So what we see here are two bipolar/JFET ~5mA I sources (or sinks). Both have an output resistance of 394Mohm. The output capacitance out to about 1-2 MHz of the one with gate to emitter feedback is about 58fF, of the other 1.7pF out to >100MHz.
When we contrive to examine the distortion associated potentially with the voltage-variable output capacitance of each, loaded with a "perfect" current source and a 1Mohm resistor, and drive with a +/- 1uA current signal, thus getting a +/-1V signal at the drains, we see THD out to the tenth harmonic of 2544ppm for the left hand one, 44.6ppm for the right. Second harmonic dominates for both, at -52.5dBr and -86.9dBr respectively. Third is -81dBr for the left circuit and -121dBr for the right. Of course these are sweet Spice lies, but maybe not too ridiculously so.
In many practical applications this performance may be irrelevant --- what can we load such a circuit with that doesn't itself contribute a boatload more of variable C etc.? Well, good question
BTW the "2SC1815A1" is roughly comparable to a 2SC1815BL (the highest-beta sort).
Modeled the right hand version. (FET on top) Yup, quite a bit better. If one was using two parts anyway, might as well. The drop across it is way too high for use in the VAS though. Must be why I see the current mirrors used there.
Wahab, your view of THD & IM levels is just fine, go for it. Personally, I have experienced far too many cases in which amps with LOVELY specs sounded poor, and I own an amp with such specs at a level where you would probably demand its quick demise, yet it manages to sound surprisingly good. I refe to Harman /Kardon 6550 integrated amp, which I purchased in December 1993.
It's spec sheet says it has a THD factor of 0.3%. Not a typo, no zeroes lost in translation, really one third of one percent. Then it adds insult to injury and delivers only 50/70 wrms into 8/4 Ohms (I guess Wayne is gathering a posse by now), but that's all one can REASONABLY expect from a SEPP designs, using a single pair of Toshiba's 2SC3281/2SA1302.
Yet, it manages to convey a lot more "air" and "space" than quite a few nominally "better" amps.
To make matters worse, they use just 17 dB of global NFB, sending the water down to Thorsen's mill.
Surely that 0.3% THD is specified at 20khz and full power.
Anyway , sounding good and sounding accurate are two different things.
Rubbish..? then you give the same answer .....
Errr isn't that what i implied .....
Precisely , they can't have the same precise TF, this is why they all sound different , even when having the same levels of THD..
No Boss ....
Should our speakers have bandwidths into the MHZ ? How does one get one without the other , they are not mutually exclusive, or did i not get that memo...?
Low THD , yes , but thanks to NFB also low IMD , high PSRR and
damping factor , all parameters that you re indirectly downplaying....
...also, harsh coughing when some sudden peak hits clipping...
edit: or, it's speed pushes some stage in the loop out of the regime that corresponds to the total 0.3% of HD...
Last edited:
Nothing of the sort if minimal care is taken , as non saturating VAS
and as low number of stages as possible.
Anyway , my amp is set to about 2 x 16W rms max power
and it never saturate.
If yours does , then either it has too low power or that the end
of the chain , that is either the speakers or your ears or both ,
have too low sensitivity...
Nothing of the sort if minimal care is taken , as non saturating VAS
and as low number of stages as possible.
Anyway , my amp is set to about 2 x 16W rms max power
and it never saturate.
If yours does , then either it has too low power or that the end
of the chain , that is either the speakers or your ears or both ,
have too low sensitivity...
No, my Pyramids use nice fast smooth tubes, solid like rocks SS voltage regulators, nested feedbacks, and optical compressors - limiters that reacts on 80% of peak power that is 100WPC.
One more evening at sim. The jfet/bjt as mentioned above, when biased with two green led's at 1.5mA, the VAS using a current mirror ccs at 3.5 ma, outputs running 110 mA.
Greens seemed to simulate better than reds. At the very low current, .5mA, is the difference in thermal tracking valid?
Remembering the Cdom was moved to a Miller comp on the VAS. Comes out with the best predicted THD at .004% full power. The original modeled at .013% Changing the Cdom increased the high order distortion, as expected, so without the ccs changes, the model was reduced to .032%. In the sim, when I upped the VAS current by 3, it reduced the distortion as it could better drive the dominant pole loading the output of it. Of course, I could never get it to be stable like that. The biggest bang for the buck is adding the current mirror to the VAS. Improving the IPS was secondary. Again, parts picked because they were in Spice.
