Yep. However, I think the best long term solution is to adjust the current sources for optimal offset, and then use the servo to steer them to get the offset to zero over time and temperature. Ideally it will be doing almost nothing almost all of the time.
If your LPT transistors are going out of balance, the distortion will also be rising. A servo hides this, so is undesirable. Besides, this process takes a very, very long time and often things will actually remain in balance.
Call this normal maintenance. You always want the best performance from your equipment within reason. Normally the LPT will remain in balance for 20 years or more. So, why does it need a servo to begin with?
Call this normal maintenance. You always want the best performance from your equipment within reason. Normally the LPT will remain in balance for 20 years or more. So, why does it need a servo to begin with?
I guess one way to find out would be to remove the servo (or just short its two inputs together), and adjust R144 and R145 for zero output offset. Then let the amp run for a while to see.
My skepticism though is based on the fact that the imbalance is a result of the two pairs of diff pairs that have fundamentally different physical behaviors. Do PNP diff pairs change their beta more or less rapidly than NPN pairs over time and/or temp? And, being Darlingtons, any variation in the junction Betas over time and temp will essentially be squared. So, suppose the beta of the NPN Darlington is 10K, and the PNP Darlington is 8K, That means that the beta of each NPN junction is 100, and the beta of each PNP junction is about 90. If the PNP beta changes by 10% over temp, an d the NPN beta changes by 5%, the resulting betas will be: NPN: 9025, PNP: 6561. So, what started as a 20% difference changes to a 37% difference. The question is, is this what actually happens. It's an easy test, so I guess we will soon know!
If this does happen, then the servo would track these variations out. That said, if the component variations ARE small, then once the basic balance is established, then these drifts may be very much second order.
My skepticism though is based on the fact that the imbalance is a result of the two pairs of diff pairs that have fundamentally different physical behaviors. Do PNP diff pairs change their beta more or less rapidly than NPN pairs over time and/or temp? And, being Darlingtons, any variation in the junction Betas over time and temp will essentially be squared. So, suppose the beta of the NPN Darlington is 10K, and the PNP Darlington is 8K, That means that the beta of each NPN junction is 100, and the beta of each PNP junction is about 90. If the PNP beta changes by 10% over temp, an d the NPN beta changes by 5%, the resulting betas will be: NPN: 9025, PNP: 6561. So, what started as a 20% difference changes to a 37% difference. The question is, is this what actually happens. It's an easy test, so I guess we will soon know!
If this does happen, then the servo would track these variations out. That said, if the component variations ARE small, then once the basic balance is established, then these drifts may be very much second order.
Well, Hafler 220's worked fine without servos (for decades). Any amplifier that used a trim pot (SAE or Phase Linear) didn't normally need frequent adjustment. If you matched the parts, amplifiers I saw a couple decades later were still close.
Marantz and many others with a single LPT and trim pot were also pretty stable.
Marantz and many others with a single LPT and trim pot were also pretty stable.
OK, so I spent a bit more time thinking about the differential current mirror, and also discovered a bunch of errors in the earlier diagram.
I was intrigued by the concept of a virtual ground that would then be offset by the servo output (I mean it is sooooo 565, right?).
So, I ditched the circuit that had the resistors in the output collector, and decided to use the servo to manipulate the control transistors for the mirrors. For a while I had a true virtual ground between the two mirrors, but then I realized that if I changed that ground, the op amp would have to sink or source the change in current, and to the extent it could not do that, the positive and negative mirrors would not change in opposite directions. So I had to get rid of (or provide) the excess current independently for each leg. That's where the idea for the criss-crossed emitter legs of Q1 and Q2 came from. By referencing these to the opposing supply rail (instead of ground or virtual ground), the current in each mirror can be controlled independently. To get them to track in a complimentary way, the virtual ground is actually in the base circuits, which are connected to the servo op amp. When the op amp output is zero (virtual ground, the bias in the Q1-Q3 leg will set the nominal current for the negative diff pair, and the current in the Q2/Q4 leg will do so for the positive diff pair.
If the offset is not zero, the bases of Q1 and Q2 will be offset positive or negative, causing one to pass more current and the other to pass less. That change will be mirrored in the current in Q5 and Q6, thereby adjusting one diff pair up and the other down in a complimentary way. I plan to breadboard this next week.
I was intrigued by the concept of a virtual ground that would then be offset by the servo output (I mean it is sooooo 565, right?).
