How important is it to flatten a tweeter impedance peak?

[JonW]

So I’m trying to make up a decent crossover for the first time and wondering about the importance of squishing impedance peaks. I’ve got a tweeter that I measured and found a decent bump as seen here:




Zaph has been using a similar tweeter in his ZRT project linked here:

http://www.zaphaudio.com/ZRT.html

He says that this ~500 Hz peak indicates ringing and that this ringing must be avoided. His solution is to use an LCR circuit. As I understand it, that’s just a notch filter. The peak is not seen in the frequency response of the tweeter alone. So how important is avoiding that peak?

If it is important to avoid this peak an LCR filter is one option. Would an equally good option be to just roll off the tweeter such that it is not playing at that frequency?

For example, here is Zaph’s plot and it looks like he is down ~25 db where the tweeter peak comes into play. That’s far down but maybe not quite far enough. So he uses a notch filter to avoid it.





Alternatively, maybe one can just get off the tweeter well before that bump. For example, here is a preliminary crossover of mine. I’m down ~35 db before the impedance bump of the tweeter comes into play. Maybe that’s off the tweeter enough (or I could get off sooner if that would be sufficient). Is that approach OK?



Maybe 2 ways to skin a... er... make a speaker.

I know that Zobels are another issue, trying to get the baseline of the impedance plots to be flat but not dealing with the peaks. I guess it’s nice to have the baseline flat as well.

I’m just trying to learn here. Kind of like learning to park by sound, I know. Thanks for any help.

-Jon

[cjd]

It probably depends on the peak an d how far into it your crossover reaches. A lot of people argue this kind of thing, because they like low parts count cheap crossovers.

On the flip side you get Jon going after every single impedance peak in the Isiris (I think that was it) on the compression driver he used.

IMO it's something you just DO when you're building a system you want to be maximised. It's something you consider leaving out when if you're looking to shave cost.

Also, if taking it out has no real impact on the net response, your'e probably good to leave it out. If, however, it noticeably alters response with vs without: I would personlaly leave it in.

With what you're building, it's probalby worth using.

C

[JonW]

Hi Chris,

Thanks for the info. :T Makes sense. So 2 more questions:

-When you say “if taking it out has no real impact on the net response…” what do you mean by “response?” Is that the impedance response, the frequency response or either?

-Do you try to flatten the woofer impedance peaks as well? My understanding is that people typically don’t bother. More just deal with the tweeter.

[cjd]

Net response = frequency response. If you can take the network in and out without noticing it, chances are response is low enough below nominal that you won't hear a change. 500Hz with an 1800Hz crossover point, it is very likely below the point where any difference can be seen/heard. So it may be a question of how much a perfectionist you are.

Same goes on woofer impedance peaks - only time I think folks really mess with them are if they're using a tube amplifier that wants near resistive loads, or flattening out impedance peaks for a midrange or a multi-way system where it will make a difference on response.

C

[JonW]

Got it. It all sounds reasonable from a common sense perspective. But it's good to hear it from someone much more learned than I am. Thanks! :T

[JonW]

OK, let me follow up here with another related question…

I’ve been modeling the crossover. I put in an LCR to suppress the tweeter impedance peak. In a circuit with only the LCR it works well to flatten the peak. But when you’ve got the whole crossover in there, the LCR really makes no noticeable difference. So, as suggested, I’ll probably just leave the LCR out.

My only remaining question, at this point, is that I end up with an impedance plot looking something like that shown below. That peak at ~1,000 Hz is not really influenced much by putting in or taking out the tweeter LCR or a Zobel on the woofer. Frequency response and phase both look good at this point.

So how good or bad is such an impedance plot?


Image not available

[chasw98]

Frequency response and phase both look good at this point.

Can you post your frequency response and phase graphs? It might also be helpful to look at the 'before' individual driver impedance plots when you flattened them out with the LCR and Zobel.

Chuck

[Paul W]

I'd check out the little "squiggles" between 100 and 300Hz. Could be cabinet panel vibration, or just noise in the measurement environment...hopefully the latter.

[Saurav]

That peak at ~1,000 Hz is not really influenced much by putting in or taking out the tweeter LCR or a Zobel on the woofer.

Try an LCR (with appropriate values) at the input to the crossover, that should take care of the peak and leave the frequency response mostly unchanged.

So how good or bad is such an impedance plot?

If you're asking "Do I need to remove that peak"... with a solid state amp, probably not. If you're using a tube amp, it may help.

[jkrutke]

Here's what happens to the filter when I remove the notch.

It's difficult for a filter to do it's job when a large impedance peak is in the way. Sometimes a tweeter like this could have the impedance peak damped sufficiently by using a traditional L-pad but I don't think it was enough in this case. Additionally, the 6600 isn't the cleanest tweeter at 500 Hz so I wanted to keep the level down.

[JonW]

Can you post your frequency response and phase graphs? It might also be helpful to look at the 'before' individual driver impedance plots when you flattened them out with the LCR and Zobel.

Coming right up, sir

So the following 3 pics show the frequency response, impedance, and phase each shown with no LCR circuit on the left. And with the LCR circuit on the right. You can see that the differences seem, at least to me, to be pretty small with versus without the LCR. Although I now see that ~500 Hz tweeter shoulder in the tweeter only output that Zaph mentioned. I didn’t see it before because it was so far down it was off the graph. But the plots below show the plots farther down. The LCR provides a teeny bit better phase tracking so I'm now inclined to keep it. Even if there's not much change in the overall impedance.

[JonW]

Try an LCR (with appropriate values) at the input to the crossover, that should take care of the peak and leave the frequency response mostly unchanged

If you're asking "Do I need to remove that peak"... with a solid state amp, probably not. If you're using a tube amp, it may help.

Interesting. I moved the LCR to the beginning of the circuit and below is what I found. Much flatter but it does dip down to lower values in general. It contrasts quite a lot with the LCR later in the circuit. Hmmm… So is this flatter impedance curve better than the less flat one? If so it comes at the expense of now seeing that tweeter shoulder 35 db down around 500 Hz. And overall lower impedance- tougher for the amps to run, as I understand it. So no LCR, an early one or a late one? Pick your poison, I guess.

I'd check out the little "squiggles" between 100 and 300Hz. Could be cabinet panel vibration, or just noise in the measurement environment...hopefully the latter.

OK, thanks for pointing that out. The pic above is actually from my filmsy test cabinet and a preliminary crossover. Not the actual one I’m using now. I was asking a somewhat general question regarding what one should look for in an ideal impedance plot.

For raw impedance measurements in my final, much more solid cabinet, I’ve got the following 2 measurements for the woofer and tweeter. They’re also a tad bumpy. Maybe it’s just noise? Like, say, my neighbor’s lawnmower that was on while I was taking measurements.

[JonW]

Here's what happens to the filter when I remove the notch.

It's difficult for a filter to do it's job when a large impedance peak is in the way. Sometimes a tweeter like this could have the impedance peak damped sufficiently by using a traditional L-pad but I don't think it was enough in this case. Additionally, the 6600 isn't the cleanest tweeter at 500 Hz so I wanted to keep the level down.

Hi Zaph,

Thanks for chiming in and posting that. Interesting. That explains what you said in the ZRT writeup about needing to stop the tweeter ringing. Makes more sense now that I see it. And I just adjusted my graphs to see down to the point where I also get that same shoulder. I don’t get that until my tweeter is maybe ~35 db down, as seen in the pics above. Whereas yours is maybe ~25 db down where it comes into play. My guess would be that at ~25 db you should deal with it. At -50 db don’t worry about it. And at -35 db… you’re on the edge. Hmmm…

[Saurav]

I moved the LCR to the beginning of the circuit

You can't just move it, you need to calculate a new LCR that's designed for the new peak you're trying to flatten. If you keep the same LCR values, it'll try to flatten the same peak. As you can see, what that did is create an impedance dip around 500Hz, since that's the peak that your LCR is designed to remove.

It looks like you're using LspCAD, I don't know how it works there, but in Speaker Workshop, I say "Insert impedance compensation -> resonant peak" and then type in 1000Hz (or wherever the peak is). SW then calculates the LCR values based on the crossover's impedance curve and sticks it into the circuit. Then I tweak by hand and watch the impedance curve until it looks good enough (flatter, but not dipping too low).

You'll probably need both LCR circuits if you want to tame the tweeter's impedance peak as well as the overall crossover's impedance peak. If you don't know how to calculate the values for the 1kHz LCR, leave it out.

[---k---]

I'd check out the little "squiggles" between 100 and 300Hz. Could be cabinet panel vibration, or just noise in the measurement environment...hopefully the latter.

Not knowing what the heck I'm talking about, I'm wondering if the box needs a little stuffing. The peek on the impedance graph is very tall and sharp. If I'm not mistaken, a little bit of stuffing will bring that down and dull the point some and get rid of the reflection causing the grudge.

[JonW]

You can't just move it.

Got it. Makes perfect sense. I'll monkey around with it more and see how things change.

You'll probably need both LCR circuits if you want to tame the tweeter's impedance peak as well as the overall crossover's impedance peak. If you don't know how to calculate the values for the 1kHz LCR, leave it out.

As you can maybe see from the plots I posted above, the tweeter LCR doesn't change the overall impedance much at all. (But it does flatten the 500 Hz peak in a circuit without the rest of the crossover.) So that one may or may not be worth keeping in. It reduces the ~500 Hz shoulder -35 db down and help the phase matching a little.


Thanks for the help.

[JonW]

Not knowing what the heck I'm talking about, I'm wondering if the box needs a little stuffing. The peek on the impedance graph is very tall and sharp. If I'm not mistaken, a little bit of stuffing will bring that down and dull the point some and get rid of the reflection causing the grudge.

Hi Ryan,

Yeah, ignore that first plot. It was the test cabinets. For the T and W raw impedance plots I posted above, those were in the real, braced cabinets. And for stuffing, there was a layer of 3" OC703 on the bottom, another smaller piece at the cabinet rear behind the woofer, and a little bit of Acoustastuf. I'm wondering if it's actually overstuffed.

[Saurav]

As you can maybe see from the plots I posted above, the tweeter LCR doesn't change the overall impedance much at all.

I wouldn't expect it to. That's not what it's for. The impedance at 500Hz will be dominated by the woofer.

(But it does flatten the 500 Hz peak in a circuit without the rest of the crossover.)

It does more than that, it improves the tweeter crossover response down to a lower frequency/level, which improves the phase matching between the drivers.

So that one may or may not be worth keeping in.

I'd suggest either (a) trusting John Krutke's judgment on this, or (b) delaying this decision until after you've built and listened to the speakers with and without the notch. It may be audible in ways that don't show up easily in the crossover simulations.

[JonW]

Saurav,

Thanks again for the help. This is very educational for me. Have a look at the plot below. I added an LCR at the beginning of the circuit. The tweeter LCR is still in there. Muuuch flatter, eh? I feel like a big lightbulb just went on.

In Massachusetts, where I grew up, there is a coastal town called Marblehead. Whenever we would see someone not so smart suddenly understand a concept, with the lightbulb lighting up over their head, we would say "The light dawns over marble head." Or, more acurately, in a Boston accent: "The lite dawns ovah mahbel hed." That's pretty much me right now.

So presumably this flatter impedance curve is better now? Maybe a more even load for the amp or something? If that's true then I'm not seeing any down side to keeping it in there. Assuming it sounds OK and I don't care about the extra parts cost. And I gather that the 60 Hz and lower peaks you pretty much have to ignore for practical purposes.

I'd suggest either (a) trusting John Krutke's judgment on this, or (b) delaying this decision until after you've built and listened to the speakers with and without the notch. It may be audible in ways that don't show up easily in the crossover simulations.

I'm happy to defer to John's expertise. But I'm trying to *understand* crossover design, not just do what someone else did. Good idea to do some listening. I can clip together some crossovers and listen with versus without the LCR's and such.

Thanks again.

[Saurav]

So presumably this flatter impedance curve is better now?

It is in my opinion, but (a) I've 'designed' maybe 2 pairs of speakers in my life, so take anything I say with a grain of salt, and (b) I use low powered tube amps, so for me, the extra parts cost to get a flatter impedance curve is worth it. I've seen many designers who leave that notch filter out, and some who put it in. I suspect that with most amps, it does not make any audible difference. I don't know that it hurts anything (other than the cost of the speakers).

[cjd]

It really depends on the amp, how much difference the impedance peak in the summed response is. Tube amps don't like big swings in impedance, and while solid state does far better with it, there may still be benefit. Not sure - I've not actually messed with this, at least so far.

Phase tracking will show up in little details - better imaging/staging would be my guess. This is information our brain processes from phase, so getting that a little bit better... I *did* say before, it depends how much of a perfectionist you are...

C

[JonW]

It is in my opinion, but (a) I've 'designed' maybe 2 pairs of speakers in my life, so take anything I say with a grain of salt, and (b) I use low powered tube amps, so for me, the extra parts cost to get a flatter impedance curve is worth it. I've seen many designers who leave that notch filter out, and some who put it in. I suspect that with most amps, it does not make any audible difference. I don't know that it hurts anything (other than the cost of the speakers).

Hey, that’s 2 more speakers than I’ve designed. I’ll give it a shot and see how it sounds with versus without all this fancy stuff. Easy to try.

It really depends on the amp, how much difference the impedance peak in the summed response is. Tube amps don't like big swings in impedance, and while solid state does far better with it, there may still be benefit. Not sure - I've not actually messed with this, at least so far.

Phase tracking will show up in little details - better imaging/staging would be my guess. This is information our brain processes from phase, so getting that a little bit better... I *did* say before, it depends how much of a perfectionist you are...

C

OK. I’ll keep in the fancy LCR bits and see if I can hear a difference or not. Still some tweaking to do before I clip together a test crossover.

If you look at this plot (also posted above) with versus without the LCR there is a little bit better phase tracking with the LCR. So that makes me inclined to keep it in there. Does that make me a perfectionist? Well, I have so much effort into this project that I might as well make it as good as possible. But more than the specifics of my current project, I’m trying to generally understand what is important for making a good speaker.

[Saurav]

Hey, that’s 2 more speakers than I’ve designed.

Well, I fudged the numbers a little... I'm still working on my 2nd one

[---k---]

Does that make me a perfectionist?

Was that rhetorical? :P

[Jed]

The more speakers you design the more realize that one particular application of a topology may or may not work with a different combination of drivers. Zobels on the tweeters can be a necessity or not depending on the other parts in the circuit, and how they handle the overall goals of the circuit. For example, moving from 3rd order electrical to 2nd order electrical may yield the same acoustic target, but different phase alignment characteristics, just like a zobel may have in another topology situation.

There are other tricks that a crossover designer can use- like flattening the impedance hump of a tweeter, but also changing the values within the zobel such that it flattens the response, creates a flat impedance, and allows the designer to hit target acoustic slopes with good phase alignment. I once tried to eliminate one shaping LCR network on my Tombstones design, but in the end the phase alignment was better with a separate Zobel in conjunction with the shaping networks. In my Lineup Maxx speakers, I have a Zobel network that is optional, but it doesn't nearly effect things as much as in my Tombstones design.

In the end, you look at your goals and how to best reach the target response. That may require a zobel on the tweeter, or it may not. It's just one piece of the puzzle. But, in an ultimate design, I think it seems to help hit target transfer functions with ease, and it's added cost offsets any compromises someone might make- especially if crossing close to the tweeter's FS.

In the graphs in some of the posts above that show a network with and without the zobel network, it's not really proving anything unless other components were modified to hit the same transfer functions.

Jed

[JonW]

The questions I was asking here in this thread were more general, just trying to learn. But in case anyone is interested, here is a link to the current design I’m in the middle of. Post 445 has the latest crossover:

https://www.htguide.com/forum/showthread.php?p=410059#post410059

Well, I fudged the numbers a little... I'm still working on my 2nd one

So that’s like 1.5 pairs of speakers. Which means you’ve made 3 speakers so far.

Was that rhetorical? :P

Heh. Anyone who works on 1 pair of speakers for over 1.5 years has got to be a little nuts. I can’t deny it.

The more speakers you design the more realize that one particular application of a topology may or may not work with a different combination of drivers. Zobels on the tweeters can be a necessity or not depending on the other parts in the circuit, and how they handle the overall goals of the circuit. For example, moving from 3rd order electrical to 2nd order electrical may yield the same acoustic target, but different phase alignment characteristics, just like a zobel may have in another topology situation.

There are other tricks that a crossover designer can use- like flattening the impedance hump of a tweeter, but also changing the values within the zobel such that it flattens the response, creates a flat impedance, and allows the designer to hit target acoustic slopes with good phase alignment. I once tried to eliminate one shaping LCR network on my Tombstones design, but in the end the phase alignment was better with a separate Zobel in conjunction with the shaping networks. In my Lineup Maxx speakers, I have a Zobel network that is optional, but it doesn't nearly effect things as much as in my Tombstones design.

