In principle, I agree, however, there are significant differences among them, meaning even they are not all the same.
Just compare 317 with 337. Anything but the same. 337 gives up on noise suppression just above 100 Hz - not very helpful, is it? Thorsten told me to dump the 337 and use the 317 instead for both +Vcc and -Vcc.
On the other hand, I honestly like the sound coming from a simple discrete regulator, referenced by a base to GND connected zener diode. I also note that zener diode manufacturing has come a looooong way since I checked them out last, they are now much more precise and much less thermal slide prone. And my beloved MJE 15030/15031 transistors are, according to Motorola/ON Semi Data Sheets, actually 60 MHz devices, so at their actual stress level, it all turns into a very widebandwidth job. Just the way I like it.
It appears Dan d'Agostino agreed when he used practically the same thing in his early 90ies Krell models.
That standard simple zener reference discrete regulator has been around since year dot, and is standard equipment in just about every audio item ever built.....and I always thought it was for economy
Dan.
Actually Max, I wouldn't be surprised if you discovered that this was by no means a cheap solution, especially in comparison with 3 point regs.
With regs, all you really need are a couple of resistors, the regs, a pair of relatively large caps before them and a few high frequency filtering low value caps, e.g. 100 nF/100V, or some such.
With this baby, the transistors cost about the same as the regs, but you use twice the caps on the input side, and an extra pair at the output side. Zeners are basically dirt cheap. And, as you probably know, high quality caps tend to be rather expensive, much more so than 3 point regs.
I have not worked it out exactly, but I wouldn't be surprised if this baby costs 2-3 times more than a similar solution using 3 point regs costs.
Also, I feel reasonably sure that Dan d'Agostino wasn't thinking in prices when he used it. His products are not exactly known for their rock bottom low prices.
I expect that you have it the wrong way around.
One transistor, one zener and one extra cap wrt a standard three pin reg...historically that would have been cheaper than the IC version, plus high volume production supply chain considerations.
Nowadays most consumer audio gear runs 3 pin regs, but this was not always so.
Dan.
One transistor, one zener and one extra cap wrt a standard three pin reg...historically that would have been cheaper than the IC version, plus high volume production supply chain considerations.
Nowadays most consumer audio gear runs 3 pin regs, but this was not always so.
Dan.
Last edited:
I got you straight, Dan, don't worry, my point was that the tables are now reversed. What was the cheaper solution once is now the more expensive version.
Given the price drops of integrated circuits in general, plus an ever increasing choice among them, I am hardly surprised.
Early Krell preamps and sister companies Aragon and Acurus up to 1994, had similar regulators since they believed in better sound of open loop regs.
But regulators output had always been heavy filtered by brute force, usually with two stage 25 ohm plus 3x1000 microfarads.
Later they started using hybrid op-amp referenced regulators as an improvement in top products. They were very simple regulators similar to well known Sulzer/Jung regs.
Today, Coda preamps and end extremely well made italian AM preams have PS regulators very similar to ones presented here.
But regulators output had always been heavy filtered by brute force, usually with two stage 25 ohm plus 3x1000 microfarads.
Later they started using hybrid op-amp referenced regulators as an improvement in top products. They were very simple regulators similar to well known Sulzer/Jung regs.
Today, Coda preamps and end extremely well made italian AM preams have PS regulators very similar to ones presented here.
No doubt. This is hardly anything new, with perhaps a small twist here and there, but basically, it's a well known circuit.
It's very possible that I like it simply because I tend to prefer wide bandwidth, low global feedback circuitry. I am well aware of both the (still) raging debate over low or high global NFB circuitry, and of most of the advantages and downfalls of each type, but historically, I have always tended to own low global NFB circuitry. It simply sounds better to me.
For example, my H/K Citation 24 power amp uses just 12 dB (3:1) global NFB, and the preamp has just a bit more. My preceeding integrated amp, H/K 680, also used 12 dB of global NFB, and the one preceeding that, H/K 6550, used 17 dB of global NFB.
However, while the 680 is an amp with an attitude, the Citation is definitely one of the most neutral power amps I have ever come across.
Only yesterday I modified some of the supply regulators in the analog stages of my current DAC. I was using TL431 shunts but after playing in LTSpice with an emitter follower after the shunt - effectively a programmable variation of the zener+tranny arrangement you've been talking about - I decided to implement this in place of the shunt. Initial indications are indeed that it has improved the sound, however the effect is fairly subtle so far.
