align it at.
(5) With all components measured, I found the latest arrival and adjusted all the other components to match that. Did iterative measurments during the process and verified that the impulses were aligned. Actually had to invert both subs and bass drivers, because there was not enough delay to align them properly without doing it.

*I actually calibrated with miniDSP in the chain, as I thought it would contribute to latency.

62835

I then measured using this configuration

62836

[pos]

By "the frequency at which I started the measurement" you mean the start frequency of the sweep?
I am not familiar with REW but I imagine you are using a sweep here. You should set the start frequency at a low value.

It is good that you removed the electrical crossovers from the chain, but you still have natural filters from the drivers+box, especially for the 2121 in the 14L sealed box, with natural 12dB/oct HP filter probably around 100Hz or so. You have to take this into account when designing your crossover (acoustical filter=electrical filter+"natural" filter).
Time-aligning a system with filters in place (be it electrical or "natural") is difficult because you cannot simply align the IR peak. What is important is phase coherency throughout the final crossover.

Here is a simple procedure that you can try:
- seal the bassreflex ports of your 4343 (not just for the measurement), as this will get you a much easier phase shift to deal with.
- measure each driver (sub, woofer, mid) in close range (~1cm) with no electrical filter
- import the response in REW and try different EQ and filter settings to obtain a clean LR 24dB/oct target for all your drivers (LP for the sub, HP and LP for the 2235, HP for the 2121)
- insert those settings into the minidsp
- insert a 2nd order Q=0.707 allpass at 80Hz in the 2121 channel
- depending if your sub is sealed or BR, insert a 1st or 2nd order allpass filter (f=f3 or BR frequency) in the 2235 channel (and 2121 channel if you have EQ points available)
- set the delay for the exact differences in distance between drivers at your listening position (the plan of the baffle should be close enough if the subs are next to each speaker...)

That way you should get phase-coherent and amplitude-complementary crossovers at 80Hz and 300Hz.

[DingDing]

So this is what happens when you ask for the mundane task of how to set the LPADS???? :D


Building a 5-way active system is no easy task!
You have to take extra care about phase coherency and the effect of one crossover point on the others: http://www.linkwitzlab.com/frontiers_5.htm#V

So for example in your case I suppose you are using different channels of your 4x10 for the sub, woofer and mid+cd+tweet.
So that means that if you are using 24dB/oct filters for the sub/woofer crossover then you should add a 80Hz 2nd order allpass filter to the mid+cd+tweet section.
If you go fully active you will have to add allpasses on all channels but, reflecting crossovers from the lower channels.

Yes, each component of the system is using an individual channel. The subs are placed close to the speakers (and each other) to gain acoustical coupling as far down as possible. Also, they are very big and 100 kg/ea, so not a lot of whiggle room for placement. Found that running the subs with L, R was better than 2xL+R even though most recordings are mono and bass < 80 Hz are non-directional. Summing the signal just made the gain structure harder to optimize, hence L, R.

Got to read up on the theory here, a lot of these concepts are new to me. I've vastly underestimated what is required, but will keep hammering at it.


By "the frequency at which I started the measurement" you mean the start frequency of the sweep?
I am not familiar with REW but I imagine you are using a sweep here. You should set the start frequency at a low value.

It is good that you removed the electrical crossovers from the chain, but you still have natural filters from the drivers+box, especially for the 2121 in the 14L sealed box, with natural 12dB/oct HP filter probably around 100Hz or so. You have to take this into account when designing your crossover (acoustical filter=electrical filter+"natural" filter).
Time-aligning a system with filters in place (be it electrical or "natural") is difficult because you cannot simply align the IR peak. What is important is phase coherency throughout the final crossover.

Here is a simple procedure that you can try:
- seal the bassreflex ports of your 4343 (not just for the measurement), as this will get you a much easier phase shift to deal with.
- measure each driver (sub, woofer, mid) in close range (~1cm) with no electrical filter
- import the response in REW and try different EQ and filter settings to obtain a clean LR 24dB/oct target for all your drivers (LP for the sub, HP and LP for the 2235, HP for the 2121)
- insert those settings into the minidsp
- insert a 2nd order Q=0.707 allpass at 80Hz in the 2121 channel
- depending if your sub is sealed or BR, insert a 1st or 2nd order allpass filter (f=f3 or BR frequency) in the 2235 channel (and 2121 channel if you have EQ points available)
- set the delay for the exact differences in distance between drivers at your listening position (the plan of the baffle should be close enough if the subs are next to each speaker...)