So, I ditched the circuit that had the resistors in the output collector, and decided to use the servo to manipulate the control transistors for the mirrors. For a while I had a true virtual ground between the two mirrors, but then I realized that if I changed that ground, the op amp would have to sink or source the change in current, and to the extent it could not do that, the positive and negative mirrors would not change in opposite directions. So I had to get rid of (or provide) the excess current independently for each leg. That's where the idea for the criss-crossed emitter legs of Q1 and Q2 came from. By referencing these to the opposing supply rail (instead of ground or virtual ground), the current in each mirror can be controlled independently. To get them to track in a complimentary way, the virtual ground is actually in the base circuits, which are connected to the servo op amp. When the op amp output is zero (virtual ground, the bias in the Q1-Q3 leg will set the nominal current for the negative diff pair, and the current in the Q2/Q4 leg will do so for the positive diff pair.
If the offset is not zero, the bases of Q1 and Q2 will be offset positive or negative, causing one to pass more current and the other to pass less. That change will be mirrored in the current in Q5 and Q6, thereby adjusting one diff pair up and the other down in a complimentary way. I plan to breadboard this next week.
Last edited:
What I don't like about op amps determining DC offset in amplifiers without speaker disconnect methods is that power up and down instabilities can cause havoc. Call it experience in seeing what can go wrong after you have a few years on something if you want.
Most transistor LPTs don't do funny things as they come up or down, but they can cause thumps for sure. Anyway, there are so many examples of great power amplifiers that did not use a DC servo that there shouldn't be any question as to whether the servo is even necessary. DC servos shouldn't be required.
Most transistor LPTs don't do funny things as they come up or down, but they can cause thumps for sure. Anyway, there are so many examples of great power amplifiers that did not use a DC servo that there shouldn't be any question as to whether the servo is even necessary. DC servos shouldn't be required.
Couldn't agree more...
That brings me to my headache...how to mod the 545 LTP so that it would be "trimmable"...
Ya learn sumpin'' new every day.
I was perusing the thread wondering what the acronym LTP was. (Long Tail Pair...Duh). In the perusal I found a short but interesting thread that described this circuit element.
Transistor Long Tailed Pair >> Electronics Notes
The Adcom manual describes the input circuit as a differential cascode, but really, the Q103/Q107 and Q104/Q108 transistors are just a current mirror.. scroll down near the bottom of the above threads nd then compare that circuit with the 565 input section
I also note that, for clarity, it is not the LTP that is responsible for any DC offset. It is the fact that there are TWO LTPs, one handling the positive half cycle and on the negative half cycle. To the extent that these are biased at two different collector voltages, the difference will show up, magnified 1000X at the output.
I was perusing the thread wondering what the acronym LTP was. (Long Tail Pair...Duh). In the perusal I found a short but interesting thread that described this circuit element.
Transistor Long Tailed Pair >> Electronics Notes
The Adcom manual describes the input circuit as a differential cascode, but really, the Q103/Q107 and Q104/Q108 transistors are just a current mirror.. scroll down near the bottom of the above threads nd then compare that circuit with the 565 input section
I also note that, for clarity, it is not the LTP that is responsible for any DC offset. It is the fact that there are TWO LTPs, one handling the positive half cycle and on the negative half cycle. To the extent that these are biased at two different collector voltages, the difference will show up, magnified 1000X at the output.
Last edited:
That's what you would call a "project", because you are really redesigning the power amplifier. But really, the design should be stable, balanced, the way it is as long as the LTP is matched carefully.
What I really didn't like about some of the Adcom amplifiers is that they were designed to be unbalanced and require the op amp to work quickly to correct that. I feel that was a massive design error.
The LTP is where the input signal is compared to the output with distortion. The only way the distortion can be effectively cancelled is for the LTP to be perfectly balanced statically. Anything at all that throws the LTP out of balance, such as a DC servo, impairs the ability to extract the difference signal (distortion components) in order to correct it. So in order to make 30 mV of DC offset into 1 mV (for example) you are allowing a higher amount of distortion. So what if you have 30 mV, 50 mV DC offset? That is not harmful to anything. Why trade distortion performance off for something that doesn't matter???
Hi cogeniac,
Yes, the LTP is a very basic circuit concept widely used. When you apply it properly the way it was intended it works extremely well. Yes, the current mirror does increase effective gain considerably, but the actual result depends greatly on the load being linear and high-ish impedance. Using a current source increases the common mode rejection a lot. However some people have used fancier current sources to increase the impedance much higher. We run into the law of diminishing returns quickly, so when I encounter a resistor based "CCS", substituting any current source circuit for the resistor increases the performance about as much as you're going to get unless your active CCS is botched. Interestingly, McIntosh used high supply voltages of 130 VDC to 150 VDC and a higher value resistor to achieve a brute force type of high impedance.
