Actually, from my testing on the MJW1302/3281, matching beta to 1 or 2% is not that hard.
It is my theory that the higher beta transistor tends to hog current. This is because the higher the beta, the less base current it draws, so less voltage drop across the base stop resistor. This in turn make the emitter voltage higher than the one with lower beta and hog current.
Actually I have been thinking about the temperature of the outer transistor is lower like you said, so if I put a transistor with say 2% higher beta on the outside, that will prevent it from hogging a little. But at the same time, since it is going to pull more current, it will balance the temperature. You put the bias spreader on top of this transistor, you'll have a better chance to have better compensation.
Yes, it is very confusing for newbie. All these time, I believed that Class A is the best, but the price paid is lower rail, which can be saturated easier. So it goes round and round.
I will have issue to raise the rail voltage because I invested in like 30 of the big caps that are 63V only. Other than that, I should be able to raise to 60V without any problem. Over that, I have to change out the driver transistor as it's only 150V.
No multiple pair "silver bullet" .
Many factors -
- vertical or horizontal extrusion and/or fins.
- Type of sensor design (single,dual,or diode) .
- C/W of extrusion.
- # of pairs. Two pair will work differently than 5 pair. And (of course ) ,a
single pair is the easiest.
All these different approaches , and the OEM's do not want to encounter
a warranty issue. So from the simple to-92 sensor on a cheap 40W
receiver to the NJL compound sensor approach , it's already worked
out for us.
There really is no "rule" , too many factors.
To convey these considerations would require references to actual working designs ....
like the two Harmon-Kardons (680/990).
A single or dual pair EF2 AB stage is much easier (but less accurate).
OS
Many factors -
- vertical or horizontal extrusion and/or fins.
- Type of sensor design (single,dual,or diode) .
- C/W of extrusion.
- # of pairs. Two pair will work differently than 5 pair. And (of course ) ,a
single pair is the easiest.
All these different approaches , and the OEM's do not want to encounter
a warranty issue. So from the simple to-92 sensor on a cheap 40W
receiver to the NJL compound sensor approach , it's already worked
out for us.
There really is no "rule" , too many factors.
To convey these considerations would require references to actual working designs ....
like the two Harmon-Kardons (680/990).
A single or dual pair EF2 AB stage is much easier (but less accurate).
OS
"averaging" already happens on the extrusion , thermally.
The Genesis stealth actually did this , they split the amp into 4 smaller
amps. Each had it's own drivers and sensor.
But , they did this for the fact they also had 4 separate extrusions.
OS
Maybe for you, having so much time in your hands. Other may have much better things to do (and tighter pockets) than matching. Spreading devices with higher beta towards the heat sink margins would be a great theme for a Red Dwarf episode.
Well, if you are hobbyist and you are here, you want to build a better amp than any of the amps on the market. You can afford to spend the time as NO engineering can compensate this.
IF you try to get into the market competing with the established brands, you better be better than them as you are not going to make it cheaper than them in large productions. Again, you cannot design to overcome this, so if you want to stand out, you need to do something else.......hand pick.
It's not even that hard, I can sort out enough transistors for say 100 amps in a day once you have it set up. You seriously think that's not worth while?
I might be a newbie in audiophile, but I've been around the block a few times. Either you are a pioneer in this, you have a patent and you have something special OR you better have a niche in the field to survive in this business. There are just too many companies out there. I question many of the established companies can get rich selling amps( say make a decent living). Case in point, if there is so much money in the field, there won't be that many established people still here talking.
Last edited:
And, if you assumed there was no asymmetric mounting or unequal cooling and also the devices are all matched...... would the lower value Re produce lower distortion? And, Rs?
[OPS only]
THx-RNMarsh
Last edited:
From my limited experience of testing the beta of transistors, my MJw1302/3281 mostly come within 4 to 5% max of beta matching. Only 4 that is like 20% difference. So if you happen not to encounter the ones that is way off, you might think it's not a problem.
But I am sure if you have one of the transistor that is 20% off, particularly on the higher side, you likely to have problem of hogging.
Larger base stop resistor can make things a lot worst from my calculations here.
I cannot emphasize, matching beta is NOT as hard as people think. I matched over 100 transistor in less than two hours. Don't tell me you have time to talk here and build you dream amp but no time to spend two hours to match the transistors. I have enough transistors to build at least 5 amps for the two hours I spent.
far as I understand, beta matching is the single most important thing to prevent current hogging unless you mess up on the heat sink department.
I don't agree with Self on bolting the bias spreader on the power transistor will prevent hogging. That's apply and orange. Beta matching is the only thing I can think of.......Unless Mr. Cordell tell me otherwise.
But I am sure if you have one of the transistor that is 20% off, particularly on the higher side, you likely to have problem of hogging.
Larger base stop resistor can make things a lot worst from my calculations here.
I cannot emphasize, matching beta is NOT as hard as people think. I matched over 100 transistor in less than two hours. Don't tell me you have time to talk here and build you dream amp but no time to spend two hours to match the transistors. I have enough transistors to build at least 5 amps for the two hours I spent.
far as I understand, beta matching is the single most important thing to prevent current hogging unless you mess up on the heat sink department.
I don't agree with Self on bolting the bias spreader on the power transistor will prevent hogging. That's apply and orange. Beta matching is the only thing I can think of.......Unless Mr. Cordell tell me otherwise.
Last edited:
AJT , are you using the "primitive" one device method ? And
large heatsinks ?
That ties in with Harry's comment. Putting the Vbe a top one of the pairs
assumes Tc for one device. You want the average of all devices.
Put the main sensor between 2 pairs , you should have an average evened
out by the extrusion. By itself , this will be a slow thermal feedback.
For the dynamic , have the second sensor on the drivers.
The two examples play out this line of thought.
- The NJL 990 has a
considerably undersized extrusion with forced air fan assistance.
The NJL/MJL 5 pair are tightly matched with .33Re. Made to be
compact and "deadly" at 200W.
-The cheaper 680 uses unmatched Sanken MT-100 Pairs and a convection
based thermal setup. It has the separate toshiba drivers and dual sensor setup.
Not as fast as the NJL , but better at handling wide thermal variations.
Each is designed specifically for the available thermal solution.
OS
Hi Bob,
Once I have my stuff back together I'll have a chance to do some poking around on this issue. I am having a problem with the idea that the servo dumping current or voltage into the feedback side of the LTP will not unbalance it. From my viewpoint, that is exactly how the servo works, by forcing the LTP to be out of balance. You might be correct, but at the moment I can't see how that is supposed to work without creating an unbalanced condition.
Bob, can you do me a favor and think about the BGW amplifier example I gave earlier? I can send you the diagram, but it is 7.3 meg in size (full manual). HiFi Engine has it if that is easier. I might need you to type slowly using single syllable words to explain this to me. I just can't see how what you are saying can be true.
Best, Chris
My knowledge has come from years of observing what goes on in the market place. I don't have a problem with DC servo setups as long as they are done properly. Most are not. I will also whole heartedly agree that getting away from that large electrolytic cap is a worthwhile enterprise.
Once I have my stuff back together I'll have a chance to do some poking around on this issue. I am having a problem with the idea that the servo dumping current or voltage into the feedback side of the LTP will not unbalance it. From my viewpoint, that is exactly how the servo works, by forcing the LTP to be out of balance. You might be correct, but at the moment I can't see how that is supposed to work without creating an unbalanced condition.
Bob, can you do me a favor and think about the BGW amplifier example I gave earlier? I can send you the diagram, but it is 7.3 meg in size (full manual). HiFi Engine has it if that is easier. I might need you to type slowly using single syllable words to explain this to me. I just can't see how what you are saying can be true.
Best, Chris
Chris,
I think that there's two ways of looking at current injection. One where a blameless type of design with a single LTP. Then adding servo current to the backside will unbalance the matched pair.
The other where you have a leach type of circuit with dual LTP. The latter will not suffer unbalancing issues from having servo current injected.
I think that there's two ways of looking at current injection. One where a blameless type of design with a single LTP. Then adding servo current to the backside will unbalance the matched pair.
The other where you have a leach type of circuit with dual LTP. The latter will not suffer unbalancing issues from having servo current injected.
Thanks Bob for your very considered reply.
But I'm not asking for debate .. only that you list whatever tests are in the public domain & their results. This is important cos the objectivists AND the Golden Pinnae have banded together to stamp out all trace of these results, heretical to both their faiths.