I'm curious as to what might be going on, and I suspect that it has to do with how the regulator handles self-generated noise on the supply. ClassB circuits generate a fair quantity of 'switching spikes' from their output stage and I have a hunch that this causes IMD in a feedback regulator as it works hard to compensate for this. So I speculate that this noise (which is primarily HF and can go way above the audio band as its rectified audio) gets smeared out across the whole frequency band by the feedback reg. So in effect the feedback reg when its loop is working to correct for this, turns into an LF noise source.
Anyone have any experience to corroborate this?
I assume you have tried this, but just in case - have you tried inserting a smaller value resistor with a shunt larger value capacitor between the TL431 and the transistor? Say, 220 Ohms and say 470 uF?
That would give you a corner frequency of around 3 Hz.
Or perhaps say 10 Ohms/2W with a say 1,000 uF cap after the transistor?
Actually, you're well inside Twighlight Zone, anything might work - or not. Depending on available time and will, you can play quite a lot here.
That would give you a corner frequency of around 3 Hz.
Or perhaps say 10 Ohms/2W with a say 1,000 uF cap after the transistor?
Actually, you're well inside Twighlight Zone, anything might work - or not. Depending on available time and will, you can play quite a lot here.
Its what I am doing yes - I have 1000uF and 62R in one case, 470uF in another case. And the emitter follower's feeding a small inductor (wound on a ferrite bead) into the opamp supply. I have this to provide more filtering and hopefully also avoid parasitic oscillations of the common-collector stage which are notorious for hating capacitive loads. And boy, does it have a capacitive load beyond that L.
Twilight zone - for sure, that's where the most fun is
Twilight zone - for sure, that's where the most fun is
Dvv . You said previously about high loop feedback amps . The argument Douglas Self uses is that crossover distortion is the universal problem with class B amps . The only weapon is feedback . The feedback tends to loose effect at higher frequencies due to lack of HF gain / stability . Thus configuring the amplifier if you like from the output stage backwards helps . Keep that feedback working as best you can .
Rod Elliot who some ridicule said something interesting . FET's allow different arrangements . The distortion although not too pretty usually is at frequencies which the class D people are happy to say don't matter ( discuss ) . Also I always found they were more amenable to vast amounts of feedback .
There is a converse argument that says that if local feedback is used then the amplifier for a certain 1 kHz distortion figure might have greater stability . It is then a debate as to what is better at 10 kHz . On paper is seems Douglas Self is right .
The sub 1 watt distortion will tell most if the previous statements are valid . Measuring a Hypex module recently it was quite good at 1 watt and below . An op amp more or less . Not easy to measure I should point out .
I think it is just possible that stability is more noticeable than moderate crossover distortion when listening ? The tests for instability commonly used might not be like speakers ?
Thinking about carbon composition resistors ( and all resistors ) . DF 96 said it is the material they are made from that is the sound and not inductance nor capacitance . I thought about it . Clay and carbon graphite is not unlike a metal except the carbon is spaced further apart . The electrons hop just like in metals . Mostly the sound of CC is known from 1950's valve amps . The resistance values usually very high . I guess the density of carbon is low and electron jump distance high if high resistance ? This might explain the sound ? The resistance might be proportional to dullness ? If so the low resistance values might be interesting . The text books say the carbon in CC touches forming a conductance path . That seems a bad description . I do feel on the whole CC are nice devices and might be part of the valve sound that is nothing to do with valves . If my idea is right and not controlling EQ or voltages then a low value CC might be a choice . I do not feel what I like about them is me preferring distortion .
Rod Elliot who some ridicule said something interesting . FET's allow different arrangements . The distortion although not too pretty usually is at frequencies which the class D people are happy to say don't matter ( discuss ) . Also I always found they were more amenable to vast amounts of feedback .
There is a converse argument that says that if local feedback is used then the amplifier for a certain 1 kHz distortion figure might have greater stability . It is then a debate as to what is better at 10 kHz . On paper is seems Douglas Self is right .
The sub 1 watt distortion will tell most if the previous statements are valid . Measuring a Hypex module recently it was quite good at 1 watt and below . An op amp more or less . Not easy to measure I should point out .
I think it is just possible that stability is more noticeable than moderate crossover distortion when listening ? The tests for instability commonly used might not be like speakers ?
