Exactly, I copied the idea from another example as I mentioned in my own thread.
Keep the speaker's impedance at or above 4 ohms at all frequencies (Hz, kHz, MHz, GHz)...
Resistors are lossy. Using a resistor throws away the amplifier's power. An inductor or capacitor is a better choice.
Ed
"Resistors are lossy. Using a resistor throws away the amplifier's power. An inductor or capacitor is a better choice."
Sorry but I don't see what you're getting at. What's being suggested is not a simple series reisistor, throwing away power. The proposal is to add a resistor in series with the notch cap, shunting the series inductor. This will do two things: (a) shape the notch and, (b) get rid of the problem of low impedance at ultrasonic frequencies. AllenB has suggested using a higher order crossover instead, but if you really want to notch out a single frequency, that seems to me like a rather blunt instrument. It may be a good solution in some situations, but perhaps not in others. The other alternative, surely, is an LRC trap, either series or parallel. (And that would also use a resistor.)
So, I just don't see what's wrong with the circuit being proposed. I'm very happy to be educated if there's a problem I'm missing.
Sorry but I don't see what you're getting at. What's being suggested is not a simple series reisistor, throwing away power. The proposal is to add a resistor in series with the notch cap, shunting the series inductor. This will do two things: (a) shape the notch and, (b) get rid of the problem of low impedance at ultrasonic frequencies. AllenB has suggested using a higher order crossover instead, but if you really want to notch out a single frequency, that seems to me like a rather blunt instrument. It may be a good solution in some situations, but perhaps not in others. The other alternative, surely, is an LRC trap, either series or parallel. (And that would also use a resistor.)
So, I just don't see what's wrong with the circuit being proposed. I'm very happy to be educated if there's a problem I'm missing.
I have been working on this crossover for nearly two weeks now, and I believe I have tried most of the configurations. I am using the tweeter in a waveguide which creates a huge bump around 2-8k (as you can imagine) that I have to deal with. Add to that the baffle step loss and the break up, you can see why I don't have that many options.
That's actually what I have done originally, but when I found out people are using a parallel cap instead, I jumped on it because I already have a notch filter for the baffle loss and adding another LCR network sounds a bit too much in my humble opinion.
In case it's not clear, most decent amps have a R+L series network on the output that separates the speaker output and loading from the amplifier feedback loop so that a "short" or high capacitance cannot short the feedback. Some may consider this part of the "Zobel network". This is a given for professional amps but perhaps not for certain DIY amps that will indeed become unstable when loaded with an ultimately capacitive OX or very long cables. If you have such an amplifier, you can add a build-out network external to the amplifier, typically being a ~2W 10 Ohm resistor in parallel with about 1 microhenry, often a coil wound around the resistor.
I should mention that 2nd order woofer filter is generally a bad idea because it presents a low impedance source at high frequencies to the woofer, which exaggerates the cone break-up resonances. A single inductor (1st order) or a 3rd order filter that finishes with a series inductor presents a high impedance at high frequencies that dampens these woofer misbehaviors. For this reason, a woofer probably sounds brighter with a 2nd order XO than with a 1st order, single inductor.
Regarding 2nd order woofer filters, I don't understand. The amplifier doesn't see the capacitor on its own - only via the inductor. So how can the amp be presented with a low impedance at high frequencies? Modelling a 2nd order woofer crossover shows the impedance rising without limit at hf. Or am I misunderstanding you?
I have been trying to get a satisfying result with a 3rd order filter in my design. When I get good response, the phase is off. When I nail the phase, FR is bad. I feel like I'm chasing my own tail.
I am surprised to hear that 2nd order lowpass filters are a bad idea given the fact that it's one of the most used (if not 'the' most used) filter types on the woofer
The phase for a 3rd order crossover is NOT the same as for a 2nd or 4th order. The drivers’ output is supposed to sum flat in quadrature as opposed to in-phase.
In general, I’ve had better results with 3rd vs 2nd order filters. And if I’m trying to make a “tube friendly“ flat-ish impedance, I go straight to 3rd order because it doesn’t have the rise around the crossover frequency(ies).
A 2nd order network on a woofer may result in 3rd or 4th order acoustic response if it has a smooth roll off and you’re pushing the limits of its useable range. But will make something with a high Q breakup rise stick out like a sore thumb.
In general, I’ve had better results with 3rd vs 2nd order filters. And if I’m trying to make a “tube friendly“ flat-ish impedance, I go straight to 3rd order because it doesn’t have the rise around the crossover frequency(ies).
A 2nd order network on a woofer may result in 3rd or 4th order acoustic response if it has a smooth roll off and you’re pushing the limits of its useable range. But will make something with a high Q breakup rise stick out like a sore thumb.
It's going to be different from creating a pseudo rolloff using a bandstop filter, however I would also check your measurements. Sometimes an incorrect measurement presents difficulties in 'fixing' the response.
If the woofer is excited at breakup by a lower frequency input that creates a harmonic, a low impedance can help that develop to give the appearance of increased distortion over what would be if the response was flat around breakup, despite your crossover filter since this is coming from the other direction.
However it's not normally something to worry about, for a few reasons.
1. The peak is usually only on axis, but that's not representative of the overall response. Don't rely on an on axis measurement to tell you the story, and don't listen on axis...
2. Speaker distortion is low order and has to be audible before it's a problem.
3. Source impedance for a second order filter looking back from the driver can be higher than first order through breakup.
Thanks AllenB. That makes sense, and the graphic shows the fall in source impedance at hf nicely for the second order filter. Two questions, though.
First, how does this reduced impedance at hf lead to increased output at hf? (After all, the frequency response graphic shows, as you'd expect, that the second order filter attenuates hf more than the first order filter, despite the lower source impedance.)
Second, if you tailor the 2nd order filter by putting a resistor in series with the cap (a very common strategy), does this help?
(Edit: two more puzzles, as I think about this. First - if cone breakup resonances are the source of this problem, surely these will not, on the whole, correlate with voicecoil movement. So how can they be fed back to the crossover? Second, connecting a driver to a low impedance source normally controls cone/coil movement, electrically damping it. So why is this low impedance source at hf a problem?)
First, how does this reduced impedance at hf lead to increased output at hf? (After all, the frequency response graphic shows, as you'd expect, that the second order filter attenuates hf more than the first order filter, despite the lower source impedance.)
Second, if you tailor the 2nd order filter by putting a resistor in series with the cap (a very common strategy), does this help?
(Edit: two more puzzles, as I think about this. First - if cone breakup resonances are the source of this problem, surely these will not, on the whole, correlate with voicecoil movement. So how can they be fed back to the crossover? Second, connecting a driver to a low impedance source normally controls cone/coil movement, electrically damping it. So why is this low impedance source at hf a problem?)
Last edited:
Anyone? @AllenB @steveu? You've said there's a potential problem with 2nd order woofer crossovers. I'd like to get my head round this better (not least because I'm currently using a system in which the woofer has a second order low pass. 
) My apologies if I'm being obtuse, but I'd be grateful for any further reading/explanation anyone can point me to.
) My apologies if I'm being obtuse, but I'd be grateful for any further reading/explanation anyone can point me to.Ok, no-one's taking me on.
My best guess is that the work done by Purifi (see here: https://purifi-audio.com/blog/app-notes-2/low-distortion-filter-for-ptt6-5x04-naa-11) is behind what AllenB and steveu were saying. Let me know if that's not so.
My best guess is that the work done by Purifi (see here: https://purifi-audio.com/blog/app-notes-2/low-distortion-filter-for-ptt6-5x04-naa-11) is behind what AllenB and steveu were saying. Let me know if that's not so.
I really feel like I'm talking to myself, now. I know this is off topic, really, and this thread is in the Solid State forum, and I'm focussing on a point about speaker crossovers. But @steveu made a very significant claim in post 26, so I'm not dropping the issue - yet. Here's what he said:
If it's true that second order filters exaggerate break-up resonances, and may sound brighter than first order filters, that would be a really important discovery for the world of passive crossovers.
But is it true? Here's where I've got to. I can't find anything which backs up those claims. My best guess is that steveu was thinking of the work done by Purifi, linked to above. But if that's the case, then I fear he's misread the Purifi work. The claim in the Purifi work is that a crossover topology which creates a low source impedance at the frequency of a break-up peak can create increased distortion levels, related to that peak. That's not the same as claiming that the resonances will be exaggerated, or the frequency response 'brighter'.
Also, I want to point out something about the Purifi work. It's very impressive work, and the distortion variation they uncover looks pretty significant. But the topology which creates the problem is not just a plain second order network, it's a network with two LC series traps (i.e. L and C in series with each other, shunting the driver), plus a shunt cap damped by a resistor. (Note, the Purifi work calls these two traps parallel notch filters, but that's using terms the opposite way round from what I'm used to, and I'm sticking with my familiar designations.) This topology results in an extremely low source impedance at 5kHz, the frequency of the main break-up peak - around 0.5 ohms. The Purifi work contrasts this with a network that uses a parallel trap to reduce the 5kHz break up peak (they call it a series notch). This creates a big peak in the source impedance, over 40 ohms at the break-up frequency.
Now, I can't really criticise Purifi for using an example that strongly illustrates the problem they've discovered. But I can't help but think that the two circuit examples they've given represent a pretty extreme case. Very few crossover topologies are going to create source impedance dips as low as 0.5 ohms. For one thing, most series notch filters include a resistor, which will limit the dip. Plain vanilla second order networks may bring the source impedance down to a few ohms, but not sub 1 ohm. When I set up Vituixcad the way that the Purifi paper suggests, to show source impedance, and use a second order woofer crossover, I get source impedance of around 10 ohms at 3 kHz (which is where my woofer has its main break-up peak). See the screenshot below. The woofer crossover is shown on the lower left side, and the source impedance appears in the bottom right hand chart of the six-pack. The rest can safely be ignored.
