Hi Traderbam,
'Audio' longevity has allowed us to pull thoughts, hypotheses, topological ideas and amplifier examinations together.
Several subscribers now have my circuit; if they are pleased with the results - then I will be pleased too.
Designs must be satisfactory to stand the test of time; this has nothing to do with my credibility.
I am not seeking and do not seek credibility, but everything I mention now, has its roots from my experience before I had a scope, let alone a simulator.
It is the circuits, the investigation methods and the interpretation of modern examination results that I am challenging; here with regard to first cycle treble distortion which can be caused by input filters as well as by the filter-like behavior of an amplifier itself.
In other words - is there a circuit better than mine that LineSource and I myself should actually be building ?
Hi Andy,
Good to have more technical thought. Thank you for posting.
Amplifiers behave like filters too, due to their propagation delay.
I entirely agree with your points (1) and (2).
The amplitude distortion arises at the leading edge of the sinewave, but everyone seems to ignore this. Why ?
It is not constant either.
The error potential wrt the original waveform increases with frequency and can be clearly audible as a lack of transient accuracy.
It is not just a SPICE artifact !
SPICE is not wrong !
It is the belief that steady state measurements are most important, which came out of early solid state fundamental nulling examinations, that is wrong.
First cycle distortion has always happened.
We actually hear that waveform loss due to exponential build up of the transient response as a loss of treble clarity.
It is an entirely natural and unavoidable occurrance.
It is audibly and scope/simulator demonstrable, and the distortion is properly calculable.
Unfortunately some simulators cannot pre-examine the circuit to set up the steady zero state conditions for the start of the first cycle calculations to have their own zero starting reference point.
They run lots of cycles to establish the steady state and then read the last cycle.
Folks who use such simulators are being misled by their software in a way they cannot understand because they don't know they lack the direct experience necessary to interpret the results.
I'm sorry to say it, but the usage of THD measuring equipment has misled the majority into believing that ultra-low THD is the ultimate goal. It is not, first cycle distortion is much more important for clean treble reproduction.
(If the past comments were anything to go by, think I'll go and find myself a decent fire-proof suit and underground bunker to hide in.)
Cheers ................ Graham.
'Audio' longevity has allowed us to pull thoughts, hypotheses, topological ideas and amplifier examinations together.
Several subscribers now have my circuit; if they are pleased with the results - then I will be pleased too.
Designs must be satisfactory to stand the test of time; this has nothing to do with my credibility.
I am not seeking and do not seek credibility, but everything I mention now, has its roots from my experience before I had a scope, let alone a simulator.
It is the circuits, the investigation methods and the interpretation of modern examination results that I am challenging; here with regard to first cycle treble distortion which can be caused by input filters as well as by the filter-like behavior of an amplifier itself.
In other words - is there a circuit better than mine that LineSource and I myself should actually be building ?
Hi Andy,
Good to have more technical thought. Thank you for posting.
Amplifiers behave like filters too, due to their propagation delay.
I entirely agree with your points (1) and (2).
The amplitude distortion arises at the leading edge of the sinewave, but everyone seems to ignore this. Why ?
It is not constant either.
The error potential wrt the original waveform increases with frequency and can be clearly audible as a lack of transient accuracy.
It is not just a SPICE artifact !
SPICE is not wrong !
It is the belief that steady state measurements are most important, which came out of early solid state fundamental nulling examinations, that is wrong.
First cycle distortion has always happened.
We actually hear that waveform loss due to exponential build up of the transient response as a loss of treble clarity.
It is an entirely natural and unavoidable occurrance.
It is audibly and scope/simulator demonstrable, and the distortion is properly calculable.
Unfortunately some simulators cannot pre-examine the circuit to set up the steady zero state conditions for the start of the first cycle calculations to have their own zero starting reference point.
They run lots of cycles to establish the steady state and then read the last cycle.
Folks who use such simulators are being misled by their software in a way they cannot understand because they don't know they lack the direct experience necessary to interpret the results.
