Charles Hansen said:
Hi Charles,
Thanks for these answers. I wasn't sure this was your 100-watt balanced amplifier, and I didn't know that you were using only a single device pair in the output stage.
With 20-20 hinsight, it sounds like it was not such a good idea to try to pack it all into one device. I don't know what you biased the devices at, but it sounds like you tried to get too much gm into a single device. This is how we all learn of course. What idle bias current did you run them at?
Using a 60V device with 25V rails is cutting it too close, in my opinion, especially given power supply variations with line and load. I take it that you only learned of the upward swing of Ic on the devices at higher voltages after the amplifier was in production. You're not the first to get burned by a manufacturers' spec sheet.
I can't explain it, but, assuming that upward swing was real, I'm guessing it was a result of whatever IR did to really push the Rds_on of the device very low; it obviously turned out to be a less-well-behaved device as a result. Here is one case where we probably agree that the device was optimized for switching at the expense of a lot of other stuff.
Given that the devices had no upward swing in Ic up to 30V, it then seems that the runaway you said the amplifiers experienced must not have occurred at idle, but rather under signal where they would be pushed into the region where the upward Ic swing occurred. Given that up there there were no other devices in parallel to hog current from, it is really curious that this happened. Can you elaborate on what you think the scenario was?
While I'm sure a labor-saving approach is tempting (using a single device and not having to match), since when is such a labor-saving approach relevant to high-end audio?
. Your amplifiers are such a work of beauty that there is little evidence of labor-saving elsewhere in them. It has not been my experience that vertical MOSFETs are all over the map, at least not in the last 10 or more years. I get a tube of them and they are usually like peas in a pod. Matching has never been difficult for me.Thanks,
Bob
Thanks Bob, for the encouragement. Actually, I once made a 100% fet power amp for a client. It still works today, I should think. However I used Hitachi lateral fets, instead of vertical fets.
I also agree with Charles that balanced bridge is a good way to go. However, Parasound has never supported the time and effort, (nor did they need to) in order for us to change from a single sided power amp design.
Now, why did I answer the question about using a low voltage Vmosfet for him? Could it be, that I thought your question a little below the belt? Condescending prehaps?
It was obvious to me, and I hope, many others
I also agree with Charles that balanced bridge is a good way to go. However, Parasound has never supported the time and effort, (nor did they need to) in order for us to change from a single sided power amp design.
Now, why did I answer the question about using a low voltage Vmosfet for him? Could it be, that I thought your question a little below the belt? Condescending prehaps?
It was obvious to me, and I hope, many others
Charles Hansen said:
Hi Charles, nice to have you back.
Referring to the measurements in that review,
http://stereophile.com/solidpoweramps/407ayre/index4.html
I noticed that the characteristic of the THD vs wattage shows a
hump in the mid-power range. I am interested in this because
I have personally always found that a monotonic character,
in which the distortion declines consistently with level, gives the
best sonic characteristic.
I have seen this hump many times in both Bipolar and Mosfet
output stages, and it was removable by increasing the bias.
Can you enlighten us as to your thinking about this phenomenon
with regards to the trade-offs (if any) that you considered?
There just aren't enough hours in the day to try everything. At some point you just have to say "that's enough" and start shipping the darned thing.
Increasing the bias isn't really an option with the current configuration. It runs pretty warm already, and I don't think consumers would find it acceptable if it ran any hotter. The only other choice would be to reduce the rail voltage, and then the bias could be goosed up. But to get a meaningful reduction in dissipation would probably mean going from 300 watts down to 100 watts. We would probably sell 1/10 as many units if it were only 100 watts.
At the end of the day it's still the best power amp I've heard in my life, and there are quite a few other reviewers and manufacturers who agree with that opinion. So if it sounded even better with more bias, that would be something special indeed.
Maybe someday we'll make an "ultimate" amp that will cost twice as much, be twice as big and have 500 watts per channel and a higher bias level (if in fact, as you propose, that is the source of the bump in the distortion curve).
PS -- Thanks for the tip, Nelson!
Increasing the bias isn't really an option with the current configuration. It runs pretty warm already, and I don't think consumers would find it acceptable if it ran any hotter. The only other choice would be to reduce the rail voltage, and then the bias could be goosed up. But to get a meaningful reduction in dissipation would probably mean going from 300 watts down to 100 watts. We would probably sell 1/10 as many units if it were only 100 watts.
At the end of the day it's still the best power amp I've heard in my life, and there are quite a few other reviewers and manufacturers who agree with that opinion. So if it sounded even better with more bias, that would be something special indeed.
Maybe someday we'll make an "ultimate" amp that will cost twice as much, be twice as big and have 500 watts per channel and a higher bias level (if in fact, as you propose, that is the source of the bump in the distortion curve).
PS -- Thanks for the tip, Nelson!
Bob Cordell said:
First of all, the link to the schematic can be found in this post:
http://www.diyaudio.com/forums/showthread.php?postid=334963#post334963
We normally ran the idle current at 550 mA.
Bob Cordell said:
No, it happened at idle. With normal line voltage the rails were at 25 volts. But when the AC line went high, the rails would climb to 28 volts or so. Now pretty much any device has an Id (or Ic) that increases with Vds (or Vce). With BJT's this is called the "Early effect". But the problem with these vertical MOSFET's went much beyond a nice straight line heading slightly upwards. It was a definite inflection point where the lines would curve upwards.
