Not sure about the holes (see the edited post above). They have about 1/2 the total area of the bigger holes in the original. I plan to fab the two-sided no-groundplane board first and see how it works. The ground plane one was just for fun.. I suppose a better way to isolate the layers would be to use a grid, At these frequencies it would appear to be a single ground plane, but it would have much less stray capacitance. Probably none of this is necessary. I am used to RF design, and Audio is much more forgiving.
I got involved in this project because I am building a DSP based audio system. I wanted something that would allow me to select HD FLAC files from a hard disk using my iPad, or stream AAC files from iTunes on the iPad, or select an analog (LP) input. The analog gets digitized at 24 bits and 96 KHz, and the 16/44.1 iTunes files get converted to 24/96 in an ASRC (rate converter). The rate converter also does the source selection under direction from the iPad via a Raspberry Pi, or via front panel controls via the RPi. The selected stream goes to an Analog Devices SHARC DSP (from MiniDSP) and that performs digital crossover and parametric equalization. The result will go out to three DACs (I am combining the low frequency channels) and that will then go out to three Adcom 565s to drive my Magnepans and a big sub.
So, there was a lot of circuit design and layout..three different boards (Analog, digital and power supply), plus the RPI.
I'll start another thread on this sometime.
The MiniDSP stuff is very cool since you can use a USB mic, and measure the system in the room, compute EQ filters, and then stuff the coefficients for the filters back into the DSP. I did this already using smaller DSP and a less elegant source selection approach, and it sounds really god. The EQ filters make all the difference.
So, there was a lot of circuit design and layout..three different boards (Analog, digital and power supply), plus the RPI.
I'll start another thread on this sometime.
The MiniDSP stuff is very cool since you can use a USB mic, and measure the system in the room, compute EQ filters, and then stuff the coefficients for the filters back into the DSP. I did this already using smaller DSP and a less elegant source selection approach, and it sounds really god. The EQ filters make all the difference.
Hi Scott,
The reduced ventilation hole area might come back to bite you. Some of those parts run warm. The grid would be much better compared to a ground plain. Don't forget that although the circuit is limited in frequency response, the individual parts aren't.
The 50R or 75R impedance at RF frequencies tends to solve some problems for you that you have to deal with in audio. Beware of all high Z points in the circuit, because that is were the details can whoop your butt pretty soundly. I don't do very much RF work, and then only up to the GPS L1 band. Below that its commercial FM broadcast and 10 MHz reference signals. Digital work tends to be more taking an analogue view to solve contamination issues. That can get pretty dicey.
Anyway, you have a really cool project running there. Our members would love to read about it.
-Chris
The reduced ventilation hole area might come back to bite you. Some of those parts run warm. The grid would be much better compared to a ground plain. Don't forget that although the circuit is limited in frequency response, the individual parts aren't.
The 50R or 75R impedance at RF frequencies tends to solve some problems for you that you have to deal with in audio. Beware of all high Z points in the circuit, because that is were the details can whoop your butt pretty soundly. I don't do very much RF work, and then only up to the GPS L1 band. Below that its commercial FM broadcast and 10 MHz reference signals. Digital work tends to be more taking an analogue view to solve contamination issues. That can get pretty dicey.
Anyway, you have a really cool project running there. Our members would love to read about it.
-Chris
I think I'll probably manually cut out the material between the holes for cooling. Those big heatsinks do dissipate some heat. As you can see in the layout, the heat sinks are slightly different, but essentially about the same total volume. They do make them taller (1.5 inches vs 1 inch) so if heat becomes an issue I may try that.
My biggest worry is the currents on the board. I guess they are not that high (20 mA), but I have been working to get the main current drawing segments (for example the entire chain from the rails to ground for the driver stages) as hefty as possible. They are currently .040, but I'll probably give them a try at .066. Most of the input stage is low current (2-3 mA) , so there the concern is more to avoid excess parasitic inductance.
The digital project was driven by my existing systems. I am using an Adcom GFA 555 to drive the Maggies. It is competent, but having seen the 565, I think that is a much better amp. The sub is powered by my old GFA 535, which is clearly too small for a 12 inch sub...(I have a small treasure trove of old Adcom gear..including the venerable GFP 565, which is very sweet, even 25 years later).
I currently use an Arcam Air DAC (24/44.1 to 24/192), which is OK, but the iPad app is pretty wonky, and then I have to swap back and forth between that and iTunes..As I said to my wife, the stereo currently requires a ride along mechanic..(having been doing audio stuff for 45 years, I am still a little surprised at how things have changed..I never thought I'd be plugging my laptop into a my stereo to set the equalization*.. Just sayin'!)...so I am aiming to solve these connectivity and control issues with a nice single unit that connects to an iPad/tablet/PC/mac via WiFi or ethernet, supports an ethernet or USB attached HDD (I have a 3 TB one now - the 24/96 FLAC files are huge!), and allows manual or iPad based switching between sources including swapping over to the phono input. I am currently using an APT Holman phono preamp. I may clone that phono stage and build it into a future version of this unit. It is sort of a super duper preamp/DAC/DSP with Phono, WiFi, Ethernet, TOS and S/PDIF inputs, and controllable from a tablet, phone or iPad. Fortunately my best friend is a software guy, so we make a great team. It helps having known him for, yikes!, 48 years..He was there when I built a double walled, sand filled, speaker cabinet..that was a mess!
Cheers,
Scott
*My college senior project in 1977 was to build a digital filter.. it was at least as big as two GFA 565s! As I write this, I have a MiniDSP SHARC on my desk. it is a single board about 3"x3" with the biggest non CPU chip I have ever seen (easily 1.25" sq. )..Supports 8K FIR coefficients. I think ours in 1977 had something like 128..yikes! Moore's law in action!
My biggest worry is the currents on the board. I guess they are not that high (20 mA), but I have been working to get the main current drawing segments (for example the entire chain from the rails to ground for the driver stages) as hefty as possible. They are currently .040, but I'll probably give them a try at .066. Most of the input stage is low current (2-3 mA) , so there the concern is more to avoid excess parasitic inductance.
The digital project was driven by my existing systems. I am using an Adcom GFA 555 to drive the Maggies. It is competent, but having seen the 565, I think that is a much better amp. The sub is powered by my old GFA 535, which is clearly too small for a 12 inch sub...(I have a small treasure trove of old Adcom gear..including the venerable GFP 565, which is very sweet, even 25 years later).
I currently use an Arcam Air DAC (24/44.1 to 24/192), which is OK, but the iPad app is pretty wonky, and then I have to swap back and forth between that and iTunes..As I said to my wife, the stereo currently requires a ride along mechanic..(having been doing audio stuff for 45 years, I am still a little surprised at how things have changed..I never thought I'd be plugging my laptop into a my stereo to set the equalization*.. Just sayin'!)...so I am aiming to solve these connectivity and control issues with a nice single unit that connects to an iPad/tablet/PC/mac via WiFi or ethernet, supports an ethernet or USB attached HDD (I have a 3 TB one now - the 24/96 FLAC files are huge!), and allows manual or iPad based switching between sources including swapping over to the phono input. I am currently using an APT Holman phono preamp. I may clone that phono stage and build it into a future version of this unit. It is sort of a super duper preamp/DAC/DSP with Phono, WiFi, Ethernet, TOS and S/PDIF inputs, and controllable from a tablet, phone or iPad. Fortunately my best friend is a software guy, so we make a great team. It helps having known him for, yikes!, 48 years..He was there when I built a double walled, sand filled, speaker cabinet..that was a mess!
Cheers,
Scott
*My college senior project in 1977 was to build a digital filter.. it was at least as big as two GFA 565s! As I write this, I have a MiniDSP SHARC on my desk. it is a single board about 3"x3" with the biggest non CPU chip I have ever seen (easily 1.25" sq. )..Supports 8K FIR coefficients. I think ours in 1977 had something like 128..yikes! Moore's law in action!
Just scored another 565.. This one is advertised and having a "hum"..I know what that means..... 🙂
So with a little electrolyte scrubbing and resoldering (or maybe a control board replacement), my small collection will be complete!
Cheers,
Scott
So with a little electrolyte scrubbing and resoldering (or maybe a control board replacement), my small collection will be complete!
Cheers,
Scott
Last edited:
Well, a nice relaxing afternoon of troubleshooting on amp #2...
I have two observations.
First, the soft start board appears to have some sort of issue. I determined this when I found that the LEDs were barely on. Out of the circuit, the forward voltage on the diodes is about 1.7 volts. In the circuit, it is about 1 volt to 1.5 volts. The really weird thing is that the voltages on the connector seem pretty wonky. With the connector unplugged, pin 3 (the middle pin) is about 40 volts. With the connector attached, this drops to 0.5 volts. This line is supposed to bias on Q119, which turns on the input diff amp current sources, so if it is not switched on all the way, the current sources are starved. I see this in a sort of half clipped signal waveform. As an interim solution, I soldered a 330 ohm resistor (I started with 10K and worked down to that) across pins 2 and 3, and it came on fine. It's a little weird now because without the soft start cutout, when you turn the amp off, the waveform takes about 10 minutes to completely die away as the big caps discharge.
I think either the transistor on the soft start board is bad, or the opto coupler is bad. I'll sort that out tomorrow. That silly little board is really hard to work on-being down in the bottom of the amp...
The second issue is a weird slow start on the positive rail. This may be due to the current source/soft start issue (I didn't want to simply short pins 2 and 3, so Q119 may not be switching all the way on yet). . Basically, when you first turn the amp on, no current flows in the positive side bias diode stack. If you give it a minute, it will eventually come on. I have not tried to sort this out. I am thinking it may be either a funky bridge, or possibly a leaky cap (although I would think that would get pretty hot if it was shunting the supply line. I'll put my DVM on the supply rail next time I turn it on. It may also be some quirky bias issue caused by the current source issue. If Q!!! is biased on full blast, I suppose the
Once it is on, everything seems balanced, but the turn-on issue means that the output is way off until the positive side comes up.
I'll do a little more on this tomorrow.
Cheers,
Scott
I have two observations.
First, the soft start board appears to have some sort of issue. I determined this when I found that the LEDs were barely on. Out of the circuit, the forward voltage on the diodes is about 1.7 volts. In the circuit, it is about 1 volt to 1.5 volts. The really weird thing is that the voltages on the connector seem pretty wonky. With the connector unplugged, pin 3 (the middle pin) is about 40 volts. With the connector attached, this drops to 0.5 volts. This line is supposed to bias on Q119, which turns on the input diff amp current sources, so if it is not switched on all the way, the current sources are starved. I see this in a sort of half clipped signal waveform. As an interim solution, I soldered a 330 ohm resistor (I started with 10K and worked down to that) across pins 2 and 3, and it came on fine. It's a little weird now because without the soft start cutout, when you turn the amp off, the waveform takes about 10 minutes to completely die away as the big caps discharge.
I think either the transistor on the soft start board is bad, or the opto coupler is bad. I'll sort that out tomorrow. That silly little board is really hard to work on-being down in the bottom of the amp...
The second issue is a weird slow start on the positive rail. This may be due to the current source/soft start issue (I didn't want to simply short pins 2 and 3, so Q119 may not be switching all the way on yet). . Basically, when you first turn the amp on, no current flows in the positive side bias diode stack. If you give it a minute, it will eventually come on. I have not tried to sort this out. I am thinking it may be either a funky bridge, or possibly a leaky cap (although I would think that would get pretty hot if it was shunting the supply line. I'll put my DVM on the supply rail next time I turn it on. It may also be some quirky bias issue caused by the current source issue. If Q!!! is biased on full blast, I suppose the
Once it is on, everything seems balanced, but the turn-on issue means that the output is way off until the positive side comes up.
I'll do a little more on this tomorrow.
Cheers,
Scott
This is a long, but hopefully informative post for folks working on these amplifiers.
I tinkered with the soft start board, and decided that it was OK...Just me not really understanding it.. But the parts seem to be working properly..
So, on the #2 amp, I removed the heat sinks, and TO-126 transistors, and cleaned them and the board some more. It is pretty darn clean at this stage. In testing there are no stray voltages anywhere on the board between the races, even at the mV level.
If you really want to follow this post, I suggest you have ready (or download from the first few posts of this thread) the schematic of the amp (I am sure Chris knows this scenic by heart....)
Since it was still not working properly I decided to isolate the positive and negative side amplifiers, and see what did and didn't work. It's very hard to tell when the + and - sides are linked. They are linked through the bias network that comes after the driver stage. Specifically through R139, and C109 (the electrolytic cap in the center of the board).
Removing Q110 and Q109 also allows you to test just the input diff amp stages.
The findings:
The input diff amps work great.
Connecting Q109 and Q110 results in a weird output on the diff amp output (i.e. at the signal sides of R106 and R107), and a similar, but higher level weird output at the collectors of the driver transistors (the big ones with the heatsinks).
I then experimented with my variable current source resistors (which go from 240 ohms to 1240 ohms). I found that by REDUCING the current levels in the diff amps, I could get the outputs to operate properly. Obviously, this is some sort of bias level issue. The only issue is that you have to reduce it a LOT! Normally each leg of the diff amps runs at about 1.5 mA each, so 3 mA for each pair. The current level that works is only about half that.
I then put the 499 ohm current source resistors back in, and started checking the bias diodes (the ones in the diode "stack" - D103, 105 and 107. These are used to set the operating point for Q111 and Q112 (the drivers). What I found here was that, with the current source set properly, I had to put in larger voltage drop diodes, and even then what I ended up with was a clipped, but not totally weird output waveform.
