(2) dirty/clean ground again [sdiy]

media.nai at rcn.com media.nai at rcn.com
Sun Aug 18 19:53:44 CEST 2002

At 6:43 PM +0200 08/17/02, jhaible wrote:
>a) SNR - an opamp has normally excellent power supply rejection
> (noise from the supply pins not showing at the output), but only
> for low frequencies. At high frequencies, you need a filter in the
> supply trace, i.e a capacitor or even an RC circuit, if noise is
> an issue.

That makes perfect sense, and why capacitor inductance could be an issue.

> Oscillation - this is related. If a signal can go from the supply pin
> to the output, and if the output of this, or another opamp in the
> circuit, has an effect on the supply rails, you can build loops of
> positive feedback that way. The latter part involves load _currents_
> - but keep in mind that stray capacitance can be enough "load" for
> frequencies in the MHz range, where the oscillations occur.
> Depending on the specific opamp you're using, internal currents
> might be enough to for musch a loop, if the power supply rails
> don't form a "shunt" (decoupling ...) to keep the loop gain low.

Is it correct that both this Mhz noise and these oscillations can be
prevented by placing a small cap in parallel with the feedback resistor??
(it seems that such caps are used in the Blacet TM).

>b) Most opamps have a voltage gain stage that forms an integrator.
>    This often means that feeding noise into one of the power
>    supply pins is feeding it in *after* the integrator. So the negative
>    feedback loop around the opamp, thru the input stage, and thru
>    the integrating voltage gain stage has a "disadvantage" against
>    HF that comes in from this power pin. The other side is usually
>    much less sensitive. So if you only want to use a single cap, it
>    can sometimes be better to go from -vee to GND rather than
>    from +vdd to -vee.
>    I'm not able to deduce this in all detail. But let this be a hint:
>    Sometimes, when you use a single supply (-vee = GND), and
>    two resistors to form a "midpoint" on the positive opamp
>    input, it can be necessary to add a cap across the resistor from
>    the midpoint to GND (= -vee). (You also need this for SNR,
>    if you are referencing your signal to GND elsewhere in the
>    circuit, but I have *also* seen this being called for because
>    of stability.)

Could you say that again, but in German??  That way it won't bother me that
I can't understand it :)

>c) Opamp manufacturers have a tendency to demand a decoupling
>    with which the opamp is stable under all circumstances. Normally
>     you are not building circuits for all circumstances, but for a single,
>     dedicated application. I have sucessfully built boards of
>     160cm x 233cm size with nothing more than one pair of aluminium
>     electrolytics and no other decoupling caps at all. But ...
>d) ... sometimes I have fought desperately to get a single stage stable,
>    even though all decoupling from the textbook had been applied.

I'm under the impression that using "extra" decoupling caps can't hurt.
For a one-time project, the additional cost seems negligible.

At 9:52 PM -0400 08/17/02, harry wrote:
>I'm gonna jump in here with some philosophy....

Oh boy...

>We "wish" that ground was an infinite sink for current... that we could
>>dump any amount of current into it and not have the voltage rise...
>and for small currents in short traces at low frequencies...we often get
>>our wish.
>But really the ground has finite resistance, and inductance.  So when we
>>pour that "shotglass" of current into it at one point... the voltage
>>rises...and it take some time for the current to reach the point where it
>>will get cancelled out (meeting its equal
>but opposite current back at the power supply.)

In other words, it is a system of equlibrium, which takes time to settle.
To put it another way, when an op-amp requires additional current (to
balance an equilibrium of its own) there will be a momentary voltage drop
at the supply pin due to the impedance of the supply rail (resistance and
inductance of the traces), and the decoupling cap compensates for that drop
by discharging.  I have heard large ( >10uF) caps described as "local power
storage", but I have not heard this description used for little decoupling
caps, although it seems that is true for them as well.  Furthermore, big
electrolytics would be insufficiently slow in this regard (as well as being
insufficient HF filters) due their high series resitance and inductance
when compared to monolithic ceramic caps.

Going back to what Roman was saying about ground being zero, even at
equilibrium without any transient currents, if PCB traces have resistance
and curent is flowing through them, then there has to be a voltage
difference between one end of a ground trace than the other, right??  If
there is an I across R then V can't be zero at both ends.  Either that, or
I'm totally screwed and I need another hobby.

So how can we solve this??  Can ground be regulated??  I've never seen a
7800 :)  However, a voltage  can be regulated, and if that regulated
voltage is held on one side of a cap, wouldn't that locally "re-zero" the
ground on the other side??  Perhaps this goes back to JH's point b) above,
by stabilizing the local ground of a single-sided amp, you stabilize the

>I prefer to think.... Hot Water Pipes.... Cold Water Pipes.... Drain Pipes.
>Now can you see why putting (lets say) separate valves from hot and cold
>>water to the drain...makes more sense than valves from hot to cold....

So the trouble is that ground is not an infinitely large pipe -- too much
water too fast and it will back up.

>Most load currents end up going from positive supply to ground... or
>>negative supply to ground... so thats where the decoupling caps belong.

I think Roman's point was that the load is across the rails for an op-amp
using a bipolar supply.

>Some rare cases have the load current going across the rails... I often run
>relays in my audio designs across the rails, rather than from one rail to
>>ground. Decoupling these from rail to rail makes sense (and I do).
>> >Isn't it more justified to use single cap just across power rails?
>> I've seen that used in synth circuits, and it is the basis of IC sockets
>> with built-in bypass caps.
>Au countaire... these sockets have the decoupling cap across pins 7 and
>>14...or 8 and 16... which screams that they are for 4000 series CMOS, or
>>HC, HCT, etc logic families with single supplies...

Good call :)  I had a feeling they weren't for op-amps because there aren't
any 8-pin decoupling sockets, and dual op-amps are very popular.  I hadn't
considered that they weren't across pins 4 and 11, either.

More information about the Synth-diy mailing list