Hi Rob.
Steffen
With only a series capacitor your CD is not protected against the pressure from the woofers going back into the horn. And I dont think you can have a passive filter with only an inductor parallel to the CD!? You need to add a series capacitor, making the filter second order, to protect the amp from shorting (through the inductor).
Steffen
Just add resistor in series first before inductor and cd and it prevents the short and amp is safe. The resistor works as a pad as well.
Here, tried it in simulator for my CD and it seems to work nicely.
Didn't buy the parts yet though so no odea how it sounds. https://www.diyaudio.com/community/...ities-of-hybrid-crossover.389129/post-7093812
Here, tried it in simulator for my CD and it seems to work nicely.
Didn't buy the parts yet though so no odea how it sounds. https://www.diyaudio.com/community/...ities-of-hybrid-crossover.389129/post-7093812
Last edited:
Hi tmuikku,
Must admit, the more i see the effort & complexity that goes into passive solutions/methods, the happier i get with my method.
Which is tuning individual drivers via min phase EQs, then adding steep complementary lin phase xovers, and then applying automatic target matching to the correcting individual drivers so they match the electrical xover curves.
Acoustic response of the individual drivers then matches textbook electric curves, as does the speakers summed response.
The method seems so dang relatively simple and easy, compared to the alternatives proposed (be they passive or even active IIR).
And as best i can tell, the method gives better results, because it provides near perfect in-band response, while solving virtually all out-of-band issues, as well as minimizing lobing.
It does come at the cost of loosing some maximum SPL in the xover regions, because the benefit of summed response over a wider frequency range is lost due to the steep xovers' small overlap range.
But that loss is a small price to pay for better SQ, ime/imo.
Just my chosen path...to each their own
Must admit, the more i see the effort & complexity that goes into passive solutions/methods, the happier i get with my method.
Which is tuning individual drivers via min phase EQs, then adding steep complementary lin phase xovers, and then applying automatic target matching to the correcting individual drivers so they match the electrical xover curves.
Acoustic response of the individual drivers then matches textbook electric curves, as does the speakers summed response.
The method seems so dang relatively simple and easy, compared to the alternatives proposed (be they passive or even active IIR).
And as best i can tell, the method gives better results, because it provides near perfect in-band response, while solving virtually all out-of-band issues, as well as minimizing lobing.
It does come at the cost of loosing some maximum SPL in the xover regions, because the benefit of summed response over a wider frequency range is lost due to the steep xovers' small overlap range.
But that loss is a small price to pay for better SQ, ime/imo.
Just my chosen path...to each their own
Hi, For sure! what DSP cannot do is affect impedance in the amplifier/driver circuit. There is possibility to add passive parts to affect impedance, assuming it reduces audible distortion, certainly THD seems to drop, and still have all the benefits of active system, do all what you just described, perfect response.
Active + passive mixed system allows to control the response and impedance rather independently. If it was passive system only then these two affect more to each other and as said active system cannot affect impedance.
I have no idea how much audible benefits there is over just active system, perhaps on some applications its worth the trouble.
Active + passive mixed system allows to control the response and impedance rather independently. If it was passive system only then these two affect more to each other and as said active system cannot affect impedance.
I have no idea how much audible benefits there is over just active system, perhaps on some applications its worth the trouble.
Last edited:
Yeppa ,
for a passive setup, I've seen how using out-of-band notch filtering can lower THD, both out-of-band and in.
And how that alters impedance to help accomplish it. Like in the purifi paper... https://purifi-audio.com/wp-content/uploads/2022/03/220211_R05-Notchfilter.pdf
What i wonder about, is the out-of-band notch needed simply because a low order xover is letting the out-of-bad resonance get excited?
Is that ou-of-band excitation the source of the increased THD? (probably i think)
It seems to me, just use a steep xover and don't excite the dang resonance to begin with....
and then who cares about response spikes or impedance issues because they have been pushed down so low in level as to be immaterial.
For example, a snip from the purifi paper showing the out-band-peaks and the passive LCR corrections to knock the down. (which did a nice job)
The target was a 2300 3rd order low-pass i think...
