Peter,
I realize that, but just a zobel won't get you very far with the complex load that is represented by the average loudspeaker. Especially if it is using a filter and has more than a single driver the load will be increasingly complex, due to the reactive components in the filter, and much harder to model and compensate for.
In my book the only approach that warrants the use of compensation at the amplifier for load variance would be to use a active crossover and connect the amplifier to the driver directly, without any passive filter components. The downside obviously is that you'll need an amplifier for each driver and that optimizing the sum of these amplifier's outputs into one coherent soundstage is a bit of challenge.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
I realize that, but just a zobel won't get you very far with the complex load that is represented by the average loudspeaker. Especially if it is using a filter and has more than a single driver the load will be increasingly complex, due to the reactive components in the filter, and much harder to model and compensate for.
In my book the only approach that warrants the use of compensation at the amplifier for load variance would be to use a active crossover and connect the amplifier to the driver directly, without any passive filter components. The downside obviously is that you'll need an amplifier for each driver and that optimizing the sum of these amplifier's outputs into one coherent soundstage is a bit of challenge.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
This is going off-topic 
I second that active filtering and a separate amp for each driver are the best way forward.
However, most speaker designers seem unaware of how far impedance correction can already go in turning passive crossover design into a science, rather than the touchy-feely black art it often seems.
A simple zobel is obviously not a full solution to linearizing a driver's impedance, but quite often it is sufficient to make the lowpass sections in a speaker crossover filter work more predictably.
To linearize the response of an enclosed driver, you need a zobel and an RLC network, the former to compensate for the series inductance of the voice coil, the latter to flatten the hump caused by the mechanical resonance.
To linearize the response of a driver in a vented cabinet, you need another RLC network because there are two resonant peaks. Several equivalent electrical connections of 2 coils, 2 resistors and 2 capacitors are thinkable that will do this.
It's possible to turn mounted speaker drivers into virtual resistors within about 10% of magnitude and a few electrical degrees over the full frequency range. Doing so with each driver in a cabinet reduces crossover design to simply making resistive terminated LC circuits that combine with the drivers' acoustical responses to create the required responses, without having to mind about impedance interactions.
Speakers built in this manner sound surprisingly similar to good active designs. The only thing such a design cannot cater for is non-linear effects which in a passive filter are free to electrically propagate through the filter where in an active design the direct low-impedance connection between amplifier and loudspeaker insures best control and thus minimum distortion.
If the aim is to make a "good" speaker, methodical passive filter design including impedance correction at each driver will get you further than anyone would dare to expect. If the aim is to make a "great" speaker, go active. But, like Sander says - the latter approach will take quite some more money and time, so there's the tradeoff.
I second that active filtering and a separate amp for each driver are the best way forward.
However, most speaker designers seem unaware of how far impedance correction can already go in turning passive crossover design into a science, rather than the touchy-feely black art it often seems.
A simple zobel is obviously not a full solution to linearizing a driver's impedance, but quite often it is sufficient to make the lowpass sections in a speaker crossover filter work more predictably.
To linearize the response of an enclosed driver, you need a zobel and an RLC network, the former to compensate for the series inductance of the voice coil, the latter to flatten the hump caused by the mechanical resonance.
To linearize the response of a driver in a vented cabinet, you need another RLC network because there are two resonant peaks. Several equivalent electrical connections of 2 coils, 2 resistors and 2 capacitors are thinkable that will do this.
It's possible to turn mounted speaker drivers into virtual resistors within about 10% of magnitude and a few electrical degrees over the full frequency range. Doing so with each driver in a cabinet reduces crossover design to simply making resistive terminated LC circuits that combine with the drivers' acoustical responses to create the required responses, without having to mind about impedance interactions.
Speakers built in this manner sound surprisingly similar to good active designs. The only thing such a design cannot cater for is non-linear effects which in a passive filter are free to electrically propagate through the filter where in an active design the direct low-impedance connection between amplifier and loudspeaker insures best control and thus minimum distortion.
If the aim is to make a "good" speaker, methodical passive filter design including impedance correction at each driver will get you further than anyone would dare to expect. If the aim is to make a "great" speaker, go active. But, like Sander says - the latter approach will take quite some more money and time, so there's the tradeoff.
Thanks Bruno for clarifying it further. (Have been there, have done that )
Even if you compensate the separate speakers itself to relax x-over design, it leaves you with one ore more impedance humps in front of the total x-over around the x-over frequencies, due to the sometimes called “the missing band pass” function. For amps with a low damping factor or high Ri (like tube amps and low/no feedback amps) it can be beneficial to compensate these humps in front of the x-over.
From your earlier reply I understand that such further compensation in front of the x-over is not of benefit with the UcD concept.
Cheers
)Even if you compensate the separate speakers itself to relax x-over design, it leaves you with one ore more impedance humps in front of the total x-over around the x-over frequencies, due to the sometimes called “the missing band pass” function. For amps with a low damping factor or high Ri (like tube amps and low/no feedback amps) it can be beneficial to compensate these humps in front of the x-over.
From your earlier reply I understand that such further compensation in front of the x-over is not of benefit with the UcD concept.
Cheers
There's a minor advantage to correcting the combined impedance of the total assembled system.
It has been claimed before by some (no doubt somewhere on this forum as well) that the "impedance in music" of a speaker is lower than the "impedance in sinewaves". A handful of papers has been written to that effect, with lots of statistical analyses using musical fragments, but apparently not much in the way of an explanation. Well, here it is:
If you have a system exhibiting a strong parallel resonance, there is a possibility of drawing more current from the amplifier than would be expected on the basis of rail voltage and impedance minimum or DC resistance alone.
To be more precise, suppose you excite the system with a sinewave at the parallel resonance and then you suddenly reverse the phase of the sinewave right at the voltage peak, you'll get a current spike roughly equal to the voltage step divided by the equivalent series resistance, which means up to twice the current you'd expect driving a resistor alone. The higher the Q of the parallel resonance, the worse the effect. So, it's quite funny to realise you can get the worst peak current at the impedance maximum...
Obviously, the impedance of a speaker is the same with music as with sinewaves, but what it shows is that one shouldn't forget phase information when estimating transient behaviour from frequency domain data.
The upshot is that when you're playing loud, you might drive an amp into current protection even though the nominal impedance of the speaker is higher than the rated load impedance of the amp. If you correct the speaker impedance to a ruler-flat line, this will no longer happen. The current spikes will flow only between the impedance correction network and the rest of the speaker.
It has been claimed before by some (no doubt somewhere on this forum as well) that the "impedance in music" of a speaker is lower than the "impedance in sinewaves". A handful of papers has been written to that effect, with lots of statistical analyses using musical fragments, but apparently not much in the way of an explanation. Well, here it is:
If you have a system exhibiting a strong parallel resonance, there is a possibility of drawing more current from the amplifier than would be expected on the basis of rail voltage and impedance minimum or DC resistance alone.
To be more precise, suppose you excite the system with a sinewave at the parallel resonance and then you suddenly reverse the phase of the sinewave right at the voltage peak, you'll get a current spike roughly equal to the voltage step divided by the equivalent series resistance, which means up to twice the current you'd expect driving a resistor alone. The higher the Q of the parallel resonance, the worse the effect. So, it's quite funny to realise you can get the worst peak current at the impedance maximum...
Obviously, the impedance of a speaker is the same with music as with sinewaves, but what it shows is that one shouldn't forget phase information when estimating transient behaviour from frequency domain data.
The upshot is that when you're playing loud, you might drive an amp into current protection even though the nominal impedance of the speaker is higher than the rated load impedance of the amp. If you correct the speaker impedance to a ruler-flat line, this will no longer happen. The current spikes will flow only between the impedance correction network and the rest of the speaker.
Hi David,
In a certain way you can say so. But in this case it is the stored energy in de moving mass and springs that was put in there by the amplifier itself. With a zobel across the speaker terminals (of the units itself) it is adsorbed by the zobel. The zobels in front of the x-over then absorb the excess stored energy in the L’s and C’s of the x-over.
Cheers
In a certain way you can say so. But in this case it is the stored energy in de moving mass and springs that was put in there by the amplifier itself. With a zobel across the speaker terminals (of the units itself) it is adsorbed by the zobel. The zobels in front of the x-over then absorb the excess stored energy in the L’s and C’s of the x-over.