As an additional experiment, I tried the amp with the pre-drivers installed (Q109 and Q110) but with the drivers removed, and a 1K resistor across the emitter base pads of the drivers. This resulted in the weird waveform again, which tells me that there is some weird issue with the bias on These pre-drivers.
To confirm this, I then tried reducing the current level from the diff amp again, and presto! high signal from the collector of Q110 (I only did this on one side of the amp).
Looking at the circuit, if the current draw on the diff amp is higher, then the voltage drop across the 1K collector resistor (R106) will be higher, and the voltage at the base of Q109 will thus be lower, so that device will be on, and close to saturation. Since we know the amp was designed to operate with about 1 mA of diff amp leg current, the drop across R106 SHOULD be about 1 volt. What this REALLY means is that the EMITTER of Q109 is at the wrong voltage, and this points to the diode stack.
How this can occur depends on some slightly not immediately intuitive (to me) observations.
The drop across these diodes is sort of fixed, but in reality it depends on the forward current of the diodes. Remember the diode turns on at some specific voltage (which varies from about 0.2 volts to about 2 volts, depending on the diode type), but the forward current increases exponentially as that voltage increases beyond the turn on voltage. If you loo closely at the diode stack, you will see that it is never referenced to ground. Instead it passes from R125 to D103 to D105, to D107 and then to R131, where it then picks up and goes to R132, and back up the negative side diode stack to R126. This means that any error in any of the diode drops or R131/132 will propagate to the other bias points, on either side of the power rails. There may be a design reason for this, but this approach also seems to be the root cause of the bias instability of the amp (thus requiring the servo), and is also, I suspect a key element of the tendency to difficult to control DC offsets.
More in the next post.
Scott
I tinkered with the soft start board, and decided that it was OK...Just me not really understanding it.. But the parts seem to be working properly..
So, on the #2 amp, I removed the heat sinks, and TO-126 transistors, and cleaned them and the board some more. It is pretty darn clean at this stage. In testing there are no stray voltages anywhere on the board between the races, even at the mV level.
If you really want to follow this post, I suggest you have ready (or download from the first few posts of this thread) the schematic of the amp (I am sure Chris knows this scenic by heart....)
Since it was still not working properly I decided to isolate the positive and negative side amplifiers, and see what did and didn't work. It's very hard to tell when the + and - sides are linked. They are linked through the bias network that comes after the driver stage. Specifically through R139, and C109 (the electrolytic cap in the center of the board).
Removing Q110 and Q109 also allows you to test just the input diff amp stages.
The findings:
The input diff amps work great.
Connecting Q109 and Q110 results in a weird output on the diff amp output (i.e. at the signal sides of R106 and R107), and a similar, but higher level weird output at the collectors of the driver transistors (the big ones with the heatsinks).
I then experimented with my variable current source resistors (which go from 240 ohms to 1240 ohms). I found that by REDUCING the current levels in the diff amps, I could get the outputs to operate properly. Obviously, this is some sort of bias level issue. The only issue is that you have to reduce it a LOT! Normally each leg of the diff amps runs at about 1.5 mA each, so 3 mA for each pair. The current level that works is only about half that.
I then put the 499 ohm current source resistors back in, and started checking the bias diodes (the ones in the diode "stack" - D103, 105 and 107. These are used to set the operating point for Q111 and Q112 (the drivers). What I found here was that, with the current source set properly, I had to put in larger voltage drop diodes, and even then what I ended up with was a clipped, but not totally weird output waveform.
As an additional experiment, I tried the amp with the pre-drivers installed (Q109 and Q110) but with the drivers removed, and a 1K resistor across the emitter base pads of the drivers. This resulted in the weird waveform again, which tells me that there is some weird issue with the bias on These pre-drivers.
To confirm this, I then tried reducing the current level from the diff amp again, and presto! high signal from the collector of Q110 (I only did this on one side of the amp).
Looking at the circuit, if the current draw on the diff amp is higher, then the voltage drop across the 1K collector resistor (R106) will be higher, and the voltage at the base of Q109 will thus be lower, so that device will be on, and close to saturation. Since we know the amp was designed to operate with about 1 mA of diff amp leg current, the drop across R106 SHOULD be about 1 volt. What this REALLY means is that the EMITTER of Q109 is at the wrong voltage, and this points to the diode stack.
How this can occur depends on some slightly not immediately intuitive (to me) observations.
The drop across these diodes is sort of fixed, but in reality it depends on the forward current of the diodes. Remember the diode turns on at some specific voltage (which varies from about 0.2 volts to about 2 volts, depending on the diode type), but the forward current increases exponentially as that voltage increases beyond the turn on voltage. If you loo closely at the diode stack, you will see that it is never referenced to ground. Instead it passes from R125 to D103 to D105, to D107 and then to R131, where it then picks up and goes to R132, and back up the negative side diode stack to R126. This means that any error in any of the diode drops or R131/132 will propagate to the other bias points, on either side of the power rails. There may be a design reason for this, but this approach also seems to be the root cause of the bias instability of the amp (thus requiring the servo), and is also, I suspect a key element of the tendency to difficult to control DC offsets.
More in the next post.
Scott
Continuing:
The operation of Q109 (and Q110) is a little hard to follow. To understand it, you have to look at how the diff amp generates the signal swing at the base of Q109 (or at the non-rail side of the load resistor R106).
When the output of the Diff amp is at its peak, the current through the conducting leg of the diff amp pair is at its minimum (the diff amp swaps the constant current back and forth between the two mirrored halves of the diff pair over the course of the signal cycle). So, this means the voltage drop across R106 is at its minimum, which means it is as close to the rail as it is designed to be. So, with the constant current sources properly set, the voltage when the Diff amp output clips at the load resistor is the voltage where Q109 should be just past its turn-on point. A transistor is pretty much like a diode in this respect, so, figure on about 1 volt across the emitter base junction at turn-on.
This means that whatever the voltage is at the emitter of Q109 at this point (i.e about 1 volt ABOVE the clipping point at R106) sets the the minimum drop across R125 (the 49.9 ohm emitter resistor at Q109). The drop will get much greater as Q109 and Q111 turn on, since they will draw a lot more current through R125 than the diode stack. I am not sure how the fact that the diode stack voltage will vary, and this will change the operating points of of Q109/111 affects things.
It is either a design feature involving some sort of temp compensating feedback, or it is a source of some small non-linearity. I'll explore this sometime...
Since Q109 is right at cutoff, the emitter current will be nearly zero, so, and this means most of this current is going down the diode stack at this point in the cycle. Since the voltages will only go down from here, this is the maximum current we will expect to see in the diode stack. We can calculate this current, sort of, by assuming some drops for the diodes, and ultimately figuring out what voltage is applied across the R131/R132 resistor pair. If we actually have the current voltage curves for the diodes (which we can get), then we can work our way back up the stack, and correct those drops, and then recompute the current through the resistors, and eventually arrive at a set of drops at some fixed maximum current, and then test the diodes for this.
I'll to this tomorrow.
The next element is to basically reverse engineer the bias design for the driver stage. This is a combination of determining the currents through R125/126 at the turn-on points for Q190/Q110, and making sure the maximum current through the diode stack fits that (plus a little). We can then figure out how the diodes are performing relative to what is expected of them.
This is for the next post an a day or so.
Cheers.
Scott
The operation of Q109 (and Q110) is a little hard to follow. To understand it, you have to look at how the diff amp generates the signal swing at the base of Q109 (or at the non-rail side of the load resistor R106).
When the output of the Diff amp is at its peak, the current through the conducting leg of the diff amp pair is at its minimum (the diff amp swaps the constant current back and forth between the two mirrored halves of the diff pair over the course of the signal cycle). So, this means the voltage drop across R106 is at its minimum, which means it is as close to the rail as it is designed to be. So, with the constant current sources properly set, the voltage when the Diff amp output clips at the load resistor is the voltage where Q109 should be just past its turn-on point. A transistor is pretty much like a diode in this respect, so, figure on about 1 volt across the emitter base junction at turn-on.
This means that whatever the voltage is at the emitter of Q109 at this point (i.e about 1 volt ABOVE the clipping point at R106) sets the the minimum drop across R125 (the 49.9 ohm emitter resistor at Q109). The drop will get much greater as Q109 and Q111 turn on, since they will draw a lot more current through R125 than the diode stack. I am not sure how the fact that the diode stack voltage will vary, and this will change the operating points of of Q109/111 affects things.
It is either a design feature involving some sort of temp compensating feedback, or it is a source of some small non-linearity. I'll explore this sometime...
Since Q109 is right at cutoff, the emitter current will be nearly zero, so, and this means most of this current is going down the diode stack at this point in the cycle. Since the voltages will only go down from here, this is the maximum current we will expect to see in the diode stack. We can calculate this current, sort of, by assuming some drops for the diodes, and ultimately figuring out what voltage is applied across the R131/R132 resistor pair. If we actually have the current voltage curves for the diodes (which we can get), then we can work our way back up the stack, and correct those drops, and then recompute the current through the resistors, and eventually arrive at a set of drops at some fixed maximum current, and then test the diodes for this.
I'll to this tomorrow.
The next element is to basically reverse engineer the bias design for the driver stage. This is a combination of determining the currents through R125/126 at the turn-on points for Q190/Q110, and making sure the maximum current through the diode stack fits that (plus a little). We can then figure out how the diodes are performing relative to what is expected of them.
This is for the next post an a day or so.
Cheers.
Scott
OK, so I modeled the diodes form the data sheets, and did an iterative computation to determine the currents and voltages we should see in the stack at various points in the signal cycle.
Point A: Q111 near cutoff, Q112 near saturation:
Point A: Q111 near saturation, Q112 near cutoff:
Point A: Q111 and Q112 at mid bias (for Class A, this is what we should see with zero input signal.
Here is the basic Diode stack iterative analysis, done assuming no current through Q109 or Q110. I started not knowing the drops across R125/R126, and guessing art the drops across the diodes based on their published specs. I then computed the current that should have flowed through that circuit, and using that, and the diode models, competed new values for the diode drops and the drops across R125/R126.
Using those values, I then recalculated the current. Noe that there is an ambiguity in the voltage value between R131 and R132. If everything is correct, these should be the same. Using the new current value, I then recomputed the diode and resistor drops, and did this a couple of times. YOU can see that in the 3rd and 4th iterations, there is hardly any change. So the values in the 4th iteration column appear to represent what we should expect to see with no conduction in Q109 and Q110 (an obviously synthetic case for a class A amp, but useful later...).
Now, in the real world, the two ends of the amp are working in opposite phase. This means Q109 and Q111 are near cutoff when Q110 and Q112 are nearly fully conducting. TO get a feel for the behavior of the bias circuit in this case I computed the current and drops across the signal chain (starting below with Q110/Q112). I se the target collector voltage for Q112 at -80 volts (See couple of posts above for this). Based on that, and the published Vce-sat voltages (about 0.15 volts) for these devices, I determined that the voltage at R126 shod be about -80.3 volts. This set the current across R126, and also changes the voltages for the negative side of the diode stack.
This is shown below below. Here you can see that the current though the diode stack has changed very slightly (2.03 mA vs 2.2 mA), but this is too small a change to change the voltage drops. So, using this new voltage at D104, we can update all the voltage values for the negative side stack, and we find, interestingly, that the junction between R131 and R132 has a slight voltage change. Form zero to about 2.3 volts. I had originally wondered why the designers had not just set this junction to ground. Now I know.. They didn't want to unbalance the currents in the negative and positive side diode stacks..so this junction floats, and over the signal cycle it is allowed to move up and down.
And here is the corresponding table for the positive peak. You can see they are symmetrical.
And finally, I worked out what we should see when the signal is at zero (either in the midpoint of the cycle, or with no signal applied). Remember, this section of the amp is operating in pure Class A, which means the zero signal bias point needs to be about midday between the two peaks (which was why I bothered to determine what those peaks looked like). Setting the zero point collector voltage ±40 volts, I then used half the signal leg current for the peak values (either of the previous tables), and was able to determine the drops across R125 and R126. I could then work backwards to update the diode table once again. So these diode drops and voltages in the circuit are, to the best of my ability to determine them, the voltages we should see in a properly operating GFA-565 with zero signal applied...
Now, it's back to the lab to see his my amp compares with these..
Cheers,
Scott
Point A: Q111 near cutoff, Q112 near saturation:
Point A: Q111 near saturation, Q112 near cutoff:
Point A: Q111 and Q112 at mid bias (for Class A, this is what we should see with zero input signal.
Here is the basic Diode stack iterative analysis, done assuming no current through Q109 or Q110. I started not knowing the drops across R125/R126, and guessing art the drops across the diodes based on their published specs. I then computed the current that should have flowed through that circuit, and using that, and the diode models, competed new values for the diode drops and the drops across R125/R126.
Using those values, I then recalculated the current. Noe that there is an ambiguity in the voltage value between R131 and R132. If everything is correct, these should be the same. Using the new current value, I then recomputed the diode and resistor drops, and did this a couple of times. YOU can see that in the 3rd and 4th iterations, there is hardly any change. So the values in the 4th iteration column appear to represent what we should expect to see with no conduction in Q109 and Q110 (an obviously synthetic case for a class A amp, but useful later...).
Now, in the real world, the two ends of the amp are working in opposite phase. This means Q109 and Q111 are near cutoff when Q110 and Q112 are nearly fully conducting. TO get a feel for the behavior of the bias circuit in this case I computed the current and drops across the signal chain (starting below with Q110/Q112). I se the target collector voltage for Q112 at -80 volts (See couple of posts above for this). Based on that, and the published Vce-sat voltages (about 0.15 volts) for these devices, I determined that the voltage at R126 shod be about -80.3 volts. This set the current across R126, and also changes the voltages for the negative side of the diode stack.
This is shown below below. Here you can see that the current though the diode stack has changed very slightly (2.03 mA vs 2.2 mA), but this is too small a change to change the voltage drops. So, using this new voltage at D104, we can update all the voltage values for the negative side stack, and we find, interestingly, that the junction between R131 and R132 has a slight voltage change. Form zero to about 2.3 volts. I had originally wondered why the designers had not just set this junction to ground. Now I know.. They didn't want to unbalance the currents in the negative and positive side diode stacks..so this junction floats, and over the signal cycle it is allowed to move up and down.
And here is the corresponding table for the positive peak. You can see they are symmetrical.
And finally, I worked out what we should see when the signal is at zero (either in the midpoint of the cycle, or with no signal applied). Remember, this section of the amp is operating in pure Class A, which means the zero signal bias point needs to be about midday between the two peaks (which was why I bothered to determine what those peaks looked like). Setting the zero point collector voltage ±40 volts, I then used half the signal leg current for the peak values (either of the previous tables), and was able to determine the drops across R125 and R126. I could then work backwards to update the diode table once again. So these diode drops and voltages in the circuit are, to the best of my ability to determine them, the voltages we should see in a properly operating GFA-565 with zero signal applied...
Now, it's back to the lab to see his my amp compares with these..
Cheers,
Scott
Last edited:
Well, that was effective!
Got my #2 amp working perfectly today.
Some useful observations for troubleshooting these amps:
If there is a large output offset, it is probably best to not stress the output stages, so I disconnect the output wires (3, and 4), and shunt those pads with a pair of 330 ohm resistors (across pads 3 and 4). This provides a return path for the currents from the driver output. Using a pair, connected together, you can round the junction between them, and isolate both sides of the amp, or you can let the junction float, which is how the amp actually operates.
I disconnect R139, which is a 280 ohm resistor that ties the two driver outputs together, and I remove C109 (the 220 uF cap in the middle of the board). The cap is only a 25 volt device, and if the driver outputs are whacked, the DC across this device can get quite high (as high as 160 volts!). I noticed the one on my #2 amp bulging slightly while I was troubleshooting it, and, not wanting to get sprayed in the face by an exploding cap, I decided to take it out until I had the outputs stable.
Since the outputs are disconnected, the feedback doesn't work, so when you drive the amp with an input signal, you need to attenuate it about 10:1 from the normal 2 volt line level.
With the two sides disconnected except by the output resistors, you can then see what the signal levels and voltages are on each side without them heavily corrupting each other, and this is very helpful in figuring out what is wrong.
As an example, today I did this so I could compare the bias voltages in my amp to the ones I had worked out a few posts above. I found that, at zero input, and with the junction of the output resistors grounded, the voltage at the collector of Q111 (the positive side driver transistor) was about 10 volts, but the voltage at the collector of Q112 (the negative side driver transistor) was -84 volts. However, inexplicably the voltage at pad 4 (the negative side output was zero. The connection between these is R136, a fusible 82 ohm resistor. Sure enough, it was open (I found nearly every fusible resistor in the #2 amp was open). As soon as I replaced that, the voltage at the collector of Q112 went to -14 volts..
I then replaced R139, and tested it with an input signal. It was prefect, so I hooked up the feedback and the output wires, and tested it with the output stages. It had good output signal, but it was offset. I then remembered that I had the servo Op Amp out, so I put that back in, and tried it. Perfect.
One other observation on these amps is that the current sources are not super well matched. This is unfortunate since any mismatch shows up as a DC offset, and that has tone swamped out by the servo. In the above case (where I had some DC offset until I installed the Op Amp) the positive side was being driven with 1.5 mA, while the negative side was cooking along at 1.8 mA. This may not seem like a big difference, but this means that there is 0.3 volts of DC offset from one side to the other at the driver inputs. The drivers have HUGE gain, so this small DC offset gets amplified significantly. it seems to me a better design would have some sort of compensation to make the two input stage currents the same, or it would include provision for adjusting this parameter.
I tried this by putting one of my precision potentiometers in place of one of the 499 ohm resistors in the current source (R144/R145), and was able to easily adjust the offset to zero with no servo at all. I may think about this a bit, and see if I can devise a better current source that is better matched, temp stable and has fine adjustability. If I do, I'll build that into my new board layout.
On to #3!
Cheers, Scott
Got my #2 amp working perfectly today.
Some useful observations for troubleshooting these amps:
If there is a large output offset, it is probably best to not stress the output stages, so I disconnect the output wires (3, and 4), and shunt those pads with a pair of 330 ohm resistors (across pads 3 and 4). This provides a return path for the currents from the driver output. Using a pair, connected together, you can round the junction between them, and isolate both sides of the amp, or you can let the junction float, which is how the amp actually operates.
I disconnect R139, which is a 280 ohm resistor that ties the two driver outputs together, and I remove C109 (the 220 uF cap in the middle of the board). The cap is only a 25 volt device, and if the driver outputs are whacked, the DC across this device can get quite high (as high as 160 volts!). I noticed the one on my #2 amp bulging slightly while I was troubleshooting it, and, not wanting to get sprayed in the face by an exploding cap, I decided to take it out until I had the outputs stable.
Since the outputs are disconnected, the feedback doesn't work, so when you drive the amp with an input signal, you need to attenuate it about 10:1 from the normal 2 volt line level.
With the two sides disconnected except by the output resistors, you can then see what the signal levels and voltages are on each side without them heavily corrupting each other, and this is very helpful in figuring out what is wrong.
As an example, today I did this so I could compare the bias voltages in my amp to the ones I had worked out a few posts above. I found that, at zero input, and with the junction of the output resistors grounded, the voltage at the collector of Q111 (the positive side driver transistor) was about 10 volts, but the voltage at the collector of Q112 (the negative side driver transistor) was -84 volts. However, inexplicably the voltage at pad 4 (the negative side output was zero. The connection between these is R136, a fusible 82 ohm resistor. Sure enough, it was open (I found nearly every fusible resistor in the #2 amp was open). As soon as I replaced that, the voltage at the collector of Q112 went to -14 volts..
I then replaced R139, and tested it with an input signal. It was prefect, so I hooked up the feedback and the output wires, and tested it with the output stages. It had good output signal, but it was offset. I then remembered that I had the servo Op Amp out, so I put that back in, and tried it. Perfect.
One other observation on these amps is that the current sources are not super well matched. This is unfortunate since any mismatch shows up as a DC offset, and that has tone swamped out by the servo. In the above case (where I had some DC offset until I installed the Op Amp) the positive side was being driven with 1.5 mA, while the negative side was cooking along at 1.8 mA. This may not seem like a big difference, but this means that there is 0.3 volts of DC offset from one side to the other at the driver inputs. The drivers have HUGE gain, so this small DC offset gets amplified significantly. it seems to me a better design would have some sort of compensation to make the two input stage currents the same, or it would include provision for adjusting this parameter.
I tried this by putting one of my precision potentiometers in place of one of the 499 ohm resistors in the current source (R144/R145), and was able to easily adjust the offset to zero with no servo at all. I may think about this a bit, and see if I can devise a better current source that is better matched, temp stable and has fine adjustability. If I do, I'll build that into my new board layout.
On to #3!
Cheers, Scott
So, today I unpacked my #3 amp, and tried it out.
It doesn't do anything bad when powered on. It has a nice clean high level output signal, but it also has about 50 Volt DC offset...
So, I decided to open it up.
It was a dusty dirty mess. Clearly never been opened up before, which is probably a good thing, given the chaos my #1 amp had seen.
Here are a few photos of it, just opened.
I then removed the control board, and found, what I suspected.. Corrosion from one leaky cap, and two other caps on the verge of leaking...
Here is the bottom of the board. You can see the corrosion on the upper left corner
The other two caps were wet and goopy on the bottoms, and left wet goopy spots on the board after I removed them.
This amp actually works better than either of the other two did when I first got them, and the corrosion isn't nearly as bad as the #2 amp, so I am pretty sure this will be a relatively easy fix.
It doesn't do anything bad when powered on. It has a nice clean high level output signal, but it also has about 50 Volt DC offset...
So, I decided to open it up.
It was a dusty dirty mess. Clearly never been opened up before, which is probably a good thing, given the chaos my #1 amp had seen.
Here are a few photos of it, just opened.
I then removed the control board, and found, what I suspected.. Corrosion from one leaky cap, and two other caps on the verge of leaking...
Here is the bottom of the board. You can see the corrosion on the upper left corner
The other two caps were wet and goopy on the bottoms, and left wet goopy spots on the board after I removed them.
This amp actually works better than either of the other two did when I first got them, and the corrosion isn't nearly as bad as the #2 amp, so I am pretty sure this will be a relatively easy fix.
Snagged yet another GFA-565 on eBay yesterday. Cheap!
So now I have four of these. I'll probably give one to my son for his sub. The reason I got #4 is that, after seeing #2 and #3, I realized just how hacked up #1 was..(cut board traces, etc. ).
I patched it up, replaced things like the power cord grommet and speaker binding posts, so it is in better shape, but it is still a little punky.
I now have #3 working through the control board output stage. Seems to be pretty stable. The negative side is a little hotter than the positive side, so I am currently running it with a pot in the current source, so I can adjust things. Getting the bias matched on both sides of these amps is critical. I think I may see if I can improve on the design with some sort of better direct bias compensation. The problem is that when the bias goes wonky the output rails, and that's a recipe for speaker disaster...
The servo is supposed to handle this, but as Anatech pointed out, it seems a shame to rely on feedback to keep the overall amp from blowing up. Better would be a bias network that can be adjusted to set everything properly, and that has internal bias feedback to keep things aligned as parts heat up and/or age.
I'll ponder that..
Cheers,
Scott
So now I have four of these. I'll probably give one to my son for his sub. The reason I got #4 is that, after seeing #2 and #3, I realized just how hacked up #1 was..(cut board traces, etc. ).
I patched it up, replaced things like the power cord grommet and speaker binding posts, so it is in better shape, but it is still a little punky.
I now have #3 working through the control board output stage. Seems to be pretty stable. The negative side is a little hotter than the positive side, so I am currently running it with a pot in the current source, so I can adjust things. Getting the bias matched on both sides of these amps is critical. I think I may see if I can improve on the design with some sort of better direct bias compensation. The problem is that when the bias goes wonky the output rails, and that's a recipe for speaker disaster...
The servo is supposed to handle this, but as Anatech pointed out, it seems a shame to rely on feedback to keep the overall amp from blowing up. Better would be a bias network that can be adjusted to set everything properly, and that has internal bias feedback to keep things aligned as parts heat up and/or age.
I'll ponder that..
Cheers,
Scott
So, I got the control board for #3 working almost perfectly. The only thing I am not happy with is that the output seems to wobble around a bit, clipping one way and then the other until it settles down. I am wondering if the other caps on the board may be unstable or something..Sort of a slow and unstable warmup, after which it performs well.
The other issue I found was that, with otherwise perfect outputs from the control board, the actually power output was a mess. Some testing showed that the positive side of the output was inoperative.
These output boards are a PITA to get to. I unscrewed the heatsink, and managed to get it free enough to probe the parts. What I found was that from the emitter of Q202 to the emitter of Q203 I showed 68 ohms. it is supposed to be 60, but that would be unlikely to render the output inoperative. I then found that the resistance from the emitter of Q203 to the output was about 860 ohms. It is supposed to be 7.5 ohms. Removing R202 confirmed that this resistor was about 2.2 megohms.
This is yet another fusible resistor, which clearly has fused.. I have found many of these devices dead in the various instances of these amps. I think I'll replace it with a 10 ohm resistor and see if the output stage works, and if so, I'll then replace it again wight he proper device.
Onward!!
Scott
The other issue I found was that, with otherwise perfect outputs from the control board, the actually power output was a mess. Some testing showed that the positive side of the output was inoperative.
These output boards are a PITA to get to. I unscrewed the heatsink, and managed to get it free enough to probe the parts. What I found was that from the emitter of Q202 to the emitter of Q203 I showed 68 ohms. it is supposed to be 60, but that would be unlikely to render the output inoperative. I then found that the resistance from the emitter of Q203 to the output was about 860 ohms. It is supposed to be 7.5 ohms. Removing R202 confirmed that this resistor was about 2.2 megohms.
This is yet another fusible resistor, which clearly has fused.. I have found many of these devices dead in the various instances of these amps. I think I'll replace it with a 10 ohm resistor and see if the output stage works, and if so, I'll then replace it again wight he proper device.
Onward!!
Scott
On replacing the resistor above I found that the output worked. for about 30 seconds until the resistor fried.. I replaced the resistor with a higher wattage device, and the amp worked again, until the 10 watt resistor died.. I touched it and burned my finger.. SO, I would hazard a guess that there is too much current flowing in that part of the circuit! (yeah, brilliant deduction, I know)...
I removed Q203 (man, that is a big transistor!) and found, as suspected, it showed 8 ohms of resistance base to collector..making for about 7 amps flowing through the chain of Q203, R202, and the load resistor.
This is the first of these amplifiers I have seen with problems in the output stage...
SO, giant transistors are on order, and now I wait.
S
I removed Q203 (man, that is a big transistor!) and found, as suspected, it showed 8 ohms of resistance base to collector..making for about 7 amps flowing through the chain of Q203, R202, and the load resistor.
This is the first of these amplifiers I have seen with problems in the output stage...
SO, giant transistors are on order, and now I wait.
S
Last edited:
OK, so I installed the big driver transistor on the output board. I fired up the amp,and everything seemed OK. I had a low signal running at the input, and the output was showing a low signal centered at zero volts offset. Every thing seemed good. I then turned up the input signal, and Poof! the 7.5 ohm emitter resistor (R202) literally went up in flames.
