Graphics
Dave and Joe, thank you for the detailed answers. It was difficult enough to understand the standard amplifier function. I have the XO and amplifiers ready with drivers, so sending it to the bin is not a practical option. The XO from ESP has phase coherence and drivers can be equalised as needed.
I will sim your amp and do more research, because the subject is intriguing. Thank you again !
Dave and Joe, thank you for the detailed answers. It was difficult enough to understand the standard amplifier function. I have the XO and amplifiers ready with drivers, so sending it to the bin is not a practical option. The XO from ESP has phase coherence and drivers can be equalised as needed.
I will sim your amp and do more research, because the subject is intriguing. Thank you again !
Your XO is active is it not? It will not interfere with a current amp since it is before the amps. You only need worry about relatively flat impedance of the drivers (in there boxes) in their passband.
It should be noted that the ESP 4th order XO is not phase coherent, only that you do not need to invert the phase of any of the drivers.
dave
Hi Dave
Ah yes, low Qms drivers, make me think back...
Before they even knew about Thiele-Small Parameters and before closed boxes became fashionable, especially in the 50's, drivers were designed to be driven unloaded - basically open back panels. If these were measured today, to get the T-S Parameters, they would have very low Qms = 1 or lower. Some (few) of these drivers survive and you can see that they use a fair amount of mechanical damping in their construction. Unfortunately today, drivers like this are rare.
Dave, what full-range drivers do you know that have low Qms - do some full-range drivers actually aim at this - can you name a few?
But there is another option:
1. Take a typical driver, say 6-8".
2. Measure its T-S Params.
3. Assuming the driver is suitable, adjust the volume of a sealed box to give a Q = 0.707 - classic 2nd order Butterworth and an admirable alignment to aim at.
4. Now drive this from a "voltage" amplifier and add various amounts of series resistance (as Richard Small pointed out to me) and we change the alignment upwards. Find the correct series resistance and you can aim for a Q = 1 (which some say is a good aim for smaller speakers). You can vary the total system Q - the Qb.
Conclusion: We can only retain the 2nd order Butterworth IF we are driving from an ideal "voltage" source of zero Ohm.
5. This is where it gets interesting: Look at the Z plot of our 2nd order Butterworth, now identify the Fb, the Z-max frequency. Construct an LCR trap where the LC component matches the tuning of the Fb. Now vary the L and C values to equate or EQ the Q and in conjunction with R we can totally cancel out the Z-max. This takes a bit of practice and using modelling techniques can save a LOT of time.
Result: Other than near Fb, the rest of the freq range will still see whatever the source impedance of the amplifier is and hence full current drive, but limiting Z only around Fb the drive impedance becomes large the R value of the LCR. Effectively the Fb does not see the amplifiers output impedance, it is largely tricked into thinking that it is being driven by the LCR.
Conclusion: As you Dave says, you could use a variable impedance amplifier, but this lower impedance will be the same (a sacrifice) over the whole frequency range - a compromise, at least on paper. But in the above example the R value of the LCR can also be used as a variable impedance source, but limiting that variability to near the Fb. Hence my view is that using a variable impedance amplifier, at least on paper, is the lesser of the two solutions and that the solution presented here is not only superior, but also far more flexible in the limited frequency/frequencies you want to be affected only.
Result Number Two: If we entire EQ the Z-max, then you have an astonishing situation, that 2nd order Butterworth is retained NO MATTER THE SOURCE IMPEDANCE.
In other words, it is ALWAYS a Butterworth alignment, whether it is driven by a "voltage" source or a "current" source.
No commercial loudspeaker currently takes advantage of this - or even the control that this approach can give you, the flexibility that it affords.
Result Number 3: While the 2nd order Butterworth example here is a mechanical high pass alignment that can NOW drive from any source impedance, does the same apply to Crossovers, which are electrical filters? If the answer is YES, then we have reversed Neville Thiele's original assertion/realisation, the mechanical filters have a mathematical equivalence in the maths of electrical filters, and using that knowledge and understanding, he adapted electrical filter maths/equations to mechanical high-pass filters. But IF our ability to lock in the same mechanical alignment with an LCR source, will that also work for Crossovers? IF Thiele's original premise proved to be true, then the answer would be the same... YES !!!
And guess what... YES !!!
Just start to think of the significance of that !!! The ramification are ENORMOUS.