Thinking about carbon composition resistors ( and all resistors ) . DF 96 said it is the material they are made from that is the sound and not inductance nor capacitance . I thought about it . Clay and carbon graphite is not unlike a metal except the carbon is spaced further apart . The electrons hop just like in metals . Mostly the sound of CC is known from 1950's valve amps . The resistance values usually very high . I guess the density of carbon is low and electron jump distance high if high resistance ? This might explain the sound ? The resistance might be proportional to dullness ? If so the low resistance values might be interesting . The text books say the carbon in CC touches forming a conductance path . That seems a bad description . I do feel on the whole CC are nice devices and might be part of the valve sound that is nothing to do with valves . If my idea is right and not controlling EQ or voltages then a low value CC might be a choice . I do not feel what I like about them is me preferring distortion .
Nige, we've been over this before.
Everybody has a right to their own opinion. D. Self thinks what he thinks and that's fine by me, I disagree so I'll be doing it the way I think I should.
I'd like to remind you that my ultimate control frequency is 50 kHz; once I get THD down to 0.7% or less, open loop, into 4 Ohms, I am happy. With 20 dB global NFB, this will become something like 0.2% or less, again, 50 kHz/4 Ohms, ref. 28.3 V at 1 kHz.
Also, it is not true that global NFB is our ONLY way of dealing with crossover distortion, there's also the matter of bias. You get your output devices to draw 100-130 mA of bias current and you will not suffer any significant crossover problems.
At the same time, this will allow you use your low power, as under typical listening conditions, in pure class A, which has no crossover distortion by default. And when you use 3 or 4 pairs of output devices, all this current tends to add up, so in fact, your output stage will be idling at 0.5 Amps or more. As an example, in my last work, 20-20,000 Hz distortion into 4 Ohms is less than 0.007% up to about 3/6W into 4/8 Ohms with just a tad above 20 dB of global NFB.
50 kHz distortion is less than 0.2% into 4 Ohms, and just a tad below 0.1% into 8 Ohms, measuring across a 100 kHz bandwidth.
Frankly Nige, what more do I want? And with a voltage slew rate of better than 110 V/uS for a nominal 28.3 Vrms output, with 0.1% THD figures reached at 175/350 W peak power output into 8/4 Ohms.
You remember my view that in order to make a great 100W/8 Ohms amp you make a good 150W/8 Ohms amp and call it 100W/8 Ohms amp. Ultimately, I am interested in peak power outputs much more than steady state. At better than 22 dBW/8 Ohms, I think I am probably doing something right.
Everybody has a right to their own opinion. D. Self thinks what he thinks and that's fine by me, I disagree so I'll be doing it the way I think I should.
I'd like to remind you that my ultimate control frequency is 50 kHz; once I get THD down to 0.7% or less, open loop, into 4 Ohms, I am happy. With 20 dB global NFB, this will become something like 0.2% or less, again, 50 kHz/4 Ohms, ref. 28.3 V at 1 kHz.
Also, it is not true that global NFB is our ONLY way of dealing with crossover distortion, there's also the matter of bias. You get your output devices to draw 100-130 mA of bias current and you will not suffer any significant crossover problems.
At the same time, this will allow you use your low power, as under typical listening conditions, in pure class A, which has no crossover distortion by default. And when you use 3 or 4 pairs of output devices, all this current tends to add up, so in fact, your output stage will be idling at 0.5 Amps or more. As an example, in my last work, 20-20,000 Hz distortion into 4 Ohms is less than 0.007% up to about 3/6W into 4/8 Ohms with just a tad above 20 dB of global NFB.
50 kHz distortion is less than 0.2% into 4 Ohms, and just a tad below 0.1% into 8 Ohms, measuring across a 100 kHz bandwidth.
Frankly Nige, what more do I want? And with a voltage slew rate of better than 110 V/uS for a nominal 28.3 Vrms output, with 0.1% THD figures reached at 175/350 W peak power output into 8/4 Ohms.
You remember my view that in order to make a great 100W/8 Ohms amp you make a good 150W/8 Ohms amp and call it 100W/8 Ohms amp. Ultimately, I am interested in peak power outputs much more than steady state. At better than 22 dBW/8 Ohms, I think I am probably doing something right.
You guys are making me think again. Ignoring crossover distortion for the moment, I am thinking about output transient accuracy. Agility and stability become mutually exclusive at some point, determined by the system itself. We can get the output to any voltage, with a certain accuracy. We can get the time rate of change of the ouput voltage to be what we want, to rapidly slew to the desired ouput voltages. But what about looking directly at the maximum time rate of change of the slew rate (and the accuracy of its rate of change)? That's where agility would have to be determined. I wonder whether or not the high-frequency distortion measurement adequately captures that performance requirement. It is a key performance spec, in my view, and so it might be best to examine it more directly. It's also directly related to stability. For example, if we have a perfect 78-degree phase margin (perfect damping giving fastest response with no overshoot), then how fast can the slew rate change? Can the voltage scream upward at a high rate and then turn (the slew rate) on a dime to make a square wave's corner, without overshoot? What limits the rate of change of the slew rate (where is the "inertia"?)? And how far can it be pushed while maintaining a 78-degree phase margin, and then failing that what is the trade-space?
Edit: Or maybe we should abandon classical control theory and the second-order system model, at this point.
Edit: intuitively, it seems like dvv's ideas about running a system at well below it's maximum power rating would be related to this, too. I.e. Running a very powerful and fast system at a fraction of its capability should result in "better" agility AND stability. So for design purposes, that idea will probably play into the trade space question.
Edit: Or maybe we should abandon classical control theory and the second-order system model, at this point.
Edit: intuitively, it seems like dvv's ideas about running a system at well below it's maximum power rating would be related to this, too. I.e. Running a very powerful and fast system at a fraction of its capability should result in "better" agility AND stability. So for design purposes, that idea will probably play into the trade space question.
Last edited:
Here is a question resulting from this . At what frequency does a harmonic becoming unimportant ? The Hypex brigade must to an extent believe this as the suppression is only about - 20 dB albeit starting with 400 kHz . Someone observed if the oscillator is stable the subharmonics would be non existent . Does this spill over in to class AB theory ?
of course even a single sinewave has derivatives of all orders
a Audio System only has to be capable of reproducing the Audio Signals
physics in musical instruments, sound wave propagation, microphones, preamp electronics, recording and reproduction signal channels all have hard “speed” limits for what we end up with as a humanly interesting, "musical" audio signal
Earthworks fastest “drum kit” mic package uses their 50 kHz mics, most legendary musical recording mics have much lower cutoff, large diaphragm vocal mics may roll off <14 kHz
workhorse analog master tape recording machines are limited by head gap, particle size, required high frequency bias – most are again rolling off by 20 kHz – yes you can do better with custom heads, faster tape speed – but only a handful of custom analog tape machines have been made for audio (for 9 figure US$ budget movie projects) that even challenge any spec of $150 sound cards, 24/96 digital audio
phonograph recording is limited to 5 kHz “power bandwidth” - few cart can track even that hot
Sony official recommendation for SACD reconstruction filter is 50 kHz fc, must be pretty high order to get ahead of the > 5th order modulator noise shaping rise to +6 dB around 1 MHz
the engineering numbers for commercial music recording distribution, reproduction today doesn't include “infinite derivative” signals – by our mainstream amp device capabilities audio signals are very slow
then there's the (lack of) speed of our audio transducers - even ES panels have stator opening/thicknes/spacing filter effects as well as the step up transformer's roll off <50 kHz
a Audio System only has to be capable of reproducing the Audio Signals
physics in musical instruments, sound wave propagation, microphones, preamp electronics, recording and reproduction signal channels all have hard “speed” limits for what we end up with as a humanly interesting, "musical" audio signal
Earthworks fastest “drum kit” mic package uses their 50 kHz mics, most legendary musical recording mics have much lower cutoff, large diaphragm vocal mics may roll off <14 kHz
workhorse analog master tape recording machines are limited by head gap, particle size, required high frequency bias – most are again rolling off by 20 kHz – yes you can do better with custom heads, faster tape speed – but only a handful of custom analog tape machines have been made for audio (for 9 figure US$ budget movie projects) that even challenge any spec of $150 sound cards, 24/96 digital audio
phonograph recording is limited to 5 kHz “power bandwidth” - few cart can track even that hot
Sony official recommendation for SACD reconstruction filter is 50 kHz fc, must be pretty high order to get ahead of the > 5th order modulator noise shaping rise to +6 dB around 1 MHz
the engineering numbers for commercial music recording distribution, reproduction today doesn't include “infinite derivative” signals – by our mainstream amp device capabilities audio signals are very slow
then there's the (lack of) speed of our audio transducers - even ES panels have stator opening/thicknes/spacing filter effects as well as the step up transformer's roll off <50 kHz
Last edited:
Settling time would be the easiest comprehensive measurement of transient accuracy. It encompasses both the rate of change and the time to a defined accuracy. Its also very difficult to measure accurately on a non-inverting amplifier. A TEK 7L13 in a 7000 mainframe can do it for most audio purposes to maybe .1%. Here is a good introduction- http://cds.linear.com/docs/en/application-note/an10f.pdf More http://www.analog.com/static/imported-files/tutorials/MT-046.pdf http://cds.linear.com/docs/en/application-note/an120f.pdf and DAC Settling Time Measurement DAC The settling time issue would be true for any amplifier, open or closed loop. I have found it to be a very good indicator of audio performance.