If you don't attempt an explanation of these heresies, surely not even the DIYaudio Inquisition can censor you for reporting facts?
__________________________
I'm not sure this holds true.
If one allows 0.01% THD20k at full power, ensuring clean overload & unconditional load stability is MUCH simpler .. and for those versed in the art, such a simple amp could easily better 5ppm THD1k at all lower levels too.
___________________________
On a sadder note, it appears that only the original CD issue of Ricky Lee Jones' Flying Cowboys has high dynamic range & crest factors.
Subsequent issues including da supa Fi formats have been sanitized and have nearly 10dB less dynamic range.
Does anyone have concrete info on this? eg compared original & new versions.
If this EVIL thing has happened, I wonder what are the legalities of sharing a copy of the untouched original?
Hi MiiB,
Yes. My world normally has but one LTP, not two opposed.
I'll have to think on your example of the Leach dual LTP. My experience on those normally shows that not only should both transistors in an LTP be matched, but things work better when all four transistors are as closely matched as possible. We would be talking about Adcom GFA-1, SAE and Phase Linear type designs.
When you have everything matched, you can read the THD difference, and you can also hear it. It took me a while to understand why customers with those amps I repaired were asking what else I did to improve the sound. What can I say? I'm a bit slow on the uptake. Anyway, again if the transistors are all matched, DC offset is lower right off. But wouldn't a corrective signal (voltage or current) affect the balance between the two LTPs? This is certainly beyond where I have gone before and I honestly do not know. Obviously, if the transistors are not matched, there is no way to measure or detect whether the DC servo is helping or hurting the situation. No question that having things matched does improve the measured performance and the subjective sound quality (or I would have stopped doing that long ago). In those circuits there are 4 electrolytic capacitors in play. That would probably be Bob's nightmare!
-Chris
Yes. My world normally has but one LTP, not two opposed.
I'll have to think on your example of the Leach dual LTP. My experience on those normally shows that not only should both transistors in an LTP be matched, but things work better when all four transistors are as closely matched as possible. We would be talking about Adcom GFA-1, SAE and Phase Linear type designs.
When you have everything matched, you can read the THD difference, and you can also hear it. It took me a while to understand why customers with those amps I repaired were asking what else I did to improve the sound. What can I say? I'm a bit slow on the uptake. Anyway, again if the transistors are all matched, DC offset is lower right off. But wouldn't a corrective signal (voltage or current) affect the balance between the two LTPs? This is certainly beyond where I have gone before and I honestly do not know. Obviously, if the transistors are not matched, there is no way to measure or detect whether the DC servo is helping or hurting the situation. No question that having things matched does improve the measured performance and the subjective sound quality (or I would have stopped doing that long ago). In those circuits there are 4 electrolytic capacitors in play. That would probably be Bob's nightmare!
-Chris
Using a servo to drive back to the input of the LTP does not really imbalance the LTP.....at least not by any significant amount. The DC open loop gain of the power amp is very high, even if you have say 200mV of offset at the output without servo, if the open loop gain is 2000, you really only drive the LTP 0.1mV out of the original balance. Even if the open loop gain is as low as 200( I hope not), the LTP is only going to be driven 1mV off balance. This is nothing to be concern with.
In the older days where low offset very high speed opamp are not easy to come by, it's a common practice to use a precision opamp to low frequency servo to make a precision high speed opamp. We called that composite opamp at the time. This technique can still be used with the CFB amp today as they are not going to be as low offset as some of the precision VFB opamps. If you want a 1.8G GBWP precision opamp, you might have to resort to that.
Hi Alan,
I'm not convinced - yet.
If you have 200 mV DC offset without the servo, the LTP is already unbalanced as that 200 mV is the best it can do. The 200 mV is either uncorrectable by the LTP, or that DC offset has been designed into the amplifier. It doesn't take that much unbalance to worsen the performance of the diff pair. This was shown in either Doug's or Bob's book and I can't remember which (sorry fellas). Have them both, read them both.
The best way to figure this out will be to take an amplifier, match the LTPs both and hang a differential probe across the feedback and signal sides of the LTPs. Then you can say with confidence whether the amount of unbalance does or does not affect the circuit. This is too interesting to pass up. I will do this at some point in the future.
I am very familiar with the composite op amp idea, and practice. It works well.
-Chris
I'm not convinced - yet.