To do this removed the various diode protection he had set up. The chance that it would do ugly things if over driven is increased. It maintains full swing of the IPS and VAS.
Of course the spec sheet says .002% full power at 1K, where the simulation is only saying .013% with perfect parts. That says something too.
I just may have to actually make these changes. At least the cm on on the VAS and red LED bias the IPS. Biggest bang for buck. Of course, I am looking at the advice I have been given for more linear transistors etc.
The big fat Motorola book came. Reading time again.
Greens seemed to simulate better than reds. At the very low current, .5mA, is the difference in thermal tracking valid?
Remembering the Cdom was moved to a Miller comp on the VAS. Comes out with the best predicted THD at .004% full power. The original modeled at .013% Changing the Cdom increased the high order distortion, as expected, so without the ccs changes, the model was reduced to .032%. In the sim, when I upped the VAS current by 3, it reduced the distortion as it could better drive the dominant pole loading the output of it. Of course, I could never get it to be stable like that. The biggest bang for the buck is adding the current mirror to the VAS. Improving the IPS was secondary. Again, parts picked because they were in Spice.
To do this removed the various diode protection he had set up. The chance that it would do ugly things if over driven is increased. It maintains full swing of the IPS and VAS.
Of course the spec sheet says .002% full power at 1K, where the simulation is only saying .013% with perfect parts. That says something too.
I just may have to actually make these changes. At least the cm on on the VAS and red LED bias the IPS. Biggest bang for buck. Of course, I am looking at the advice I have been given for more linear transistors etc.
The big fat Motorola book came. Reading time again.
If time permits, a new schematic incorporating all of these changes would be helpful for us.
Hi,
No, they do not need MHz bandwidth. Should our Amplifiers have bandwidth into the MHz?
I normally aim most of my Amplifiers to have -3dB points of 10Hz and 40KHz at full power. My current tube Amp is a bit wider bandwidth, but not much. And it has high THD (it has no global loop feedback, only local output stage feedback to screengrid and cathode) and a damping factor of a smidgen over 2...
I use linear circuits with an emphasis on low high order products (that is audibility and evidence based, not green table based) and as little NFB as I can get away with. For example I find Class AB PP Amp's without loop feedba ck problematic, but for tubes I only use 6-12dB.
The result is usually very little except 2nd & 3rd HD, with higher order rapidly rolling off. Oh and good sound.
Ciao T
No Boss ....
No, they do not need MHz bandwidth. Should our Amplifiers have bandwidth into the MHz?
I normally aim most of my Amplifiers to have -3dB points of 10Hz and 40KHz at full power. My current tube Amp is a bit wider bandwidth, but not much. And it has high THD (it has no global loop feedback, only local output stage feedback to screengrid and cathode) and a damping factor of a smidgen over 2...
I use linear circuits with an emphasis on low high order products (that is audibility and evidence based, not green table based) and as little NFB as I can get away with. For example I find Class AB PP Amp's without loop feedba ck problematic, but for tubes I only use 6-12dB.
The result is usually very little except 2nd & 3rd HD, with higher order rapidly rolling off. Oh and good sound.
Ciao T
Hi,
Here we go again.
IMD 9in a competently designed amplifier are just different ways of describing the same underlying nonlinearity.
High PSRR is necessary if the powersupplies are designed to have low noise.
Damping factor is a synthetic measurement unrelated to anything real. A damping factor of 10 or so suffices to minimise the frequency response variations due to varying speaker impedance and the so called "damping"of the woofer is limited by the Speaker DCR. Hence the true Damping factor rarely reaches 2.
I do not care about synthetic benchmark performance (not just in Audio - for example database server TPS benchmarks are another such scam), unless these synthetic benchmarks are demonstrated to predict actual performance.
The common audio benchmarks (THD, IMD, Damping Factor, Noise) are all promoted to prominence by marketing departments, they are in essence marketing scams, plain and simple and have zero use as predictor of sound quality (once certain absolute minimums are met), their only use is for Advertising.
Ciao T
Here we go again.
IMD 9in a competently designed amplifier are just different ways of describing the same underlying nonlinearity.
High PSRR is necessary if the powersupplies are designed to have low noise.
Damping factor is a synthetic measurement unrelated to anything real. A damping factor of 10 or so suffices to minimise the frequency response variations due to varying speaker impedance and the so called "damping"of the woofer is limited by the Speaker DCR. Hence the true Damping factor rarely reaches 2.
I do not care about synthetic benchmark performance (not just in Audio - for example database server TPS benchmarks are another such scam), unless these synthetic benchmarks are demonstrated to predict actual performance.
The common audio benchmarks (THD, IMD, Damping Factor, Noise) are all promoted to prominence by marketing departments, they are in essence marketing scams, plain and simple and have zero use as predictor of sound quality (once certain absolute minimums are met), their only use is for Advertising.
Ciao T
Hi,
Yet Amplifier distortion is usually rated at full rated power, which for my amp (appx 22dBW) and the Speakers (90dB/2.83V/1m) it will drive gives 112dB/1m and around 109dB at the listening position for a stereo pair (my tube Amp is 16dBW BTW, so it only manages 106dB/1m for one speaker).
If we want to set other reference points they must be explicitly stated.
Ciao T
Yet Amplifier distortion is usually rated at full rated power, which for my amp (appx 22dBW) and the Speakers (90dB/2.83V/1m) it will drive gives 112dB/1m and around 109dB at the listening position for a stereo pair (my tube Amp is 16dBW BTW, so it only manages 106dB/1m for one speaker).
If we want to set other reference points they must be explicitly stated.
Ciao T
a. wayne,
I have heard of only a few.
All,
If you believe that sound quality could be related to transient response accuracy, and to not having large images of the audio transients on your rail voltages, then you might want to take a look at the relatively simple math involving transient current demand, which relates: maximum transient current amplitude change through a device, the expected minimum time to be able to do that, the desired maximum rail voltage disturbance to be caused by a transient event, the needed capacitance, and the distance/inductance between an active device's "point of load" connection pair and the capacitance already mentioned.
Alternatively, or additionally, you can look at coming up with a target impedance for the power rails (max desired voltage disturbance divided by max current change), as seen across the pins or power/gnd connections of the device or local circuit in question. But you still have to make sure that it will also meet the transient speed/timing requirements.
[Or don't even worry about the math and just fit as many caps in parallel as possible, as close as possible across each output device's PSU point-of-load.]
First, please let me issue a disclaimer, and say that I am only dabbling in this stuff, and am not and never have been an audio professional, and don't have nearly as much time to spend on it as I would like.
A few months ago, I was wondering how to actually calculate the minimum required values for decoupling caps for audio power amplifiers' output devices (and for any active devices, in any circuit). The only related material I could find was in papers and books about high-speed digital circuits, and it was all focused on the capacitors that would be sprinkled-around on power/ground planes (Henry Ott, Bruce Archambeault, and others), for chips like digital ball-grid array types. So I decided to apply the basic concepts to audio power amplifiers, and to specifically focus on cases where power and ground planes were not available (although it applies there, too). I am sure it has been done before, many times. But I have not seen it, and mainly I have not seen it emphasized as much as I think it ought to be, for audio.
I haven't actually finished that development effort, yet. <grin>. But I got far-enough to convince myself that for any particular audio power amp, even those with local decoupling caps, it is very easy (and probably typical) to get the decoupling capacitance either too small or too far away or both. Of course, those conditions can be acceptable, depending on the requirement for max rail-voltage disturbance amplitude, and on the transient equivalent-frequency response's maximum frequency requirement (as in f = 1 / (π ∙ trise) ).
Steady-state performance and THD and IMD (and even HD) tell so little of the "performance" story. It's a lot like looking at Statics but not Dynamics, if you were a mechanical engineer. It can't have much to do with soundstage image quality, for example. And it couldn't make the difference between a great amp and an exquisite amp. (IMO, of course.)
If we use the simplest systems-theory view, the only thing besides steady-state response is transient response. So transient response has to be the key. And like steady state (as measured with HD, THD, IMD, etc), it should be as accurate as possible, up to the maximum output levels and speeds. For the best performance, the reproduced timing and tracking of the wildy-changing mixed "real music" signal must be precise and accurate.
Why do many designers (apparently) simply assume that it will be so, and focus instead on the simple steady-state metrics like THD? How would that ever help to ensure that (or even take into account the desire that) anything that CHANGES will be accurately reproduced?
It won't. Wavebourn and some others see it. But maybe not too many want to think outside of their comfort zones.
Sorry to be so damn long-winded. Brain slows way down at end of day. Anyway, the still very-rough, probably-error-laden beginnings of describing and understanding/appreciating "the problem" with the few nano-Henries of power and ground rail self-inductance per centimeter of conductor, and how to know how to possibly overcome it with decoupling capacitors, in order to enable even the possibility of having something like a maximum slew-rate signal from zero to the rail be reproduced accurately (and thus, we hope, also everything else we could throw at it), is linked-to at http://www.diyaudio.com/forums/power-supplies/208579-30vdc-10a-psu.html#post2942537 , with the main link and its most-relevant post numbers in post 5, but also some other relevant ramblings in post 8.
After seeing that the problem looks like it could be significant, and how difficult it can be to get the inductance low-enough when using simple construction techniques, I first looked at paralleling smaller capacitors, where it turned out that even the leads or connections used might often have to be paralleled, all the way to the point-of-load connections, in order to get the full parallel inductance-reducing effect (which doesn't occur if there is any mutual inductance).
THAT made me realize that we could benefit from paralleling multiple copies of each power and ground rail, all the way from the PSU to each active device.
If space was unlimited, that would theoretically enable us to lower the power/ground rails' impedance, AS seen by each active device, to be as low as desired, since the inductance and resistance would both be basically divided by the number of rails used, and the capacitances would be multiplied by that number. Ribbon cables with many interleaved copies of the rails come to mind.
So, I'm wondering if maybe everyone here but me already knows all of this. Anyway, if not, the equations are very simple (maybe too simple; they are only valid for very short time periods). Plug in your numbers and see what happens. I have simulated it but not built it. I guess the best case outcome would be if someone can point out why decoupling caps are not so important, after all.
These are good statements, whole waveform contains everything that could in principle be listened. All the Fourier transforms are restricted approaches, as was mentioned in John Curl's thread, useful for education purposes mostly. Proper measurement approaches, really related to sound perception, can not be based only on the Fourier transforms.
One must compare a whole output waveform to input waveform, and use a penalty coefficient for differences between low-level signal constituents.
Gootee, the effects you are speaking about are listened easily. Good cure for them are shunt power supplies arrangement, where shunting device is located JUST NEAR active device. One can hardly measure the difference as compared to standard serial PS (probably could measure with more sophisticated techniques), but subjective sound impression is very different.
This approach is a must in my recent DIY designs. However, I admit, that industry designers would hardly be able to justify such a PS arrangement against their managers.
The bad practice circle is closed like this: no good measurement approach - no justifications against managers - similar quality mass products - double blind tests for those buyers who doubts and not satisfied.
Last edited:
Unless your waveform contains an infinite number of finite discontinuities or any infinite continuities you can use Fourier and it will perfectly preserve the waveform. All music waveforms are bandwidth-limited (if only by the microphones) so have no discontinuities, therefore a Fourier transform fully contains the information in the waveform. Whether this is useful or not is a different issue, because a problem is best tackled in the way which works best, but let's not have any loose talk about Fourier not capturing things like envelope. It might not do it in a useful way, but it does do it.
The only problem here, Thorsten, is that commerical devices with small caps usually also have small transformers - NOTHING like the Goldmund. Then again, for one Goldmund, you could buy like 10+ of such devices.
Let's not mix cranberries with watermelons.
Which is why you plan to use 144,000 uF instead of 13,600 uF.
Simple. Unless the manufacturer is is HK of Denon (at the high end of their line), those "50W" on others are usually only for one channel driven and short bursts of power at 1kHz frequency (nothing lower like 40-60Hz bass)
Gee, thanks. I have seen The Light!!!
Come on, let's be real, shall we? FTC standards are pretty clear on the subject. Both channels simultaneously driven into 8 Ohms, 20...20,000 Hz, at or below rated distortion.
Yes, the rating is as per the IHF standards - 20-20,000 Hz, at rated power.
Agreed on the second comment. Just one small note - if you're interested in just how really powerful an amp is, just put on The Blue Man Group first audio CD, song No.8. At a point, one of them strikes really hard a bass drum with a diameter of 8 feet. A slam like you're not likely to find on many other CDs.
It sure separates the boys from the men. Both for amps AND loudspeakers.
- Status
- Not open for further replies.