"Rotala" RA-820BX3 Offset Stability Problem (contd.)
Aaaargh – Yes, Chris. Something simple – and then again maybe not.
While being so focused on the DC offset issue, I violated the rule of looking at the possible consequences of changes and also - to always take the time for a second critical look of the schematic.
Why hadn’t I noticed the 120k R685 in the negative rail? What is it doing there? Yes, it is a voltage dropping resistor, something you would find in a valve amp design - but here? (And yes, it forms a 13Hz LP filter with C619 - but that is not important right now).
If you look at the schematic's original printed voltage notations (from the +/-41V rail design) the negative rail to the first LTP is reduced to -18.6V. A current of 82uA will give another 1.8V drop in the 22k resistors.
So, each LTP transistor has only 16.8 + 0.6V =17.4Vdc across it. Why?
Well my old dusty brain suddenly remembered that back in 1973 the preferred (or just about only) low noise transistors for audio input duty was BC109/BC179. And I bet that was what Otala originally used.
BC179 is a 60’ies Philips design in metal TO-18 cans with surprisingly great specs for the time. I vaguely remember the original 20 page datasheet with every bl**dy h-model parameter measured and graphed out. As a uni student, I had the pleasure of having to understand the lot. Of course, I have long since forgotten everything.
But the drawback was that these were only 20V Vce devices and if you exceeded that – you would have some dead (and not cheap) silicon in the can.
Rotel put in the more modern 2SA1016s with a max Vce of over 100V, but still kept the 120k dropping resistor in?? Why?
And if you look at my scribbles for the +/-31V amp I have, the voltage across the LTP is only 7.67 + 0.6 = 8.27V.
When I then increased the LTP current this further decreased this voltage - to the point where the LTP transistors were effectively choked off.
No wonder that further freeze spraying didn’t give much offset reaction! Duh! (hand smack on forehead, indeed)
Instead of progressing more carefully as I usually (should) do, I decided to brutally short the 120k - and the amp immediately came back to life. 😀
Further, I had originally observed some rounding of the sine test tone tops at high output volume settings, indicating that some transistor(s) were probably at the edge of their linear range and starting to distort.
With the 120k gone the sine output now looks perfect on the scope👍 although with a hefty -6.5V output offset that eh, does not appear to be stable 👎
One step forward and two back.
Aaaargh – Yes, Chris. Something simple – and then again maybe not.
While being so focused on the DC offset issue, I violated the rule of looking at the possible consequences of changes and also - to always take the time for a second critical look of the schematic.
Why hadn’t I noticed the 120k R685 in the negative rail? What is it doing there? Yes, it is a voltage dropping resistor, something you would find in a valve amp design - but here? (And yes, it forms a 13Hz LP filter with C619 - but that is not important right now).
If you look at the schematic's original printed voltage notations (from the +/-41V rail design) the negative rail to the first LTP is reduced to -18.6V. A current of 82uA will give another 1.8V drop in the 22k resistors.
So, each LTP transistor has only 16.8 + 0.6V =17.4Vdc across it. Why?
Well my old dusty brain suddenly remembered that back in 1973 the preferred (or just about only) low noise transistors for audio input duty was BC109/BC179. And I bet that was what Otala originally used.
BC179 is a 60’ies Philips design in metal TO-18 cans with surprisingly great specs for the time. I vaguely remember the original 20 page datasheet with every bl**dy h-model parameter measured and graphed out. As a uni student, I had the pleasure of having to understand the lot. Of course, I have long since forgotten everything.
But the drawback was that these were only 20V Vce devices and if you exceeded that – you would have some dead (and not cheap) silicon in the can.
Rotel put in the more modern 2SA1016s with a max Vce of over 100V, but still kept the 120k dropping resistor in?? Why?
And if you look at my scribbles for the +/-31V amp I have, the voltage across the LTP is only 7.67 + 0.6 = 8.27V.
When I then increased the LTP current this further decreased this voltage - to the point where the LTP transistors were effectively choked off.
No wonder that further freeze spraying didn’t give much offset reaction! Duh! (hand smack on forehead, indeed)
Instead of progressing more carefully as I usually (should) do, I decided to brutally short the 120k - and the amp immediately came back to life. 😀
Further, I had originally observed some rounding of the sine test tone tops at high output volume settings, indicating that some transistor(s) were probably at the edge of their linear range and starting to distort.
With the 120k gone the sine output now looks perfect on the scope👍 although with a hefty -6.5V output offset that eh, does not appear to be stable 👎
One step forward and two back.
Cleaning up rail voltage for the input pair? Simple RC filter on rails, albeit high resistance and low capacitance, usually it is vice versa.
Hi Per,
It (R685) did something else too. It is a cheap and crappy, device dependent voltage regulator depending on the average current draw of the diff pair.
Okay, take the gloves off now and do it really right. Enough doing things in a stupid way. Use either LND150 or DN2540 devices as a current source. Set the current so some value enough to keep a zener diode rated at 10 VDC or whatever works. The emitter of Q609 will follow the collector voltage of Q605, so you need to look at voltage changes there too. So start at 10 VDC. (Hint: you can use a shunt voltage reference, that makes things extremely quiet and stable.) Once you have this working you can play with the zener voltage, but if you end up around 10 VDC, there is the reference option. 🙂
Having this supply regulated isolates the front end from anything the line voltage or voltage drops or noise on that supply suffers. This is exactly what I would do. Why is C621 there? That couples the positive supply noise to the collector supply. I'd return that to common or pull it entirely.
-Chris
It (R685) did something else too. It is a cheap and crappy, device dependent voltage regulator depending on the average current draw of the diff pair.
Okay, take the gloves off now and do it really right. Enough doing things in a stupid way. Use either LND150 or DN2540 devices as a current source. Set the current so some value enough to keep a zener diode rated at 10 VDC or whatever works. The emitter of Q609 will follow the collector voltage of Q605, so you need to look at voltage changes there too. So start at 10 VDC. (Hint: you can use a shunt voltage reference, that makes things extremely quiet and stable.) Once you have this working you can play with the zener voltage, but if you end up around 10 VDC, there is the reference option. 🙂
Having this supply regulated isolates the front end from anything the line voltage or voltage drops or noise on that supply suffers. This is exactly what I would do. Why is C621 there? That couples the positive supply noise to the collector supply. I'd return that to common or pull it entirely.
-Chris
"take the gloves off now and do it really right"
Chris, I won't repeat what I occasionally have had the notion to do with this amp - gloves or no gloves.
But I very much also want to understand why this amp (mis-)behaves like it does.
In my younger days, Otala was regarded as one of the audio Gurus
, and you would need a lot of 'hair on the chest' to counter or criticize his papers. Further, this amp design was apparently copied by most of the major players in the 1970-1990ies: Pioneer, Kenwood, Sony, Nikko, Sansui, Yamaha and of course Rotel.
So I have decided to tread carefully. Firstly, I can zero the output by re-adding a 50R trimmer to the first LTP emitters.
But it is not stable, and it is most definitely a temperature sensitivity issue - not a current source stability problem.
Just one gentle fingertip to one of the LTP transistors immediately gives a + or - 200mV offset. It is that hyper sensitive.
So I tried to replace the white tubing with a BNC connector dust cover that I happened to have, and that actually improved things.
So the main part of the problem definitely lies in the first LTP - not unexpectedly. .
Chris, I won't repeat what I occasionally have had the notion to do with this amp - gloves or no gloves.
But I very much also want to understand why this amp (mis-)behaves like it does.
In my younger days, Otala was regarded as one of the audio Gurus
, and you would need a lot of 'hair on the chest' to counter or criticize his papers. Further, this amp design was apparently copied by most of the major players in the 1970-1990ies: Pioneer, Kenwood, Sony, Nikko, Sansui, Yamaha and of course Rotel.So I have decided to tread carefully. Firstly, I can zero the output by re-adding a 50R trimmer to the first LTP emitters.
But it is not stable, and it is most definitely a temperature sensitivity issue - not a current source stability problem.
Just one gentle fingertip to one of the LTP transistors immediately gives a + or - 200mV offset. It is that hyper sensitive.
So I tried to replace the white tubing with a BNC connector dust cover that I happened to have, and that actually improved things.
So the main part of the problem definitely lies in the first LTP - not unexpectedly. .
Hi Per,
I'm hardwired to solve problems. Design problems included. Locking the collector voltages simply remove a variable and actually helps you troubleshoot.
After many years in warranty service, and helping correct design issues (official service bulletins), I don't just jump in without carefully considering what changes may do. When I see a flakey design, I look really closely. This is a flakey design.
Really, what you want to do is make the LTP as stable and noise-free as possible. Sadly many early respected designers were later found to not be perfect and some made really large goofs. To make matters worse, some designers copy early "golden designs" and make changes without understanding the design. Notice I suggested keeping the expected collector values close to the original design? A legend is only human. It is possible for them to make errors.
One thing you will always see in a good, solid design is variables removed. When a designer gets "cute" and saves money - that's when things get weird and performance falls.
My suggestion did not involve cutting any traces, and it is 100% reversible. It may help you figure things out - or not. Yes, the LTP is suspect, it should behave like any other. One question. Does this thing have gain at DC? If it does, then there is one giant issue. Look at the negative feedback to ground path by the LTP. No cap? It has gain at DC. So at least temporarily stick a cap in series with that connection and see if it tames the issue a bit.
I'm hardwired to solve problems. Design problems included. Locking the collector voltages simply remove a variable and actually helps you troubleshoot.
After many years in warranty service, and helping correct design issues (official service bulletins), I don't just jump in without carefully considering what changes may do. When I see a flakey design, I look really closely. This is a flakey design.
Really, what you want to do is make the LTP as stable and noise-free as possible. Sadly many early respected designers were later found to not be perfect and some made really large goofs. To make matters worse, some designers copy early "golden designs" and make changes without understanding the design. Notice I suggested keeping the expected collector values close to the original design? A legend is only human. It is possible for them to make errors.
One thing you will always see in a good, solid design is variables removed. When a designer gets "cute" and saves money - that's when things get weird and performance falls.
My suggestion did not involve cutting any traces, and it is 100% reversible. It may help you figure things out - or not. Yes, the LTP is suspect, it should behave like any other. One question. Does this thing have gain at DC? If it does, then there is one giant issue. Look at the negative feedback to ground path by the LTP. No cap? It has gain at DC. So at least temporarily stick a cap in series with that connection and see if it tames the issue a bit.
I must be missing something - why is the NFB DC wise (via R653) provided to the input transistor base (Q603)?
Usually it is provided to the "other transistor" (Q605) base? Mistake in schematics?
Also R653 is connected after the fuse F601 - when the fuse blows (or is removed) there will be no NFB and amp may become unstable?
Usually it is provided to the "other transistor" (Q605) base? Mistake in schematics?
Also R653 is connected after the fuse F601 - when the fuse blows (or is removed) there will be no NFB and amp may become unstable?
Correct, who would tell us is it an inverting or non inverting amplifier?
Tried to simulate 1kHz (0.1V input voltage) in Cabirios LTSpice model - output is oscillating highly.
Just look at how many times the signal goes from base to collector, that's an inversion. Also the collector of the LTP the signal comes from, collector Q603 is inverted, Q605 is in phase. Some designs are more difficult to figure out, but you can use the basics to follow the signal.
See my simplified drawing in post#1038 - this is how you couple an inverting (op)amp. No schematic mistake.
If you send in the signal on the + input and ground what I scribbled as 'Vin', you get a non-inverting (op)amp.
If you send in the signal on the + input and ground what I scribbled as 'Vin', you get a non-inverting (op)amp.
At least in simulation, this is what works best. In the stock circuit, if I adjust the offset to <1 mV and then unbalance the LTP temperatures by setting one of the transistors to 32º (the other is 27º by default, so just 5º difference) I get ~110 mV offset. If I do the same with a cap in series with R603 (the 6k8 to ground before the input), I get ~12 mV. I've fiddled with the LTP in other ways (e.g. CM load, different CCS, different current, fixed voltage at the bottom of the load) and it makes very little difference, this seems to be the only thing that really lowers the offset dependence on temperature.
Hi Cabirio,
That is interesting.
I would have thought that putting a capacitor in series with the 6k8 would block the DC bias current and quickly send the input transistor into cutoff.
But I am of course willing to try it out in real life, what cap value did you use?
That is interesting.
I would have thought that putting a capacitor in series with the 6k8 would block the DC bias current and quickly send the input transistor into cutoff.
But I am of course willing to try it out in real life, what cap value did you use?
It's just like a standard inverting opamp with a DC blocking cap at the input, the bias current into In- comes from the output through the feedback resistor. In your simplified diagram of post #1038, you wouldn't be surprised if the 6k8 resistor wasn't even there, right? I used a 10u cap but as far as the simulation goes the value doesn't make a difference, since we're just looking at DC offset.
Of course, you're right. Sorry, I was still in the non-inverting amp world.
I just measured the gain of the first stage and found it to be 22, which is unusually large. I think that Otala used a gain of 13 in this stage, so I was planning to reduce the two collector resistors from 22k to say, 12 or 10k? Your spice is faster than my soldering iron, what do you get with this change?
I just measured the gain of the first stage and found it to be 22, which is unusually large. I think that Otala used a gain of 13 in this stage, so I was planning to reduce the two collector resistors from 22k to say, 12 or 10k? Your spice is faster than my soldering iron, what do you get with this change?
I get ~17 with 22k, ~10 with 12k, ~8 with 10k.
So your spice model uses a hFE of about 300 for the KSA992? The ones I got have a hFE of 400. So i think I'll put in two 10k for R611/613.
But I actually meant the temp sensitivity with a lower gain, all things equal it should help things?
So your spice model uses a hFE of about 300 for the KSA992? The ones I got have a hFE of 400. So i think I'll put in two 10k for R611/613.
But I actually meant the temp sensitivity with a lower gain, all things equal it should help things?
Are we not putting the horse behind the carriage here?
In ideal situation the DC feedback path should effect the input transistor base enough to correct the DC drift - correct?
So if it is failing to do its job perhaps the impact of NFB should be increased?