If you have 200 mV DC offset without the servo, the LTP is already unbalanced as that 200 mV is the best it can do. The 200 mV is either uncorrectable by the LTP, or that DC offset has been designed into the amplifier. It doesn't take that much unbalance to worsen the performance of the diff pair. This was shown in either Doug's or Bob's book and I can't remember which (sorry fellas). Have them both, read them both.
The best way to figure this out will be to take an amplifier, match the LTPs both and hang a differential probe across the feedback and signal sides of the LTPs. Then you can say with confidence whether the amount of unbalance does or does not affect the circuit. This is too interesting to pass up. I will do this at some point in the future.
I am very familiar with the composite op amp idea, and practice. It works well.
-Chris
I posted earlier in this thread about the offset caused by the imbalance of the two current sources of the complementary LTP type IPS. I even posted the simulation. I think in that configuration, there are more than one cause of offset. I am just too lazy to go back 50 pages to look for that.
In my Acurus, I have over 50mV offset, the two LTP are monolithic pairs that the two transistors should match very well (from my knowledge of IC design, 3mV max). So if the amp has a DC gain of unity( evil electrolytic in feed back loop). The offset should be only 3mV at the output. So I don't think you can blame it all on the LTP, depends on the IPS.
There are definitely short comings of the complementary IPS, it might work very well if it's all contained in one piece of silicon where everything is matched. But a discrete complementary IPS is a different story.
If you have a change to actually do some measuring on a complementary IPS, please look at the two current sources, maybe intentionally imbalance them and see whether I am right. I had a hunch, that prompt me to do the simulation, but I did not have a real circuit to play with. Let us know the result.
JMHO
Edit:
I just remember, having a larger emitter degeneration resistors in the LTP made it worst as the the gain of the LTP being reduced. Mr. Cordell suggested the resistor value should be 10 times the r'e, that make the gain of the LTP low and has to be picked after the LTP. that's when the offset problem gets worst when the two current sources are made imbalanced. When I reduced the emitter resistors, the offset got better in the simulation.
In my Acurus, I have over 50mV offset, the two LTP are monolithic pairs that the two transistors should match very well (from my knowledge of IC design, 3mV max). So if the amp has a DC gain of unity( evil electrolytic in feed back loop). The offset should be only 3mV at the output. So I don't think you can blame it all on the LTP, depends on the IPS.
There are definitely short comings of the complementary IPS, it might work very well if it's all contained in one piece of silicon where everything is matched. But a discrete complementary IPS is a different story.
If you have a change to actually do some measuring on a complementary IPS, please look at the two current sources, maybe intentionally imbalance them and see whether I am right. I had a hunch, that prompt me to do the simulation, but I did not have a real circuit to play with. Let us know the result.
JMHO
Edit:
I just remember, having a larger emitter degeneration resistors in the LTP made it worst as the the gain of the LTP being reduced. Mr. Cordell suggested the resistor value should be 10 times the r'e, that make the gain of the LTP low and has to be picked after the LTP. that's when the offset problem gets worst when the two current sources are made imbalanced. When I reduced the emitter resistors, the offset got better in the simulation.
Last edited:
i built a lot of leach super amps on fairly large heat sinks, this my experiences with them...
i installed thermal cut-outs with about 85*C to switch on fans as needed, fins are running vertical...
this was in the mid 80's and now i am more into tubes than SS, but i am build a big one n the coming days...
Either unbalancing the current sources by 2:1 or
a 2:1 degeneration difference of the genders will increase THD by X10 on a symmetric amp.
The amp would be over 300mv at this point, but the servo keeps offset at zero.
At a standard <50mv offset un-servo'ed mismatch , all is nice at <20ppm thd-20.
THD rose to a horrible .01% with a gross unbalance as described above.
This is with a symmetric IPS. The blameless with current mirrors behaves
differently. The mirrors "force" the balance.
OS
a 2:1 degeneration difference of the genders will increase THD by X10 on a symmetric amp.
The amp would be over 300mv at this point, but the servo keeps offset at zero.
At a standard <50mv offset un-servo'ed mismatch , all is nice at <20ppm thd-20.
THD rose to a horrible .01% with a gross unbalance as described above.
This is with a symmetric IPS. The blameless with current mirrors behaves
differently. The mirrors "force" the balance.
OS
Last edited: