mixer oscillating after recap.

JohnRoberts said:

I don't have a very clear mental image of your system, but allow me to scatter shoot.

Are power supply + and - rails swinging relative to channel grounds when everything is singing?
If yes, what do you see between channel ground and these PS grounds?

Click to expand...

Yes, rails swinging wrt to chan grounds, I can measure entire oscillation. All + an - power rails, on all 8 regulator boards, are swinging relative to channel grounds. All 8 regulator boards have same levels. cca 400mVpp @ 1.8MhZ. This is relative to three grounds. Relative to fourth ground level is 200MHz Vpp.

Actually, even if I measure between channel grounds, and regulators 0V, there is same oscillation. It is 1.8MHz 400mVpp at 3 grounds, and fourth ground is cca 200mVpp.

So entire regulators scheme is swinging relative to channel grounds. And even channel grounds inside mixer are swinging relative to common ground point between smoothing caps in external PS box.

If the PS ground is clean wrt channel ground, while PS + and _ are swinging up and down, the PS needs to get stiffened up wrt to it's own ground.

If the PS ground is swinging wrt the the channel ground then that path needs to get stiffened up. (easier to say than do if it is a length of wire, with inductance, etc is involved).

Click to expand...

Both PS ground and rails are swinging wrt to the channel grounds. Or opposite. And even channel grounds inside mixer are swinging wrt to remote channel grounds in external transformer/rectifier/smoothingcap box.

There is a compliance somewhere between the PS ground and channel ground that is allowing the opamps to make the rails swing up and down (same as a common emitter amplifier stage The opamp power pins are like the collectors and the opamp output is like the emitters).

FWIW you should be able to make the who thing work with 10 ohm resistors in series with every PS line and even the grounds, as long as there is adequate local PS decoupling (you need to let the current coming out of the opamp output pins, get back to the PS rails locally).

Click to expand...

All channels already have 10ohm resistors in series with PS rails. For test I inserted 10ohm resistor in reference to ground for regulator boards - nothing changed.

Basically I have two power supplies, first one feeding second. First one is in remote box, 3m of cable, and is made of transformer, rectifier diodes, 4x10000uF smoothing caps and 100nF bypas caps. All channel grounds and reference ground for regulators are tied back to a point in between 4x smoothing caps.

Second power supply are 8 regulator boards, which are getting ground reference and +/- voltage from first external supply.

I dont use regulator reference as channel ground. Channel grounds go all the way back to first PS. Thus second PS reference (0V) can swing relative to channel grounds. Second PS (8 regulators) reference for 0V is tied back to first supply, but through 3m cable. I tried to connect second PS's reference directly to channels within mixer, but nothing changed, osc remains.

It is always logical, after you find it...   been there, done that... you are not the first or last...
JR

Click to expand...

Thnx for encouragement. I'm totally lost.

No, you just identified it... the entire power supply (ground included)  is swinging relative to the channel ground. That is the problemo.

To visualize what is going on... your channel opamps are coupling that HF AC current into the local ground, and sucking it from the + and - supplies that aren't effectively grounded to the local ground so the whole supplies just swing up and down.

The resistors in the +/- rails don't matter because it is drawing current not voltage through those leads.

OK, one more thing to try. Strap a pair of 100uF capacitors from the channel side of those 10 ohm resistors in series with the -/- rails, maybe some .1uF too.  Now connect the ground side of those 100uF caps to the local channel ground. If this doesn't stop the oscillation I will be surprised.

You may only need the .1uFs but it seems I suggested that a few days ago.

JR


 

JohnRoberts said:

No, you just identified it... the entire power supply (ground included)  is swinging relative to the channel ground. That is the problemo.

To visualize what is going on... your channel opamps are coupling that HF AC current into the local ground, and sucking it from the + and - supplies that aren't effectively grounded to the local ground so the whole supplies just swing up and down.

The resistors in the +/- rails don't matter because it is drawing current not voltage through those leads.

OK, one more thing to try. Strap a pair of 100uF capacitors from the channel side of those 10 ohm resistors in series with the -/- rails, maybe some .1uF too.  Now connect the ground side of those 100uF caps to the local channel ground. If this doesn't stop the oscillation I will be surprised.

You may only need the .1uFs but it seems I suggested that a few days ago.

JR

Click to expand...

Right now I have it like this. Please see attached picture. (entire original schem http://84.255.203.119/smsm16-2schem.pdf )Such decoupling is already on every channel. Are these 220uF positioned the way you suggested 100uF's to be? I could add 0.1uF in parallel. I'm out of supply of 0.1uF atm, waiting for delivery.

If you look at schematics, decopling 220uF caps are connected to 0P ground. Those should stabilyze supply, but they don't, because 0P is swinging arround too, together with other 3 grounds. I need to tie all four grounds inside mixer to ground in external PS box, where main ground star point it, but don't know how, because it is already connected there with thick 3m cable. I can only think of shortening that ground cables, but this may just reduce osc a bit, not remove it.

Something in mixer is forcing oscillation. I will find that sucker. In following days I will redo entire bypassing on all boards, to have it from rails to ground. Then I will take it from there on.

Now I have 100nF between rails, not to ground. Can I keep those caps between rails, and just add two more for each IC, each from rail to ground? Thus I will have 3 caps on each IC. Is that ok, or do I better remove cap inbetween rails, and just keep two caps from rail to ground?

Let me guess are all those grounds with different names kept separate? Do they all have their own wires going off to somewhere to eventually meet up?

If yes, no wonder the sucker is singing... at HF the wires have inductance so rise in impedance...

Yes your cap between the rails, but not to the local ground is making it worse.

Just jumper your OP ground to one or more of those other grounds and see if it stops.

In the final design you probably can stabilize it with .1Ufs or so.  Putting the big caps on the board but then bringing their ground lead back somewhere else kind of defeats the purpose of putting them there close to the circuit.

The purpose for power supply caps spread around a design is to provide a path, for ground current to get back to the power supply rail without a long circuitous path...

That ground design looks like one optimized to single mindedly avoid hum, while ignoring everything else.

JR

PS: If you are so in love with the multiple circuitous ground design, you can do what is called a hybrid ground often used around RF, where HF capacitors provide multiple parallel ground shunts while remaining a relative open circuit at mains frequency. For example a .1uf between the OP ground and audio ground at the board might be enough to stabilize it.

John,
yes, grounds separate, over 3 meters of wire each, then joining inside external supply.

Jumping 0P to any other ground does almost nothing. But connecting all 4 grounds together at mixer side makes a change. Finally!

I connected all 4 grounds at the end of ribbon cable (1m long). Thats practically at chennel level. Oscillation changed in level in frequency. With jupered grounds, it is down to 180mVpp (from 530mVpp with unconnected grounds) and osc frequency is up to 3.1MHz (from 1.8MHz with unconnected grounds).

Then I connected (jumpered) all grounds at the middle of ribbon cable (0,5m from both ends), and osc went almost away, just 20mVpp @ 7MHz. I guess if I tied grounds togeter in several places along the length of ribbon cable, possibly at every channel, osc would be gone.

So something is happening!!! Finally I was able to make a change, at least. It seems separate long grounds are a bad idea, at least in my system.

Probably as a solution I could bypass grounds at HF with 100nF (to each other, or to some other common???), on each channel at ribbon cable connector. What do you think about this? Is it a solution, or is it more a clumsy patch. Should I seek for proper solution somewhere else, within channels, or is linking grounds at HF a proper way to do it?

My doubt is, why having separate grouds then, if I tie them together already at channel level? Why four leads on ribbon cable, if one is better.

Any ideas, second thoughts, warnings, anyhing...?

...

gnd said:

John,
yes, grounds separate, over 3 meters of wire each, then joining inside external supply.

Click to expand...

There be impedance.

Jumping 0P to any other ground does almost nothing. But connecting all 4 grounds together at mixer side makes a change. Finally!

Click to expand...

QED

I connected all 4 grounds at the end of ribbon cable (1m long). Thats practically at chennel level. Oscillation changed in level in frequency. With jupered grounds, it is down to 180mVpp (from 530mVpp with unconnected grounds) and osc frequency is up to 3.1MHz (from 1.8MHz with unconnected grounds).

Then I connected (jumpered) all grounds at the middle of ribbon cable (0,5m from both ends), and osc went almost away, just 20mVpp @ 7MHz. I guess if I tied grounds togeter in several places along the length of ribbon cable, possibly at every channel, osc would be gone.

Click to expand...

The goal is not to get oscillation to just go away but banish it entirely. You want a margin of stability beyond just not oscillating by itself. Marginal circuits can alter transient signals or signals with high edge rates.

So something is happening!!! Finally I was able to make a change, at least. It seems separate long grounds are a bad idea, at least in my system.

Click to expand...

Long grounds are not the problem,,, different grounds are.

Probably as a solution I could bypass grounds at HF with 100nF (to each other, or to some other common???), on each channel at ribbon cable connector. What do you think about this? Is it a solution, or is it more a clumsy patch. Should I seek for proper solution somewhere else, within channels, or is linking grounds at HF a proper way to do it?

Click to expand...

Yes, as I already suggested HF Hybrid ground approach, where small caps shunt out the long wire impedance at HF. But rails to ground, not rail to rail.  

My doubt is, why having separate grouds then, if I tie them together already at channel level? Why four leads on ribbon cable, if one is better.

Any ideas, second thoughts, warnings, anyhing...?

...

Click to expand...

Good question, It sounds like a misguided star ground fixation.

For a slight veer but perhaps instructional, have you ever heard of "driven rail" audio power amps? They work the same way your power supply is misbehaving. In a driven rail power amp, the normal output of the power amp is tied to ground, and then the loudspeaker is connected between the power supply common and the amp ground. Instead of driving the loudspeaker directly, the amp drives the whole power supply up and down, driving the speaker that is connected to the PS common. (This is an attractive low cost approach for budget amp makers because all the amplifier circuitry, except for the very last power transistors can be cheap low voltage parts.)


OK design philosophy time... You can not have significant compliance between a separate power supply ground, and where all your opamps are sending their output current (or you get your unintended driven rail amp).

For this design invest in a bunch of .1uF (maybe .01uF if oscillation high enough frequency) disc caps and couple all the grounds to each other at HF. For future designs look at a simpler (but not simple) grounding scheme. A very powerful concept outside the chassis is balanced/differential 3 wire signal transfer... This way ground is not a signal conductor. Inside the chassis , it it too expensive and impractical to treat everything balanced (while some inclined to excess will try for simple products). What we do instead is convert our balanced outside signal, into an unbalanced internal signal relative to some local ground. The local ground for the input receiver, the eq, or fader stage, are each treated as separate nodes with signal treated differentially wrt to these different nodes. The output stage is treated relative to its local ground (often bonded to the chassis there).  

Now for the leap... These different grounds can all be the same connected PCB trace, but we don't expect the voltage on this ground trace to be the same in different places on the PCB. We don't care as long as we treat the audio signal differentially when we pass it between the different circuit sections. The ground over at the input could even have some shield garbage on it, and we could ignore it because the signal comes in differential. When we pass it to the next stage, we use a differential stage to subtract any ground difference between these two sections. etc. Note: this local differential approach, will even ignore a ground loop, since it ignores any and all ground voltages. (Lifting grounds is a sign of poor product design. IMO)

Generally this doesn't require adding a lot of differential stages since all opamps have both + and - inputs, it's often a matter of adding a couple resistors at opamp  + pin to either reference the opamp's output to where it's going or where it's coming from.

From appearances your best approach for now is the hybrid ground, for future designs think about where the current is coming from and where it is going...

And remember.. "ground is a concept not a single voltage".

JR

PS: Using multiple lines in a ribbon cable for ground is commonly done to realize lower DCR and some extra reliability, but in general treating power and signal grounds differently can lead to problems in large consoles (I know).
 

Ok, I'm back to local bypassing with 100nF from each IC rail to ground.

I have four local grounds. To which one to bypass each IC, to make is most effective? Looking at http://www.analog.com/static/imported-files/application_notes/135208865AN-202.pdf (page3), it is important to which ground I bypass. If I understand properly figure 3b on page 3, to make it efficient, bypass must go to reference ground of each IC. Am I understanding it properly? I cannot figure 3c though, where bypass is going rail to rail.

In my schematics ( http://84.255.203.119/smsm16-2schem.pdf ), to where do I bypass each channels tage ic1a, ic1b, ic2a and ic2b to make it effective?

...

If you feel you must have 4 different grounds, maybe you need to bypass every one of them.

I repeat, think about where the opamp output current is going...

Have you tried my earlier suggestion of coupling the grounds together with .1uf caps? 

You need to give the high frequency current coming out of the opamp and into the audio ground a low impedance local path back to the power supply without traveling over the long ground wires. If you already have decoupling caps from PS rails to power ground a cap between power ground and audio ground(s) should help.

Are you still unclear about why this is happening?  Your grounds have 2x the length of wire in series between them. So its a long trip for HF current to go all the way down the audio ground wire and back up the power ground wire to get to the common of the bypass caps and back to PS rails. 

JR

John, I'm uncertain, because I don't know electronics theory and don't have experience. I have hard time visualizing where currents are going. Sorry if I don't quite get it.  :-\

I will connect all grounds together, with bypass caps on HF, or even keep just one ground in the end - I will check this later, after I fix all channel boards, and see what is quieter.

First I want to bypass locally at each IC. This means taking 27 boards out. It is plenty of work and I want to do it right.

At each channel board, I have decoupling caps 220uF from PS rails to 0P. I can put 100nF cap parallel to 220uF, from rail to 0P. Then I can connect 0P to audio ground. Problem is, I don't know which one is audio ground.

On boards, there are four grounds, 0IN, 0A, 0P and 0F. I need to decide to which of those to bypass ICs on board. I'd say that 0P is power ground, and other three are audio grounds? Most gain stages on channels are referenced to 0F (on output boards gain stages are referenced to 0A), so I'd say I need to:
1. first, bypass all ICs on channel beards to 0F (and to 0A on L/R output boards)
2. and second, place a 100nF cap between 0F and 0P, possibly somewhere near ribbon cable connector on each board. On output boards i put cap between 0A and 0P.

Does this sound correct?

...

gnd said:

John, I'm uncertain, because I don't know electronics theory and don't have experience. I have hard time visualizing where currents are going. Sorry if I don't quite get it.  :-\

Click to expand...

Perhaps invest a little time into understanding Ohms law, which predicts current flow if you know voltage and resistance. Simply X volts across something like the pots, will generate Y mAmps of current flow. You don't even need to know how much to grasp the general flows.

I will connect all grounds together, with bypass caps on HF, or even keep just one ground in the end - I will check this later, after I fix all channel boards, and see what is quieter.

Click to expand...

OK, your time to manage.

First I want to bypass locally at each IC. This means taking 27 boards out. It is plenty of work and I want to do it right.

Click to expand...

This is still a little uncertainty as to which ground to decouple to if you have 4.

At each channel board, I have decoupling caps 220uF from PS rails to 0P. I can put 100nF cap parallel to 220uF, from rail to 0P. Then I can connect 0P to audio ground. Problem is, I don't know which one is audio ground.

Click to expand...

If you don't know I can't tell you... All of the grounds need to be bonded together locally at HF, thus my repeated request to couple together with at least .1 uF.

On boards, there are four grounds, 0IN, 0A, 0P and 0F. I need to decide to which of those to bypass ICs on board. I'd say that 0P is power ground, and other three are audio grounds? Most gain stages on channels are referenced to 0F (on output boards gain stages are referenced to 0A), so I'd say I need to:

Click to expand...

0in looks like a dedicated input ground
0p looks like power ground
0f indeed looks like most of the immediate audio loads
0a is nominally "audio" ground but they connect things like insert jack grounds, and pan pot wiper. Note: I would be a little concerned about connecting insert shields to a clean audio ground, but I guess opinions vary.
0v is a LED ground, that doesn't appear to connect off the board.

1. first, bypass all ICs on channel beards to 0F (and to 0A on L/R output boards)

Click to expand...

Yes 0f is one obvious destination for audio current.
I am not sure it makes that much of difference which local ground you decouple to, as long as they are all low impedance at HF to each other.

2. and second, place a 100nF cap between 0F and 0P, possibly somewhere near ribbon cable connector on each board. On output boards i put cap between 0A and 0P.

Does this sound correct?

...

Click to expand...

The order I would follow is to see if oscillation stops when I hard jumper all the grounds together. If yes then disconnect one at a time to find the problem ground, and then look at bonding it to the others.

This design apparently made sense to someone at one time. To me, not so much, but as I like to say, consoles are one of the most complicated simple products to design (power amps are the other, simple but hard).

JR

Don't think of the long wires as wires but as inductors!  The higher the noise frequency the more difficulty they have getting through the long wires.

For now I was able to remove oscillation by tying some grounds together at different points, mostly by bringing star point closer to channels, and then some local bypassing. I also removed some 5532's and replaced them back with 072's. It takes some experimenting, tieng ground together in different places has different results. But I'm getting there, probably mostly it is ground problem.

John, thank you for most valuable input.

While dealing with oscillation, I noticed weak points in my grounding scheme. A bit too much hum/buzz. I'm redoing some internal connections. Atm I disconnected all channels and am working just on mix bus grounding, trying to minimize influence of garbage in it. Power regulators need some work too. All about grounding now, trying for optimum.

...