Big peak at about 5kHz to tame, and another one at 10k
Ok, a look at a 2300Hz 96 dB/oct lpf (16th order)......to visualize how much of the 5k and 10 peaks would be attenuated by a steep lpf.
5kHz is down almost 110dB !
If we call the 5kHz peak +15dB relative to desired response (looking at top purifi trace), it would be down 95dB! with the steep low pass in place. (110-15)
10k is so out of sight, it's out of mind LOL
Anyway, if there's a reason for a hybrid approach, other than trying to hang onto low order xovers, I have open ears and mind.
Otherwise, it seems to me like a lot of wasted effort/attention.
On the Purifi graphs why does the peak at 10k go from 48dB to 62dB after a 5k notch is applied ?
Rob.
the source of distortion in this case is the motor and the peak just amplifies it. The distortion shows up at fpeak/3 for the 3rd harmonic because the cone breakup peak amplifies the 3rd harmonic. The fpeak/3 would be in the pass and of any steep low pass. The trick to attenuate it is to provide a high impedance loading of the driver at the peak frequency as the app note is explaining.
cheers Lars/Purifi
a hybrid approach can reduce distortion in a way that is not possible by an active only filter as the app note explains. that is the reason.
that is how notch filters work: the attenuation is up to infinite at the notch frequency but worse above.On the Purifi graphs why does the peak at 10k go from 48dB to 62dB after a 5k notch is applied ?
Rob.
Sorry, I'm struggling to understand. If I applied a 5k notch filter to the 2nd graph with my digital xo it would not boost the response at 10k by 14dB.
Hi Lars, the paper is very interesting!
Is this example scenario correct interpretation of what you wrote above? :
Lets say we have DSP and 1kHz is our lowpass point on a woofer. Problem in motor makes blip in driver impedance around 1kHz at some excursion and resulting distortion current, for example 3rd order one, flows. If we have passive network to raise impedance for driver around 3kHz less error current flows as result. Even though the problem causing impedance blip is still there at 1kHz we just reduced the effects of it some at 3kHz, low passed the distortion as well.
So, in addition to low pass the actual content at 1kHz with DSP we can also low pass distortion products originating from the driver by inreasing driver load impedance with passive network, right? My intuition says this could be very audible improvement because the distortion products that fall outside of the woofers passband there is reducing masking as we have lowpass, also directivity changes, cone resonances amplify further and to various directions. This would be most audible distortion I think, that falls out of band and sticks out plain on sight. If masking is involved then steep lowpass filter would be even worse I guess, less masking.
The paper shows this happen in a passive speaker application, but work similarly for active systems.
ps. on a multiple entry horn woofers are behind acoustic lowpass and distortion is lowpassed as well. Perhaps no audible difference in this application.
Last edited:
you are correct if it was just the addition of a separate notch. the three filters have all been optimised to follow the target response. The first filter rolls off too fast (undershoots the target) initially and the peaks stick above. the second filter follows the target beyond 5k but raised the 10k. the last filter has enough degrees of freedom to follow the target almost perfectly. The optimiser in VCAD was used.
Hi Lars, thx for commenting.
The different effects of the parallel vs series notch filters on impedance and THD is very interesting. I am a baby at understanding passive LCR corrections / xovers, so much to digest still.
If passive impedance correction as part of an active xover strategy offers sonic improvement, I'll want to explore it.
My simple solution so far for minimizing THD, has been to use drivers powerful enough they are loafing at desired listening levels.
That said, I have to admit I'm not convinced THD is a big problem...seems to me it just mainly shifts timbre/tonal balance while not really effecting clarity, and it's been hard for me to hear audible differences in THD measurements that differ by 10-20 dBr. Compared to starting with THD about -60dBr across the spectrum.)
I've been focusing on IMD, under the theory it can/does reduce clarity itself. This thread has been about that effort.
I have experienced real sonic improvement in clarity, by going from a 3-way that covered 100Hz up, to a 4-way. (this is counting a coax CD as two ways).