The question which the Purifi work leaves unanswered is about how much extra distortion will be generated if the source impedance at the breakup peak is higher than 0.5 ohms, but less than 40? Say, 10 ohms? I can't really answer that, for the moment, but I'd guess the problem would be far smaller. So, to sum up: panic averted.
If it's true that second order filters exaggerate break-up resonances, and may sound brighter than first order filters, that would be a really important discovery for the world of passive crossovers.
But is it true? Here's where I've got to. I can't find anything which backs up those claims. My best guess is that steveu was thinking of the work done by Purifi, linked to above. But if that's the case, then I fear he's misread the Purifi work. The claim in the Purifi work is that a crossover topology which creates a low source impedance at the frequency of a break-up peak can create increased distortion levels, related to that peak. That's not the same as claiming that the resonances will be exaggerated, or the frequency response 'brighter'.
Also, I want to point out something about the Purifi work. It's very impressive work, and the distortion variation they uncover looks pretty significant. But the topology which creates the problem is not just a plain second order network, it's a network with two LC series traps (i.e. L and C in series with each other, shunting the driver), plus a shunt cap damped by a resistor. (Note, the Purifi work calls these two traps parallel notch filters, but that's using terms the opposite way round from what I'm used to, and I'm sticking with my familiar designations.) This topology results in an extremely low source impedance at 5kHz, the frequency of the main break-up peak - around 0.5 ohms. The Purifi work contrasts this with a network that uses a parallel trap to reduce the 5kHz break up peak (they call it a series notch). This creates a big peak in the source impedance, over 40 ohms at the break-up frequency.
Now, I can't really criticise Purifi for using an example that strongly illustrates the problem they've discovered. But I can't help but think that the two circuit examples they've given represent a pretty extreme case. Very few crossover topologies are going to create source impedance dips as low as 0.5 ohms. For one thing, most series notch filters include a resistor, which will limit the dip. Plain vanilla second order networks may bring the source impedance down to a few ohms, but not sub 1 ohm. When I set up Vituixcad the way that the Purifi paper suggests, to show source impedance, and use a second order woofer crossover, I get source impedance of around 10 ohms at 3 kHz (which is where my woofer has its main break-up peak). See the screenshot below. The woofer crossover is shown on the lower left side, and the source impedance appears in the bottom right hand chart of the six-pack. The rest can safely be ignored.
The question which the Purifi work leaves unanswered is about how much extra distortion will be generated if the source impedance at the breakup peak is higher than 0.5 ohms, but less than 40? Say, 10 ohms? I can't really answer that, for the moment, but I'd guess the problem would be far smaller. So, to sum up: panic averted.
Last edited:
Sorry for the late reply.
This really doesn't have anything to do with breakup, harmonics happen across the spectrum and the cone responds as a source. Some people have gone as far as to use a current source to drive a speaker. The reason breakup gets singled out could be because of the general expectation that those frequencies have already been filtered and shouldn't be produced.
On top of this a breakup peak has additional gain for the harmonic in the direction of the peak, in the amount of the difference in response of the driver at the two frequencies, and this looks untidy on a plot.
This really doesn't have anything to do with breakup, harmonics happen across the spectrum and the cone responds as a source. Some people have gone as far as to use a current source to drive a speaker. The reason breakup gets singled out could be because of the general expectation that those frequencies have already been filtered and shouldn't be produced.
On top of this a breakup peak has additional gain for the harmonic in the direction of the peak, in the amount of the difference in response of the driver at the two frequencies, and this looks untidy on a plot.
Indeed, thanks AllenB.
I've read the stuff in the Elsinore thread about current drive etc. The most relevant post seems to be this one: https://www.diyaudio.com/community/threads/the-elsinore-project-thread.97043/post-7316848
In it, tmuikku says: "If you were to make it a second order filter by adding a shunt capacitor, the capacitor would provide very low impedance path for the high frequency backEMF current and the distortion would be back up."
I have to say, I'm not convinced. For one thing, his comments about back EMF are a little puzzling, to me. But also, the source impedance values in actual woofer crossover circuits just don't seem very low, based on the modelling I've done, using the method suggested in the Purifi paper. Am I doing it wrong? You can see what I've done on the screenshot above.
Anyway, I'm not convinced that there's a problem with second order woofer crossovers.
I've read the stuff in the Elsinore thread about current drive etc. The most relevant post seems to be this one: https://www.diyaudio.com/community/threads/the-elsinore-project-thread.97043/post-7316848
In it, tmuikku says: "If you were to make it a second order filter by adding a shunt capacitor, the capacitor would provide very low impedance path for the high frequency backEMF current and the distortion would be back up."
I have to say, I'm not convinced. For one thing, his comments about back EMF are a little puzzling, to me. But also, the source impedance values in actual woofer crossover circuits just don't seem very low, based on the modelling I've done, using the method suggested in the Purifi paper. Am I doing it wrong? You can see what I've done on the screenshot above.
Anyway, I'm not convinced that there's a problem with second order woofer crossovers.
Last edited: