Only fair he sends one out , i'm sure you could make it the quietest best amp in the world ...
I must say as well that the HFN/RR measurements, while better than nothing, make me appreciate Atkinson's ones in Stereophile all the more.
Well, that is mean. Plus that style of bathroom scale is usually not very accurate
A line of early aftermarket autosound amps I worked on a little for JBL, circa 1990, looked a lot like carpet sweepers, one of those old Bissell jobs, something noted as well by Tom Jacoby iirc.
I remember those , that was you ...
Actually the main man was Darrell Bennett, who inherited the project from some others. I got involved a little based on some problems, including excess distortion in the switchable Sallen-Key three-pole lowpass, which was to allow a channel to be used for a woofer alone.I remember those , that was you ...
It was actually one of my early atta-boys. The filter had three 18k resistors in series to the noninverting input of a unity-gain buffer, and switched the three capacitors to the junctions of the resistors and input in and out. There was excess distortion at high frequencies with the caps switched out for full bandwidth. I conjectured that the BiFET opamp was suffering from common-mode distortion, associated with the PFET input stage and substrate capacitance. Since active drive of the power rails (really only the negative rail drive was needed) was out of the question, I experimented with a compensatory impedance in series with the inverting input and tuned it for minimum distortion. It was enough to bring the product into spec. My bosses were impressed, having in particular vocally discounted my conjecture until I showed that the fix solved the problem
Brad, so what if the noise is a little above some others?
Hopefully, you and I do not listen to them for noise and distortion, but for the music. If the music comes out better to much better than from other, nominally technically superior, then that's all there is to it.
As I see it, Dan d'Agostino was one of the first designers to recognize the simple fact that low THD and IM distortion values mean very little - if anything? - in terms of the final result, the sound. Of course, this should not be taken literally, but I'd like to remind one and all that for the last two decades, Krell products had THD values of 0.3% no problemo, yet always came out on top, if not as No.1, then as one of the top 3 products in any group.
Personally, I agree completely with Thorsten especially and Nigel in good part, the 20 Hz to 20 kHz THD is just fine if it's 0.1% or so, so long as its say 50 kHz performance is what it should be, meaning just a little higher than its 20 kHz performance, not 5 or more times that, as is the most common case.
Hopefully, you and I do not listen to them for noise and distortion, but for the music. If the music comes out better to much better than from other, nominally technically superior, then that's all there is to it.
As I see it, Dan d'Agostino was one of the first designers to recognize the simple fact that low THD and IM distortion values mean very little - if anything? - in terms of the final result, the sound. Of course, this should not be taken literally, but I'd like to remind one and all that for the last two decades, Krell products had THD values of 0.3% no problemo, yet always came out on top, if not as No.1, then as one of the top 3 products in any group.
Personally, I agree completely with Thorsten especially and Nigel in good part, the 20 Hz to 20 kHz THD is just fine if it's 0.1% or so, so long as its say 50 kHz performance is what it should be, meaning just a little higher than its 20 kHz performance, not 5 or more times that, as is the most common case.
Yes but...
The spectrum of the distortion is of course important. Low-order 1000ppm is not the worst thing in the world. Could it be that the amount of noise is intentional?? Is it an attempt to bury the distortion below a noise floor at low to moderate levels? Some signal-averaged residuals would be very interesting to see. Has Stereophile reviewed this yet? I'd like to see JA's sidebars. I don't recall it as being in there yet, although I see D'Agostino is advertising there.
What bothers me is how the noise can be that bad. I can't conceive of tradeoffs that would enhance some other performance parameters and entail that much noise! Especially for linear amps, getting substantially better noise performance should be pretty trivial. I was managing about 30uV rms output noise unweighted with a somewhat-smaller class D amp a while back, and that at higher overall gain.
The fairly high output impedance (and I am trusting the published HFN/RR measurements, perhaps unduly) suggests a low global feedback design. And that's o.k. if one is persuaded that this enhances the sound quality. It does not by itself say much of anything about noise.
I guess it is conceivable that he is massively degenerating the input devices, to have plenty of local feedback linearization, and that this is accounting for a considerably higher e sub n there.
I'm just puzzled. For many sources and nominal-efficiency loudspeakers, that amount of noise is not necessarily going to be intrusive. For high-efficiency ones in a quiet room at typical distances, you are likely to hear audible hiss if the roughly 400uV is predominantly high frequency. Again, actual listening and/or more data is needed.
