QSC Waveguide project, XOs (active and passive)

[augerpro]

He should be able to move that shunt resistor after the cap then recalculate the values to arrive at the same frequency response. You can't always do that when moving the series resistor, so if its current placement is the only way to the better response than that doesn't have to move. There should be a way in the software to use the current frequency response as a target, then move the shunt resistor, and have the optimizer bring the response back to where you started.

[Saurav]

There should be a way in the software to use the current frequency response as a target, then move the shunt resistor, and have the optimizer bring the response back to where you started.

That's a good idea. I know how to do this in SW, but it's not the easiest interface to describe

Doug, try something like:

* Right click the calculated FR, rename it to something else
* Re-configure the network, move the shunt resistor to after the cap
* Go into the network optimizer, click on the target button, give it the renamed FR that you saved off earlier. Select both resistors and the cap as things to try optimizing

Then re-run the optimizer and see what happens.

[penngray]

He should be able to move that shunt resistor after the cap then recalculate the values to arrive at the same frequency response. You can't always do that when moving the series resistor, so if its current placement is the only way to the better response than that doesn't have to move. There should be a way in the software to use the current frequency response as a target, then move the shunt resistor, and have the optimizer bring the response back to where you started.

I have tried many values but the best simmed response is always with the cap after the Lpad.

I couldnt get the optimizer to fix that problem either.

[Saurav]

Post the driver frequency and impedance responses, and a screenshot of your current crossover, and I can take a stab at Speaker Workshop

[penngray]

I couldnt attach the FRD files because they are too big so here are the links

http://www.penngray.com/htroom/measurements/TD12M_FRD.txt

http://www.penngray.com/htroom/measurements/TD12M_ZMA.txt

http://www.penngray.com/htroom/measurements/BMS4550_FRD.txt

http://www.penngray.com/htroom/measurements/BMS4550_ZMA.txt

[augerpro]

Can you post a plot comparing both filter responses?

[penngray]

I will later tonight Augerpro, I have been playing around with it more today but its family time again.

Thanks for all the input guys, Its tedious and fun, fun and fustrating to do it right....Man, the DCX is calling out to me saying stop doing on that messing around

[augerpro]

Did you ever compare the 1425 with the 1.5khz versus 2.0khz crossover? How did it sound?

[Saurav]

OK, first step - with the network in post 35, here's what I get for Speaker Workshop's predicted response:





There's a big impedance spike in the BMS response at 870Hz, which I think explains the bump in response at that frequency. That's probably also causing the negative summation around that region. You'll need a well-tubed notch filter to tame that. IIRC, Jon had 3 notch filters in his Avro Part or Isiris crossover to tame the tweeter impedance peaks.

Overall impedance looks like a 4 ohm speaker, not too bad.

So, all that tells me is that Doug did his SW sim correctly, because my pictures match his I'll play with the optimizer next and see what I can come up with. Caveat emptor, I suck at passive crossovers

Brandon, what was that topology again that you were playing with, I want to try for a 3rd order rolloff on the tweeter, by combining 1st and 2nd order Butterworths?

[Saurav]

After 30 minutes of Speaker Workshop, this is the best I can come up with, and frankly I think Doug's circuit has better response. I might be able to come up with something better if I switched to LspCAD at this point (since that optimizer can work on the whole circuit, not just one half at a time), but I'm trying to play by the 'rules'.







The impedance peaks at around 150 ohms at around 2.5kHz. Would probably be worth putting in an LCR leg to bring that down.

[augerpro]

I think your new model looks better actually. It's flatter overall. While phase integration exactly at the crossover point is not quite as good, it is still fine and it is better above and below that. Any reason you used two caps instead of a more standard cap and inductor? Any way I think your new crossover is looking more promising overall, just need to maybe address that CD resonance a bit more. What slopes were you using?

The impedance shows exactly what I was talking about. In Doug's post 35 crossover it is half all the way down in the bass. That's a lot of power to be grounding. Given the Re of the woofer Doug has more power going to ground than he will have going through the speaker!