We can now lock in Crossovers the same way that we can lock in a 2nd order Butterworth alignment or any other sealed box alignment aimed at.
(Aside: Yes, it also works for vented boxes, but will leave that for now).
Dave, if the LCR is the new source impedance, then the LCR is now the variable source Z that can be used across the whole frequency range selectively and also incorporate Crossovers that also track well. We now have a level of control that a variable impedance would not - and just aim at the highest impedance possible.
If I may, a closer look at the Elsinore Mk6 Crossover shown here, this applies all of the above, all the factors balanced against each other.
As stated, it was the Elsinore Mk5 and Mk6 speakers that made me - actually forced me as a necessity - to construct this 40 watt "current" amplifier of this thread. The modeling proved it would work and I wanted it to be proven in REAL LIFE. The two are totally in agreement.
Modeled:
It is of course the electrical measurement, actually measured and NOT modeled, that is significant.
Actual measurement:
There is a LOT more work to be done - it is interesting that here in Sydney, the home of Thiele-Small Parameters, that a small number of us are looking at the above. What is certainly true, that the Elsinores. compared to LR4 Crossovers that dominate a large number of commercial speakers with Crossovers, that they have a pristine quality, a beguiling clarity that is so easy on the ear, so sweet and yet capable of a high resolution that some lighteners have likened it to ribbons (which draw quite linear current, as I indeed pointed out to Allen as they were his preference), so do most headphones (and hence some listeners prefer them over to speakers? - not me) and elctro-statics which don't track current but are voltage devices. Why are we designing speakers as IF they were voltage devices when in fact they are current devices - and Crossovers and their designers loose sight of that and make horrible sounding Crossovers and don't even know it because THAT is all they have ever heard.
Down here we are starting up a serious and proper research project that will explore these things - we have a Discussion Paper already and that sets the scene. It will be two-fold as I see it, the importance of current contaminations on Voice Coil linearity and the role that Crossovers that cause or at the very least exacerbate the problem.
And of course, third-fold, this will also have a LOT of meaning to those like you who are in the full-range camp and in fact will justify what you are doing in that area. A form of exoneration against those who pooh-pooh the use of single drivers - not me of course, as single drivers will always be the ideal driver, period. It always will be, only practicality notwithstanding.
As our head academic says, with a funded research background, no such study has been conducted and as he is getting nearer retirement age, this is something he wants to be part of and we are looking at something that will prove to be of legacy value.
Whew! That was a LOT, so Dave, over to you.
Cheers, Joe
.
Ah yes, low Qms drivers, make me think back...
Before they even knew about Thiele-Small Parameters and before closed boxes became fashionable, especially in the 50's, drivers were designed to be driven unloaded - basically open back panels. If these were measured today, to get the T-S Parameters, they would have very low Qms = 1 or lower. Some (few) of these drivers survive and you can see that they use a fair amount of mechanical damping in their construction. Unfortunately today, drivers like this are rare.
Dave, what full-range drivers do you know that have low Qms - do some full-range drivers actually aim at this - can you name a few?
But there is another option:
1. Take a typical driver, say 6-8".
2. Measure its T-S Params.
3. Assuming the driver is suitable, adjust the volume of a sealed box to give a Q = 0.707 - classic 2nd order Butterworth and an admirable alignment to aim at.
4. Now drive this from a "voltage" amplifier and add various amounts of series resistance (as Richard Small pointed out to me) and we change the alignment upwards. Find the correct series resistance and you can aim for a Q = 1 (which some say is a good aim for smaller speakers). You can vary the total system Q - the Qb.
Conclusion: We can only retain the 2nd order Butterworth IF we are driving from an ideal "voltage" source of zero Ohm.
5. This is where it gets interesting: Look at the Z plot of our 2nd order Butterworth, now identify the Fb, the Z-max frequency. Construct an LCR trap where the LC component matches the tuning of the Fb. Now vary the L and C values to equate or EQ the Q and in conjunction with R we can totally cancel out the Z-max. This takes a bit of practice and using modelling techniques can save a LOT of time.
Result: Other than near Fb, the rest of the freq range will still see whatever the source impedance of the amplifier is and hence full current drive, but limiting Z only around Fb the drive impedance becomes large the R value of the LCR. Effectively the Fb does not see the amplifiers output impedance, it is largely tricked into thinking that it is being driven by the LCR.
