Yes, it's for a power amp. about 200-300 watts AB.
Here's my thinking, which is a bit arbitrary. I'm looking to design an amp that can provide as transparent a window into the audio spectrum as perceivably possible. I've read that audio system that can recreate sound accurately to 100khz have been chosen over less able systems in documented listening as more more pleasant more realistic to listen. I'm vague on the biological foundation for this result, although it has to do with somehow the transients above the audio spectrum at least to 100khz are perceivable and there absence and be missed in A/B tests where one system has that extended range and one system lacks the extended range.
This is kind of moot at the moment because commercial music fails to encode information that high. Although that's just a matter of time, and one can still designing amps that will produce that range now. Hence the motivation. So, I was thinking 140V/us would allow a fairly good 100khz square wave.
I've been reading both the Self and Cordell books on amp design and they seem to lack any todo about current feedback topologies. Do you know of one?
Here's my thinking, which is a bit arbitrary. I'm looking to design an amp that can provide as transparent a window into the audio spectrum as perceivably possible. I've read that audio system that can recreate sound accurately to 100khz have been chosen over less able systems in documented listening as more more pleasant more realistic to listen. I'm vague on the biological foundation for this result, although it has to do with somehow the transients above the audio spectrum at least to 100khz are perceivable and there absence and be missed in A/B tests where one system has that extended range and one system lacks the extended range.
This is kind of moot at the moment because commercial music fails to encode information that high. Although that's just a matter of time, and one can still designing amps that will produce that range now. Hence the motivation. So, I was thinking 140V/us would allow a fairly good 100khz square wave.
I've been reading both the Self and Cordell books on amp design and they seem to lack any todo about current feedback topologies. Do you know of one?
Nelson Pass's F5 is a good example of a fast "current feedback" amp. Be careful though, current feedback isn't automatically a ticket to high speed. Strictly speaking, what gives the high speed is a push-pull input stage that can operate in class AB to give high output currents in either direction. That kind of input stage is more common with "current feedback" designs, especially opamps, but can be done with either kind of feedback.
For example power amps like the Quad 303 and JLH's 1969 10W class A design both use "current feedback" but have a single transistor input stage, so don't have any speed advantage.
OTOH, a lot of John Curl's designs use complementary differential input stages which allow high peak currents, and thus high slew rates, even though he uses voltage feedback.
That said, 140V/uS isn't that fast and should be quite possible with more "normal" topologies. For 200-300 watts, a bridged amp may be the best option, in which case you only need 70V/uS on each side, which would be even easier.
I haven't read any of Doug Self or Bob Cordell's books but I am aware that Doug Self thinks high slew rate is a bad idea. AFAIK, Bob doesn't have any such reservations though. I'd be surprised if none of his designs offered high slew rates.
I've seen the Krell KSA schematic and I like the dual differential input. I happen to like symmetry. I also noticed that Cordell details how to have a push-pull VAS with a single differential input stage. I like that too.
In my sim, which started out as a copy of AmpLabs C200, then modified with details from the design books and bits and pieces from threads here, I've already got the slew rate up to about 70V/uS and pretty symmetrical. I had to migrate to a two pole compensation scheme, which allows a remnant of over shoot to persist, although the side bands of the 20khz-19khz fundamentals and the peak at 39khz of the TID test are now 117 dB lower than the fundamentals. Without the two pole compensation, it was more like 80dB down. Reportedly, this would be a noticable improvement.
The differences in opinion from these pillars of the audio community is quite impressive. Being eclectic, I find a diversity of opinions to be a good thing. It just gives me more to explore, and more from which to choose what I like.
I have a 2x20V 225VA toroidal power transformer that I bought a long time ago for a old project that's been sitting around. I can use that to build a 2x30 watt class A amp with which to do some tests. I lack any speakers that would do it justice though, just some Stageworks PA monitors, with low efficiency drivers that have a noticable coloration in their tone.
Last edited:
My recollection is that the maximum slew rate required to reproduce audio band signals is on the order of a few V/uS. There are, however, two factors which argue for more than just a few V/uS slew-rate. Should the input spectrum be wider than that of the audio band, such as with DAC analog stages, then we need a greater slew rate than would be required to simply reproduce audio band signals. This will likely also be true for any practical gain stage not featuring effective band limiting either at it's input, or in it's feedback loop.
The other factor is that, should the load present an significant shunt capacitance to the gain stage, current delivered by that stage, needed to charge and discharge that shunt capacitance, must be commensurately significant. Increasing the gain stage's current delivery capability consequently increases it's slew-rate in to a given shunt load capacitance. So, here, we're not attempting to reproduce a wider input spectrum, we are increasing the current delivery capability for the purpose driving some amount of shunt capacitance. An increased stage slew rate is incidental, in this case.