The filter I'm using is a cascaded BW1+BW2 with an Fc at the frequency where the wavelength is 4x the AC offset of the tweeter and woofer. For these drivers it should be around 1300hz, but I think that was what I calculated for a backmounted woofer. Might be lower with this setup. but still a good place to start. Also the tweeter is reverse polarity.

[Saurav]

I think your new model looks better actually.

Heh, OK, if you say so. Shows how much I know about looking at FR simulations

Yeah the 2 caps looks silly doesn't it. I put a 2nd order filter ahead of the L-pad, the optimizer calculated 3.5mH for the shunt inductor, so I took it out and it didn't really make much of a difference to the response. I could probably combine the caps and come up with something equivalent.

I was working with LR4 1500 targets, since that's where Doug's XO ended up.

I'll take a stab at your BW1 + BW2 approach, and put a little more effort into the impedance. So is that symmetric, the low-pass on the woofer is the same slope at the same frequency?

[Saurav]

What are my options for extracting minimum phase from these FR measurements? I've tried Response Modeler, and on multiple occasions, it seems to have trouble with these FR plots that are presumably exported by HOLM or ARTA (not sure what Doug used). It gets to about 50% and then locks up, and an error file starts growing in the directory with thousands of entries that look like:

At 4/10/2010 5:20:57 PM inside object Phase.GetPhaseSample
--> Error Number: 9 Subscript out of range
------> CheckPoint Number: 1

If I look at the FR file in Notepad, the phase is wrapping all over the place, it goes down to -400 and up to 1000. Not sure exactly why that causes the spreadsheet to lock up. I don't have the DLL so I'm using the slower method, not sure if that might make a difference.

I'll see if exporting it back out from SW results in something that Response Modeler can handle.

[augerpro]

Can you plot the responses over each to compare them? Without the individual driver responses? I have hard time looking at those plots, had to print them in black and white to make it easier to read.

Yes both slopes are symmetric

[penngray]

Its the text file export from HOLM.

[Saurav]

Can you plot the responses over each to compare them?

Sure, which 2 do you want to see? Doug's post 35 final vs. my final?

[penngray]

Did you ever compare the 1425 with the 1.5khz versus 2.0khz crossover? How did it sound?

I did but it was with only one speaker and not with both. There was something not right with the 1.5K response vs the 2K response but that could be all subjective opinion considering I didnt spend a lot of time with it also, Im convinced the 1425 isnt good choice to cross below 2KHz. Zilch and I talked for a whole week on the 1425 problems below 2Khz (in another project thread, Reece's waveguide line array build).

I then started to focus more on the Celestion 1745 because it has a nicer impedance below 2KHz. I also took a 180 and became excited about passive XO design so Im hear at this point. Many drivers, partial solutions.

[augerpro]

Sure, which 2 do you want to see? Doug's post 35 final vs. my final?

Yes, summed response versus summed response

Doug> so it did audibly improve with the 2khz crossover?

[penngray]

The impedance shows exactly what I was talking about. In Doug's post 35 crossover it is half all the way down in the bass. That's a lot of power to be grounding. Given the Re of the woofer Doug has more power going to ground than he will have going through the speaker!

I was using the same XO Pete/Zilch used on their woofer. I just wanted to see what it did and it seem to work out okay. Why would it drop the impedance so much?

Im not sure how to even see the changed impedance in speaker workshop. I tried to create an impedance curve from my reponse plot but it was just a blank screen.

[penngray]

Yes, summed response versus summed response

Doug> so it did audibly improve with the 2khz crossover?

Yes, there was a little less what I call harshness. It was recommended to listen to vocals on both so I picked some eagles songs that had great harmony (no instruments) and with one speaker I switched back and forth. Its really a crude test.

[augerpro]

Pete's crossover won't have the impedance issue, the problem arises when you moved the shunt resistor before the rest of the filter.

[penngray]

Can you post a plot comparing both filter responses?

Here is the cap before and after comparison

Image not available

When the cap is behind the Lpad it maintains its flatter response and doesnt start sloping until 2K, before it starts sloping closer to 3K. That early slope creates a dip with my woofer in intial sims.