Conclusion: As you Dave says, you could use a variable impedance amplifier, but this lower impedance will be the same (a sacrifice) over the whole frequency range - a compromise, at least on paper. But in the above example the R value of the LCR can also be used as a variable impedance source, but limiting that variability to near the Fb. Hence my view is that using a variable impedance amplifier, at least on paper, is the lesser of the two solutions and that the solution presented here is not only superior, but also far more flexible in the limited frequency/frequencies you want to be affected only.
Result Number Two: If we entire EQ the Z-max, then you have an astonishing situation, that 2nd order Butterworth is retained NO MATTER THE SOURCE IMPEDANCE.
In other words, it is ALWAYS a Butterworth alignment, whether it is driven by a "voltage" source or a "current" source.
No commercial loudspeaker currently takes advantage of this - or even the control that this approach can give you, the flexibility that it affords.
Result Number 3: While the 2nd order Butterworth example here is a mechanical high pass alignment that can NOW drive from any source impedance, does the same apply to Crossovers, which are electrical filters? If the answer is YES, then we have reversed Neville Thiele's original assertion/realisation, the mechanical filters have a mathematical equivalence in the maths of electrical filters, and using that knowledge and understanding, he adapted electrical filter maths/equations to mechanical high-pass filters. But IF our ability to lock in the same mechanical alignment with an LCR source, will that also work for Crossovers? IF Thiele's original premise proved to be true, then the answer would be the same... YES !!!
And guess what... YES !!!
Just start to think of the significance of that !!! The ramification are ENORMOUS.
We can now lock in Crossovers the same way that we can lock in a 2nd order Butterworth alignment or any other sealed box alignment aimed at.
(Aside: Yes, it also works for vented boxes, but will leave that for now).
Dave, if the LCR is the new source impedance, then the LCR is now the variable source Z that can be used across the whole frequency range selectively and also incorporate Crossovers that also track well. We now have a level of control that a variable impedance would not - and just aim at the highest impedance possible.
If I may, a closer look at the Elsinore Mk6 Crossover shown here, this applies all of the above, all the factors balanced against each other.
As stated, it was the Elsinore Mk5 and Mk6 speakers that made me - actually forced me as a necessity - to construct this 40 watt "current" amplifier of this thread. The modeling proved it would work and I wanted it to be proven in REAL LIFE. The two are totally in agreement.
Modeled:
It is of course the electrical measurement, actually measured and NOT modeled, that is significant.
Actual measurement:
There is a LOT more work to be done - it is interesting that here in Sydney, the home of Thiele-Small Parameters, that a small number of us are looking at the above. What is certainly true, that the Elsinores. compared to LR4 Crossovers that dominate a large number of commercial speakers with Crossovers, that they have a pristine quality, a beguiling clarity that is so easy on the ear, so sweet and yet capable of a high resolution that some lighteners have likened it to ribbons (which draw quite linear current, as I indeed pointed out to Allen as they were his preference), so do most headphones (and hence some listeners prefer them over to speakers? - not me) and elctro-statics which don't track current but are voltage devices. Why are we designing speakers as IF they were voltage devices when in fact they are current devices - and Crossovers and their designers loose sight of that and make horrible sounding Crossovers and don't even know it because THAT is all they have ever heard.
Down here we are starting up a serious and proper research project that will explore these things - we have a Discussion Paper already and that sets the scene. It will be two-fold as I see it, the importance of current contaminations on Voice Coil linearity and the role that Crossovers that cause or at the very least exacerbate the problem.
And of course, third-fold, this will also have a LOT of meaning to those like you who are in the full-range camp and in fact will justify what you are doing in that area. A form of exoneration against those who pooh-pooh the use of single drivers - not me of course, as single drivers will always be the ideal driver, period. It always will be, only practicality notwithstanding.
As our head academic says, with a funded research background, no such study has been conducted and as he is getting nearer retirement age, this is something he wants to be part of and we are looking at something that will prove to be of legacy value.
Whew! That was a LOT, so Dave, over to you.
Cheers, Joe
.
Last edited:
The variable transimpedance was an example. Mostly FR speakers around here. No XOs at all. The resonant trap trick could be quite useful thou.
The whole XO vrs current amps issue is eliminated if you go active.
I think what Joe & crew are exploring is very important, the mainstream of the speaker world is pretty conservative and needs some poking. EnABL for instance.
dave
The whole XO vrs current amps issue is eliminated if you go active.
I think what Joe & crew are exploring is very important, the mainstream of the speaker world is pretty conservative and needs some poking. EnABL for instance.
dave
More than just useful, since you can control the Z and it's fully variable and only limited to the part where you want to reduce the Z. Basically I don't need a variable Z amp and it works beautifully with FR.
Indeed, and maybe the real reason that active speaker that are voltage driven, the real reason many find the sound better is not just because there is no Crossover and that gives them the idea that the amplifier somehow has greater control, hah, when in fact the real evil of the Crossover is that it contaminates the current purity that the Voice Coil deserves. So active is not superior for the reason they think.
What we need is to find a measurement that shows and represents current contamination before it gets to the VC (indeed the VC's inductance likely plays a part, but to a lesser degree). It will be observable as noise, indeed phase noise which is also the term used when describing jitter - and while jitter is thought of in a digital context, jitter is in fact a totally analog concept. This is about noise being generated, but noise NOT in the random sense though, but programic. Hence it should be definable in mathematical terms IMO.
True, true...
Cheers, Joe
.
Audition this TODAY, July 5, 2015, in Sydney, Oz, at 2pm !
If you are in the Sydney area (Australia), the Audiophile Society of NSW is showcasing this revolutionary DIY transconductance power amp, driving a pair of the superb DIY Elsinore Mk6 current-compatible speakers
Not to be missed if you can possible get there.
Here is the link for details:
ASoN - Sunday July 5th 2015 - An outstanding lineup of hi-fi components and music - Audiophile Society of NSW - StereoNET
Cheers (with apologies for the late notice),
Warren
If you are in the Sydney area (Australia), the Audiophile Society of NSW is showcasing this revolutionary DIY transconductance power amp, driving a pair of the superb DIY Elsinore Mk6 current-compatible speakers
Not to be missed if you can possible get there.
Here is the link for details:
ASoN - Sunday July 5th 2015 - An outstanding lineup of hi-fi components and music - Audiophile Society of NSW - StereoNET
Cheers (with apologies for the late notice),
Warren
Member
Joined 2009
Paid Member
I'd like to understand this better, with regards single driver full range speakers. You are saying the better approach is to use an amplifier, directly coupled to the speaker driver voice-coil, with a high output impedance a.k.a. current driver. Then, an additional RCL series network is placed across the voice-coil and tuned to the L.F. resonance peak of the speaker (if vented there are two peaks and two networks). Then, where the speaker has a high impedance peak due to resonance the series network has a compensating low impedance. In measuring the system impedance the L.F. resonance peak will be much diminished. But does this imply better control over the cone at this frequency?
It's really quite simple, using Thevenin we can determine that if the response of the alignment does not change, then neither does the Q of the system - same response, same alignment, same Q, simple.
Cheers, Joe
.
Hi, Joe
Nice to see you are working with the concept.
I just wanted to bring to your attention some notes on your schematics that may also have to do with the bass issues mentioned in the link above.
The usual supply bypass capacitors for the IC have been omitted. Is this deliberate? I also find the reservoir capacitors fairly small for this power level; as the offset cancellation current is also taken directly from the supply, this can cause some distortion (as you have measured).
As I see it, the output impedance at 20 kHz should be about 80 ohm due to the 0.1uF capacitor.
In the Elsinore, the lower impedance peak seems to be uncompensated (post #143) which is understandable as it requires large component values. This may, however, lead to large excursions on current feed if the signal has still some content at 20 Hz.
Nice to see you are working with the concept.
I just wanted to bring to your attention some notes on your schematics that may also have to do with the bass issues mentioned in the link above.
The usual supply bypass capacitors for the IC have been omitted. Is this deliberate? I also find the reservoir capacitors fairly small for this power level; as the offset cancellation current is also taken directly from the supply, this can cause some distortion (as you have measured).
As I see it, the output impedance at 20 kHz should be about 80 ohm due to the 0.1uF capacitor.
In the Elsinore, the lower impedance peak seems to be uncompensated (post #143) which is understandable as it requires large component values. This may, however, lead to large excursions on current feed if the signal has still some content at 20 Hz.
Last edited:
Hi Esa
A few things, I made this amplifier mainly to prove a point. It was really more about the Elsinore speakers than the amp. I needed to physically prove that the concept not only worked on paper, but by making this amp I was able to prove and demonstrate it works in the real world.
