Improve a Rotel amp THD by 20dB!

Do we really want to go through it all again?
I would personally appreciate it - usually the way of solving problems and describing the progress improves in time and each next round is better then the previous one.
Especially when this amplifier seems to need some more thorough redesign following which would be educational probably for a lot of forum members (same techniques and methdos can be applied to approaching also other amplifiers).

There is a saying that you cannot step into the same river twice 🙂

But I would separate this amplifier into a dedicated thread otherwise the sub-topic will become quite difficult to follow in the end.
 
Well, I have managed to step into the same 💩 river more than twice.😊

But otherwise I do agree that this thread has been going on for a very long time, has covered a lot of different things - and gone on a number of tangents.
(Although I honestly think that Chris and I over the years have mostly managed to keep it civil and on the subject - well ok, sort of).

Thread_Jacking.gif


So, no I don't agree that this RA-820BX3 stuff need to go onto a new thread - if anything, it is probably closer to the original title than many other things that has been covered.
I do take your comment about the educational aspect and will try to better explain or provide a reference to the things that are being changed in the upgrade.

In the end I guess it is up to the senior moderators here on diyAudio to decide when the activity has dried out and it is time to end this old thread.
Which after all seems to have been of some interest to quite a few people - I mean, the thread has now had over 215k views!
 

Attachments

  • Thread_Jacking.gif
    Thread_Jacking.gif
    6.3 KB · Views: 44
  • Like
Reactions: Smartdriver
As long as the thread helps people, why close it? You're doing an excellent job.

Offset adjust. Diff pairs must be balanced, but manufacturing cannot take the time. Therefore they "fix" it, but matching always improves things. This circuit looks like it still has gain at DC (0 Hz), not great. So that amplifies any DC imbalance.
 
  • Like
Reactions: Smartdriver
In case it helps if/when LTSpice is eventually used, I have tweaked models for the small signal and VAS/driver transistors, based on actual measurements of the ones in my 820AX, here:

https://www.diyaudio.com/community/threads/tweaked-bjt-ltspice-models.404783/

Actually I also have models for the output transistors, but I'm a bit weary of sharing them since the DCA Pro has limited current capability (10 mA I think). I'm away from my computer now but I'll have another look later and see if they look reasonable enough to share.
 
  • Like
Reactions: madis64
Hi cabirio,
That's a great start for simulating. What we also really need is a random factor for transistor parameters. Give limits, and vary beta, Vbe and other factors so that your sim is realistic with real parts. Same for resistor values. It would be nice to see voltage coefficients for resistors also.
 
Ok, I've checked the 2SD1047/2SB817 models I have and they seem to be accurate enough. Although I could only match the actual measurements at low currents, I also matched to the datasheet quasi-saturation, fT(Ic) and beta droop at high currents, as well as the on-chip ballast resistors. Please find attached the updated file with those models included.
Hi cabirio,
That's a great start for simulating. What we also really need is a random factor for transistor parameters. Give limits, and vary beta, Vbe and other factors so that your sim is realistic with real parts. Same for resistor values. It would be nice to see voltage coefficients for resistors also.
I've tried this many times, attempt to make the simulation more "realistic" by introducing random variations in a number of parameters and/or component values. You'd be surprised how little of a difference it makes in most cases. The main "problem" with simulation (in quotes because it doesn't have to be if you're aware of the limitations and interpret the results accordingly) is not that all LTPs and CMs are perfectly balanced or that all resistors have 0 voltage coefficient, but the fact that the transistors models themselves, complex as they may seem, are much, much simpler than the real thing. If a circuit is half-competently designed (i.e. to not depend critically on certain beta, Vbe and/or other component values), the simulation isn't going to change much if all transistors of type X have exactly the same parameters or they haven't.

With all those caveats (plus the additional one that, unlike a real PCB, LTSpice connections have no parasitic L, C and R), I think the rather interesting compensation scheme we have here deserves at least a stability simulation, just to see what we get. Stay tuned...
 