At least, that's my understanding.
Last edited:
Hey Ken, I just realized that if I doubled the input stage and made a pushpull VAS then maybe the slew rate would approach 2x what it is now. That would get it in the neighborhood.
Well, I'm looking at expanding the the output spectrum to a reasonable amount beyond the audio spectrum as a good thing, with the goal being to be able to reproduce all material withing that spectrum in a consistent manner throughout it, meaning with approximately the same amount of distortion regardless of frequency.
Nyquist theory has it that a sampling rate of 2x the highest frequency is all that is required to reproduce that frequency or lower --in a perfect world. As it turns out electronics are far enough away from perfect that to get good digital effect IMHO, one has to go to at least a sampling frequency of 5x the highest frequency. 96khz is the lowest I like to use, and I'd prefer to have 192khz sampling.
From a quick sine wave differentiation, It looks like the maximum rise rate of a 20khz sine wave to 40V is 5V/uS. So for 100khz that would be 25V/uS. Although audio material is going to have lots of transients that rise significantly faster than that. I view the square wave testing as a way to reveal an amps capacity to handle such non-sine wave material.
If I comprehend you correctly, I noticed that also depends on the compensation method employed. I tried just raising the bias current in the input stage, that did increase the slew rate, although it also required an increase in the Miller compensations capacitor to reestablish stability, an increase of such magnitude that for all practical purposes it nullified the gain in slew rate from the current increase. That's what motivated my switch to two pole compensation. I have a few different schematics of current feedback designs now. I'll have to build some new models.
Well, I'm looking at expanding the the output spectrum to a reasonable amount beyond the audio spectrum as a good thing, with the goal being to be able to reproduce all material withing that spectrum in a consistent manner throughout it, meaning with approximately the same amount of distortion regardless of frequency.
Nyquist theory has it that a sampling rate of 2x the highest frequency is all that is required to reproduce that frequency or lower --in a perfect world. As it turns out electronics are far enough away from perfect that to get good digital effect IMHO, one has to go to at least a sampling frequency of 5x the highest frequency. 96khz is the lowest I like to use, and I'd prefer to have 192khz sampling.
From a quick sine wave differentiation, It looks like the maximum rise rate of a 20khz sine wave to 40V is 5V/uS. So for 100khz that would be 25V/uS. Although audio material is going to have lots of transients that rise significantly faster than that. I view the square wave testing as a way to reveal an amps capacity to handle such non-sine wave material.
The other factor is that, should the load present an significant shunt capacitance to the gain stage, current delivered by that stage, needed to charge and discharge that shunt capacitance, must be commensurately significant. Increasing the gain stage's current delivery capability consequently increases it's slew-rate in to a given shunt load capacitance. So, here, we're not attempting to reproduce a wider input spectrum. We are increasing the current delivery capability for the purpose driving some shunt capacitance. An increased stage slew rate is incidental, in this case.
At least, that's my understanding.
If I comprehend you correctly, I noticed that also depends on the compensation method employed. I tried just raising the bias current in the input stage, that did increase the slew rate, although it also required an increase in the Miller compensations capacitor to reestablish stability, an increase of such magnitude that for all practical purposes it nullified the gain in slew rate from the current increase. That's what motivated my switch to two pole compensation. I have a few different schematics of current feedback designs now. I'll have to build some new models.
If I understand what you have described, it sounds to me that, perhaps, the increased input stage current has also increased the input stage transconductance. This would in turn increase the open loop gain, which would increase the feedback, which would increase the capacitance needed for the Miller comp. cap. The larger Miller cap. may then be consuming the increased bias current, preventing you from gaining the expected improvement in slew rate from the current increase. I'm surmising here.
I suggest trying to reduce the input stage transconductance by increasing local degeneration, so that you do not have to increase the Miller cap. to maintain closed-loop stability. Hopefully, that will return the missing slew rate increase.
Last edited:
look over some schematics of products from manufacturers that participated in the "slew rate wars" - late 1970s to mid 1980s (kenwood and sansui come to mind right away).
just watch out for some of the problems that resulted: output stage common mode conduction, rail sticking when clipping, etc. i remember doing more than a few tweeter replacements around those times.
there are consequences for everything.
if anatech is around, he might comment with more details.
also, the driver stage from cordell's 50w mosfet amp from a while ago would definitely be worth studying ...
good luck!
mlloyd1
just watch out for some of the problems that resulted: output stage common mode conduction, rail sticking when clipping, etc. i remember doing more than a few tweeter replacements around those times
.there are consequences for everything.
if anatech is around, he might comment with more details.
also, the driver stage from cordell's 50w mosfet amp from a while ago would definitely be worth studying ...
good luck!
mlloyd1
Last edited:
If I understand what you have described, it sounds to me that, perhaps, the increased input stage current has also increased the input stage transconductance. This would in turn increase the open loop gain, which would increase the feedback, which would increase the capacitance needed for the Miller comp. cap. The larger Miller cap. may then be consuming the increased bias current, preventing you from gaining the expected improvement in slew rate from the current increase. I'm surmising here.
