THAT 1646 noise in PSU

[saint gillis]

  Hey there!

  I'm using a 1646 output balanced driver in a circuit, directly driven from an op amp without any resistive component between them.
  The 1646 chip is properly decoupled with a pair of 100n caps close to vcc and vee pins.
  I noticed that when I put the 1646 in the circuit it brings something like 20mV (peak to peak) HF noise in both PSU rails (+/-15v)
  When I plug a second 1646 it brings 20mV additional noise in the PSU rails (40mV total noise  in both PSU rails when both chips are engaged) ..

  Does it sound normal?

 

[saint gillis]

Noise is ~20GHz

 

[Monte McGuire]

Probably not 20GHz, maybe 20MHz? Regardless, HF noise like that could be caused if the circuit is oscillating. Also make sure that you're probing the circuit properly - a wideband scope probe attached to any sort of wire will pick up all sorts of RF garbage, including some that is not actually in your circuit. The ground sleeve near the probe tip is a great way to ground a probe for HF measurements.

 

[saint gillis]

Yes, I'm using a PSU for tests, I was connecting the probe to the PSU's ground , and the circuit is connected to the PSU with something like four 20cm wires (+15, -15, audio ground, decoupling ground).

When I put the probe at the 1646 bypass caps pin which are tied to the ground i can't see any noise at the PSU. But when I put the probe at the PSU ground, the noise appears only when I insert the THAT chips in the circuit. I guess without enclosure all of this can be expected...

 

[audiomixer]

are you inserting the chip into the powered circuit? what do you mean by when I insert the that 1646 chip.... without you don't really have a circuit anyway... are you at least comparing with a similar current draw. measure your (idle?) current draw and put an equivalent resistor in the circuit instead of the that chip when measuring your supply rails.

- michael

 

[saint gillis]

It is a small circuit with a ne5534 buffer, a 10k pot , a non inverting ne5534 with a gain of ~2,2, and a that 1646, the ICs are on sockets, so I can remove some of them..
The PSU I use for testing can easily provide 1,5A / rail

 

[ricardo]

Do not under any circumstances try to telescope 1646 pin 3 out to some so-called "clean" ground.

If you do, put a 1-100 nF on pin 3 to a local ground or Vee. Right at the pin.  If you need to feed pin 3 from a voltage source for single-supply operation, make sure there's a 47-100R in series and a C to a local ground or Vee.

Not sure if this warning appears in the present THAT 1646 datasheet but its important to avoid oscillation and / or marginal stability.

 

[abbey road d enfer]

When I put the probe at the 1646 bypass caps pin which are tied to the ground i can't see any noise at the PSU. But when I put the probe at the PSU ground, the noise appears only when I insert the THAT chips in the circuit. I guess without enclosure all of this can be expected...

Voltage is a two-legged thing. Where is the "ground" of the probe connected? Can you draw your set-up?
If there was an oscillation, I guess you should be able to see it at one of the outputs; I've never seen an oscillation occurring inside a building block and not being revealed outside it.
BTW, what's the BW of your scope? When you say 20GHz, is it really that, or 20MHz? How can you tell? Is it seen as a stable waveform that you can count on the screen?

 

[saint gillis]

Hello, here's the complete set-up :
http://hpics.li/e4cdd66

  And yes of course you were all right, noise is 20MHz (~50ns period oscillation)
  I bypass the ne553x op amps rail to rail, and I bypass the 1646 rail to ground with a dedicated ground path

  When I put the ground of the probe on C3 or C4 grounded pin no noise measured at the psu, but when the ground of the probe is connected to "gnd" (and not "gnd1") on the pcb, noise appears, and if I connect the probe's ground at "PSU_0" noise is measured at the PSU rails.

 

[ricardo]

http://hpics.li/e0ff4a5

  I bypass the ne553x op amps rail to rail, and I bypass the 1646 rail to ground with a dedicated ground path

You need to use ELECTROLYTICS AT each 5532/4 and 1646 (within 1cm) and they have to be from rail to ground.

Otherwise, even if your shebang doesn't oscillate, you'll have excessive THD.

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

 

[saint gillis]

Thank you I'll do!

 

[Monte McGuire]

Hello, here's the complete set-up :
http://hpics.li/e0ff4a5

  And yes of course you were all right, noise is 20MHz (~50ns period oscillation)
  I bypass the ne553x op amps rail to rail, and I bypass the 1646 rail to ground with a dedicated ground path

  When I put the ground of the probe on C3 or C4 grounded pin no noise measured at the psu, but when the ground of the probe is connected to "gnd" (and not "gnd1") on the pcb, noise appears, and if I connect the probe's ground at "PSU_0" noise is measured at the PSU rails.

Having that many grounds is a good recipe for having no ground at all. Also, having the power supply regulators that far away from the circuits they power is a bad idea. The inductance of a 15-20cm wire is pretty high. Sure, you have bypass caps on the load side, but they only do so much. Put the regulators right next to their load. You can move the rectifier bridge and regulator input filter caps further away, if you're worried about rectified currents sneaking into a sensitive signal reference, but the regulators can be right next to their load - why not take advantage of the low impedance output they produce.