Cheers
soren said:
I like your attitude. If I'm not mistaken you are the primary author of the papers I referenced in the beginning of this thread. If so, I would very much appreciate it if you would email me the values used in the unreadable simulation schematics included in your papers that I referenced. You must have noticed that Bruno more or less blew off the claim made in the paper with regard to the superiority of a linearized hysteresis self oscillating class d amplifier as compared to the UcD style class d circuit. I would like to attempt to verify this claim using LTspice (hence the need for the values).
By the way, have you read the thread on the leapfrog method of switching amplifier control loop design? If not, considering your work, you may find it an interesting read.
Regards -- analogspiceman
Bruno Putzeys said:
I suppose the other side must be design inaccessibility due to licensing fees or patent protection.
Seriously, you make a valid point about the comparator seeing naught but the slope at the crossing of its inputs. The question is, how do the carrier's dynamic slope variations under the influence of frequency dependent feedback of output, power supplies, etc. affect distortion?
The Poulsen papers boldly claim that, when it comes to self-oscillating modulation schemes, linearized hysteresis outperforms UcD's near sinewaves. Personally, I find the inscrutable nature of the supporting arguments to be less than convincing, but the lab data is pretty hard to dismiss (especially since the paper is so coy with many of the important details).
classd4sure said:
The paper is very intriguing, but good luck reading the 2 point font on the schematics.
Regards -- analogspiceman
Analogspiceman,
Now I'm truely wondering about that paper, and I seem to have a neck for decyphering unreadable text, so feel free to fwd the paper, I have a prototype class-D running on the testbench as is, so I might as well give it a go.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Now I'm truely wondering about that paper, and I seem to have a neck for decyphering unreadable text, so feel free to fwd the paper, I have a prototype class-D running on the testbench as is, so I might as well give it a go.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Ehh no, it's output impedance and frequency response. If people didn't want that, protecting it with a patent wouldn't be very economical.analogspiceman said:
1) Noticeably.analogspiceman said:
2) In a manner which eludes simple algebraic analysis. So far only a complete time run on spice will do.
Point is of course, they don't have the full modulator that I use, which I have put serious effort into optimising. If you take the modulator from the patent, well you can see the distortion on a scope...analogspiceman said:
Bruno,
I'm guessing you mean something like this (modulator from the patent used, carrier frequency waveform shown)?
UcD carrier frequency waveform
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
I'm guessing you mean something like this (modulator from the patent used, carrier frequency waveform shown)?
UcD carrier frequency waveform
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
No, I'm talking of the distortion of an audio signal, not of the carrier. What you've got there is called a layout problem. More precisely, it looks like you've parallelled a smaller ceramic cap with the filter cap in an attempt to reduce the 100MHz hash. Unfortunately, the two caps form a parallel resonant circuit.
Bruno,
Point taken, but you mentioned the following:
No such thing, just a honest Epcos MKP 100V- 470nF capacitor, the rest you see is the layout acting up, but since I'm using perforated board that's a given.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Point taken, but you mentioned the following:
No such thing, just a honest Epcos MKP 100V- 470nF capacitor, the rest you see is the layout acting up, but since I'm using perforated board that's a given.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Alright, so the full modulator used on the UcD modules is of a better design with less distortion than the one described in the patent. Being the hands-on kind of guy I am I figured that this should show clearly when feeding a UcD180 module a 10kHz square wave and have it output around 10Vpp into a 8-ohm resistive load and compare that to my UcD prototyp that uses the simple modulator schema as seen in the patent.
UcD180 10kHz square wave, 10Vpp out into 8-ohm
UcD prototype 10kHz square wave, 10Vpp into 8-ohm
I have difficulty finding higher distortion, of course the spikes in my prototype are evident, as is the higher carrier residue, but they're a layout issue. The rise and fall times of the prototype however look better than the UcD180 and so does the overshoot at either side of the square wave? Or am I missing something here?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
UcD180 10kHz square wave, 10Vpp out into 8-ohm
UcD prototype 10kHz square wave, 10Vpp into 8-ohm
I have difficulty finding higher distortion, of course the spikes in my prototype are evident, as is the higher carrier residue, but they're a layout issue. The rise and fall times of the prototype however look better than the UcD180 and so does the overshoot at either side of the square wave? Or am I missing something here?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.