@gootee, jcx, 1audio, nigel
All of your points are valid topics which must be addressed.
I cannot speak for others, but here's how I do it.
I have no particular target frequency I aim for in my open loop response, except that my feeling is that it must hit 50 kHz at ts -3 dB point, ref, nominal output at 1 kHz. My reasining is simple enough - if the amp can reproduce the highest required frequency in an auduio system, assumed to be 20 kHz, in open loop conditions, then it does not have to wait for the feedback signal for it, as it would if its open loop -3 dB point was 1 kHz or less. Consequently, I expect its transient response to be a non-issue, however only if we work with a fair safety margin (where "fair" is a factor of at least 50% above the absolute minimum).
As Tom put it, we can park that frequency point almost anywhere we choose to. While technically true, in practical terms it's not quite so simple without making some serious compromises with stability, and that's the one thing I will not play games with.
To prove that point, I usually get the amp to hit something like 100 kHz under oopen loop conditions. Then I start making it work into real world loads, as I have a small collection of crossovers and drivers from real world speakers (e.g. my own AR94, several Infinity models, a few JBL models, etc). They are far from being perfect, however they beat the hell out of a lab resistor usually used for simulation and free me from dreaming up combinations every which way.
Then I start to compensate, including Miller compensation. I believe in a distirbuted compensation model, meaning that I never ever eave it all to just one or two Miller caps. Rather, I compensate each stage as much as I can. As a result, the Miller cap will always have a smaller value than it otherwise would, although it will still be the first limiting factor on the list, but rather than use values like 33,47 or even 100 pF, it will likely be between 15 and 22 pF. Because it is in real life also influenced by parasitic capacitance of the PCB, its ultimate value will be determined in situ, ie on the board.
Eventually, this will lead to a closed loop response of at least 350 kHz, but more often over 500 kHz, with more global NFB (with me that's like 26 dB), this might well approach 1 MHz. It is this value which will determine the ACTUAL real world voltage slew rate. Then I install a low pass filter at the input, with -3 dB at 200 kHz; after experimenting quite a bit with that, I realized that reVox and Sony are quite right in placing it there (or at 150 kHz, in some Sony models), as those are points which do not cause excessive phase shift at 20 kHz, and anyway, they are first order filters.
Hence, I have two voltage slew rates: one is the internal value, what the amp can do on its own, and the other is the effective, real life value, which is limited by the 200 kHz mark. The fact that amp will do like 3 or 4 TIMES the imiting value of the input filter eliminates oddities, possible RF breaktrhough and whatnot junk which a not so good CD player might spew out.
Now we come to the other side of the coin of wide bandwidth design. While its rise time can be surprisingly short, the not at all unusual even is the square wave overshoot, with very possible ringing, and of course, the settling time. The ringing is usually all but eliminated by the input filter, or at the very least, it's reduced to tlearble levels. If not, I can always adjust the local compensation to obtain a reasonable operformance; a wide bandwidth amp will ALWAYS have some overshoot, but it should not be large and should have no ringing after it, much less - God forbid! - signs of protracted ringing even at low level. It takes patience and work, but can be done.
Settling time can be a bitch, though. I never had any particular problems with it because early on I got lucky and met that German made amp (LAS, I posted its schematic here at least twice), which had the problem solved. A simple collector to base fast diode on the "lower" cascode transistor will shorten the settling time by a factor of 10:1, which is, trust me on this, damn significant. Fortunately, these days, we have a selection of very fast diodes to choose from, in the original LAS amp, its makers used a germanium AA132 diode for its low 0.2V voltage drop and good speed - but that was in 1977.