I'm sorry to say it, but the usage of THD measuring equipment has misled the majority into believing that ultra-low THD is the ultimate goal. It is not, first cycle distortion is much more important for clean treble reproduction.
(If the past comments were anything to go by, think I'll go and find myself a decent fire-proof suit and underground bunker to hide in.)
Cheers ................ Graham.
Graham Maynard said:
it may have but it certainly isn't reflected in your posts so far.
Graham Maynard said:
you should first prove that it is clearly audible (I have not seen any such mentioning from reputable sources).
2ndly, what jcx's "experiment" proves is that because spice is dumb, it will actually measure THD when there is none (in jcx's example, an rc network (something that cannot generate thd) is simulated with thd figures.
Graham Maynard said:
did you mean "you actually hear"? I certainly didn't hear it.
Graham Maynard said:
Simple answer to this is I don't know and don't have the
gall to suggest something else may be better without
having some personal experience with it.
GM's circuit takes a particular design approach, that I'm
sure all JLH fans would approve of and GM has developed
the circuit within that approach.
At the moment LS does not have a better option.
Different design approaches would suggest other solutions :
e.g. 3 stage or seperate rails with an Op-Amp and FET's.
But at the moment there is no other circuit in the running.
My experience of British class A amplifiers built along these
lines (e.g. Sudgen) is that they sound very good.
I wonder if there's some mileage in converted to a single rail
design because for treble unit the coupling capacitor could be
a high quality film type and this would protect the (presumably)
expensive treble ribbon under fault conditions and also simplify
design of a regulated or capacitive multiplier power supply
for the amplifier.
sreten.
LineSource asked:
He later gave us this:
Graham asks me this:
My answer:
Looking at the Visaton driver it is very clearly said, that the impedance is flat. Unfortunately LineSource also tells us, that "The inductance is very small". That's why I wish to see the impedance curve.
Because my thought is: Why not build an amplifier that gives out current instead og voltage?
He later gave us this:
Graham asks me this:
My answer:
Looking at the Visaton driver it is very clearly said, that the impedance is flat. Unfortunately LineSource also tells us, that "The inductance is very small". That's why I wish to see the impedance curve.
Because my thought is: Why not build an amplifier that gives out current instead og voltage?
Graham Maynard said:
I've re-read the thread, and I'm still not clear what you mean by 'first cycle distortion'. Is this a new distortion mechanism, or simply a different way of measuring?
The basic reasoning behind steady-state THD measurement is surely that any plausible non-linear behaviour by the circuit will necessarily create harmonic artefacts at some level when fed with a continuous sinewave input. You may need to measure at very high or very low frequencies, at high or low signal levels, but sooner or later it will show up in the THD figure.
Cheers
IH
AndyC's comments and Graham’s own equation of filter and amp performance under his 1st cycle distortion confirm my first intuition that that “1st cycle distortion” is simply a measure of bandwidth
The specific formulation of 10 KHz 1st cycle distortion of < 0.01% equates to a 1 MHz open loop bandwidth requirement (at least for a 1st order roll off)
While I did use “ridiculous” in characterizing the implied high bandwidth I am not in fact directly critical of such a requirement as being a necessary condition on loop gain-bandwidth for achieving low distortion with global negative feedback
However after designing a feedback stage that fast I would then filter the input to prevent MHz signals of even mV levels from reaching the input stage, particularly with undegenerated bipolar input transistors in a low noise design
Ian,
I believe that significant distortion can “hide” from single tone distortion measurements, a partial solution is to use multitone tests to characterize the distortion; in general nth order nonlinearities need to be probed with n tones swept over the n-dimension “box” formed by each tone being considered an independent axis – this allows the complete characterization of the Volterra kernels that describe the nonlinear system – up to the nth order
The specific formulation of 10 KHz 1st cycle distortion of < 0.01% equates to a 1 MHz open loop bandwidth requirement (at least for a 1st order roll off)
While I did use “ridiculous” in characterizing the implied high bandwidth I am not in fact directly critical of such a requirement as being a necessary condition on loop gain-bandwidth for achieving low distortion with global negative feedback
However after designing a feedback stage that fast I would then filter the input to prevent MHz signals of even mV levels from reaching the input stage, particularly with undegenerated bipolar input transistors in a low noise design