As you have seen, the real critical thing is that the two transistors used in a LTP are both matched and the circuit allows them to be balanced. Only then will the circuit perform as expected.
Yes, the LTP is a very basic circuit concept widely used. When you apply it properly the way it was intended it works extremely well. Yes, the current mirror does increase effective gain considerably, but the actual result depends greatly on the load being linear and high-ish impedance. Using a current source increases the common mode rejection a lot. However some people have used fancier current sources to increase the impedance much higher. We run into the law of diminishing returns quickly, so when I encounter a resistor based "CCS", substituting any current source circuit for the resistor increases the performance about as much as you're going to get unless your active CCS is botched. Interestingly, McIntosh used high supply voltages of 130 VDC to 150 VDC and a higher value resistor to achieve a brute force type of high impedance.
As you have seen, the real critical thing is that the two transistors used in a LTP are both matched and the circuit allows them to be balanced. Only then will the circuit perform as expected.
My suggestion for this was to adjust the bias on the LTP so as to move the collector of the LTP output up or down slightly.
From what I can tell, the 545 basically uses the output DC bias of each previous stage to set the bias of each subsequent stage (together with other components.)
The single ended signal finally gets split through an inverting unity gain stage and applied to the output stages. If the DC level of the two signals feeding each side of the Class B output stages is off, then the outputs will reflect that with a DC offset. I think the VR602 POT in the circuit is supposed to control that.
Prof: have you tried adjusting that pot?
From what I can tell, the 545 basically uses the output DC bias of each previous stage to set the bias of each subsequent stage (together with other components.)
The single ended signal finally gets split through an inverting unity gain stage and applied to the output stages. If the DC level of the two signals feeding each side of the Class B output stages is off, then the outputs will reflect that with a DC offset. I think the VR602 POT in the circuit is supposed to control that.
Prof: have you tried adjusting that pot?
Hi The Prof,
The transistor matcher design can be put together on breadboard (that's how I designed it, and made my first one). Finally I designed a PCB for it, then others did as well. cogeniac designed a PCB that works very well.
That is one critical thing you need to fix. Maybe look up modern matched transistor pairs in one case to see if they will work. Linear Systems makes some wonderful stuff.
The transistor matcher design can be put together on breadboard (that's how I designed it, and made my first one). Finally I designed a PCB for it, then others did as well. cogeniac designed a PCB that works very well.
I used my matcher the other day and got some Darlington pairs within 1 Mv of one another.
I also bought 100 of each for the next round, so I have a good population to choose from.
I found it is easier and faster to just pick a transistor, and then go through all the others to see if any of them match. You can tell pretty quickly if the pair is drifting apart, or converging. If they are drifting apart, just go to the next device. if they are converging, then sometimes they settle out close, and sometimes they go past one another and diverge the other way. If that happens I just move on. So you can avoid all the waiting on every pair, and focus only on the ones that actually match.
Prof; I'd definitely look up the matcher thread. and build up a matcher board. You don't need a real PCB. Just hand wire it.
AnatechL: Did the forum store ever do a buy on that board to sell them?
I also bought 100 of each for the next round, so I have a good population to choose from.
I found it is easier and faster to just pick a transistor, and then go through all the others to see if any of them match. You can tell pretty quickly if the pair is drifting apart, or converging. If they are drifting apart, just go to the next device. if they are converging, then sometimes they settle out close, and sometimes they go past one another and diverge the other way. If that happens I just move on. So you can avoid all the waiting on every pair, and focus only on the ones that actually match.
Prof; I'd definitely look up the matcher thread. and build up a matcher board. You don't need a real PCB. Just hand wire it.
AnatechL: Did the forum store ever do a buy on that board to sell them?
Hi cogeniac,
There didn't seem to be any interest at the time. You did a good job and those that wanted one seemed to be serviced by you just fine.
At this point, I am perfectly fine if you offered your board to them. The design I gave to everyone here, but I think it would be a great service for everyone if the store did offer it. You worked at this as well, and that board is your design. So if they offered you anything to use it, I am totally cool with that. I would simply like to see it easily available to members and maybe some good service techs might pick one up. It would allow them to do better work.
-Chris
There didn't seem to be any interest at the time. You did a good job and those that wanted one seemed to be serviced by you just fine.
At this point, I am perfectly fine if you offered your board to them. The design I gave to everyone here, but I think it would be a great service for everyone if the store did offer it. You worked at this as well, and that board is your design. So if they offered you anything to use it, I am totally cool with that. I would simply like to see it easily available to members and maybe some good service techs might pick one up. It would allow them to do better work.
-Chris