[Saurav]

I was using the same XO Pete/Zilch used on their woofer. I just wanted to see what it did and it seem to work out okay. Why would it drop the impedance so much?

It's the L-pad on the tweeter, ahead of everything else, that causes impedance issues across the whole frequency range. Impedance and passive XOs can get tricky sometimes. This is one of the big reasons I find active easier

[penngray]

It's the L-pad on the tweeter, ahead of everything else, that causes impedance issues across the whole frequency range. Impedance and passive XOs can get tricky sometimes. This is one of the big reasons I find active easier

Ah...that is a great point. Thanks!

I wonder if we bi-wired then things would be fine. Keeping the passive XOs seperate. EDIT: I mean bi-amp.

Oh, I still love the DCX but its nice to understand the passive design. One thing is for sure its giving me even a great appreciation for active designs how they simply do not have all these minor issues.

[Saurav]

Black is Doug's original, red is mine from the earlier post. Smoothed to 1/16th octave to make it easier to compare.



I'm working on something that should hopefully be better, I'll try to post something later tonight.

[Saurav]

I wonder if we bi-wired then things would be fine. Keeping the passive XOs seperate.

No, because the amp would still see the full impedance. Running separate wires to the tweeter and woofer portions won't change that. If you bi-amp, that's a different story.

[penngray]

Heh, OK, if you say so. Shows how much I know about looking at FR simulations

Im still confused on why he thinks yours is flatter...I will do both and compare them.

Your XO on the woofer creates more of a hump, I guess that doesnt mater.

[penngray]

No, because the amp would still see the full impedance. Running separate wires to the tweeter and woofer portions won't change that. If you bi-amp, that's a different story.

Oops, I meant bi-amp!!

[penngray]

Black is Doug's original, red is mine from the earlier post. Smoothed to 1/16th octave to make it easier to compare.



I'm working on something that should hopefully be better, I'll try to post something later tonight.

Here is to me being a noob, I like the black better in the XO region I only have a peak in the 3K range to deal with.

[penngray]

For some reason I do not get your response with the same values...hmmm.

Image not available

[augerpro]

That's easier to read. So what I saw was smoother response in the crossover region, even though it is a bit drooped there. Kind of a toss up, but I'd prefer a slight broad dip then some peaks. Then above that the treble overall was flatter relative to the rest of the response. It looks like the treble may be a little high in level, even though it is flatter.

[augerpro]

Drop that shunt resistor to like 5 ohms

[Saurav]

For some reason I do not get your response with the same values...hmmm.

Looks the same to me, what are you seeing that's different? My response curve is smoothed 1/16th octave, maybe that's what you're seeing?

Edit: I see, mine dips down in the woofer range towards the XO frequency, yours doesn't.

And yeah, the tweeter level needs to be dropped some more.

[penngray]

I will check what data I used with mine. I have to believe I used the same data....arrg, frustration sets in.

Time to go to bed.

Thanks for the education guys, Im getting it.....slowly!

[Saurav]

There's something else that's weird. I took the crossovers and dropped in the FR data that I had from when I was doing the DCX sims for the TD12M-4550. The response changes quite significantly, which means the relative phase information in the two sets of measurements is different. I can't tell you which is correct. So the question is, did you calculate the FR curves from impulse responses? If you did, what did you do for the gating and the zero-point settings?

You'll have to build one of these passive XO versions and see if the measurements match the sims. Unlike with an active XO where a phase error can be corrected with a simple delay, on a passive XO it pretty much means going back to the drawing board and starting from scratch.

[penngray]

There's something else that's weird. I took the crossovers and dropped in the FR data that I had from when I was doing the DCX sims for the TD12M-4550. The response changes quite significantly, which means the relative phase information in the two sets of measurements is different. I can't tell you which is correct. So the question is, did you calculate the FR curves from impulse responses? If you did, what did you do for the gating and the zero-point settings?

You'll have to build one of these passive XO versions and see if the measurements match the sims. Unlike with an active XO where a phase error can be corrected with a simple delay, on a passive XO it pretty much means going back to the drawing board and starting from scratch.