Ian,
I believe that significant distortion can “hide” from single tone distortion measurements, a partial solution is to use multitone tests to characterize the distortion; in general nth order nonlinearities need to be probed with n tones swept over the n-dimension “box” formed by each tone being considered an independent axis – this allows the complete characterization of the Volterra kernels that describe the nonlinear system – up to the nth order
I have proceeded with a somewhat more conventional grounded supply variation on my original op amp/mosfet circuit (instead of doing my tax returns)
U2,3 are IRF logic-level fets rated 98W by IRC – meaning that they can probably handle ~30 W power dissipation without water cooling
Q1 with R1,2 forms a feedback current servo controlling U2 to ~4A Iq, taking the output from the center of the R1,2 current sense gives a push-pull class A output (Q4,5,8 provide 50 mA bias to Q1, the parallel Q4&8 are necessary for worst case power dissipation with t0-92s)
U3 is a common source output fet with Q2 providing low Z gate drive
The LT1115 is a low noise, high bandwidth op amp that provides large gain, the compensation is somewhat complicated by the additional gain and inversion from U3
The most unusual feature of this circuit is the bais for Q2 which partially compensates the huge gate C swing of U3 at low Vds, Q6,7 feedback current sink is set at 50mA, at higher U3_Vds R13,Q3 shunts some of the current, reducing the bias and increasing the incremental R of Q2 – increasing damping as U3 gate C decreases
The circuit’s compensation is trimmed to reasonable margins that still give enough audio frequency gain for most distortion components to be below –120dB, but the whole effort could in fact be a pipe dream given the likely accuracy of the subcircuit model of the IRL3103 with respect to analog behavior
Less aggressive compensation could likely tame real mosfets with some distortion penalty
U2,3 are IRF logic-level fets rated 98W by IRC – meaning that they can probably handle ~30 W power dissipation without water cooling
Q1 with R1,2 forms a feedback current servo controlling U2 to ~4A Iq, taking the output from the center of the R1,2 current sense gives a push-pull class A output (Q4,5,8 provide 50 mA bias to Q1, the parallel Q4&8 are necessary for worst case power dissipation with t0-92s)
U3 is a common source output fet with Q2 providing low Z gate drive
The LT1115 is a low noise, high bandwidth op amp that provides large gain, the compensation is somewhat complicated by the additional gain and inversion from U3
The most unusual feature of this circuit is the bais for Q2 which partially compensates the huge gate C swing of U3 at low Vds, Q6,7 feedback current sink is set at 50mA, at higher U3_Vds R13,Q3 shunts some of the current, reducing the bias and increasing the incremental R of Q2 – increasing damping as U3 gate C decreases
The circuit’s compensation is trimmed to reasonable margins that still give enough audio frequency gain for most distortion components to be below –120dB, but the whole effort could in fact be a pipe dream given the likely accuracy of the subcircuit model of the IRL3103 with respect to analog behavior
Less aggressive compensation could likely tame real mosfets with some distortion penalty
Attachments
Graham Maynard said:
Actually, I agree with your last sentence above. One weakness I see in the "pure objectivist" argument is too much dependence on frequency domain and steady state arguments and not enough attention payed to transient behavior. But that's a flame war for another thread
.Let's first look at why this "early cycle distortion" goes to zero as the bandwidth of the circuit goes to infinity for a first order filter. For such a filter, we can compute the exact response to the sine wave pulsed on at t=0 as I described in my previuos post. The transfer function approach taken assumes zero initial energy in the system, so if the filter had a series R and shunt C, the voltage across that C is assumed to be 0 at t=0. Thus the output voltage at t=0 is 0. You get the sum of an exponential and a phase shifted sine as the output. The amplitude and phase of that sine is exactly what you'd get with a steady-state analysis (AC analysis in SPICE). At t=0, the amplitude of the exponential and the initial value of the phase shifted sine must add up to 0, since that's the initial output voltage. So that says that as the phase shift of the sine approaches 0 (increasing bandwidth), the amplitude of the exponential must also approach 0 since the initial value of the phase shifted sine must add with the initial value of the exponential to produce 0. The reason the distortion approaches 0 here is not so much because the transient is decaying quickly (though I'm sure this helps), but because the amplitude of the transient component is approaching 0 as the bandwidth gets large due to the requirements on initial conditions.