If you are using a Darlington configuration in your VAS, again, a fast recovery diode from output transistor to base of the buffer will do the same trick. Believe me, the boys from Marantz were not stupid when they started using it. I believe this is one of the key factors why their amps usually sound very well defined, even if the miss out on overall tonality on occasion.
Lastly, I believe all this is much more easily done using as stiff as possible power supplies. I know the industry generally loves loose PSUs because they cost less and allow for unnaturally high voltages to iron out the specs, but if there's one thing I fear it's an amp with sliding supplies. My way in this case means that you define how much power you want at the output and into which impedances, and then dimension your PSU to cater for your worst case scenario. Heck, I KNOW that's expensive, that means something like using a 500VA toroid for a nominally 100W/8 Ohms power amp - per side. Any amp will sound better when fed power like that.
The Yankees on this forum should be comfy with that idea, that's how Detroit used to and still does build its car engines, big blocks with torque to die for, all the while chanting the mantra "There's no replacement for displacement". Well, here, the horse power is volts, and the torque is the current. Now, remembver the explanation what's horse power and what's torque, also originating from USA I am told - horsepower (volts) tell you how fast (high in power into 8 Ohms) you can hit a wall, and torque (current) tells you how deep (to what impedance down) you'll get into the wall.
Sorry for the picturesque analogies, but I find them fascinating and much clearer than a bunch of formulas. And anyway, Nige and I do it all the time.
JCX, what you are saying sounds to me like saying, hey, the speed limit is 70 mph, so that's all we really need. While that may be absuloutely so, I think we can agree that if a car is capable of doing say 120 mhp, it will handle the allowed 70 mph much more easily and leisurely than if it can do just 80 mhp.
Same thing with audio amps. There is some doubt that bandwidth limiting at very low levels will impair our preception of the high frequencies; I don't know if that is really so or not, but I do know that the vast majority of Bitish designed and manufactured amps do just that, and to this day, I haven't heard one which did not sound at least a little shut in its treble range to me. They may otherwise be even outstanding, but I will eventually find myself thinking that when the drummer hits the brass full force, it does not bite my ears as it would in real life.
Obviously, this has much to do with out personal perception of sound, hearinh, loudspeakers and rooms, which is wha I stress "to me".
All of your points are valid topics which must be addressed.
I cannot speak for others, but here's how I do it.
I have no particular target frequency I aim for in my open loop response, except that my feeling is that it must hit 50 kHz at ts -3 dB point, ref, nominal output at 1 kHz. My reasining is simple enough - if the amp can reproduce the highest required frequency in an auduio system, assumed to be 20 kHz, in open loop conditions, then it does not have to wait for the feedback signal for it, as it would if its open loop -3 dB point was 1 kHz or less. Consequently, I expect its transient response to be a non-issue, however only if we work with a fair safety margin (where "fair" is a factor of at least 50% above the absolute minimum).
As Tom put it, we can park that frequency point almost anywhere we choose to. While technically true, in practical terms it's not quite so simple without making some serious compromises with stability, and that's the one thing I will not play games with.
To prove that point, I usually get the amp to hit something like 100 kHz under oopen loop conditions. Then I start making it work into real world loads, as I have a small collection of crossovers and drivers from real world speakers (e.g. my own AR94, several Infinity models, a few JBL models, etc). They are far from being perfect, however they beat the hell out of a lab resistor usually used for simulation and free me from dreaming up combinations every which way.
Then I start to compensate, including Miller compensation. I believe in a distirbuted compensation model, meaning that I never ever eave it all to just one or two Miller caps. Rather, I compensate each stage as much as I can. As a result, the Miller cap will always have a smaller value than it otherwise would, although it will still be the first limiting factor on the list, but rather than use values like 33,47 or even 100 pF, it will likely be between 15 and 22 pF. Because it is in real life also influenced by parasitic capacitance of the PCB, its ultimate value will be determined in situ, ie on the board.
Eventually, this will lead to a closed loop response of at least 350 kHz, but more often over 500 kHz, with more global NFB (with me that's like 26 dB), this might well approach 1 MHz. It is this value which will determine the ACTUAL real world voltage slew rate. Then I install a low pass filter at the input, with -3 dB at 200 kHz; after experimenting quite a bit with that, I realized that reVox and Sony are quite right in placing it there (or at 150 kHz, in some Sony models), as those are points which do not cause excessive phase shift at 20 kHz, and anyway, they are first order filters.