So rightly or wrongly, I've been attributing the clarity improvement to reduced IMD.
But your paper and comments have made me realize I'm also using the drivers entirely within their well behaved impedance curve ranges, when running 4-way.
Maybe using drivers only in their well behaved impedance ranges is a big factor in the clarity improvement ??
Certainly dunno, food for thought, thank you
I'm guessing your THD plots given are using the fundamental frequency as reference. If so, it's easy to that anything that knocks down response at 5kHz will also reduce THD at the fpeak/3. (1650Hz)
I'm still left wondering if a steep low pass (like the 16th order i proposed) wouldn't accomplish the same thing.
If you measure THD using the harmonic frequency as reference, like per the option in REW, would 1650Hz still show the same level of reduction?
I guess I'm most curious what would THD for a 1650Hz fundamental as reference look like, for the solution you presented in the paper, and for a simple steep low pass like i use.
It seems that would be a pretty good test. Any holes in that idea? thx
I'm no Lars so pardon writing, but trying to help me and you all imagine the situation. Why wouldnt DSP help with distortion due to driver having issues in the motor?
there is kind of simple idea behind all this: the distortion is emitted by the driver yes, and its made by the driver! Amplifier doesn't (directly) produce or amplify this distortion, the driver makes it and emits it by itself.
Thought experiment:
What makes the driver emit anything at all acoustically is current in the voice coil making the cone assembly move in precense of magnetic field. Now, in order for the distortion originating in driver motor be measurable/audible acoustically it needs to alter the current that flows through the voice coil in order to the distortion be emitted by the driver, and measured by our ear or equipment, to realize its audibility. Audible distortion is made by current originating from some nonlinearity in the electromechanical driver. To reduce distortion we could make better driver that makes less distortion (current), but if we are stuck what we have then we can reduce driver generated distortion by reducing current originating from the driver, in other words raise impedance of driver load on frequencies where the distortion current flows and current (distortion) is reduced.
Current doesn't flow if driver terminals are not connected to anything so no distortion but we need to connect the driver to an amplifier.
Driver is hooked to a circuit with the amplifier, whose output impedance is part of the circuit. Now if the amplifier is voltage amp with ~0ohm output impedance, this is almost 0ohm load for driver and maximum current flows in the circuit including the distortion current originating from the driver and gets emitted acoustically. We could reduce distortion current either by increasing amplifier output impedance, or by using thin long cabling for added resistance, or apply some coils and caps and resistors in the circuit to increase impedance seen by the driver selectively, to reduce current generated by the driver. DSP cannot do this as its not part of this electrical circuit, it cannot change the driver or impedance of circuit the driver is hooked to.
Well, at least this is how I imagine the situation and subject and would like to be corrected if its wrong, thanks! trying to increase understanding.
there is kind of simple idea behind all this: the distortion is emitted by the driver yes, and its made by the driver! Amplifier doesn't (directly) produce or amplify this distortion, the driver makes it and emits it by itself.
Thought experiment:
What makes the driver emit anything at all acoustically is current in the voice coil making the cone assembly move in precense of magnetic field. Now, in order for the distortion originating in driver motor be measurable/audible acoustically it needs to alter the current that flows through the voice coil in order to the distortion be emitted by the driver, and measured by our ear or equipment, to realize its audibility. Audible distortion is made by current originating from some nonlinearity in the electromechanical driver. To reduce distortion we could make better driver that makes less distortion (current), but if we are stuck what we have then we can reduce driver generated distortion by reducing current originating from the driver, in other words raise impedance of driver load on frequencies where the distortion current flows and current (distortion) is reduced.
Current doesn't flow if driver terminals are not connected to anything so no distortion but we need to connect the driver to an amplifier.
