New Regulator

[PigletsDad]

Very hard doing valid measurements inside running digital system, as probe wires act as antenna to pick up digital signals - check what same probe arrangement gives when measuring ground.

To make more or less valid measurements, you have to minimise loop area between scope ground and signal - try winding scope earth round probe. For best measurements, use coax with BNC at scope end, soldered to board under test with minimum lead lengths - you probably want to put say 100Ohm resistor with very short leads to isolate circuit from coax capacitance in signal line.

 

[Bemused]

Ah ha maybe I had a moment.
Just gone further upstream with the scope and found a load of noise.

LM7812 output below
Scales 5mV and 1uS per division.

Gyrator output below
Scales 5mV and 1uS per division.

Supply to board below
Scales 100mV and 1uS per division.

 

[Bemused]

Very hard doing valid measurements inside running digital system, as probe wires act as antenna to pick up digital signals - check what same probe arrangement gives when measuring ground.

To make more or less valid measurements, you have to minimise loop area between scope ground and signal - try winding scope earth round probe. For best measurements, use coax with BNC at scope end, soldered to board under test with minimum lead lengths - you probably want to put say 100Ohm resistor with very short leads to isolate circuit from coax capacitance in signal line.

Many thanks for the advice PD, couple of questions please.
Should I use a specific type of coax and where should the resistor go?

Scope shot below is same arrangement as previous but on regulator 0v output terminal.
Scales 5mV and 1uS per division.

 

[PigletsDad]

OK, the scope shot shows your measurement noise floor.

Coax type doesn't matter much; something reasonably thin will be easier to work with. 100Ohm resistor goes between co-ax core and measurement point. Coax shield soldered to nearby earth point on ground plane.

 

[Bemused]

OK, the scope shot shows your measurement noise floor.

Coax type doesn't matter much; something reasonably thin will be easier to work with. 100Ohm resistor goes between co-ax core and measurement point. Coax shield soldered to nearby earth point on ground plane.

Great your a superstar.

 

[Bemused]

Thinking the 797 output has no decoupling to ground, it all happens after the pass transist.
On a positive note my DAC has NEVER sounded this good so something is right.

 

[RyanSoundLab]

Tony, despite those scope waveforms at the input to the whole circuit and the gyrator and LM outputs, my guess is that the problem is still with the basic regulator circuit. It's oscillation is affecting all your readings.

That said, I don't call your scope readings "noise". It's oscillation but interestingly it's not a smooth oscillation which would give you a sine wave. The circuit is trying to do something (like correct a small voltage error at the output) but because the correction circuit is much faster than the output circuit itself, it overcorrects, then backs off momentarily in the other direction - repeated ad nauseum. I would suspect that the AD797 is much faster than the output transistor/22uf cap. The 2SD669 has an output impedance which will interact with the 22uf cap to cause a phase shift which may be at the heart of the problem. Hate to suggest it again, but slowing down the AD797 further with an even bigger feedback cap (maybe 470pf) might help.

Kit

 

[Bemused]

Thanks again for your input Kit, no shortage of ideas yet!

My bad terminology with noise, yes it a very uniform oscillation.

An early version had a Darlington output stage and it's exaggerated phase shifts played havoc.

For a momemt I suspected 3 pin reg trouble but I've left that thought behind and agree early stage noise (oopps oscillation) is output transients being pushed back right to my transformer.
And any transient hiting the voltage reference will be phase shifted so the loop starts correcting at the wrong time. (Just talking aloud here)

If it slows I can always speed it up again.
I can also move the 797 to run at less than unity gain.

With my industrial digital stuff I find the loop needs opening on occasion in order to fully understand the nature of a particularly nasty beast.

Maybe I try no output capacitor again or at least move back to a small one again. I have 7343 & 3216 pads down.

This is all fun but takes a great deal of time for each trial trial, especially when I can't find the dammed cap I need

 

[Bemused]

On a positive note I put a chunky heatsink on Q6 and its gone from an 'ouch' touch test to barely warm.

 

[Bemused]

Good news and bad news and some so so news.

Lets get the crappy stuff out of the way first, I killed my DAC again.
Two of the four right hand side DACs dead one on the left so its just two a side for now. Been testing with Martins crowbar on the output on every new arrangement but had a fractured base joint in the crowbar transistor so it didn’t fire...

Tried also sorts of capacitor values today, it needs an output cap of around 22uF, 2.2uF IS too small and 100uF is silly and it was the 100uF that caused a start-up overshoot and nailed the DAC.

Tried a 1 and 10uF tantalum in most bypass positions I could think of to no gain in performance.

