New Posts New Posts RSS Feed - 1970s Design Indulgence
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

Welcome to the Graham Slee Audio Products Owners Club

 

Open to all owners plus those contemplating the purchase of a Graham Slee audio product wishing to use our loaner program: join here
Subscribe to our newsletter here (Rules on posting can be found here)


1970s Design Indulgence

 Post Reply Post Reply Page  <1 4647484950 62>
Author
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 8388
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 13 Mar 2019 at 4:59pm
Parting Shots: 3

10/3/19 version is given 1k6 for R7 and 150 pF (again) for C5.

(if you think I'm having problems understanding this configuration you should read Electronics For Vinyl...)

This makes the VAS "take note of what it is being asked to do" by C5. Otherwise T1 makes it roll on more just where the output stage's transition frequency approaches unity.

Amazingly the old commercial circuits used lower frequency power transistors which intuition says should have been oscillating merrily.

No wonder so many abandoned this configuration: when the going gets tough... At least I've given it my best shot.

Then you look at the highly praised Williamson (tube jobby) with zero phase and gain margins. Funny how subjectivism works...
Not simple enough for Google-Bot to understand...
Back to Top
Lucabeer View Drop Down
Senior Member
Senior Member


Joined: 24 Jan 2009
Location: Torino, Italy
Status: Online
Points: 419
Post Options Post Options   Thanks (0) Thanks(0)   Quote Lucabeer Quote  Post ReplyReply Direct Link To This Post Posted: 13 Mar 2019 at 5:49pm
Originally posted by Graham Slee Graham Slee wrote:

Parting Shots: 1

The bode plot according to the expert behaviour models provided by On-Semi, Central Semiconductor and Bob Cordell.

I and the computer can't lie, and can only take what they all provide and number crunch it.



An open loop bandwidth of almost 4 Mhz with such a large phase margin?????? Now I understand what you mean when you speak of "wideband design"!!!!




Edited by Lucabeer - 13 Mar 2019 at 5:53pm
Back to Top
Lucabeer View Drop Down
Senior Member
Senior Member


Joined: 24 Jan 2009
Location: Torino, Italy
Status: Online
Points: 419
Post Options Post Options   Thanks (0) Thanks(0)   Quote Lucabeer Quote  Post ReplyReply Direct Link To This Post Posted: 13 Mar 2019 at 5:57pm
One thing strikes my curiosity, seing such a Bode plot. Have you got a SACD player that does pure DSD (no conversion to PCM)? I think it would be very insightful to use one for testing, since SACD is infamous for its noise shaping which pushes all kind of crap to very high frequencies under the assumption that in that range it can't be heard. But I am sure it would be a naughty way to excite oscillations in equipment which doesn't have all this gain and margin... or which happily oscillates in spite of them due to "bad" components.


Edited by Lucabeer - 13 Mar 2019 at 5:59pm
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 8388
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 17 Mar 2019 at 7:29am
Having had enough with trying to coax the output stage into behaving by tweaks to the voltage amplifier, I sat down and analysed what had been happening.

The sequence of events was mainly 1. fit new Cdom; 2. switch on; 3. enjoy for up to three days; 4. fit new Cdom; 5. switch on ...; 6. ...; 7 fit new Cdom ...

So once the new Cdom burned-in (or possibly out but without getting hot?) that too good to be true sound went.

And if it's too good to be true, it's got to be a lie? Hasn't it?

My life's experience thus far with amplifiers is those who can't, write (or set up in hi-fi marketing); and those who can, struggle.

Every power transistor runs into mischief more than once in a while, and it would seem the higher the voltage (to get lots of watts) the worse it gets. Vceo would seem to be part of the problem, and a trip down fake-lane tends to support that assumption (http://sound.whsites.net/fake/counterfeit-p3.htm).

Not that I have counterfeit MJL3281a transistors: when I showed mine which differ from the old MJL3281a to On-Semi, they said they'd investigate, and for back up Farnell, the distributor is doing the same.

I'm sure they'll come back with the answer that they're OK.

So it has to be parasitics. The capacitance I can live with and design round. The inductances are invisible, they exist, but are invisible because the manufacturers choose not to include the parameters in simulation models.

When you decide to make a right old mess of your simulation schematic by adding inductances to every component where inductance exists, the bode plots start to reveal what the imagination has been telling you.

SPICE is great! The models used are not!

The low-ish voltage of the Proprius must be its saving grace. The higher voltage here must be having its influence, but I'd chosen 260 volt transistors for use on a 77 volt (max.) rail!