Of course Krell has been noted for behemoths that handle low-Z loads and very high currents, so such amps would probably not be the first place to turn for driving high-efficiency speakers to begin with.
It's only a few days ago that I purchased a network analyser and one of the main reasons for it is to properly measure paralleled caps, supply impedances and stuff like that... once I find enough time for that besides the day job (where it will be useful, too). Today I made a first try with 1000uF/63V electro plus 100nF ceramic... a wiggle around 2Mhz was visible but not high-Q and still well below 1 ohm (if I did the normalizing right). I expect to find more severe ringing with bigger can size electros plus "just the wrong" (too small) bypass ceramic/film even when sitting as close as posible.
For gross resonance problems, injecting a current step and looking at the response offers insight with less expensive tools (see Measuring Capacitor ESR and Inductance for the concept).
Last edited:
As for the ratio of main vs bypass caps, you'll run into problems once the main cap already is fully inductive and the bypass is still fully capacitive at the intersect of the impedances, especially when there is neither resisitive component in the coupling connections nor parallel resistive damping with snubbers.
So a rule of thumb would be to select caps that have their turnover points seperated by less than one decade of frequency and also chose the bigger one to have just enough ESR to give a broad valley instead of a narrow notch at its turnover.
For caps with same order of ESL (including stray) this means staying lower than about 100:1 capacitance ratio as turnover freq is proportional to 1/sqrt(L*C).
So a rule of thumb would be to select caps that have their turnover points seperated by less than one decade of frequency and also chose the bigger one to have just enough ESR to give a broad valley instead of a narrow notch at its turnover.
For caps with same order of ESL (including stray) this means staying lower than about 100:1 capacitance ratio as turnover freq is proportional to 1/sqrt(L*C).
Last edited:
If the bypass is too small then what you're looking at is the inductance of the can overwhelming any impact of the bypass at frequencies lower than the resonant frequency of the ceramic/film.
Frank
@Brad
I understand your musings and, make no mistake, I am definitely not opposing your views.
On the other hand, like Krell sound or not, no-one can possibly say that Dan d'Agostino doesn't know his trade too well. Meaning, that if he left it as is, there is very probably some reason for that. Possibly he thinks this is low enough to be inoffensive, or solving that problem may cause problems appearing elsewhere, I don't know.
Let me remind you of what I think could be an amp designer's worst nightmare, he has no idea what his amp might be connected to, and people have been noted to come up with the strangest of ideas in time.
I've noted a few oddities like that in my time as well. In 2001 or 2002, I was desperately trying to make Graham Slee from UK to dump his NE based headphone amp for an AD 826. Eventually I did it, and he did admit it sound quite a bit better with the AD, but he also noted that it had an unexplained offset. Didn't bother him, he has a cap in series with the output, but it sure bothered me, this 0.6V offset.
Solving the problem was no bid deal, just a simple 1N4148 diode from output to the minus Vcc line, but that offset shouldn't have been there in the first place, it flew against their own Data Sheet.
I sent them a mail outlining the problem, with a schematic I was using. Got no reply, BUT the next batch of these op amps, purchased about 4 or 5 months later on had the problem resolved. It was almost pointless using a servo, it was that good.
These things happen. Just looking at the internal topology of modern op amps it's hard to imagine that a mistake should never somehow creep in, after so many checking and rechecking.
In 1990 or 1991, a friend and I were doing a text for a local PC magazine related to numeric coprocessor (remember those?) effciency and speed. The notm was, of course, Intel, but another company was making a lot of waves with their take on the matter, for the life of me I can't remember their name now. Anyway, I supplied the hardware, and he supplied the software, which was a massive architectural program he wrote, based on MS Fortran (among other tests).
During the test, we discovered that the alternative had serious problems understanding numbers 0 and 1 - believe it or not. Consequently, it kept churning out wrong results on Fortran based programs (and I hasten to add, in those days, MS Fortran was THE standard against which all others were compared).
We got in touch with the company via a BBS board (all the rave in those days) and explained the problem. They were quick to answer, tahnking us for bringing their attention to a problem, and they would get back to us. And they did, about two months later, sending us a new version of the chip, with a request that we try it out. We did, and it worked just fine.