Like I said, we tried dozens of potential replacements looking for something that would work better and plug into the existing circuit. We wanted something that was rated at least at 60 amps and 60 volts, had the same pinout, and the same package. We tried parts from a half-dozen manufacturers. ALL of them suffered from the same problem to some degree or another except a Samsung-designed part that Fairchild had purchased (and then discontinued). (For whatever reason, the P-channel parts had no trace of this problem.)
Which just points out that adding a couple of extra zeros does not make a better sounding amp, especially when you MUST use negative feedback to get those extra zeros.
I've been trying to point that out for months.
Let's give Halcro a gold star for making one of the best measuring amps in commercial production today. This does NOT mean that they make the best sounding amps, even though they do sound pretty good to my ears. However, measured distortion is NOT the only or even the major criterion as to making the best sounding amp.
For me, at this time, it is the generation of 7th harmonic distortion that worries me the most, after the overall distortion is 'low enough', the slew rate is 'fast enough' and the peak current is 'more than enough'.
Charles may have another major concern that he has addressed successfully, such as RFI infiltration through the power supply, or something else.
I've been trying to point that out for months.
Let's give Halcro a gold star for making one of the best measuring amps in commercial production today. This does NOT mean that they make the best sounding amps, even though they do sound pretty good to my ears. However, measured distortion is NOT the only or even the major criterion as to making the best sounding amp.
For me, at this time, it is the generation of 7th harmonic distortion that worries me the most, after the overall distortion is 'low enough', the slew rate is 'fast enough' and the peak current is 'more than enough'.
Charles may have another major concern that he has addressed successfully, such as RFI infiltration through the power supply, or something else.
Jan, I went to single sided drive, BECAUSE I thought that I could get away with it! IT IS CHEAPER. I wanted a simple amp to replace the Marantz 7. The model 9 would come later. Power mosfets LACKED the dreaded second breakdown that forced me to use a bridged amp in the first place. I knew that I could make a 25-50W amp with a single side with bipolar transistors, or lateral fets. Why not 100W with these brand new and exciting Vmosfets? Cost me plenty.
Now why don't we ask Charles what his maximum practical supply voltages were for his bigger all fet power amps?
Now why don't we ask Charles what his maximum practical supply voltages were for his bigger all fet power amps?
john curl said:
The V-1 was our second generation all-FET power amp. The overall topology was similar to the V-3 (schematic linked in a previous post), but we got rid of all the vertical parts.
The inputs were changed to JFET's, while the folded cascodes and the outputs were changed to lateral parts. The entire design was much better sounding, and also had a much wider bandwidth and much lower distortion, especially at high frequencies and high powers. (Stereophile's tests can be seen at):
http://stereophile.com/solidpoweramps/635/index5.html
Please note that this amp was tested when John Atkinson lived in Santa Fe and his AC line voltage was a very low 110 to 115 or so. This affected the power ratings of his tests considerably. Now that he lives in Brooklyn, he has the opposite problem and his line voltage is running around 123 to 126 and he routinely gets very high powers from the amps he tests.
The lateral output FET's are rated at 160 volts, so they could be used in a VERY powerful amp. We paralled 8 devices in each leg of the "H-bridge" to get 200 watts per channel (32 matched devices per channel!). We built one pair of prototypes with 12 paralleled devices (48 matched devices per channel!) that gave 300 watts per channel and sounded amazing.
There are two big problems -- the cost of the parts and labor required to match them, and the heat dissipation. MOSFET's sound better and better as the bias is increased (without limit, in my experience). For a while we had the 300 watt prototypes idling at 600 watts per channel. Sounded great, but was stupidly impractical.
In contrast, the current MX-R using bipolar transistors has the same output power but idles at about 120 watts per channel (40 watts of this is just excitation current for the E-I transformers). The amp runs MUCH cooler and sounds quite a bit better than the big prototype FET design did.
Anyway the point is that with enough of the lateral devices, you could easily make a 1000 watt bridged design. It's just that to get good performance you would need to use a boat-load of them and the idle power would be completely impractical.
Sorry, Charles, I did not know that you had already switched to lateral devices. I have never had any real problem with them, either, except the usual current and drive limitations.
I was actually implying a conversation that we had years ago about vertical mosfets, and the upward limiting powersupply voltage, due to their implied defects.
I was actually implying a conversation that we had years ago about vertical mosfets, and the upward limiting powersupply voltage, due to their implied defects.
Just to support the V-mosfet option . Here is the link to Creek 5350SE measurements by Stereophile.
http://www.stereophile.com/integratedamps/327/index7.html
I've used there 4 vertical "logic level" N-mosfets per channel (HUF76639)- in a "totem pole" configuration. That arrangement helped to reduce the voltage per each transistor and share the dissipation at the same time using only a single pair for amplifing the signal. No source resistors, fairly simple VAS, idle current 70 mA.
Alex
. Here is the link to Creek 5350SE measurements by Stereophile. http://www.stereophile.com/integratedamps/327/index7.html
I've used there 4 vertical "logic level" N-mosfets per channel (HUF76639)- in a "totem pole" configuration. That arrangement helped to reduce the voltage per each transistor and share the dissipation at the same time using only a single pair for amplifing the signal. No source resistors, fairly simple VAS, idle current 70 mA.
Alex
See the sharp edges on distortion picture of the Creek amp? This is NOT good. Please remember folks that 1000's of amplifier circuits have been developed over the last 50 years. Only some are exceptional, most are forgetable. The very worst is a distortion picture that looks like a series of pulses or square waves. Triangle waves are not too good either. This is because higher order distortion is being added to the musical signal. This WILL change the sound for the worse.