The lack of by-pass caps? The compactness of the circuit does not really come across in the diagram, the reservoir caps are very close to the power supply pins and low ESR.
Yes I can see you thinking that, but as I remember it, and I would need to redo it again to confirm, it didn't measure that way when you calculated the difference in current versus voltage. I aimed for 100mVAC in for 250mA output and when you track the error it comes out like that. As I said, I really need to revisit it, but it would be an academic exercise more than anything else.
Not in practice.
In reality the peak (lower F) is much higher than the sub-20 Hertz shown. The LCR has a Q that can be varied and can be surprisingly accurately modeled (would need to do a workshop and demonstrate it rather than as a topic here), and carefully varying the LCR, the lower F peak is suppressed both in frequency and amplitude. In other words, the upper F EQ/Q affects and suppresses the lower F, but without actually eliminating it.
The real question is, how does it work in practice? Actually amazingly well. This quick bit of modeling shows why:
Note the vertical scale here is 10dB and we are looking at the 'effect' and it is down about -20dB, and note there is barely a hint of it lipping upwards, rather it looks more like a small plateau.
Trust me, there are no excessive excursions from the drivers, and this speaker is remarkably easy to drive (I would say more so than any commercial multi-way loudspeaker). It will go very loud indeed and sounds totally in control and not the slightest signs of discomfort. (Anybody near here is happy to come around for a demo.)
I also did a smaller speaker that was 4 Ohm and less sensitive, here the lower F was 21 Hertz and 19 Ohm (around -15 to 17dB down). Again a plateau formed from around 30 Hertz and a tiny lip at 21 Hertz before dropping off quickly. This time, playing loud, the only problem is that the amp was quicker to clip. No erratic excursions.
What is remarkable though is the quality of the bass with both speakers. It just completely shatters the idea of Damping Factor - something that Richard Small dismissed way back 1975 in a conversation I had with him. To him it was nonsense and all that mattered was where the alignment ended up. And here the modeling (based on T-S driven software) is very accurate and revealing - and the results confirm it.
Re 4700uF being small value? That made me smile as we were making Gainclones with as little as 1000uF caps as this was copied across from what 47 Labs were doing with their Gaincard Amplifier based on the LM1875 chip. That was the fad insisted upon and was expected. But most Gainclones ended up with the LM3875 used here.
Cheers, Joe
A few things, I made this amplifier mainly to prove a point. It was really more about the Elsinore speakers than the amp. I needed to physically prove that the concept not only worked on paper, but by making this amp I was able to prove and demonstrate it works in the real world.
The lack of by-pass caps? The compactness of the circuit does not really come across in the diagram, the reservoir caps are very close to the power supply pins and low ESR.
Yes I can see you thinking that, but as I remember it, and I would need to redo it again to confirm, it didn't measure that way when you calculated the difference in current versus voltage. I aimed for 100mVAC in for 250mA output and when you track the error it comes out like that. As I said, I really need to revisit it, but it would be an academic exercise more than anything else.
Not in practice.
In reality the peak (lower F) is much higher than the sub-20 Hertz shown. The LCR has a Q that can be varied and can be surprisingly accurately modeled (would need to do a workshop and demonstrate it rather than as a topic here), and carefully varying the LCR, the lower F peak is suppressed both in frequency and amplitude. In other words, the upper F EQ/Q affects and suppresses the lower F, but without actually eliminating it.
The real question is, how does it work in practice? Actually amazingly well. This quick bit of modeling shows why:
Note the vertical scale here is 10dB and we are looking at the 'effect' and it is down about -20dB, and note there is barely a hint of it lipping upwards, rather it looks more like a small plateau.
Trust me, there are no excessive excursions from the drivers, and this speaker is remarkably easy to drive (I would say more so than any commercial multi-way loudspeaker). It will go very loud indeed and sounds totally in control and not the slightest signs of discomfort. (Anybody near here is happy to come around for a demo.)
I also did a smaller speaker that was 4 Ohm and less sensitive, here the lower F was 21 Hertz and 19 Ohm (around -15 to 17dB down). Again a plateau formed from around 30 Hertz and a tiny lip at 21 Hertz before dropping off quickly. This time, playing loud, the only problem is that the amp was quicker to clip. No erratic excursions.
What is remarkable though is the quality of the bass with both speakers. It just completely shatters the idea of Damping Factor - something that Richard Small dismissed way back 1975 in a conversation I had with him. To him it was nonsense and all that mattered was where the alignment ended up. And here the modeling (based on T-S driven software) is very accurate and revealing - and the results confirm it.
Re 4700uF being small value? That made me smile as we were making Gainclones with as little as 1000uF caps as this was copied across from what 47 Labs were doing with their Gaincard Amplifier based on the LM1875 chip. That was the fad insisted upon and was expected. But most Gainclones ended up with the LM3875 used here.
Cheers, Joe
Last edited:
There is one thing about this type of amplifier, the unusual distortion profile, where the odd order distortions are suppressed on a lower level than odd order. I find that fascinating:
10 Watt into 8 Ohm. THD 0.023% & THD+N 0.028%
(Ignore the 16KHz peak, that was inherent in the measurement.)
Cheers, Joe
.
10 Watt into 8 Ohm. THD 0.023% & THD+N 0.028%
(Ignore the 16KHz peak, that was inherent in the measurement.)
Cheers, Joe
.
Last edited:
I hope you didn't get me wrong somehow, Joe. I was only helping to locate the possible weaknesses to make the system perform still better, as we have a common interest.
Simulations on the displacement behavior and measurement of the amp distortion at low frequency could also be helpful here.
Simulations on the displacement behavior and measurement of the amp distortion at low frequency could also be helpful here.
As a rule of thumb, in a 1000uF reservoir cap the voltage drops 10 V per ampere of current drawn during the 10 ms discharge period (50 Hz systems). Of course the amp still works somehow, but to me this is not a place to save.Joe Rasmussen said:
No worries. This amp was put together quite quickly and performed rather well real world wise. So why not post it for the full-range guys primarily. And they don't need a lot of power anyway. But what is surprising is that this power chip is common DIY fare for making cheap Gainclones, and yet configured as transconductance amps they both measure and sound very different indeed. In fact it is the sound that is really surprising and provided speakers are capable (and my Elsinore Mk6 can be driven by any source impedance - real world wise) that shows this up. I will never build a Gainclone amp, that's for sure.
Cheers, Joe
After lots of tests with series resistors on voltage source amps to simulate current drive, and after seeing the dramatic effects in terms of distortion reduction, I would like to use a real current source in my active setup.
This design seems to be the most accomplished and usable one, and 50W will be enough for my needs.
I have some questions though, please forgive me if these were already answered before :
What happens with this amp in case of short or open circuit?
Is it on/off pop free ?
Beside the higher gain (32dB?), is it suited for a 16 ohm nominal load?
Is there a PCB available somewhere for that amp?
... or a ready-made version? (one can dream)
This design seems to be the most accomplished and usable one, and 50W will be enough for my needs.
I have some questions though, please forgive me if these were already answered before :
What happens with this amp in case of short or open circuit?
Is it on/off pop free ?
Beside the higher gain (32dB?), is it suited for a 16 ohm nominal load?
Is there a PCB available somewhere for that amp?
... or a ready-made version? (one can dream)
Interesting for the readers of this thread:
http://www.diyaudio.com/forums/software-tools/279762-layouting-free.html
http://www.diyaudio.com/forums/software-tools/279762-layouting-free.html
Member
Joined 2009
Paid Member
Hi
JBL C500G 5" mid:
http://www.diyaudio.com/forums/multi-way/277997-break-up-frequency-distortion.html#post4410302
JBL 2020H and TAD TM1201H 12" midwoofers:
http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-403.html#post3026510
I also got similar results with woofers (but these needs specific EQ when used around Fs...) and tweeters.
The only devices that did not show improvements in my tests were compression drivers.
JBL C500G 5" mid:
http://www.diyaudio.com/forums/multi-way/277997-break-up-frequency-distortion.html#post4410302
JBL 2020H and TAD TM1201H 12" midwoofers:
http://www.diyaudio.com/forums/vendors-bazaar/190434-hypex-ncore-403.html#post3026510
I also got similar results with woofers (but these needs specific EQ when used around Fs...) and tweeters.
The only devices that did not show improvements in my tests were compression drivers.
What about the one developed by JPS1964 in the "Layouting for free" thread?
It seemed to look really well thought out. Only thing I would have changed is the position of the LM3875 chip to make it protrude outside the board so it can more easily be attached to a heatsink...