When the smoke cleared, I found, after some probing, that all of the base resistors on the output board are open.
I am not quite sure what might have caused that (all of the transistors check out properly), but looking at the circuit, the flamed R202 makes sense. If the output transistors are disconnected (Because their base resistors are open), then the only thing driving the output load is Q203 (the big transistor I replaced) through the 7.5 ohm emitter resistor. When the signal is low, this transistor can drive the load (I saw this on the scope). But, when you turn up the volume, the driver is trying to put about 80 volts peak across the load resistor (which is a big 350 watt monster, and this little 1/4 watt 7.5 ohm emitter resistor.
Doing the math, if we have 80 volts across 15.5 ohms (8 ohms for the load, and 7.5 for the emitter resistor, we are pulling about 5 amps through this little 1/4 watt resistor. Under normal operation, the load on this is rather small, because all the current is being generated by the output transistor bank, and this raises the voltage at the load resistor, and the voltage drop across the 7.5 ohm resistor is just the voltage drop across the base resistors and the base emitter junctions in the output.
Said another way, if the base resistors are open, then the full signal voltage across the driver emitter resistor is about 40 volts. 40 volts over 7.5 ohms is about 5 amps.
So, tomorrow, I'll clean up the fire residue, and replace all the base resistors.
Cheers,
Scott
When the smoke cleared, I found, after some probing, that all of the base resistors on the output board are open.
I am not quite sure what might have caused that (all of the transistors check out properly), but looking at the circuit, the flamed R202 makes sense. If the output transistors are disconnected (Because their base resistors are open), then the only thing driving the output load is Q203 (the big transistor I replaced) through the 7.5 ohm emitter resistor. When the signal is low, this transistor can drive the load (I saw this on the scope). But, when you turn up the volume, the driver is trying to put about 80 volts peak across the load resistor (which is a big 350 watt monster, and this little 1/4 watt 7.5 ohm emitter resistor.
Doing the math, if we have 80 volts across 15.5 ohms (8 ohms for the load, and 7.5 for the emitter resistor, we are pulling about 5 amps through this little 1/4 watt resistor. Under normal operation, the load on this is rather small, because all the current is being generated by the output transistor bank, and this raises the voltage at the load resistor, and the voltage drop across the 7.5 ohm resistor is just the voltage drop across the base resistors and the base emitter junctions in the output.
Said another way, if the base resistors are open, then the full signal voltage across the driver emitter resistor is about 40 volts. 40 volts over 7.5 ohms is about 5 amps.
So, tomorrow, I'll clean up the fire residue, and replace all the base resistors.
Cheers,
Scott
Last edited:
OK, I replaced all of the resistors on the positive side output board.
For some reason just as I did this, the soft start board relay appears to have failed.. Sigh...
So, I jumpered across that and carefully applied power. Another instant zap and puff of smoke!
It turned out that in my soldering of these resistors (which are very hard to access between the output boards), I managed to bridge a circuit trace. This basically shorted the emitter and base of Q203, which unceremoniously popped when I applied power. I found and fixed the short, replaced the transistor, and voila!
It works!
Generates about 350 watts of output power, clean signal with no oscillations or other observable distortion. The output is centered on zero volts, and the signal clips symmetrically at full power.
Now I need to remove and clean and troubleshoot the soft start board. I suspect the relay has failed. I found that the big 4 ohm 20 watt resistor had failed. I replaced it, and the new one got very hot but the relay didn't switch. I determined that the 4 ohm resistor gets 120 VAC across it at startup, just for a moment until the relay effectively shorts it out. If the relay doesn't close, the resistor just cooks until it fails.
So off on a new troubleshooting adventure! At least this one doesn't involve cleaning electrolyte!
Cheers,
Scott
For some reason just as I did this, the soft start board relay appears to have failed.. Sigh...
So, I jumpered across that and carefully applied power. Another instant zap and puff of smoke!
It turned out that in my soldering of these resistors (which are very hard to access between the output boards), I managed to bridge a circuit trace. This basically shorted the emitter and base of Q203, which unceremoniously popped when I applied power. I found and fixed the short, replaced the transistor, and voila!
It works!
Generates about 350 watts of output power, clean signal with no oscillations or other observable distortion. The output is centered on zero volts, and the signal clips symmetrically at full power.
Now I need to remove and clean and troubleshoot the soft start board. I suspect the relay has failed. I found that the big 4 ohm 20 watt resistor had failed. I replaced it, and the new one got very hot but the relay didn't switch. I determined that the 4 ohm resistor gets 120 VAC across it at startup, just for a moment until the relay effectively shorts it out. If the relay doesn't close, the resistor just cooks until it fails.
So off on a new troubleshooting adventure! At least this one doesn't involve cleaning electrolyte!
Cheers,
Scott
Hi Scott,
If you ever have one shorted output transistor, you have to replace them all, and the driver transistors. No ifs, ands or buts. We are talking about on one channel.
Why?
Because all the opposing transistors were operated beyond their maximum ratings. Changes have most likely occurred in those. So, how about the same bank as the shorted one? They go because those transistors were also operated beyond absolute maximum too. The transistor that shorted merely died before the others. It should also be easy to understand that the drivers also operated beyond reasonable limits (reasonable limits being abs max), so they go too.
The damage the other parts sustained may range from failing later on to being very non-linear compared to the usual characteristics. The beta can change, normally down. Leakage can certainly increase and they can become noisy too. This matters more in the signal stages. If you had a DC offset (you did), then the diff pair transistors broke down reverse E-B. That tends to make the victim transistor noisy with reduced beta. In other words, it is no longer a matched pair of low noise transistors.
The general rule is to replace semiconductors until you find a good stage, then replace that as well. Doing that should keep your work safe and sound.
-Chris
If you ever have one shorted output transistor, you have to replace them all, and the driver transistors. No ifs, ands or buts. We are talking about on one channel.
Why?
Because all the opposing transistors were operated beyond their maximum ratings. Changes have most likely occurred in those. So, how about the same bank as the shorted one? They go because those transistors were also operated beyond absolute maximum too. The transistor that shorted merely died before the others. It should also be easy to understand that the drivers also operated beyond reasonable limits (reasonable limits being abs max), so they go too.
The damage the other parts sustained may range from failing later on to being very non-linear compared to the usual characteristics. The beta can change, normally down. Leakage can certainly increase and they can become noisy too. This matters more in the signal stages. If you had a DC offset (you did), then the diff pair transistors broke down reverse E-B. That tends to make the victim transistor noisy with reduced beta. In other words, it is no longer a matched pair of low noise transistors.
The general rule is to replace semiconductors until you find a good stage, then replace that as well. Doing that should keep your work safe and sound.
-Chris
All of the outputs appear to be working fine.
I think the issue was with a hard offset, the bases of the output bank were overdriven, and that caused all of the base resistors to fail. Once they failed, then the emitter resistor in the second stage of the output failed.
When I replaced that, the 2nd stage output transistor (Q203) failed (so clearly it had been stressed). There is no evidence that the output transistors failed or were overstressed.
However, having fixed the source of the offset, and fixed the output base resistors, the amp appears to be working properly.
The failure of the 2nd stage transistor was due to the fact that the entire output current was being passed through that one transistor. The output devices were protected by the open base resistors.. I note that is didn't fail at turn-on, but failed when I increased the input signal level.
So, I honestly think the system is OK as is.
It is also important to note that by my observations, high output offsets are not caused by heavy bias issues at the Darlington inputs. Rather, small offsets in the bias currents (a few hundred micro-Amps) can generate offsets at the driver stage that will rail the outputs. These input offsets are not high enough to damage the input transistors, but they result in DC biases in the Q109/Q111 and Q110/Q112 stages that will produce substantial DC offsets at the outputs.
It is precisely for this reason that I am working on a balanced current source. The original design uses two "related" current sources, but the fact that they are actually only related means that there can be significant differences in the bias currents in the two diff pairs. To the extent that the currents through the LEDS differ, the tail currents in the diff pairs will also be different. Look at pretty much any 565 and the LEDs will be glowing with different brightnesses, indicating different currents, and therefore different forward voltage drops, and thus different tail currents for the positive and negative diff pairs. And a built-in offset for the servo to correct.
I have observed these differences at the 500 uAmp level, and that relates to about 0.5 volts difference in input bias at the driver stages. The drivers have very high gain (about 20-25 dB), so a 0.5 volt offset in the input bias will result in a 50+ volt offset at the output (this is also why the servo is usually effective in compensating for these differences - since it requires small input offset to adjust it out).
A much better solution is to create a bi-polar current source where the two legs have currents derived from the same voltage source, so if one changes the other changes. Significant changes will then move the operating points of both banks together, so you may get some clipping at higher power levels, but you should never get an offset between the outputs.
You are actually the person who inspired this idea, since you noted once that it is a shame that the Adcom folks relied on a servo to control this issue. I agree, and I am now working on a design mod to eliminate this problem. I'll name it the "Anatech Upgrade"... 🙂
Cheers,
S
I think the issue was with a hard offset, the bases of the output bank were overdriven, and that caused all of the base resistors to fail. Once they failed, then the emitter resistor in the second stage of the output failed.
When I replaced that, the 2nd stage output transistor (Q203) failed (so clearly it had been stressed). There is no evidence that the output transistors failed or were overstressed.
However, having fixed the source of the offset, and fixed the output base resistors, the amp appears to be working properly.
The failure of the 2nd stage transistor was due to the fact that the entire output current was being passed through that one transistor. The output devices were protected by the open base resistors.. I note that is didn't fail at turn-on, but failed when I increased the input signal level.
So, I honestly think the system is OK as is.
It is also important to note that by my observations, high output offsets are not caused by heavy bias issues at the Darlington inputs. Rather, small offsets in the bias currents (a few hundred micro-Amps) can generate offsets at the driver stage that will rail the outputs. These input offsets are not high enough to damage the input transistors, but they result in DC biases in the Q109/Q111 and Q110/Q112 stages that will produce substantial DC offsets at the outputs.
It is precisely for this reason that I am working on a balanced current source. The original design uses two "related" current sources, but the fact that they are actually only related means that there can be significant differences in the bias currents in the two diff pairs. To the extent that the currents through the LEDS differ, the tail currents in the diff pairs will also be different. Look at pretty much any 565 and the LEDs will be glowing with different brightnesses, indicating different currents, and therefore different forward voltage drops, and thus different tail currents for the positive and negative diff pairs. And a built-in offset for the servo to correct.
I have observed these differences at the 500 uAmp level, and that relates to about 0.5 volts difference in input bias at the driver stages. The drivers have very high gain (about 20-25 dB), so a 0.5 volt offset in the input bias will result in a 50+ volt offset at the output (this is also why the servo is usually effective in compensating for these differences - since it requires small input offset to adjust it out).
A much better solution is to create a bi-polar current source where the two legs have currents derived from the same voltage source, so if one changes the other changes. Significant changes will then move the operating points of both banks together, so you may get some clipping at higher power levels, but you should never get an offset between the outputs.
You are actually the person who inspired this idea, since you noted once that it is a shame that the Adcom folks relied on a servo to control this issue. I agree, and I am now working on a design mod to eliminate this problem. I'll name it the "Anatech Upgrade"... 🙂
Cheers,
S
Last edited:
Hi Scott,
I can only comment on what I have learned over the years from testing and failure analysis. The problem is that damaged parts can appear to be okay - otherwise they would have been replaced. I saw a lot of equipment that had been repaired by other shops, blown again. More often than not, they only replace one or a few outputs and generally not the drivers. Most "bad amplifiers" were only suffering from poor service work, not from a problem in design. I'm not being over-cautious, more realistic. But then I repair equipment for a living and the end result is very important to me. That includes the sound quality. I'll leave that subject where it is now. If you ever do work for another person, please consider what I have said.
As for the modification to use current sources, please credit yourself. Call it "servo freedom" or some such. I do firmly believe that amplifiers shouldn't need DC servos and the only reason they exist is to remove a step on the assembly line that involved a human and instruments. Yes, I feel the Adcom servo has too much authority, and that the circuit needs it. I'll add that DC offset control is applied in the wrong place and has been for years. But that involves more technician time during manufacture. Many technicians would be too lazy or in a hurry to follow the proper steps.
In short, you should be using matched pairs (closely matched) and balance out the rest of the circuit so that it "wants" to sit at the zero point. Doing things the way most are done today is much like driving a car that pulls to one side.
You did the work, so take the credit. 🙂
-Chris
I can only comment on what I have learned over the years from testing and failure analysis. The problem is that damaged parts can appear to be okay - otherwise they would have been replaced. I saw a lot of equipment that had been repaired by other shops, blown again. More often than not, they only replace one or a few outputs and generally not the drivers. Most "bad amplifiers" were only suffering from poor service work, not from a problem in design. I'm not being over-cautious, more realistic. But then I repair equipment for a living and the end result is very important to me. That includes the sound quality. I'll leave that subject where it is now. If you ever do work for another person, please consider what I have said.
As for the modification to use current sources, please credit yourself. Call it "servo freedom" or some such. I do firmly believe that amplifiers shouldn't need DC servos and the only reason they exist is to remove a step on the assembly line that involved a human and instruments. Yes, I feel the Adcom servo has too much authority, and that the circuit needs it. I'll add that DC offset control is applied in the wrong place and has been for years. But that involves more technician time during manufacture. Many technicians would be too lazy or in a hurry to follow the proper steps.
In short, you should be using matched pairs (closely matched) and balance out the rest of the circuit so that it "wants" to sit at the zero point. Doing things the way most are done today is much like driving a car that pulls to one side.
You did the work, so take the credit. 🙂
-Chris