It happens. No more, no less.
I understand your musings and, make no mistake, I am definitely not opposing your views.
On the other hand, like Krell sound or not, no-one can possibly say that Dan d'Agostino doesn't know his trade too well. Meaning, that if he left it as is, there is very probably some reason for that. Possibly he thinks this is low enough to be inoffensive, or solving that problem may cause problems appearing elsewhere, I don't know.
Let me remind you of what I think could be an amp designer's worst nightmare, he has no idea what his amp might be connected to, and people have been noted to come up with the strangest of ideas in time.
I've noted a few oddities like that in my time as well. In 2001 or 2002, I was desperately trying to make Graham Slee from UK to dump his NE based headphone amp for an AD 826. Eventually I did it, and he did admit it sound quite a bit better with the AD, but he also noted that it had an unexplained offset. Didn't bother him, he has a cap in series with the output, but it sure bothered me, this 0.6V offset.
Solving the problem was no bid deal, just a simple 1N4148 diode from output to the minus Vcc line, but that offset shouldn't have been there in the first place, it flew against their own Data Sheet.
I sent them a mail outlining the problem, with a schematic I was using. Got no reply, BUT the next batch of these op amps, purchased about 4 or 5 months later on had the problem resolved. It was almost pointless using a servo, it was that good.
These things happen. Just looking at the internal topology of modern op amps it's hard to imagine that a mistake should never somehow creep in, after so many checking and rechecking.
In 1990 or 1991, a friend and I were doing a text for a local PC magazine related to numeric coprocessor (remember those?) effciency and speed. The notm was, of course, Intel, but another company was making a lot of waves with their take on the matter, for the life of me I can't remember their name now. Anyway, I supplied the hardware, and he supplied the software, which was a massive architectural program he wrote, based on MS Fortran (among other tests).
During the test, we discovered that the alternative had serious problems understanding numbers 0 and 1 - believe it or not. Consequently, it kept churning out wrong results on Fortran based programs (and I hasten to add, in those days, MS Fortran was THE standard against which all others were compared).
We got in touch with the company via a BBS board (all the rave in those days) and explained the problem. They were quick to answer, tahnking us for bringing their attention to a problem, and they would get back to us. And they did, about two months later, sending us a new version of the chip, with a request that we try it out. We did, and it worked just fine.
It happens. No more, no less.
When ESL1, ESR1 are the parasitics of the bigger cap,
and ESR2 is the parasitic of the smaller one,
the optimum capacitance is (smallest one that will have good effect) :
C = 4*ESL1 / (ESR1+ESR2)^2.
It's also clear that ultra low ESR of the main cap is not a good idea if you want to add smaller ones (and bypasses) remotely.
For a true flat impedance profile the bypass ESR should equal the main cap's one, then the bypass circuit is the complex conjugate of the main cap. Of course at HF the bypass' ESL dominates the picture again.
LTspice is very helpful to illustrate the effect if there is severe mismatch (as mentioned by Frank)
Find attached a sim file for it (plot V(out), or if you want correct units in linear view, plot V(out)/1A). Rename to *.asc before opening.
Last edited:
Oh man, this is becoming wild.
I do not doubt KSTR's words, but please consider - first just about everybody drills you that you need low ESR capacitors for best performance, and now KSTR tells us that you should not use low ESR caps if you want to bypass them?
KSTR, let me ask you outright - if you have two say 10,000 uF/63V caps in parallel per supply line feeding your power amp, how would YOU bypass them? If at all?
10,000 uF/63V caps by Fisher & Tausche.
I do not doubt KSTR's words, but please consider - first just about everybody drills you that you need low ESR capacitors for best performance, and now KSTR tells us that you should not use low ESR caps if you want to bypass them?
KSTR, let me ask you outright - if you have two say 10,000 uF/63V caps in parallel per supply line feeding your power amp, how would YOU bypass them? If at all?
10,000 uF/63V caps by Fisher & Tausche.
the 10m+10m are separated from the bypass (local decoupling) caps by some milliohms of cable and trace resistance. That helps attenuate the tendency to ring.
The decoupling ceramic also needs some damping, this implies that a hiQ cap will be underdamped. So we choose a NON C0G/NP0 cap. The X7R, etc. have a lower Q due to the higher esr.
