ectoplasm wrote:
(snip)
As a side issue, I came across following page:
http://www.dself.dsl.pipex.com/ampins/webbop/5532.htm
Quote:
"DECOUPLING & STABILITY. 5532 and 5534 type opamps require careful
supply-decoupling if they are to remain stable; otherwise they appear
to be subject to some sort of internal oscillation that degrades
linearity without being visible on a normal oscilloscope.
Right.
The essential requirement is that the +ve and -ve rails should be
decoupled with a 100nF capacitor between them, at a distance of not
more than 2 inches. It is NOT necessary, and often not desirable to
have two capacitors going to ground; every capacitor between a supply
rail and ground carries the risk of injecting rail noise into the
ground.
This is a bit muddled, I think. For stability, the chip may
need bypass capacitance, both rail to rail, and from each
rail to signal common. It all depends on where signal
current is going.
If you unhook the opamp output from everything but its
intended feedback paths (minimal load) then there is little
output current, though there may still be significant
changes in supply current through the part during some parts
of the signal waveform. If those supply current changes
cause changes in the supply voltage (due to distribution
trace inductance) and that supply voltage variation leaks
into the signal stream, either through the internals of the
chip, or through some external bias arrangement, then the
signal purity is compromised. If the effect is large enough
and the phase shift right, the whole circuit may achieve
oscillation.
Then hook the output back to its intended load. Now, you
have all the previous potential problems (pun intended) but
also new currents from the supply rails through the output
load to ground and/or back to the supply rails, depending on
how the load is connected (see output stage that drives
signal into a rail supplied bias network). So, now, the
signal swing causes bounce in the ground rail, as well as
new causes of bounce in the supply rails. All these new
bounces might contaminate the signal purity or even cause
oscillations.
I don't think it is a good idea to just bypass rail to rail
and ignore the load current and its coupling between the
power rails and the ground rail. I prefer to use two
capacitors in series connected to form the smallest possible
loop between the power pins of the chip, and their common
node, then connected to the local signal ground through a
common path. Alternatively, grounding the two capacitors to
two different parts of the ground rail, but adding a jumper
directly between their grounded ends works, too.
I hate to see designs with a chip between two ground rails
that are connected together at some distance point/s, with
one bypass capacitor going to one of them and the other
bypass capacitor going to the other of them. This puts an
unnecessarily large inductance in the path of the rail to
rail current changes.
If you are worried that this bypass arrangement is going to
inject noise from the supply rails into the signal ground,
then you need to think about where that noise is coming
from, and how to reduce it. The power rails must inject
current into the ground rails through any grounded load, so
you must figure out how to contain that supply to ground
current, locally, with bypass capacitors, so it doesn't
wonder all around the system. Having clean supply rails is
a separate but related problem.
The main rail decouple electrolytics can be used to do the job
for several 5532/4 packages nearby, and this cost saving is an
important layout point.
I disagree with this, for the above stated reasons. I think
large supply electrolytics have one function... to store
energy during the rectifier peaks, so that they can supply
power between them. And the charge paths have to be
separated from the discharge paths, so that they don't
inject ripple frequency voltages into the ground.
Likewise, it is not normally necessary to
decouple each package individually.
It may not be strictly necessary, but, unless you are
willing to determine the effects, experimentally, why risk it?
One capacitor every few inches is
sufficient if the power tracks are of reasonable thickness. (ie 50
thou)"
Length matters more than thickness, in this case, since we
are dealing with inductive effects, mostly. Sharing bypass
capacitors across several inches of trace length ( in
precision analog circuits, not TTL logic) is a recipe for
cross talk and instability. The whole concept of bypassing
is to contain the highest frequencies of supply and load
currents to the smallest possible (lowest impedance at the
highest frequencies) part of the system, so that the DC
rails do not act as high frequency signal paths.
So according to the author there must be a 100nF (electrolytic, he
says) capacitor between the Vcc and gnd pins of the 5532 nearby the
5532.
I think you misunderstand. The electrolytic part of the
discussion referred to the large capacitors (main rail
decoupling caps), not the individual chip decoupling
capacitors, which are not shown, at all. The 100 nF caps
should be low ESR ceramic or film capacitors. I avoid Z5U
and Y5V ceramics, and use X5R or X7R types as surface mount
bypass, because, for the same capacitance and voltage
rating, they generally have lower ESR. I like the Panasonic
V series of stacked film capacitors for film bypass in
through hole applications, though there are lots of good
multilayer ceramic caps too.
Has anyone ever heard of this requirement and is it a necessity? I am
just surprised it is missing in the headphone amp's circuit (there are
100nF's in the power supply).
Schematics gives no guidance (or only hints) as to exactly
where any of the bypass capacitors go. The power supply
schematic for this project hints that there should be a pair
of 100 nF caps closely wrapped around the voltage
regulators. Take that hint.
I would add a pair of 100 nF caps for each opamp chip, and
mount the pair of 10 k resistors that bias the output stage
so that they connect to the supply rails right where the
bypass caps for the last opamp connect to those rails.
In addition, I would use the pair of 1000 uF storage caps,
for the 22 volt rails in the supply, as bypass capacitors
for the output stage (since that stage is their main load),
connecting them directly between each of the output
transistor collectors and the ground return point for the
headphones. This means that the current from rectifiers to
these capacitors must have separate traces than the load
current leaving them, especially on the grounded side. That
headphone ground return point then becomes the star ground
reference point for the rest of the circuit, since all other
ground currents are tiny, compared to the headphone
currents. Route one extra ground trace for 2 grounded
points in the output driver stage, (and their bypass
capacitors) and another trace back to the first stage for
the 6 places ground is needed there, and to its bypass
capacitors. If you want to limit any possible noise
injected into this branch via the bypass capacitors, add
resistance between each of the supply rails and the opamps
and their bypasses. 100 ohms should do it. Precise voltage
regulation is not at all important, there, but injected
signal voltage from the output stage, at high frequencies,
where the supply rejection ratio of the opamp is not high,
is important.
It is also quite possible the the circuit would work
acceptably, if you did most of this sub optimally. I just
went through the concepts for your education. Lots of ratty
stuff works well enough that nobody notices the imperfections.