Yep, that's about how it went.
Easy enough, just increase the emitter resistors ton the differential pair transistors to balance out the increase in transconductance from the increased bias current. I'll give it a try.
Bonsai,
Can you describe what you mean by a stand-off voltage to the input stage?
As I see it the available slewing current while in class-a is limited by the idle current exactly the same was a VFA is limited, except of course that a vfa will clip at this current whereas a cfa will just enter class ab (a very attractive feature in many cases, just wouldnt think an audio power amp is one of them as I would assume that both cfa and vfa would be designed to comfortably avoid this condition).
Thanks
-Antonio
Last edited:
Perhaps you might like to get a PC soundcard, connect it up in loopback mode and run RMAA to verify that it has output up to 20k, and then connect up a pair of IEMs from a reputable source like Shure and listen to a rising tone output from one of the many pieces of software which can do this. This will give you some insight into the audibility of high frequency sound and what it truly contributes to the appreciation of music.
Any performance parameter of a piece of electronic equipment can be boosted, but usually at the expense of some other parameter, if nothing other than cost.
Many amplifiers are designed to provide 'as transparent a window into the audio spectrum as perceivably possible', certainly those that claim to be hifi. Power in music certainly drops off at higher frequencies. Perhaps a small effort to conduct a personal experiment might be preferable to expending a lot of effort to achieve a performance to meet a requirement only justified by theories with no experimental evidence to support them and generally conceived only to support the most arcane excuses for the failure of A/B testing to reveal any statistically significant results.
Any performance parameter of a piece of electronic equipment can be boosted, but usually at the expense of some other parameter, if nothing other than cost.
Many amplifiers are designed to provide 'as transparent a window into the audio spectrum as perceivably possible', certainly those that claim to be hifi. Power in music certainly drops off at higher frequencies. Perhaps a small effort to conduct a personal experiment might be preferable to expending a lot of effort to achieve a performance to meet a requirement only justified by theories with no experimental evidence to support them and generally conceived only to support the most arcane excuses for the failure of A/B testing to reveal any statistically significant results.
I offer the following simple non-audiophile calculation:
In a single pole system, Bandwidth * Rise Time = .35
300W at 8 Ohms is about 70V peak, 140V peak to peak.
140V/uS at 300W would be a Rise Time (10%-90%) of about 0.8uS.
Therefore, the Bandwidth needs to be 435kHz (at full power). Don't be suprised when your Zobel smokes. Note that if power is reduced, the bandwidth will have to be increased to maintain 140V/uS.
Slew rate is not a good indicator of overall amplifier quality. If the goal is transparency, our time would be better spent reducing open loop distortion.
In a single pole system, Bandwidth * Rise Time = .35
300W at 8 Ohms is about 70V peak, 140V peak to peak.
140V/uS at 300W would be a Rise Time (10%-90%) of about 0.8uS.
Therefore, the Bandwidth needs to be 435kHz (at full power). Don't be suprised when your Zobel smokes. Note that if power is reduced, the bandwidth will have to be increased to maintain 140V/uS.
Slew rate is not a good indicator of overall amplifier quality. If the goal is transparency, our time would be better spent reducing open loop distortion.
Of course, Slew rate is not good indicator of overall power amplifier quality. But, it'll make the amplifier more perfect.
Low distortion, high SR, high BW - this is the goal of solid state amplifier.
Antonio,
In my sx-Amp, I run the TIS at 60mA standing current (note, I do not use a conventional mirror TIS). The front end never leaves class A mode either.
This type of trick is not really suited to IC opamps where current consumption is a key parameter.
Take a look at the sx-Amp write up on my website - page 12 shows the circuit diagram. R33 and R36 provide the stand off voltage and set the TIS current.
http://hifisonix.com/wordpress/wp-content/uploads/2012/10/The-sx-Amplifier.pdf
In my sx-Amp, I run the TIS at 60mA standing current (note, I do not use a conventional mirror TIS). The front end never leaves class A mode either.
This type of trick is not really suited to IC opamps where current consumption is a key parameter.
Take a look at the sx-Amp write up on my website - page 12 shows the circuit diagram. R33 and R36 provide the stand off voltage and set the TIS current.
http://hifisonix.com/wordpress/wp-content/uploads/2012/10/The-sx-Amplifier.pdf
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.