So what happens when the initial voltage on the capacitor is not zero? The amplitude of the transient will no longer approach 0 as the bandwidth gets large, since the initial output voltage at t=0 is no longer 0. Thus the "early cycle distortion" will increase. What would happen if SPICE used a cosine instead of a sine? Same thing, the "early cycle distortion" would increase because the amplitude of the transient component must increase from 0 to match the initial output voltage condition.
If one were to do a spectral analysis of this "early cycle distortion" in SPICE, one would see harmonics of the fundamental present with a simple single pole filter. Yet when doing the Laplace transform analysis, which is an exact solution in the case of a linear circuit, no harmonics will be present at all. The reason the harmonics show up in the SPICE analysis is the properties of the DFT that were discussed in the part of the thread that is now split off. Though I suppose this metric might be useful for indicating how much of the transient solution is sneaking in to the output, it's misleading in the sense that it predicts harmonics that aren't there. In my view, its a misapplication of the DFT.
Hi sreten,
I would agree with your suggestion for series 'C' between amplifier and ribbon.
____________________________________________________
Hi boholm,
Current amp instead of voltage ?
Good question, but can a current amp be stable at supersonic frequencies when the cables have inductance ?
My own thoughts are that the amp should be right behind the tweeter anyway, not 12ft away.
____________________________________________________
Hi Ian,
Yes most of us have measured THD, but most folks measure the distortion that arises after the first cycle has established a group delayed sinusoidal steady state.
With linear amplifier operation at high frequencies the first cycle is always the most distorted.
____________________________________________________
Hi jcx,
You found that 0.01% first cycle distortion equates to approximately 1MHz of bandwidth.
I have never checked, but the original 1969 JLH class-A did have a 500kHz bandwidth, which they did not realise at the outset because this was higher than the measuring equipment itself.
That's a mighty good first cycle distortion figure for Mosfets !
Nice to see something different.
I note your use of (borrowing your word) agressive pull-down of gate voltages. It is inverting; could the other 500 ohm R3 be used as input ?
If you have 0.002%, is that circa 5Mhz on the simulator ? I wonder what the stability margin would be like out there ?
____________________________________________________
Hi Andy,
Nice discussion.
I follow all that, and we both understand the nature of the exponential waveform start-up.
I note you have used the term 'early cycle distortion', but there is no way of quantifying that wrt the whole first cycle which is being examined.
A fourier examination will do this for any sinewave, whether first or Nth.
Yes, the higher the bandwidth the lower the apparent 1st cycle distortion, so how far do we go ? Big question.
I have been disheartened at hearing the dulled (distorted)transient (treble) responses from designers who have told us that their amplifiers cannot possibly alter the waveforms they amplify because the THD figure is say 0.001%.
We have to start somewhere in comparing 'early cycle distortion' between designs, and the only way currently available to us is by measuring distortion for the entire first cycle from t=0. This will cover everything along the signal path between source and speaker, and 10kHz is a decent reference.
I sent a simulation of the distortion waveform into EW a couple of years back, but I'm blowed if I can find it tonight. It was complex to derive and took the form of one falling exponential wrt input up to approx the group delay period, followed by a rising exponential charge balance up to approx 4.5x group delay period. These two curves out of the subsequent first cycle cannot fail to generate low value harmonic subtraction. Can we hear harmonics of 10kHz ? - no; but we can hear the effect upon high frequency transients, as I have checked through introducing buffered filters in real life.
Should appreciate your further thoughts.
Cheers .............. Graham.
I would agree with your suggestion for series 'C' between amplifier and ribbon.
____________________________________________________
Hi boholm,
Current amp instead of voltage ?
Good question, but can a current amp be stable at supersonic frequencies when the cables have inductance ?
My own thoughts are that the amp should be right behind the tweeter anyway, not 12ft away.
____________________________________________________
Hi Ian,
Yes most of us have measured THD, but most folks measure the distortion that arises after the first cycle has established a group delayed sinusoidal steady state.
With linear amplifier operation at high frequencies the first cycle is always the most distorted.
____________________________________________________
Hi jcx,
You found that 0.01% first cycle distortion equates to approximately 1MHz of bandwidth.
I have never checked, but the original 1969 JLH class-A did have a 500kHz bandwidth, which they did not realise at the outset because this was higher than the measuring equipment itself.
That's a mighty good first cycle distortion figure for Mosfets !
Nice to see something different.
I note your use of (borrowing your word) agressive pull-down of gate voltages. It is inverting; could the other 500 ohm R3 be used as input ?
If you have 0.002%, is that circa 5Mhz on the simulator ? I wonder what the stability margin would be like out there ?
____________________________________________________
Hi Andy,
Nice discussion.
I follow all that, and we both understand the nature of the exponential waveform start-up.
I note you have used the term 'early cycle distortion', but there is no way of quantifying that wrt the whole first cycle which is being examined.
A fourier examination will do this for any sinewave, whether first or Nth.
Yes, the higher the bandwidth the lower the apparent 1st cycle distortion, so how far do we go ? Big question.
I have been disheartened at hearing the dulled (distorted)transient (treble) responses from designers who have told us that their amplifiers cannot possibly alter the waveforms they amplify because the THD figure is say 0.001%.
We have to start somewhere in comparing 'early cycle distortion' between designs, and the only way currently available to us is by measuring distortion for the entire first cycle from t=0. This will cover everything along the signal path between source and speaker, and 10kHz is a decent reference.
I sent a simulation of the distortion waveform into EW a couple of years back, but I'm blowed if I can find it tonight. It was complex to derive and took the form of one falling exponential wrt input up to approx the group delay period, followed by a rising exponential charge balance up to approx 4.5x group delay period. These two curves out of the subsequent first cycle cannot fail to generate low value harmonic subtraction. Can we hear harmonics of 10kHz ? - no; but we can hear the effect upon high frequency transients, as I have checked through introducing buffered filters in real life.
Should appreciate your further thoughts.
Cheers .............. Graham.
Graham Maynard said:
Okay, I think I see where you're coming from now. BTW, when I said "early cycle distortion", I really meant "the first N cycles of the waveform" as opposed to the normal SPICE approach of using the last N cycles. I shouldn't have used this pet phrase without explanation. I wasn't trying to suggest using some fraction of a cycle. I see that you're taking N = 1 and just using the first cycle. If I understand correctly, you're looking for a single number that represents the "goodness" of the transient response of a circuit, and the harmonic distortion of the periodic extension of the first cycle of the 10 kHz pulsed sine is what you're using to get that single number.
I can see in the case of a single pole low pass filter that the residual would look like a high-pass filtered square wave. Since there's a decaying exponential summed with a phase shifted sinusoid, one would expect to see this exponential at the beginning of the first cycle of the residual. At the beginning of the second cycle of the periodic extension, one would expect a discontinuity of roughly the amplitude of the exponential. So the appearance of the residual would be very much like a high-pass filtered square wave for this reason.
I think some of the confusion that's arisen relates to the potential for ambiguity in interpreting what you're doing with the FFT. Since what you're measuring is a kind of harmonic distortion, the reader is likely to assume that you're strictly trying to measure the nonlinearity of the circuit under transient conditions. However, it appears that the dominant source of the "distortion" can be simply attributed to the purely linear effects of finite rise and settling time, as well as the phase shift of the frequency response. It's helpful to clarify these issues early in the game.