Attachments

AngelP, R613 is an erratum, it should be 2k2 rather than 22k, no? Assuming this and after adjusting for offset and Iq, I get these currents, mostly close to what you've measured except for the VAS:

Q603/Q605 (LTP): ~85uA
Q617/Q619 (VAS): ~13mA
Q623/Q625 (drivers): ~35mA
Output transistors: ~11mA

Attached the ac simulation for your perusal, models file posted above.
 

Attachments

Sorry, you're right. I downloaded a clean service manual to copy the schematic and didn't pay attention to your notes, Per. To clarify we're talking about R611/R613, the LTP load, could you please confirm? While you're at it, take a look at R673/R675, the VAS emitter resistors, I'm still getting ~13 mA vs. your measured ~23 mA. See the new voltages below, but I won't post the corrected asc till we're sure it all checks out...

BX3-V.png
 
That wouldn't make much of a difference I think, there's very little voltage between the emitters of Q617 and Q619. Other than the transistors being damaged, which is a possibility, my bet would be on R673/R675 being something like 47R instead of 100R, or even those being correct but R625/R627 (R16/R17 in mine) being something like 3k instead of 1k8. Both those options would result in the measured ~23 mA. Typos and errata aren't uncommon in Rotel service manuals...
 
Hello all,
Nice to see so much activity on the thread and thank you for your kind words.

Yes, I can confirm that that the collector resistor R611 is not 2k2 as per the schematic, but actually 22k – same as R613.
And yes, I miscalculated the VAS current, it is indeed 13mA - which still makes the un-cooled VAS trannies run awfully hot.

Just as Bob Peace said: My favourite simulation software is - soldering.
So I guess I'll just crack on.
 
  • Like
Reactions: anatech
Rotel RA-820BX3 Circuit Origin and Stage 4 Changes:

Well, it appears that this amp is not Stan Curtis’ work. It is (almost) a carbon copy of a renowned design by Lohstroh and Otala first published in an IEEE paper in 1973 - with the modest tile: An Audio Power Amplifier for Ultimate Quality Requirements.

I haven’t yet read this original paper, but Doug Self has a nice review of this Four Stage Amp topology at the start of the Amp Architecture, Classes and Variations chapter in his excellent book, Audio Amp Design.

So, the Rotel engineers simply took this design and did not dare or want to alter it, and to my knowledge nor has anyone else (although I will probably be solidly corrected on this). So then why should I start to mess with this classic design?

Well, there is the overheating issue that needs to be rectified if this amp is supposed to last another couple of decades. The FR-2 pcb will continue to blacken and crumble if nothing is done about it, so that is the first job.


First, on to Stage 4 and the big 910R 2W feed resistors to the driver collectors. I believe that the original idea was that these would keep the stage in Class A for longer with currents to better cut off the output transistors. That ‘better’ has later been debunked – a single 47R or 100R ½W resistor between the driver emitters does the same job, with far less excess heat produced.

And as it will only have 1.2V (two Vbe’s) across it, so the driver current can be reduced from 35mA to 12mA – which will make the driver transistors and heatsinks run much cooler as well.

1721823846581.png


I then decided to increase the output bias setting from 2.5mV to a more 'normal' 5mV, giving about 24mA in each power transistor, as they are the ones doing all the hard work. With ±31V rails, the heat sink can easily cater for the heat produced, even with the extra set of output transistors compared to the RA-820AX. The heat sink doesn’t even get warm to the touch.

I would say that’s the stage four heat sorted.:xfingers:
 
I think you've made some pretty solid, intelligent decisions.

I think that looking at previous designs from the standpoint of what we know now is wise. We have components and techniques they didn't have to use back then. If you have a good design, pay close attention to what they did and never assume you know better. Apply todays knowledge after thinking about it - then test it afterwards.