Driver is hooked to a circuit with the amplifier, whose output impedance is part of the circuit. Now if the amplifier is voltage amp with ~0ohm output impedance, this is almost 0ohm load for driver and maximum current flows in the circuit including the distortion current originating from the driver and gets emitted acoustically. We could reduce distortion current either by increasing amplifier output impedance, or by using thin long cabling for added resistance, or apply some coils and caps and resistors in the circuit to increase impedance seen by the driver selectively, to reduce current generated by the driver. DSP cannot do this as its not part of this electrical circuit, it cannot change the driver or impedance of circuit the driver is hooked to.
Well, at least this is how I imagine the situation and subject and would like to be corrected if its wrong, thanks! trying to increase understanding.
Last edited:
I guess if we had sufficiently sophisticated model of the driver properties so that the driver generated distortion current was well predictable then DSP could in principle have transfer function that negates some of the driver genrated current? Is this what Klippel Controlled Sound system does?
Hi Tmuikku
From post 68
Just commenting on your "Thought experiment" in post 75. It looks like you look only at the motor!? Does cone-break-up generate error-current? Is the "model" too simple?
Fortunately the way the woofers and midranges are mounted/radiating in a MEH attenuates harmonic distortion. It´s actually kind of a very different story what happens in a MEH and in a normal direct radiator, and needs to be taken into account I think.
I have an old friend, who´s done hundreds of passive filters (Duelund filters), and he often reports, that it makes a difference to linearize the impedance peak at the resonance-frequency of the woofer in closed boxes. He has even claimed that he could hear improvements on vented boxes, when linearizing the upper of the two impedance-peaks, almost making the vented box sounding as a closed box!? This happens, would I guess, because it makes the life easier for the amplifier!? My friend reported improvements with a NLE1000 amplifier that produced 2x500W.
What I am trying to convey here is, that maybe it is worth optimizing the load for the amplifier before DSP!? Make things right in the hardware before applying software.
Have a nice day. Have to go to work.
Steffen
Steffen
From post 68
Just commenting on your "Thought experiment" in post 75. It looks like you look only at the motor!? Does cone-break-up generate error-current? Is the "model" too simple?
Fortunately the way the woofers and midranges are mounted/radiating in a MEH attenuates harmonic distortion. It´s actually kind of a very different story what happens in a MEH and in a normal direct radiator, and needs to be taken into account I think.
I have an old friend, who´s done hundreds of passive filters (Duelund filters), and he often reports, that it makes a difference to linearize the impedance peak at the resonance-frequency of the woofer in closed boxes. He has even claimed that he could hear improvements on vented boxes, when linearizing the upper of the two impedance-peaks, almost making the vented box sounding as a closed box!? This happens, would I guess, because it makes the life easier for the amplifier!? My friend reported improvements with a NLE1000 amplifier that produced 2x500W.
What I am trying to convey here is, that maybe it is worth optimizing the load for the amplifier before DSP!? Make things right in the hardware before applying software.
Have a nice day. Have to go to work.
Steffen
Steffen
Hi Steffen,
for sure its simple model, trying to understand whats happening to be able to connect the dots, what affects what, imagination, understanding, and if its correct enough then we can adapt and use the knowledge in our applications. I already have very clear idea on this in my own head but hard to say if its right thinking or not yet as not much feedback is returned to my writings other than what I've been interpreting from Lars responses here on this thread recently, so thank you and thanks to Lars and mark and others making the discussion happen. Hopefully understanding on the subject increases as result
I think Cone breakup makes acoustic effect immediately, its separate to what voice coil does, different parts of the cone are not following voice coil movement. As mentioned by Lars, cone break up amplifies what voice coil does (to some directions), if there is breakup peak at 5kHz it will boost distortion at that frequency, which originated lower down in the passband. Perhaps some of cone breakup reflects to voice coil and current as well and then its relieved similarly as motor distortion, by increasing load impedance for the driver. Even if we cut cone breakup peak with DSP down in the original sound fed by amplifier so that it won't excite the breakup but the distortion (current) introduced by the driver will. We want to tame the break up peaks passively and so that driver load impedance is increased.