The simulation had been showing a slight response bump at around 20MHz for a while (it has been mentioned before), but when parasitics were modelled in, it really took off.

Sometimes you just have to fight fire with fire, or in this case, inductance with inductance. I've been here before, and that was the only way of curing an always-on studio amplifier design.

According to the simulator a value of 150 - 200 nH thwarts the parasitics, and going on the number of turns and diameter used in the studio amp, isn't far away from what I have now. Did that use the MJL3281a? No, it used TIP33 and TIP34 power transistors in a DC coupled design having a LTP input stage.

This says straight away the problem is nothing to do with the voltage amp I'm using, although I've tried to get it to make the output stage behave.

Now, let's take a look at Cdom: why do we use a ceramic multilayer? What did we use before they came on the scene? Polystyrene is what was used.

The VAS transistor (as it is called) swings the entire signal which can be from 3V to 15V (42V p-p) at normal listening room volumes, and on top of the signal is the parasitics which are all the way out to 20MHz in this case.

I'll not go into the math but the capacitor needs to be fast - a pulse cap maybe? Polystyrene fits the bill, but also a polypropylene pulse cap does.

On fitting one before adding the inductors the sound was awful. A result at last! It was revealing how hard it was working to control the parasitics.

In went the inductors. These go in series with the output emitter resistors. The awful sound was no more.

The too good to be true sound had also gone. In its place was the run-of-the-mill you'd expect, which might develop into something quite palatable given time.

So are we nearly there yet? Just over the next hill... Wink
Not simple enough for Google-Bot to understand...
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 8388
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 25 Mar 2019 at 7:54am
In a departure from convention the power amp is now stabilised passively.

Actually there still is some active compensation but only of a mild nature, and which encloses the whole class-A stage.

The vintage class-A stage here should not be seen as current feedback verbatim. It mimics the technique used in many a valve amplifier and I cannot recall the term current-feedback being used for those. Current and voltage are mutually connected functions so one could argue for one or the other, but argument seldom solves problems (unless a mathematical argument of course, which only leads to one answer).

The input stage is a voltage amplifier as is the subsequent stage which gets called the voltage amplifier stage or VAS, and it is quite wrong to call it that here. Both T1 and T2 are the voltage amplifier, with input presented to T1.

Treated as IPS and VAS Cdom is used to "soak up excess phase", except that it doesn't. Cdom does however stabilise it to the extent of not resulting in a blown amplifier, and the "free negative feedback" it provides, often regurgitated in books on amplifier design, might help achieve a good measured specification. However, having tried every value and tweak imaginable, it does not lead to good sound.

When driving Cdom, T1 collector behaves well throughout the audio spectrum, but upward of that exhibits its own instability, and we expect Cdom to soak that up too. It won't.

How to stabilise T1 is a difficult question to answer, as using negative feedback collector to base would be highly dependent on source impedance. The only other option is to shunt its collector which can use a capacitor directly to ground. The right value sees it rolling off at 20dB/decade with a kink in its response due to its series feedback, which is the necessary position to attach the global negative feedback loop.

So we have a pole followed by a zero and then a pole, which is much better than expecting Cdom to deal with the opposite, which it can only make a mess of.

Now we look at T2. Cdom is there (was there) to swamp Miller capacitance. Collector-base capacitance is magnified by the substantial gain of T2, and Cdom works to fulfil this function. What it also does is distort the upper waveform of T1.

An LTP input stage might not suffer such distortion by Cdom, but if you model an OP27 or OP37 you will see its high frequency residue, which demonstrates that Cdom is not the magic solution we might think it is.

Therefore we need to isolate the effect Cdom has on T1. Shunting T2 collector to ground via a suitably sized capacitor has a similar effect to Cdom, but has to be gain times larger. It also reduces T2's gain with increasing frequency, and with less signal current flowing its input resistance increases with frequency (Re = 26mV/Ic and with Ic falling Re increases and R base is beta times Re - or that's what the simulator indicates).

When plotted we see the result of two poles. T2 compensation is still dominant but we have T1 rolling off higher in frequency. The net result is 20dB/decade for T2 and slightly above the 0dB crossing, in comes T1's pole making the downward slope 40dB/decade. This indicates instability which we will come to in a minute. And by the way, these steps in the curve are only seen in isolation, and globally it is only 40dB/decade.