I'm afraid I've strayed off topic, so I'll limit my comments to the above.
While I have explored R3 as a non-inverting input the main reason for going inverting is that R3 must be well grounded for MHz AC signals, not something to rely on at an input
R6 is shunted by the noise gain R9,C3 and is less sensitive to high freq input impedance of the driving circuit – which I would still like to be very low at MHz, perhaps a ef buffer inside the op amp feedback loop of the (assumed) preceding active filter stage – I wouldn’t expect 500 Ohm input Z to go over well in a general purpose audio amplifier
Reversing driver wires is easy for biamping diyers if the inversion isn’t canceled elsewhere
Stability was checked (and compensation adjusted) by looking at 1 Vpp steps at 0 and +/-5 V out, leading to the realization of the need for some compensation of U3 gate C variation with bias point
.ac analysis with V7 gives the loop transmission, evaluating gain and phase margins at the differing bias points gives >40 degrees phase margin at 5 MHz and 10 dB min gain margin at 23 MHz
if I even halfway trusted the mosfet model I would suggest that a skilled builder might have a chance at getting the circuit to work, but I would be surprised if further serious reengineering wasn't necessary too
R6 is shunted by the noise gain R9,C3 and is less sensitive to high freq input impedance of the driving circuit – which I would still like to be very low at MHz, perhaps a ef buffer inside the op amp feedback loop of the (assumed) preceding active filter stage – I wouldn’t expect 500 Ohm input Z to go over well in a general purpose audio amplifier
Reversing driver wires is easy for biamping diyers if the inversion isn’t canceled elsewhere
Stability was checked (and compensation adjusted) by looking at 1 Vpp steps at 0 and +/-5 V out, leading to the realization of the need for some compensation of U3 gate C variation with bias point
.ac analysis with V7 gives the loop transmission, evaluating gain and phase margins at the differing bias points gives >40 degrees phase margin at 5 MHz and 10 dB min gain margin at 23 MHz
if I even halfway trusted the mosfet model I would suggest that a skilled builder might have a chance at getting the circuit to work, but I would be surprised if further serious reengineering wasn't necessary too
Actually MOSFETs seem an ideal solution. We're talking 6A RMS through 1 ohm, or 6v RMS, however you want to look at it. That's 8.4V peak to peak. Using N-channel vertical MOSFET devices having .02 ohm on resistance, there is negligible drop across the MOSFETS. A supply of +- 9V @10A would suffice for the MOSFET output stage, versus perhaps +-11 V for bipolar.sreten said:
Gate drive voltage would have to be higher than the output stage voltage. Let's say it would be 6-10 volts higher, but at a few milliamps of current, so that is hardly a show stopper. This is done in many current MOSFET amp designs, including some I have designed and built. To be fair, you could do the same thing with bipolars in EF configuration (as in the famous JBL T circuit from 1967), but you would have a higher Vsat drop than with MOSFETs, typically 2v as opposed to 0.12v for a single MOSFET.sreten said:
I'm surprised to hear that. The MOSFET dissipation would be 20% lower, because you can use lower rails than on a bipolar design, assuming would use the same standing current for either, whatever Class A standing current you want to use. For instance, using the +- 9v rails and 5A standing current, you dissipate 90 watts steady state (45 watts per device), whereas the bipolar design using +-11V rails would dissipate 110 watts. In my personal opinion, MOSFETs would be far more reliable in this case, since they do not have SOA problems.sreten said:BJT's are the only sensible option here, a Mosfet output
stage would have at least 3 x the static dissapation,
i.e. > 200w for 25W output.
This is probably why so many car amps use MOSFETs. The loads are typically ridiculously low impedance, 1-2 ohms from paralleled woofers. Incidentally that is a logical proof by contradiction that MOSFETs are well suited to this application.