Yeah, the ports make acoustic low pass so the distortion products are attenuated acoustically. You are right MEH is not most suitable application to test this stuff as its very low distortion already with acoustic low pass and multiple drivers and what not. Perhaps wrong thread eventually for this discussion but its important discussion nevertheless and I hope it continues. Perhaps relevant for MEHs as well (until some one tries it
part of good sound of MEH could be just this, acoustic low passing the distortion products. And low distortion in general. Perhaps we could make our direct radiating speakers close to MEH performance with some of this passive / active mix, low pass (some) distortion electrically.Yeah, this is another thing, separate issue, we can address with passive network on active system. Both amplifier and the driver are on the circuit and we can "optimize" impedance load for both, with passive parts. I posted demonstration / breakdown on this in the Elsinore thread https://www.diyaudio.com/community/threads/the-elsinore-project-thread.97043/post-7099694 Both impedances, amplifier load impedance And driver load impedance (for the distortion current originating from driver) can be adjusted somewhat separately especially in active multiway speaker context. At least so it looks like to me currently, I haven't done any so perhaps there is applications we cannot optimize both and need to choose one.
I'm not sure however, how audible this portion is to make the amplifier work easier by flattening its load impedance, perhaps it is important. This is what Joe talks about in the Elsinore thread and says it is. Its separate thing from driver motor distortion though what the Purifi paper and recent discussion in this thread is about. You can do both with passive parts, or just the other, or neither as we usually do not have passive parts mixed with active system.
Last edited:
I don't understand why breakup at 5kHz, has anything to do with origination down in the passband.
I get back to wanting to use a single sine or stepped-sine to compare alternatives for handling breakup response, and their effect on passband THD.
I always wonder how much peaky out-of-band response spills into the typical short log sweep measurement, as THD artifacts lower in the passband.
On the graphs in the linked paper, it looks like response at 5kHz breakup is about 20 dB down in level, after inserting its low pass and notch filers.
I'd like to see the driver's THD with a lin phase 96dB/oct low-pass in place...where 5kHz is down over 100dB.
Another simplfied textual example then 5kHz breakup resonance. The breakup is there on the cone as physical property. Its 6db peak on manufacturer datasheet on otherwise nice frequency response. We plan to crossover below the breakup.
We have 16th order lowpass filter in DSP at 1.7kHz, not much of the sound gets to amplifier or to driver at 5kHz anymore, nice. But, full sound goes to amp and driver at 1.7kHz, which the driver motor adds harmonics to due to non-linearities in the motor and 3rd order harmonic now lands on the 5kHz and excites the resonance and gets extra 6db boost from it, comes out louder than without the resonance, or other harmonics around the resonance. The 5Khz sound did not come from source, or from the amp, but was generated in process in the driver.
Now, the original sound we feed to the amp and to driver is heard only up to 1.7Khz where steep low pass cuts it out, but all the harmonics generated in the driver (on its passband) emit all the way up to many kHz, nice soup of harmonics spiced with cone resonance well past 1.7kHz. Unless are filtered out as well, like the direct sound was in the DSP. Acoustic low pass or passive components added into the circuit reducing current in the driver.
5kHz breakup resonance. The breakup is there on the cone as physical property. Its 6db peak on manufacturer datasheet on otherwise nice frequency response. We plan to crossover below the breakup.We have 16th order lowpass filter in DSP at 1.7kHz, not much of the sound gets to amplifier or to driver at 5kHz anymore, nice. But, full sound goes to amp and driver at 1.7kHz, which the driver motor adds harmonics to due to non-linearities in the motor and 3rd order harmonic now lands on the 5kHz and excites the resonance and gets extra 6db boost from it, comes out louder than without the resonance, or other harmonics around the resonance. The 5Khz sound did not come from source, or from the amp, but was generated in process in the driver.
Now, the original sound we feed to the amp and to driver is heard only up to 1.7Khz where steep low pass cuts it out, but all the harmonics generated in the driver (on its passband) emit all the way up to many kHz, nice soup of harmonics spiced with cone resonance well past 1.7kHz. Unless are filtered out as well, like the direct sound was in the DSP. Acoustic low pass or passive components added into the circuit reducing current in the driver.
Last edited: