[update: sadly, it appears the Italian Tuneups I've been performing don't have long lasting effects. My current plan is to workout a warmup strategy, that each time gets over an apparent thermal trouble spot that seems to be the crux of the problem, but once the amp gets into a funk it seems non-trivial to keep it going long enough to warm up without shutting down first. Now I've come close to what originally worked with the first Italian Tuneup, and it doesn't seem to work into the 8 ohm load I figured I could keep connected in parallel with the switchbox, making it a trivial procedure to do the warmup. So do I try different power levels, different frequencies, different waveforms, ramping up or down, or a different load? What gives the best "kick" to get started?
So, the glow has faded, but the fact that it worked has indeed opened up a window of understanding and a window of opportunity I'm currently still trying (and tiring) to figure out, so this is still a work in progress on December 12, 2018.]
It had gotten to rock bottom: the amplifier wouldn't even idle, with shorted inputs, and nothing connected to the outputs (which, with no signal, is fine for this amplifier). It would just shut down for no apparent reason, in as little as 5 minutes after having been turned on.
The problem had started just one month after it had been serviced for the 3rd time in two years to fix this problem, twice by the Krell factory, which had finally refused to work on it anymore, and then The Service Department, recommended to me by Krell, after which it had worked more perfectly than ever before. For one month.
All the capacitors have been replaced in these last two years. Hundreds of capacitors. And many other parts, many other repairs, and a few upgrades. I approved every repair and all but one suggested upgrade. (BTW, it might not be a good idea to perform Italian Tuneup if the amplifier needs other service.)
After I confirmed it was working perfectly after a month of simply listening to music, I brought on the ABX amplifier switcher, to allow me some ad hoc comparisons with the Hafler 9300, designed by the creator of the Acoustats. I did a few casual nonblind AB comparisons, the Krell seemed fine, and the winner by a tiny margin (if I were actually hearing a difference at all). But then a week later, when I wasn't doing any AB comparisons, I played a high resolution digital recording, the HRx sampler, at maximum "standard" level (defined in an earlier post). And curiously it shut down during a softer passage just after a louder one.
Over the next month I performed a few dozen elimination tests, to try to determine if the problem was in the amp or something else, and then which channel of the amp, and what conditions caused the shutdown. Several times before, it seemed like I had cured the problem, I even played the very same recording that made it shut down the first time since the last service, and it sailed through just fine. But then it shut down later playing something else.
Now, with the amp only able to idle for a few minutes before shutting down, I thought certain it would need professional repair again. I'd never be able to sell it for a decent price like this, and I didn't really want to sell it, but I might wait until next year so as not to be putting all my money into this amplifier, just my "surplus."
Or maybe, I'd try to fix it myself. I was beginning to think about that again.
When I was in my teens (many decades ago), my best friend had a Pontiac GTO. He loved it for a few years. When I was riding with him on the freeway one day, we were taking an ad hoc trip from Los Angeles to San Francisco, he floored it for a few seconds. "Blasting out the carbon," he said, "keeps the engine running smoothly."
I'm not sure if that was the proper technique, but there is something just like that called an "Italian Tune Up" which consists of something like full power operation for some period of seconds or minutes. And done correctly, this does help remove carbonized deposits from valves and other things. It's a widely recognized technique, especially applicable to cars that are only driven gingerly on a regular basis.
I decided last week I would try something like this on the Krell. I waited for the weekend to get it all set up properly.
Into the living room I moved my Sound Technology 1700B, a digital Rigol oscilloscope, and a never-before-used "2000W" 4 ohm audio brake load from China.
The connections from the Krell merely needed to be moved around from the ABX box over to the Sound Technology analyzer, and merged with connections to the load which was newly terminated with Furtech locking bananas. So it was two sets of locking bananas plugged it to each of the two balanced signal inputs of the analyzer, externally loaded with 4 ohms.
I first ran the apparently (and now proven!) dysfunctional left channel (overheating, anyway, I did not previously know this channel was wrong, but it was clearly running hot, and depending on the experiment the shutdowns seemed to predictably occur at a particular warm up temperature...around 135F).