Either type of output device could be made to work well in this application. MOSFETs would cost more but be more reliable. It's the usual tradeoff of conflicting values, designer's judgement call in my opinion.
slowhands said:
The "on resistance" value is only pertinent in switching applications. In linear applications, it really doesn't mean much (except in a very indirect secondary way). What's relevant here is transconductance. In this case (3.5 amps of idle current), a bipolar output stage will have a transconductance of around 150 or so. This gives an output impedance of around 10 milliohms before any feedback is applied. This cannot be matched by any practical MOSFET output stage.
So for this application (very low impedance loads), I would concur with Graham that bipolars would be better suited. (My perspective is from someone who has designed several commercially successful MOSFET amps, so I am not biased.)
As far as car amps go, the last time I checked (and it has been a number of years so I could be wrong), they all used bipolars for the audio circuitry and MOSFETs only for the switching power supplies. They always called these products "MOSFET" amps purely for marketing reasons.
Nelson brings up a good point. The output impedance of a bipolar output stage will also have an additional impedance equal to the driver stage output impedance divided by the beta of the output transistors.
In my previous post I recommended a triple emitter follower (JBL T-Circuit developed by Bart Locanthi in the mid-sixties, and as used in the Leach amplifier). If you run the drivers at a reasonable current (25 - 50 mA) then the contribution from this source is quite small -- another 10 milliohms or less.
My original point remains unchanged. The output impedance of a MOSFET output stage under equivalent operating conditions (3.5 amps idle current) will be at least an order of magnitude higher than for bipolars.
In my previous post I recommended a triple emitter follower (JBL T-Circuit developed by Bart Locanthi in the mid-sixties, and as used in the Leach amplifier). If you run the drivers at a reasonable current (25 - 50 mA) then the contribution from this source is quite small -- another 10 milliohms or less.
My original point remains unchanged. The output impedance of a MOSFET output stage under equivalent operating conditions (3.5 amps idle current) will be at least an order of magnitude higher than for bipolars.
My post was a reaction to the comment: "BJT's are the only sensible option here, a Mosfet output stage would have at least 3 x the static dissapation". That just seemed incorrect to me, and I presented analysis to show that MOSFET outputs could have lower static dissipation.Charles Hansen said:
Perhaps you missed my point: the supply voltage will directly affect power dissipation in the output stage in Class A, and supply voltage can be lower for a MOSFET amp - due to low on IR drop at clipping as compared to the Vsat of bipolars.
Charles Hansen said:
The original poster asked: 'What is the best sounding circuit topology to drive a 1 ohm pure resistive load with 25-35 watts of class A power? The "first watt" is the most important.'
This is probably relevant for many amp designs, because many loudspeakers dip down to the 1-2 ohm range near resonance. So it's a great question for this forum. But "Best sounding" means different things to different people. For instance, I have acute hearing at high frequencies and training as a musician. I want to hear music as close as possible to live performance quality -- and I can hear the difference!
Minimal distortion at high frequencies and excellent transient behavior seem to be some objective criteria that mark the amps and speakers that I most enjoy listening to. Every watt is important. I run 500 watt amps at 0.1 watt so the 500 watt peaks will be clean. But that is me.
We as designers should determine what criteria the listener most values, and deliver that. At this point, I would ask the original poster, Linestage, for the list of qualities, with weightings, that determine best sound in his opinion.
The triple emiiter follower is good idea. For the first two stages You can use higher voltage, say +/-20V, which is also good for the input and VAS. And for the output devices go down to +/-11V.
Right now, I'm working on hibryd mosfet class A amplifier, This will be bridged, to get higher bias current over the output devices. I found that vertical mosfets (I use IRF520, and 530, from ST) sound better if the bias is higher. 3-4A is necessary to get good sound without nfb.
I plan to use some +/-8V, and 4A.
sajti
Right now, I'm working on hibryd mosfet class A amplifier, This will be bridged, to get higher bias current over the output devices. I found that vertical mosfets (I use IRF520, and 530, from ST) sound better if the bias is higher. 3-4A is necessary to get good sound without nfb.
I plan to use some +/-8V, and 4A.
sajti
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.