But the saving grace is that the signal continues to fall at 40dB/decade (and steeper for a while) and only "tries" to flatten at T1's "zero", which if you remember, then hits another pole and continues to roll down. No matter what stimulus (because it won't be audio up there) it is way down in the microvolts and has little chance of propagating around the circuit to produce the distortions not covered by the textbooks.

This is in complete contrast to the usual pole splitting compensation for which Cdom is used. That allows T1 to distort its output into Cdom producing all sorts of spurious fits and starts, when all we really want is the curve to die away.

So how do we deal with the rather inconvenient 40dB/decade? Making T2's shunt capacitor larger will only encourage T1's output to rise (by unloading it - a loose form of negative impedance), so we now must use overall negative feedback compensation around the complete class-A stage. We must not include the phase shifts and complex impedances of the output stage as this would be counterproductive.

This indicates the use of a capacitor between T2 collector and T1 emitter, but it must instead be taken to the global NFB return point, which is the resistor drop below T1's emitter.

This pushes the dominant pole downwards in frequency and causes the roll-down to be 20dB/decade right down to 16dB below crossing, which is its gain margin. We now have 76 degrees of phase margin with output stage included.

76 degrees is a little above the "half way house" between the guarantee of no ringing (90 degrees) and the minimum stable point (45 degrees). It is doubtful a power amp will ever give 90 degrees unless we accept the spurious contributions of a negative feedback Cdom.

I also expect we won't have such a good high frequency distortion result. Because we don't have any negative feedback afforded by a NFB Cdom the THD at 20 kHz is bound to rise more steeply.

But what we might have achieved is better sound. Is this a contradiction? It would appear so, but the distortion of Cdom on T1 resulted in spurious noise, which although far and beyond audio frequencies, causes the output transistors to use power doing things which are not needed, and beyond their transition frequency, causes that to happen in the drivers.

A test for the above is to listen to the mains transformer (toroid’s being sensitive). It is obviously quiet when doing nothing, and if idling you expect some buzz, but if noticeable then it isn't really idling.


Edited by Graham Slee - 25 Mar 2019 at 11:27am
Not simple enough for Google-Bot to understand...
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 8388
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 26 Mar 2019 at 6:55pm
I thought you'd like the results of the latest stabilisation technique?

Well, although I thought the sound quality was reasonable, the measured distortion was embarrassing to say the least.

Capacitor burn-in was evident with the sound cycling between thinning and beefing up again a number of times.

The type of stabilisation was published by the AES, and had been written by Sansui designer Takashi, and that's about all I know.

Reverting to NFB stabilisation, it would seem the larger the value of Cdom the worse the sound gets. The Proprius gets away with only 22pF and sounds fine powered-on all the time.

The extra gain required to do more output power requires a larger value of Cdom. Cutting it fine I tried 47pF again and soon found it needed the additional small capacitor between T2 collector and T1 emitter. This time 15pF gave good stability along with the 47pF above. There is also the local NFB of 390k between T2 collector and the NFB input.

I will also mention that I swapped output transistors to Toshiba 2SC5200.

Phase margin with the 2SC5200's is 88 degrees, but with only 10dB gain margin.

Phase margin with the MJL3281a's is 85 degrees with 14dB gain margin.

And it should also be explained that once every conceivable parasitic is modelled in, every amplifier design I've simulated results in the narrowest of phase and gain margins that it's a wonder that any sound remotely like music.

So may as well keep mucking about with it a little longer.

What makes matters worse is not knowing what I'm working with. The MJL3281a has moved production 2,400 km from Manila to Singapore, and now looks like a "fake" IMO. And my supplier who has a quality policy longer than most folk's arms gives the country of origin as China!

Below are smashed open images of the Manilla and Chinese Singapore versions.



Not simple enough for Google-Bot to understand...
Back to Top
Dave Friday View Drop Down
Senior Member
Senior Member


Joined: 07 Apr 2011
Location: Spain
Status: Offline
Points: 156
Post Options Post Options   Thanks (0) Thanks(0)   Quote Dave Friday Quote  Post ReplyReply Direct Link To This Post Posted: 27 Mar 2019 at 9:54am
Morning,what does the waveform look like ( the 12 meg one)..
I just had a thought about anode bend detectors !
If the 12 meg peak is nonlinear the negative feedback won't cope/work?
Kr.

lp12,oc9mk3,ca610p,krimson40watt pa,kef105.4
Back to Top
 Post Reply Post Reply Page  <1 4647484950 62>
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.01
Copyright ©2001-2018 Web Wiz Ltd.

This page was generated in 0.078 seconds.