That way you should get phase-coherent and amplitude-complementary crossovers at 80Hz and 300Hz.

Holy macaroni this is not simple and will certainly not yield instant gratification! :D

Yes, in REW "all" measurements are done with a sine swipe of your choosing (0-x Hz) where x is limited by the sampling rate obviously. "All" because you can use the RTA and run averages on content playing back or pinknoise, whitenoise etc too. All measurements with info on phase, decay so on so forth are done by swipes however.

Thank you very much for the procedure. Will be working through that and reading up on xover theory going forward. Can post some screenshots of the measurement results if anyone is interested to see and maybe want the filter settings I come up with for the close range measurements.

Sealed sub(s) btw. You will probably laugh, but here's the setup so you have a visual.

62837
Driver: LMS Ultra 5400 in well braced, large sealed enclosures, no damping.

If you're interested in seeing my current miniDSP settings (http://www.hifisentralen.no/forumet/akustikk-og-rom/75034-hjelp-til-oppsett-av-hardware-tidskorrekt-impulsrespons-i-rew.html#post1858538), I posted them here. It's in a Norwegian forum, but the screenshots are self explanatory. Nobody has reacted to the settings, but I realize that you guys are probably bigger on xover designs than the group over there.

Again, really appreciate the help.

---

How low would you set the starting freq. for the swipe when measuring the mid+hf+uhf close range? Could do 0-24kHz all the time if need be. A little concerned about 1cm distance from the high excursion subs though, it will bang into the mic at low frequencies, so have to take it further away from the subs.

[DingDing]

I keep looking for the "Like"-button to show my gratitude towards posts on this forum all the time. It should seriously be implemented, the CMS driving the forum has it as an addon/module I'm sure.

[pos]

Wow that is some serious subs right there!
I think you should probably cross them lower than 80Hz: the 2235 are capable woofers!

Regarding the allpass filter thingy, in fact you would be better off with plain filters, and it will be easier to explain:
So imagine you want to simulate a cascading filter like in Linkwitz' drawing: the signal is filtered for the sub and the high-passed signal goes to the woofer, which in turn passes the high-passed signal to the mid...
As you cannot do such cascaded filters with your miniDSP you just have to "replicate" the filter setting of the lower channels on each channel.
So for example if you woofer has a 24dB/oct acoustical HP filter at 80Hz then you just have to dial an electrical 24dB/oct HP filter for your mid...

Regarding the sweep, yes make them start as low as possible for the driver you are testing. So that would be something like 10Hz for the subs, and maybe 50Hz for the mid...
No need to push the volume, especially with a close range measurement: you should not have excursion problems.

Do post those measurements on the forum in txt or wav format (do not forget to plug the vents on the 4343), I will try to simulate some EQ and filter settings.

[DingDing]

pos, they are excellent drivers, high linear excursion, xmax = 33.7mm and xmech = 44mm with the big enclosure the system is quite efficient too. I've pumped about 70V into each (measured at the binding post) and the maximum power draw at the socket has been only 1381w for both(!) when driving them really hard with music. They have continous power handling of 2kW and are made to be excellent in small enclosures of 100L and even less. It's important with some space behind the pole vent for airflow. You can read and see measurments of this driver in 100L closed box @ data-bass.com (http://www.data-bass.com/data?page=system&id=3&mset=35) if you're interested.

This is what I've been spending my past 4-6 months on, estimating in WinISD, learning about building materials, tools and closed designs. They are well braced, but oscillate a little bit when I run 10 Hz bursts @ 70V due to the share force. They sound really good too. Not like a max spl type woofer. Problem is they're expensive, and I've strained my budget hard by getting them. Started with one in 104L, but ended up with two of these with about 155L net internal working volume to get more efficiency and less distortion due to compression (not that it was an issue with the 104L design).

Have used plenty of EQ on the low end through close range measurements and linkwitz-transform through biquad in MiniDSP. This was my first venture into DIY, I'm very proud but have not done any finish work as I'm in the experimental phase.

62838

62839

62840

62841

62842

http://www.hifisentralen.no/forumet/attachments/diy-og-utvikling-ha-yttalere-forsterkere-etc/270965d1405949150-ti-tommler-og-18-tommere-ma-let-om-reda-vende-bass-uten-mye-forvrengning-cmuniq3.jpg

[DingDing]

I think you should probably cross them lower than 80Hz: the 2235 are capable woofers!

Regarding the allpass filter thingy, in fact you would be better off with plain filters, and it will be easier to explain:
So imagine you want to simulate a cascading filter like in Linkwitz' drawing: the signal is filtered for the sub and the high-passed signal goes to the woofer, which in turn passes the high-passed signal to the mid...
As you cannot do such cascaded filters with your miniDSP you just have to "replicate" the filter setting of the lower channels on each channel.
So for example if you woofer has a 24dB/oct acoustical HP filter at 80Hz then you just have to dial an electrical 24dB/oct HP filter for your mid...

Regarding the sweep, yes make them start as low as possible for the driver you are testing. So that would be something like 10Hz for the subs, and maybe 50Hz for the mid...
No need to push the volume, especially with a close range measurement: you should not have excursion problems.

Do post those measurements on the forum in txt or wav format (do not forget to plug the vents on the 4343), I will try to simulate some EQ and filter settings.

The 2235h are excellent. I absolutely love how they sound (which is why I've been wondering about 4355! My brother has a very large listening space and might end up getting them (he loves my 4343B a lot and hardly talks about anything else, hehe), but for my current room they're to big :(

I've experimented with different crossovers as these Ultra woofers can actually be crossed quite high too. As we know, my xover settings has not been perfect, so I will continue to experiment with different settings within the 4x10HD. With miniDSP we can have up to four different configurations live in the device at any given time, and you can also import settings. It's just a wonderful device. More advanced filters can be implemented through biquads, see:

62843
One biquad for the HPF and one for the LPF. (This screenshot is not from any of my active filters).

miniDSP has this huge spreadsheet out to find the coefficients for different filters (http://www.minidsp.com/images/fbfiles/files/All_digital_coefs_v1-20101026.zip) (link to spreadsheet), I've used it to find the coefficients for the Linkwitz-Transform (http://www.minidsp.com/applications/advanced-tools/linkwitz-transform) on the subs. Maybe I can use that to construct something more advanced.

Thank you, I'll try to do it the simple way w/o all pass filters first then, and try to improve upon that when I've found something which works well.

The reason why I'm using two measurment setups (UMIK-1 and all of what I showed you earlier with the diagram) is because REW does not allow for time coherent measurements when using USB microphones such as UMIK-1. To get proper timing info in REW, you need to have a loopback signal to discount the latency through the signal chain. As it stands, this is not possible with REW w/USB devices. I'm hearing Holm has capabilities for proper timing with USB so I should probably give it a go, as the dB-meter and everything is really complex in comparison to the excellent UMIK-1.

I can put the raw data from everything on Google Drive and link it from here if you or anyone else would like to look. Can also do txt and wav, np. Will show room response later too, you guys will probably laugh, because it's far from linear, but I like it. :D

Also, I'm hearing that the lead man on the Equalizer APO open source project has said that they will incorporate delays and more into the software, so it will probably turn into a super flexible DSP solution for those who are willing to build something with many audio I/O.



Unfortunately, I didn't have much time to work on the next version of E-APO
recently, but I will get back to it soon. This next version will contain a Delay
command and it will also be possible to copy audio to additional "virtual"
channels so that you can apply processing to them and mix them back later.

So you might not be able to directly adjust the delay
of frequency ranges but with some lines in the config file you can split the
audio into virtual channels for specific frequency ranges, apply the delay to
them and then add them together to form the output. You can ask me for an
example of that when the next version is released if you want.

Phase correction will not be possible yet, however.

So no phase correction soon, but these are exciting times considering the computational power of computers nowadays.

[pos]

Very impressive construction, I bet it sounds fantastic.
I wish I had your woodworking skills :)