Increased C(kit) my new C8, pin 6 to pin 2 from 47pF to 100pF to 200pF to 680pF
Bingo (This is the good news part) 680pF nailed it and output noise has dropped to scope noise floor.

Pertinent values below
R14 IS 0RO
Output cap is 22uF
C12 is 680pF

Next steps are Buffalo repair or treat myself to a new one this time

Then take PDs advice and improve the scope measurement procedure and have a more involved look at C8.
Thinking of defiantly adding some series R for C8 and see if I can clearly define and control this pass band when armed with better scope measurements.

Not an expert with high speed regulators for sure but I can now solder like a demon

Tony

 

[martin clark]

Tony,

Ouch. Oh well...

Well done on the fix. Comp cap at 680pF /100R is 3dB down at 2.4Mhz - bingo for squashing the 5Mhz oscillation. Good result, even with AD825s I took to bolting them down like this around 1-2Mhz (and they are generally easier to apply than the 797!)

I don't think playing with output transistors further will improve on this usefully. There's no need for a 10Mhz oscillator and in any case - this is already well into the region where you want local decoupling to be doing the work, not the regulator. Once you're up and running you might want to play with C6 at 1nF even.

 

[Bemused]

Cheers Martin, well they do say "if you can't fix it"
Have to expect a casualty or two else spend an inordinate length of time checking and bug builds have always been my Achilles heel. In this case the crowbar has been disconnected and connected every 1/2 hour for a few days so in hindsight... Nevertheless your crowbar has lit up on a few occasions and saved a dac or two so thanks for the tip.

I had not intention to play with the output transistor, sorry if I implied I did.
Mostly satisfied with it the whole regulator at this voltage and current.

Some board changes needed.
Delete R14, add series R to C8
I have already got some layout improvements done, saved three layer jumps and a little tidy up.
Need some real estate around Q6 to fit a largish heat sink.

C6 at 1nF even

C6 the reference filter surly you had a typo.

Tony

 

[martin clark]

C6 the reference filter surly you had a typo.

Sorry, yes: I meant C8 from pin6 on your schematic!

BTW, at the values you settled on, now C12/R14 are doing nothing (pole at 3.4Mhz). You can maybe play with increasing R14 back up again; or removing them entirely (link-out R14) which is probably the better route to try since they'll no longer add anything useful, your increased C8 dominates all.

 

[RyanSoundLab]

The joys of board layouts! I must have spent 1/3 of my life the last couple of years doing them. Ever tried "autorouter"? Worthless for analog circuits. Your board seems to have a good bit of space available overall - just need to shift and squeeze around Q6 to get the space you want for the heat sink. Enjoy and be sure to post your new artwork for us to admire. (the current layout is nice indeed).

Kit

 

[Bemused]

Sorry, yes: I meant C8 from pin6 on your schematic!

BTW, at the values you settled on, now C12/R14 are doing nothing (pole at 3.4Mhz). You can maybe play with increasing R14 back up again; or removing them entirely (link-out R14) which is probably the better route to try since they'll no longer add anything useful, your increased C8 dominates all.

Thanks and understood.

 

[Bemused]

The joys of board layouts! I must have spent 1/3 of my life the last couple of years doing them. Ever tried "autorouter"? Worthless for analog circuits. Your board seems to have a good bit of space available overall - just need to shift and squeeze around Q6 to get the space you want for the heat sink. Enjoy and be sure to post your new artwork for us to admire. (the current layout is nice indeed).

Kit

I use Eagle and tried auto routing, its a fun thing but as you say "worthless" for what we do.
Its always the last few traces that are the hard ones.

I'll pop up the latest works in progress,

This thread is far from over BTW
I've other tricks in mind!

 

[martin clark]

Thanks and understood.

Instructive, though.

When I suggested the RC (100r/22nF) combo it has a -3dB point of 72Khz from the output, intended to prevent the output oscillation ever happening (hey, that's a decade or so below where you might expect the inclusion of a fast pass transistor to start affecting things) In fact it still rang like a bell, and by trial you've found you have to shut down the inner loop - but by a rather smaller amount. That's interesting

- And well done!

 

[Bemused]

Instructive, though.

When I suggested the RC (100r/22nF) combo it has a -3dB point of 72Khz from the output, intended to prevent the output oscillation ever happening (hey, that's a decade or so below where you might expect the inclusion of a fast pass transistor to start affecting things) In fact it still rang like a bell, and by trial you've found you have to shut down the inner loop - but by a rather smaller amount. That's interesting

- And well done!

Wasn't the unpredicted behaviour OF THE 100r/22nF combo because we created a complex nest of poles with R14/C12R11/C9 ?

 

[Bemused]

Current schematic marked up.
I marked the poles on

 

[Bemused]

Next board (works in progress)