The decoupling ceramic also needs some damping, this implies that a hiQ cap will be underdamped. So we choose a NON C0G/NP0 cap. The X7R, etc. have a lower Q due to the higher esr.
Those are good ones. But they will go inductive in the lower kHz range (eg 5Khz for 100nH), no way around it.
For DIY (for commercial use it would infringe patents) I'd try to use real inductance cancellation (google for Timothy Neugebauer papers about that) and no bypass, then feed the supply via coax or stripline to the point of load (will be OK for some 10cm). Then check experimentally if there is ringing and try install snubbers if needed. Amps are current sinks and DON'T damp the supply, the transformer side also does not damp most of the time (when diodes don't conduct) and shock-exites any tank-circuit at the diode on/off times.
That's what I keep saying, F & T are my standard, but I'll gladly take BC Components as well. I have 20+ on stock at all times.
On the power amp side, I tend to use 100 uF in parallel with 2.2 uF Wima metalalized, and lastly 1 Ohm+470 nF Wima in series to the ground. That takes care of most residual capacitor inductance, but 470 nF is a tentative value only, it could be anything from 680 nF to 220 nF. No way of knowing, you just have to try and see.
Anyway, thank you for responding.
We are also supposing that the HFN/RR measurements are accurate, and that the amplifier was not oscillating into some strange load, or as the result of some bogus connections. Note that the only inputs are balanced, and perhaps the output is also. No mention is made in the sidebar of the test equipment used, although as I'm only a sporadic reader of the magazine maybe they disclose this sometimes, and it is always the same. Again, although Stereophile isn't perfect, it does make me appreciate them by comparison.
I haven't been a close reader of Krell specs in previous reviews, so I have no idea how they tended to run in terms of signal-to-noise. I only picked up anecdotally that they are powerhouses in terms of high current --- in fact weren't they initially developed to support some very-low-Z and inefficient ribbon speakers --- were they ones from Apogee?
Actually I find a lot of highly-touted equipment to have disgracefully high noise, so I am not singling out this latest effort from D. d'A. When six of us were still cooperating in a speculative class D amp effort and about to go to a tradeshow, it was impossible in the short term to find a digital source low-enough-noise to have the power amplifier noise dominate. People remarked as to how the noise at the fairly-efficient speakers was virtually inaudible, even with the ear pressed up against, and I reminded one of the consortium that it was strongly dominated by the Bel Canto DAC, loaned to us along with the big speakers by Andrew Jones. That itself had replaced the (iirc) Simaudio piece, which was much worse for noise.
As I say with respect to the reviewed amp, I'm just puzzled and curious. I am even tempted to speculate about the conscious use of stochastic resonance, although the effectiveness of that will be very loudspeaker-efficiency-dependent.
Krell ......!
Krell Full Power Balanced 350mc monoblock amplifier Measurements | Stereophile.com
This is a very quiet amplifier, with an A-weighted S/N ratio of more than 95dB. This is referenced to 2.83V output; this is equivalent of more than 130dB referenced to the amplifier's clipping point into 8 ohms! Even without the A-weighting network and extending the measurement bandwidth to 500kHz only worsened the S/N ratio to 84dB (again re. 2.83V).
Levels of harmonic distortion were also very low, particularly into higher impedance loads. The bottom trace in fig,3, for example, was taken at 2.83V into 8 ohms, and doesn't vary from 0.005% across the band. Reducing the load to 4 ohms increased the THD figure slightly to 0.007%, though into 2 ohms, the distortion went up to 0.03%.
Krell Full Power Balanced 350mc monoblock amplifier Measurements | Stereophile.com
This is a very quiet amplifier, with an A-weighted S/N ratio of more than 95dB. This is referenced to 2.83V output; this is equivalent of more than 130dB referenced to the amplifier's clipping point into 8 ohms! Even without the A-weighting network and extending the measurement bandwidth to 500kHz only worsened the S/N ratio to 84dB (again re. 2.83V).
Levels of harmonic distortion were also very low, particularly into higher impedance loads. The bottom trace in fig,3, for example, was taken at 2.83V into 8 ohms, and doesn't vary from 0.005% across the band. Reducing the load to 4 ohms increased the THD figure slightly to 0.007%, though into 2 ohms, the distortion went up to 0.03%.
- Status
- Not open for further replies.