Hence, I have two voltage slew rates: one is the internal value, what the amp can do on its own, and the other is the effective, real life value, which is limited by the 200 kHz mark. The fact that amp will do like 3 or 4 TIMES the imiting value of the input filter eliminates oddities, possible RF breaktrhough and whatnot junk which a not so good CD player might spew out.
Now we come to the other side of the coin of wide bandwidth design. While its rise time can be surprisingly short, the not at all unusual even is the square wave overshoot, with very possible ringing, and of course, the settling time. The ringing is usually all but eliminated by the input filter, or at the very least, it's reduced to tlearble levels. If not, I can always adjust the local compensation to obtain a reasonable operformance; a wide bandwidth amp will ALWAYS have some overshoot, but it should not be large and should have no ringing after it, much less - God forbid! - signs of protracted ringing even at low level. It takes patience and work, but can be done.
Settling time can be a bitch, though. I never had any particular problems with it because early on I got lucky and met that German made amp (LAS, I posted its schematic here at least twice), which had the problem solved. A simple collector to base fast diode on the "lower" cascode transistor will shorten the settling time by a factor of 10:1, which is, trust me on this, damn significant. Fortunately, these days, we have a selection of very fast diodes to choose from, in the original LAS amp, its makers used a germanium AA132 diode for its low 0.2V voltage drop and good speed - but that was in 1977.
If you are using a Darlington configuration in your VAS, again, a fast recovery diode from output transistor to base of the buffer will do the same trick. Believe me, the boys from Marantz were not stupid when they started using it. I believe this is one of the key factors why their amps usually sound very well defined, even if the miss out on overall tonality on occasion.
Lastly, I believe all this is much more easily done using as stiff as possible power supplies. I know the industry generally loves loose PSUs because they cost less and allow for unnaturally high voltages to iron out the specs, but if there's one thing I fear it's an amp with sliding supplies. My way in this case means that you define how much power you want at the output and into which impedances, and then dimension your PSU to cater for your worst case scenario. Heck, I KNOW that's expensive, that means something like using a 500VA toroid for a nominally 100W/8 Ohms power amp - per side. Any amp will sound better when fed power like that.
The Yankees on this forum should be comfy with that idea, that's how Detroit used to and still does build its car engines, big blocks with torque to die for, all the while chanting the mantra "There's no replacement for displacement". Well, here, the horse power is volts, and the torque is the current. Now, remembver the explanation what's horse power and what's torque, also originating from USA I am told - horsepower (volts) tell you how fast (high in power into 8 Ohms) you can hit a wall, and torque (current) tells you how deep (to what impedance down) you'll get into the wall.
Sorry for the picturesque analogies, but I find them fascinating and much clearer than a bunch of formulas. And anyway, Nige and I do it all the time.
JCX, what you are saying sounds to me like saying, hey, the speed limit is 70 mph, so that's all we really need. While that may be absuloutely so, I think we can agree that if a car is capable of doing say 120 mhp, it will handle the allowed 70 mph much more easily and leisurely than if it can do just 80 mhp.
Same thing with audio amps. There is some doubt that bandwidth limiting at very low levels will impair our preception of the high frequencies; I don't know if that is really so or not, but I do know that the vast majority of Bitish designed and manufactured amps do just that, and to this day, I haven't heard one which did not sound at least a little shut in its treble range to me. They may otherwise be even outstanding, but I will eventually find myself thinking that when the drummer hits the brass full force, it does not bite my ears as it would in real life.
Obviously, this has much to do with out personal perception of sound, hearinh, loudspeakers and rooms, which is wha I stress "to me".
Even though this seems a very reasonable idea, I find the greatest gains are achieved when I deliberately use a system at rated capabilities. For a long time now I have always pushed an audio setup to run hard up against straightforward physical constraints - these are, for cheap systems speaker related such as carcase rattling, power supply limits for delivering continuous strong bass, overheating cutouts, or inbuilt protection circuitry injecting audible noises - classic clipping has not really been a factor.
The advantages are that all aspects of the chain are thoroughly conditioned, and it teases out remaining weaknesses; and, when you turn up the wick for those special moments, when you're in the mood, there are no unexpected surprises - there's no running across to turn down the volume ...
- Status
- Not open for further replies.