I ran the channel with a full power signal, just below where visible change to the sine wave seemed to be occuring, which turned out to be around 705 watts into 4 ohms.
I measured distortion as 0.18%, a bit higher that I'd hoped. Everything seemed fine for 4 minutes, and the left channel was barely heating up more than the right. Then the shutdown occurred. The temperature had risen a mere 14F from 78F to 92F. Like nothing.
Now that test, I had presumed, would exercise the output function maximally, the emitter say (I'm not really sure...in fact I'm not sure this makes any difference, the current through the transistor always occurs in certain proportions, regardless of whether any flows into the load. This is something I don't know.
But I decided to try a different approach. After the amp had apparently cooled back down to ambient (I completely shut down the amp at the breaker), which required only a hour or so, I had planned to wait another day, but I was so excited by the prospect, and seeing the amp was now back to ambient, I plunged ahead.
I figured the "worst case" for heating the transistors is the famous 1/3 power output. Into 4 ohms, the maximum power (before true clipping) is 900+ watts. So 1/3 power is 300 watts into 4 ohms.
I miscalculated this as being sufficiently close to 30V to just go ahead with 30V, which is full scale on the 30V range of the analyzer.
Anway, that's what I did, I ran at 30V. And I saw the left channel heat up faster than I've ever seen it heat up before. More than I imagined it could do. I was afraid it might shut down at any moment. But it kept going and going and reached 175F (from ambient 78F) in 15 minutes. I chose that moment to turn the amplifier off. I did not want this test to potentially cause problems (though, actually, I've seen it go to 207F without apparently changing anything, and I believe it has a "kickdown" program at 180F which reduces the bias automatically to reduce the temperature).
Well, this was clearly different. It had clearly blown past the old 135F where shutdowns were occurring.
I let the amp cool down to 160F and decided to just let it run at a maximum sustainable level for hours, partly as a test and partly because this would be another kind of exercise which might help.
I chose the level of 10V or 25W. At that output level, the temp actually rose a bit from 160F to 170F or so, but I just let it keep running, and it ran for 3 hours without issue.
At both 30V (225W) and 10V (125W) the THD measured 0.07%, which appears to be what other people have measured into 4 ohm loads. The specification is 0.03%, which appears to be for 8 ohm loads and using the balanced inputs (which I was not doing since my ST has only unbalanced signal output).
I then let the amp idle and cool down idling overnight. It was still running ok.
The next day I played music at loud and medium levels for 2 hours, and again in the morning for 2 hours, and it still ran OK.
It is, apparently fixed. And just as important, if the problem comes back, I know what sorts of things I may do to get it going again.
I DO have some ideas as to how to prevent the problem from coming back, as well as what ultimately causes it. Before this "repair" I wasn't even exactly sure what channel was causing the shutdown. I suspected the left channel because it runs hotter mostly, and shutdowns seemed to occur at certain temperatures in that channel. Now, since I only treated the left amplifier channel to make the problem go away, it appears quite conclusively that it was/is the left channel of the amplifier at fault, and nothing else. And somehow, certain kinds of operation cause it to change in some way that makes it start failing.
If all I do is Italian Tuneups, I don't really need to know the underlying cause. But my most intuitively satisfying theory is that a certain power transistor on the left side is prone to something like carbonization. It may have some cracks or grooves which get carbonized under some circumstances, causing an internal short which makes it misbehave. This carbonization begins to vaporize or disperse when the heatsink temperature is around 135F. If the amplifier isn't being played hard enough, the computer inside decides this looks like a serious problem and shuts the amplifier down. WHEN the amplifier is being played hard enough, it quickly blasts through this carbonization before it has time to look like a serious problem to the computer, setting the stage for perfect operation until the carbon builds up again.
I suspect the computer program is being too paranoid, and be more tolerant of slight deviations from ideal transistor behavior to allow the normal decarbonization to occur. But this suspicion is irrelevant for the forseeable future, because re-programming the computer is nothing I am going to have the time or expertise to do for a long time into the future, if ever, and it is fraught with risks as well. It is quite likely the "paranoid" computer has saved the amplifier from certain destruction many times, and would be unsafe to operate with much less paranoia. So meanwhile I simply have to live with the paranoid computer program, perform Italian Tuneups as needed, and try to figure out how to slow down the carbonization (or whatever) build up process so I don't have to do an Italian Tuneup often. Though, in principle, I could do it ever day before serious listening.