Yes it was done in HOLM, Yes I did the zero-point settings and the gating was about > 4 ms.

I know I removed the 36 uf Cap from the tweeter on this measurement.

I will double check everything again and run new measurements. Having my sim look different then yours using the same values worries me.

I do have a question about other software packages, do they reverse engineer a crossover to DCX values? Say I copied your posted XO above can I create the values needed for the DCX?

[Saurav]

do they reverse engineer a crossover to DCX values

Not directly. What you'll have to do is export the woofer and tweeter response from Speaker Workshop. Then you'll need to use the LspCAD Pro demo, which can optimize to an arbitrary target function. Import those FR curves in, set up a DCX XO with the required functions (some filter and some EQ), and play with it until the response matches what you had in Speaker Workshop.

[Saurav]

I imported the .txt files into HOLM to check. The first thing I noticed is that the woofer response appears inverted, the impulse goes downwards. The woofer impulse also shows up before t = 0. I'm not really sure what that means since I imported FR and not impulse data. The gap between the two seems OK.

Yeah, I'd suggest re-checking those measurements.

Edit: I think I know what happened. Looks like you used the tweeter to lock t=0, then measured the tweeter and woofer. Since the tweeter is further away, that put the woofer impulse at t < 0. The relative time delay looks right, so just repeat the same thing but start with the woofer, and use that to lock t=0. Then the tweeter will be to the right, and the phase should look right. Also check the wiring on the woofer and make sure it's not backwards.

[Dennis H]

The first thing I noticed is that the woofer response appears inverted, the impulse goes downwards.

I was wondering about that. In Pete's crossover, he has the tweeter inverted relative to the woofer but Penn got decent modeled response with them wired the same. Most likely the woofer was accidentally inverted during the measurement.

[soho54]

FWIW, I can get the same thing as Saurav's FR/IMP graphs in post 55 with the tweeter inverted.

[penngray]

My felt arrived (100% wool).

I cut a hole the size of a dime, placed it between the horn and the CD.

Here are some test measurements with my Celestion 1745/QSC 152i waveguide.

EDITED, need to fix measurements.

[penngray]

I imported the .txt files into HOLM to check. The first thing I noticed is that the woofer response appears inverted, the impulse goes downwards. The woofer impulse also shows up before t = 0. I'm not really sure what that means since I imported FR and not impulse data. The gap between the two seems OK.

Yeah, I'd suggest re-checking those measurements.

Edit: I think I know what happened. Looks like you used the tweeter to lock t=0, then measured the tweeter and woofer. Since the tweeter is further away, that put the woofer impulse at t < 0. The relative time delay looks right, so just repeat the same thing but start with the woofer, and use that to lock t=0. Then the tweeter will be to the right, and the phase should look right. Also check the wiring on the woofer and make sure it's not backwards.

Im sure I screwed up somewhere in the testing since my main goal this time was just to get some data and learn how to build XOs. I probably didnt check something properly.

[soho54]

Please post the final measurement files when you get them. It's something interesting to pass the time with. I have half a dozen different layouts already. I'm really liking these drivers with a quasi-series xover inspired by the NatPs. It's a lot of parts, but they are smallish values. :B

[soho54]

Oh yeah, could you add ~90dB to the HOLMImpulse info for the posted files. Some of the FRD based programs laugh at negative SPL levels. It isn't that bad to convert, but I'm lazy. :T

[Saurav]

I found an interesting limitation in HOLM Impulse yesterday. The SPL scale cannot be set to go higher than 20dB. Also, HOLM cannot scale a dataset by more than 99dB.

My measurements are usually in the -ve range as well, but I had one set where I was messing with the calibration so the numbers were up in the 140dB range. So when I pulled that into HOLM, I could not change the vertical scale to show me the full response. It was mildly frustrating - I don't see why those limits should be built it, the math shouldn't be any harder if you're scaling by 50dB or 100. It just seemed like the designer had a 0dB reference system in mind, instead of an absolute-SPL system.

Speaker Workshop works slightly better for this kind of manipulation. I'm sure the paid version of ARTA would do this as well.