Best of luck debugging this! I'd suggest a "dead-bug" layout over a piece of un-etched FR-4, using that un-etched board as one big ground. From there, move to a PCB layout, but realize that all of the copper you etch away making isolated grounds could have been used to make one big, low impedance ground that is high quality, instead of just adding series inductance to each of your 'isolated' grounds.

 

[bruno2000]

It seems a bit odd that the major console mfgs (SSL, Neve) only put in 100nF ceramics next to 5532/4 chips.  Comments?
What kind of values are you talking about?  4u7F?  Bigger?
Thanks so much for sharing your experience!
Best,
Bruno2000

http://hpics.li/e0ff4a5

  I bypass the ne553x op amps rail to rail, and I bypass the 1646 rail to ground with a dedicated ground path

You need to use ELECTROLYTICS AT each 5532/4 and 1646 (within 1cm) and they have to be from rail to ground.

Otherwise, even if your shebang doesn't oscillate, you'll have excessive THD.

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

 

[saint gillis]

You need to use ELECTROLYTICS AT each 5532/4 and 1646 (within 1cm) and they have to be from rail to ground.

Otherwise, even if your shebang doesn't oscillate, you'll have excessive THD.

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

Thanks Ricardo I've read everything and also the pdfs from AnalogDevice and others. I've also read and interesting article pointing out the differences between "Decoupling" and "Bypassing"

Well I have found the cause of the oscillation noise in the PSU, it could be interesting to analyse it a bit :

As it is : http://hpics.li/e4cdd66

  -  When I remove the 1646 chips, the PSU rails are ok, when I put them into the circuit it brings some 20MHz noise in booth rails, and when I put the gnd of the scopes probe at the PSU's ground ("PSU_0") I can also measur noise at C3 and C4 pins connected to "GND1".
  - Now I connect together "GND1" and "GND" on my pcb. The noise disappears

  The explanation must be simple, but I don't get it...

 

[abbey road d enfer]

It seems a bit odd that the major console mfgs (SSL, Neve) only put in 100nF ceramics next to 5532/4 chips.  Comments?

There are always(?) electrolytics in addition, generally 47-220uF.
IMO Ricardo's recommandation is not the only possible scheme.
There are two aspects that govern the choices: stability and minimization of distortion due to improper current circulation.
Stability is relatively easy to achieve with the typical pair of 100nF ceramic close to the opamp. minimization of distortion takes significant capacitance, but also requires careful analysis of current circulation, in particular how the current that is injected into the load comes back to the decoupling caps.
Ricardo's arrangement offers excellent local bypass instead of centralized bypass, but very often most of the loads are rather light, except for the output stage (that would be typical of many mixers). In that case, a centralized set of caps is perfectly adequate, provided the return path is correctly routed.

 

[ricardo]

I'm quite happy to argue theory with anyone but why don't you look at kingston's thread that I linked to.

He actually TRIES STUFF OUT AND MEASURES THE RESULTS.

MUCH better to look at test results than believe some guy pontificating on the internet.  Stability and good reliable THD performance are intimately related in most circuits using good OPAs.

You don't show where GND & GND1 are connected.  This EXACT physical point is important

You have also ignored my post #6

As an ex-Calrec guy, don't get me started on Neve & SSL  8)

Of course if you prefer your OPAs to all sound different, or are happy with a bit of oscillation here & there ... just ignore everything I say.

 

[abbey road d enfer]

I'm quite happy to argue theory with anyone but why don't you look at kingston's thread that I linked to.

It seems there is some confusion as to whom you are answering to. I'm the one who "argues" (although I'm not arguing, just giving a widerperspective), the OP doesn't. Please note that I was quite active in kingston's thread.

He actually TRIES STUFF OUT AND MEASURES THE RESULTS.

I actually did that in the course of my professional activity.

MUCH better to look at test results than believe some guy pontificating on the internet. 

Agreed.

Stability and good reliable THD performance are intimately related in most circuits using good OPAs.

I beg to differ. Indeed, unstability is definitely a source of distortion, but distortion can result from many other mechanisms, that do not necessarily correlate with unstability, particularly, as I mentioned, improper arrangement of return current flow.

You don't show where GND & GND1 are connected.  This EXACT physical point is important

You have also ignored my post #6

As an ex-Calrec guy, don't get me started on Neve & SSL  8)

Of course if you prefer your OPAs to all sound different, or are happy with a bit of oscillation here & there ... just ignore everything I say.

This should be addressed to the OP, not me.

 

[ricardo]

Abbey, I don't think we disagree at all.  And yes.  You were a important contributor to Kingston's thread.

 

[ricardo]

  - Now I connect together "GND1" and "GND" on my pcb. The noise disappears

  The explanation must be simple, but I don't get it...

Duu.uh!  Do you mean GND & GND1 WEREN'T CONNECTED in your initial tests? 

I hang my head in shame for not noticing this  :-[

 

[audiomixer]

I think we all missed that point. the 'sewer' ground needs to go to the common ground (I deliberatly did not use the word ground here...) of course, otherwise the filter caps are just rail to rail.

Michael