I have many anecdotal experiences that suggest that playing very high frequencies, such as from high resolution recordings, or from FM tuner outputs digitized at 24/96, is an issue. I noticed this long ago (because of shutdowns occurring while playing FM tuners at modest levels, and even the recent series of shutdowns which started when I played a high res recording), and decided that since I use supertweeters highpassed at 18 or 20kHz, I might as well lowpass the Krell at something like those frequencies. And in one set of experiments, years ago, I decided the system sounded better this way (with the Krell lowpassed) and even measured better. Now I never have much confidence in the reliability of my subjective "tests", and later measurements showed no measurable advantage (the Acoustats roll off steeply at 18kHz anyway, so it "doesn't need" to be lowpassed electronically to blend with supertweeters starting at 18kHz), so I had not been doing this recently. But there is also little point to trying to drive the Acoustats higher than 18kHz because they have a strongly capacitive load which reduces impedance down to less than one ohm at the highest frequencies. This is problematic for an amplifier which tries to run Class A for everything, and has a bandwidth of 300 kHz.
Another line of resoning with fewer supporting anecdotes is that operating with the ABX tester, and running the Krell with no load while still getting an input signal (the way I have been doing it) so the other amplifier can play the speakers, is not a good thing. I was told (by Steven at The Service Department) that you don't have to connect speakers to the Krell. But I did not specifically ask about continuing to run an input signal into it. With many Class AB and true Class A amplifiers, this would not be an issue. But it appears the Krell computer (the so called "anticipator") sets the bias level in large part based on the input signal, not just (or at all) how much actual current the amplifier is delivering. I have seen some evidence that it runs hotter with no load attached than with the speakers attached when there is signal input, as any engineer would expect of a Class A amplifier because if energy isn't being dissipated in the speakers it has to dissipate within the amplifier itself. Now with a real Class A amplifier this is not a problem, as the amplifier normally idles at the highest bias and has to be engineered for that. But with a plateau bias amplifier, it depends on the programming of the computer--was it programmed to allow sufficient thermal margin in case the speakers are disconnected? Now, I've actually done this many times, and the Krell did not seem immediately to be adversely affected. It might even be, in a sort of Italian Tuneup kind of way, beneficial. But because it is a highly suspect unknown, with a few supporting anecdotes as well, I've decided that before doing more ABX or similar blind testing, I'm going to have to modify the tester to switch in a dummy load when the amplifier is not playing the speakers. I believe this is dooable, there is already a 1k or similar load switched across the unused amplifier by the switcher. I just have to rewire that 1k load as a connection to a external dummy load.
I'm going to fly with those two "preventative" ideas for now, and try to remember relevant playing details when the next shutdown occurs so I can more and more reduce the causes of the carbonization or whatever is occurring.
UHOH
All was fine on Friday evening, I started off playing music with gusto, quickly cranking up the right side to 145F.
Then, I wanted to do two things. I wanted to take the historic photo showing the Italian Tuneup setup. I'd cheat by actually measuring the right channel.
Now, just before, the amp had been heated up. But I wasn't paying attention to the cooldown.
I got the right channel going, and was just about to measure, when, the Krell shut down.
I was thinking, maybe the right channel needs the Italian tuneup too??? (At the end of the day, there is no clear evidence for this.) But as it turned out, the now idling left channel was right at that troublesome, for idling, temperature of 135F, where I'd seen the lions share of shutdowns and most of them while idling.
So I switched to measuring to the left channel, pounding it with 225 watts for a 175F warmup, then a slight cool down, then did a lot of interesting measurements. The distortion remains at 0.07% at most levels at 1kHz, until very high levels. At 20kHz, 1 watt or less do measure 0.07% as well (all into four ohms btw). But cranking up to 100w the 20kHz distortion begins to rise. At 600w, 20kHz distortion has risen to 0.4%. (Recall, the Pass Labs XA200.8 has 1% distortion into 8 ohms at 20kHz at lower voltages. I'm testing the Krell with higher power and tougher load.)