In the end, you look at your goals and how to best reach the target response. That may require a zobel on the tweeter, or it may not. It's just one piece of the puzzle. But, in an ultimate design, I think it seems to help hit target transfer functions with ease, and it's added cost offsets any compromises someone might make- especially if crossing close to the tweeter's FS.

In the graphs in some of the posts above that show a network with and without the network, it's not really proving anything unless other components were modified to hit the same transfer functions.

Jed

Thanks for the perspective, Jed. That all makes perfect sense. I’m still in the middle of my first design so I’m trying to learn these things. You’ve made, what, 85 designs. So you should have a good handle on these things by now.

In my current xover I have 2 LCR’s, a tweeter padding resistor, and 2 third-order slopes. That’s 13 components. I know some people shy away from too many parts “getting in the way of clean sound.” But to my novice eyes, each one of those segments looks like they help the overall situation. It’s interesting. I've still got lots to learn.

[Jed]

I know some people shy away from too many parts “getting in the way of clean sound.” But to my novice eyes, each one of those segments looks like they help the overall situation. It’s interesting. I've still got lots to learn.

And some people believe a fullrange Fostex (1mm xmax) with super tweeter and 1uF cap crossed at 15k sounds better than a low distortion multidriver system- so to each his own.

[Jed]

In my current xover I have 2 LCR’s, a tweeter padding resistor, and 2 third-order slopes. That’s 13 components. .

Make sure you measure off axis to make sure you are not causing a dip in the power response with those LCR shaping networks.


For example, I decided to leave a few peaks in the Tombstones design on axis because off axis it turned into a diffraction dip, so the power response ended up flat after summing the on axis and off axis response. If I had notched the response down a few db onaxis, it would have ended up a bit lifeless sounding in that area because the power response would have dipped.

You probably did that check already but I just thought I'd mention it.

Jed

[JonW]

And some people believe a fullrange Fostex (1mm xmax) with super tweeter and 1uF cap crossed at 15k sounds better than a low distortion multidriver system- so to each his own.

He he. I’m going to the fall Iowa DIY speaker meet. Never been to one. Maybe I’ll get to hear some crazy designs (as well as excellent ones). Should be lots of fun.

Make sure you measure off axis to make sure you are not causing a dip in the power response with those LCR shaping networks.


For example, I decided to leave a few peaks in the Tombstones design on axis because off axis it turned into a diffraction dip, so the power response ended up flat after summing the on axis and off axis response. If I had notched the response down a few db onaxis, it would have ended up a bit lifeless sounding in that area because the power response would have dipped.

You probably did that check already but I just thought I'd mention it.

Jed

Thanks for the tip. No, I didn’t really look into this so it’s good that you mentioned it. I think I understand the concept of power response (just googled it) but not sure how to see that in LspCAD. There is a section of the manual that shows a box to click for calculating power response but I don’t see the box there on the computer. I’ve got the standard version of LspCAD ($200). Maybe that’s a feature only on the pro version.

At any rate, here are the 2 things I did do to consider off axis response:

-Have LspCAD simulate the frequency response at 30 degrees off axis. Nothing much changes except for the expected tweeter roll off above ~10,000 Hz.

-I took driver measurements in the cabinet each at 30 degrees off axis. I then put those measurements into the crossover. The phase shifts a little such that there is not as good of an overlap when looking above the crossover point. With regard to frequency response, the whole plot drops down by around 2 db or so. The tweeter rolls off above ~10,000 Hz as to be expected. There is a small dip at the crossover point of maybe 1 db. I might not have noticed that dip if I wasn’t comparing the off-axis plot to the on-axis plot. But I’d say that the overall plot is pretty flat, still to within +/- 3 db or so (except for the tweeter loss above ~10,000 Hz).

So what else should I be doing?

[Jed]

So what else should I be doing?

forget the sims... measure at different off axis points and see what happens. But it sounds like you are doing that and not noticing anything major.

[JonW]

forget the sims... measure at different off axis points and see what happens. But it sounds like you are doing that and not noticing anything major.

Got it. Thanks. :T