So now the left channel had been tuned again, so I switched back to the right channel (a process requiring turning off the amp) and now, to keep things going, I also hooked up the left channel for music playback.
Still, I got to shutdown before I could get a good run at 1/3 power. Once again the left channel was at the suspicous 135F.
I'm thinking to do this right I'm going to need to run both channels simultaneously. I have more dummy loads on order.
And maybe the thing is also, no power cycling when hot, and messing around. Just blast to high temperatures by playing robustly, then keep playing--never pause. Pausing to mess around messes whatever up.
And so, the waste of time continues...
Back to the Hafler 9300, sounding better and better. I kick it off at exactly the same level I had the Krell at (with my 0.1dB level matching) and it sounds...wonderful in every way.
But later, once the Krell had cooled down to 85F, I decided to try exercising and measuring the right channel of the Krell. I could pop off a quick 1/3 power burn and then measurment before the left channel gets back up to 135F, I thought optimisically.
So I ran the right channel at 225w for 15 minutes. The back right heatsink was 163F while the power regulator heatsink in front was 173F. Strange. I proceded to do distortion measurements at different powers and frequencies.
The right channel seemed a bit lower in distortion, 0.06% at 1/3 power at 1khz, and basically 0.06% at other powers. When I tried 20k, I got half the distortion in the other channel, 0.2%, at high powers. At lower powers, 1W, distortion at 20kHz was only 0.065%, and at 10W it was barely higher. So, slightly better than the left channel (despite running cooler) and especially so at 20kHz.
I kept testing for awhile, and strangely the left channel was not pressing upwards to the tricky 135F. I decided to let the amp idle overnight.
Overnight, something amazing happened. The problematic left channel is idling identically to the right, at 115F instead of rising to 140F as I had been seeing recently (when it didn't shut down at 135F). I had only seen that after the last repair by The Service Department when the amp first returned. Somehow, it seems to have fixed itself through the latest exercise, and even more strange, it seems to have been the Italian Tuneup on the right channel that has fixed the left channel, despite the channels having their own separate computers.
This is still a work in progress, I hope, and not a clean slate as might be hoped, but it continues to seem like the Italian Tuneup is effective medicine that does gets the Krell working better. For now, I've decided to let the perfect 115F idle in both channels run for 24 hours or more so that it gets remembered. It's remarkably free of clinking also.
Sunday Nov 18
The Krell has been idling without fail since Friday at 11pm after my last set of full heat stressing and measurments starting from cold (because otherwise I was hitting shutdowns, seemingly still left channel related because the suspicious 135F temperature there, but maybe not).
After that, curiously now, the idling has returned to the wonderful performance first achieved a few months ago when I got the amp back from The Service Department. That was the first time ever (after 3 previous servicings by Krell factory) that the left channel idled normally at at temp below 150F. Going way back, even after the first Krell service, the left channel shot up to high temperature quickly, hitting some stop it seemed at 180F, and then bouncing back down, then back up slowly, and so on. I figured there was a bit of noise or something that kept the bias going up until the kickdown, presumably at 180F as with the earlier KSA "S" models like the KSA-200S, after which plateau 2 or one was enforced, until the temperature cooled enough to re-enable all bias plateaus, then off it went, and so on. That's the way it was from day one in 2008, predictably making clinks on each kickdown.
Actually, though, Krell in their second servicing fixed the endless rising, but the heat was still too high at idle, nearly 160F.
But isn't this strange...exercising the right channel seems to have fixed the bias level holding in the right. After 38 hours of idling, the left channel temperature is 121F in the "backmost" (frontmost in my current cooling optimized backwards setup) heatsink panel, 123F in the middle panel (where airflow is slightly reduced due to the sack of digital components upstream), and 117F in the front "power regulator" heatsink. When the middle panel is warmer in the current setup, things are working good. When the "back" panel is way hottest...that's what the weirdness used to look like, the back panel weirdly hot and causing the weird thermal cycling.
So it looks like both channels need the Italian tune up, even if it looks like only one is bad, they seem to have some weird connection, or perhaps did this time. Actually, this narrative ignores that I did do some more testing on the right channel, just not the full warmup thing. Maybe that's what cleared up the carbon. But, anyway, I think two channel tuneup and testing is best and plan to do it that way in future.
The right channel panels, turned to have the lesser airflow, are nonetheless all 116F (uniformly for both back panels and regulator panel). Note that the cooler regulator panel would be cooler too if on the other side. So this channel is drawing less bias somehow.
My best theory that the funky transistor which sometimes goes weird around 135F otherwise has a slighly different characteristic that the DC feedback of the other balanced half fights against through the servo. Given such a thing is happening, it's a wonder that distortion still measures about the same, 0.07% instead of the 0.06% of the right channel, and a bigger difference at 20kHz, 0.4% vs 0.2%, all into 4 ohms.
After playing some loud music for 60 minutes, I let the Krell idle for two hours. It did so without issue, with the left heatsinks returning to 141F. This suggests there is a bit of problem left, but if generally the Krell is started with some robust playing, to heat to 140F or so, it will run OK after that, and if not, it needs another Italian Tune Up.
Why Keeping the FPB Running is Worth It
For what it's worth, no amplifier I've ever tried in my Acoustat system has sounded better to me, in sighted tests, than my Krell FPB 300. In blind tests, my results so far align with the literature of blind testing of amplifiers: I have not been able to provably hear the difference between the Krell and my other very fine transistor amplifiers like the Hafler 9300 and the Aragon 8008BB. Nevertheless, I do not take these blind tests as conclusive either, and I accept that I like the Krell sound best, just not provably. My audio hobby is emphatically not only about listening to music, it's about learning about many things, including audio amplifiers and how they sound, and even if they can be distinguished sonically at all. That last bit is actually Audio Science, something I confess I do very little of. But I like having the equipment for doing so. And one place to start is with the comparison amplifiers of similar quality, but very different in design and construction, so my top shelf collection, of which my Krell is the crown jewel, the most expensive and exotic--and with lots of stuff to show for that. It's nice having at least one of those--and I believe mine is still up among the best.
I do want the best sounding amplifier for my system to listen to music anyway, even if it's only that I think it might be the best sounding amplifier, without any proof it actually is. Then with this "best sounding amplifier" I can compare with other amplifiers and those comparisons become meaningful and relevant.
It has these objective design attributes:
1) Class A operation (through Plateau Biasing anyway), so no crossover notch distortion and less distortion generally.
2) Very low distortion, including into 4 ohms or less, and up to the highest frequencies. The spec is 0.03% at 8 ohms. I myself have measured 0.07% at high power levels into 4 ohms, which is excellent and far superior to most power amplifiers.
3) Regulated power supply for the output stage--VERY rare among even high end transistor amplifiers. Ensuring low dynamic modulation, another issue never measured but easy to understand.
4) Wide bandwidth: 0.5 Hz to 300kHz.
5) DC coupled, so no capacitor sound.
6) High power which doesn't sag, 450 measured watts into 8 ohms, 900 into 4 ohms, even more into 2 ohms. My electrostatic speakers are very inefficient and have impedance that runs to less than 2 ohms at the highest frequencies. They are also very revealing of distortion.
7) It's generally a nice amplifier to use. It has a nice soft pushbutton to start and the slow start causes no noise or light dimming. The connections are very nice. It looks very nice as well as Impressive.
I've often thought, if I had the money and space, it would be nice to have a newer and more highly regarded Pass Labs amplifier, such as an XA 200.8. Now these monoblocks require more space than I have, and also consume 680W continuously each, for a total of 1340W. That's about double my average power consumption with the Krell, which is already getting to be a bit much. So they wouldn't be very practical for me, let alone the $44,000 pricetag. But what if I had the money, the space, etc, wouldn't it be wonderful?
But it turns out, you can find on many blogs, many people do NOT find the Pass Labs amplifiers to be superior to their favorites, which may be old Krell, Levinson, and Threshold amplifiers.
And there may be objective reasons the Pass Labs amplifiers and other current favorites (including mine, in the lustworthy category) like Threshold and D'Agostino and Levinson might not be as good sounding as my FPB 300, as well:
1) They generally do not have the regulated power supply. The XS 300 does. The very top Levinson usually does. Those may cost way double what even the XA 200.8 costs and consume even more power, just for this pretty obvious feature.
2) They do not have as wide a bandwidth, or as low distortion, especially at higher frequencies. The XA 200.8 reaches 1% distortion at 20kHz and 8 ohms and conservatively rated power. The Krell only gets up to 0.1% or so.
Those same considerations apply to most amplifiers ever made by Threshold, Levinson, and even Krell.
3) Output impedance. SoundStage measured the FPB 300 output impedance as 0.07 ohms. That compares very favorably with the 0.14 ohm output impedance of a Pass Labs XA 200.8, more favorably with the D'Agostino Momentums at 0.2 ohms, and the D'Agostino Progressions at 0.44 ohms. These newer amplifier just aren't doing the damping factor like some older ones.
4) Power into 4 ohms, and 2 ohms. Neither the XA 200.8's nor the D'Agostino amps produce more than 600 watts into 4 ohms. My FPB easily cranks out 700 watts into 4 ohms still with still very low distortion as I measured, but measurements by others show over 900 watts into 4 ohms with less than 1% distortion. It was measured by Martin Collums as having over 1500W into 2 ohms (which isn't specified).
Scanning the measurements of amplifiers, only a few stand out as having possibly better measurements for me (with a very inefficient 2 ohm minimum speaker).
1) The Threshold SA12e. This is the pinnacle of Threshold designs by Nelson Pass and the most desirable (if you can handle the power consumption heat), which has far lower distortion and output impedance than other Threshold models. It bests my FPB by having output impedance of 0.02 ohms, and possibly even lower distortion. Nevertheless, I've seen reviews in which FPB 300 and SA12e are subjectively compared with the FPB being declared the winner.
2) The Levinson 33H is right in there also. Newer Levinsons with hybrid "digital" output stage produce more noise at lower levels like 1 watt. Lesser Levinsons don't have the 4 ohm power, etc.
3) The Soulution has lower distortion, as does the Benchmark, and the top Halcro, but not enough 4 or 2 ohm power.
4) No doubt many others I've never heard of, but generally less known.
Now, another amplifier which might be still better than the FPB is the Krell KRS-200, said to be a D'Agostino personal favorite. And it ought to be, given the history, build, original price, and power consumption. It's a true 200W Class A amplifier with no fiddling around. There's also the KAS 1 and 2, which are plateau bias amps with regulated power supplies, much fancier and pricier and overbuilt than the FPB amps. Actually, there may be a multitude of other amplifiers I don't know of, that would have the esseitnal features of regulated power supply, low distortion beyond 500W in 4 and 2 ohms, even at 20kHz and above. But, of all the ones I read and hear most about most of the time, there aren't many.
All these amplifiers are incredibly more expensive than what I've paid for the FPB 300, even including all my servicing adventures. I've read negative reviews on Levinson service which no doubt is very costly, Pass Labs gets the highest recommendations for servicing, which they hardly ever seem to need anyway.
Now I don't know, I haven't had those amplifiers in house, and possibly never will. Even if I did, I still probably couldn't have a firm (and proven) subjective opinion. But it appears, I have "a contender" and that's what I want. I want an amp about as good as it gets, but not so much my finances are collapsing. And I want to be able to compare nearly as good amps of radically different designs, and find if I ultimately can provably hear the difference. My Hafler is an example of a radically different approach: a very simple design, using MOSFETS and almost all FET circuitry. My Aragon is a simpler amplifier than the Krell also, but bipolar outputs like the Krell, with massive conventional power supply. These amplifiers are all objectively excellent, which is a good starting point.
Despite limited power, the Hafler does wonders with high peak power and current, and indefatiguability. The 9500 has considerably worse measurments, and the 9505 has way more complexity. I feel the 9300 is the goldilocks of Strickland TransNova designs. It has the lowest distortion, even lower than the Krell. But it lacks the regulated power supply, and higher continuous power of the Krell.
