Bob Cordell's Power amplifier book

With a careful layout, it is easy to limit signal ground to very localised areas such that a plane is not required for signal ground and it can be starred. The plane can then be used for HF-critical things like decoupling and zobels. Star grounding is not suitable for decoupling and zobels as it has high impedance at HF.

I like to pick a point on the ground plane to be the "one true audio signal ground” and to here connect speaker return and a twisted pair carrying the feedback signal and signal ground. The other end of the feedback twisted pair then connects to a small plane around the input stage, to which the input signal & ground is connected.

The only thing I haven’t figured out yet is the best way to route the ground signal for loop-gain compensation (if a loop-gain compensation configuration that involves ground is being used).

Harry, I don't necessarily disagree. But this pretty sophisticated stuff and not always adhered to by diy-ers.
It's much easier to avoid all that and terminate the zobel for instance at the same point that you terminate the speaker return (assuming you do that sensibly).
After all, you wouldn't terminate the speaker return at the ground plane either, right?

Jan
 
Harry, I don't necessarily disagree. But this pretty sophisticated stuff and not always adhered to by diy-ers.
It's much easier to avoid all that and terminate the zobel for instance at the same point that you terminate the speaker return (assuming you do that sensibly).
After all, you wouldn't terminate the speaker return at the ground plane either, right?

Jan
Not the same signals.

The Speaker Return carries audio signals and maybe a bit of RF interference. The PSU supplies the Speaker Current. The Speaker Current, eventually, MUST RETURN to the PSU.

The Zobel Return does not carry audio signals, It carries the harmonics created by the half sine waves starting and stopping as the amp crosses over from +ve to -ve current flows and back again. These HF and VHF signals NEED a low impedance route to flow.

The Speaker Return and the Zobel Return do not NEED the same type of node termination.
 
That's a pretty normal behavior (short of the low 1nF that triggers the oscillations).

A large load capacitance may be seen as a shunt frequency compensation, the same way as a voltage regulator is compensated by the output cap (and usually requires an electrolytic with rather large ESR and ESL to remain stable). Such a large cap is limiting the amplifier ULGF, effectively moving the dominant pole at the output cap. As such, the phase stabilility condition is largely met, and all you get is the amortized transient response of the output cap and the inherently inductive output impedance in any global feedback amplifier.

Now, when the output cap value is decreased to some 10-100nF, this no longer creates a dominant pole, but just another pole that falls in before the ULGF. As such, the gain drops at 12dB/oct and the phase may go beyond the stability condition. Please note that this output pole is NOT splitted by the negative feedback network, because it is NOT in the feedback loop!

This would explain why most amplifiers don't like a 10-100nF cap at the output, being in fact much more tolerant to 1-2uF. The open loop output impedance plays also an important role in this behavior.

It would though not explain why your amp starts oscillating with 1nF of output cap. I would think that your amp has an unusually high ULGF, and it does not properly separate/split the open loop poles. Or the gain margin is to low at HF. As such, another pole at HF is just enough to trigger a local or global oscillation condition. One to another, you should review your amp frequency response around and beyond the ULGF and see how a 1-10nF affects it. This could be difficult to simulate, given the current devices stock models validity.

This explanation is confusing and seems incorrect if I understand well.


In a regulator, the dominant pole is made by the output capacitor of the regulator circuit. But the loop gain of the regulator ( if based on an opamp) includes the closed loop gain of the opamp itself acting as an error amplifier. This opamp is cabled in the loop as a positive voltage amplifier (closed loop) with large bandwith. The opamp shoud contribute no poles to the loop gain of the regulator feedback loop for obvious stability reasons.

In this case, if the amplifier do not oscillate with a large capacitor it means that the pole created by the load capacitor with the output (small) impedance of the open loop amplifier is still higher than the crossover frequency ( gain bandwith) of the loop gain. If ( and that is reality) the output impedance of the open loop amplifier is not zero then the load capacitor IS PART of the loop for stability analysis.

This amplifier is supposed to be a miller compensated dominant pole topology where the dominant pole is created by the compensating miller capacitor which is itself in a minor feedback loop in the overall loop. The load capacitor will not created a dominant pole which is still created by the pole splitting effect of the minor feedback loop.

Beside the overall loop, there are two other internal feedback loops where oscillation can take place: the miller minor loop and the common collector class ab output buffer.

The minor loop can oscillate for example if a resistor is included in serie with the miller capacitor to create a zero cancelling the output pole created by capacitive loading; A rapid small oscillation riding on the step response of the amplifier is a sign. The oscilloscope trace becomes fuzzy. The cure is a small capacitor in // with the miller RC network but this cure is load dependend.

What happens here seems to be output buffer oscillation.
Most buffers are two or three stages of common collectors in cascade ( drivers and output transistor).

Due to beta dropping with frequency, the output impedance of a common collector is inductive if there is a resistor in the base. The common collector is an impedance transformer. If there is an inductance in the base, it is easy to show that the output impedance becomes an inductance in serie with a negative resistor. This inductance in the base of the output transistor is created by the driver common collector output impedance.

In a passive one pole network( resistance, cap, L ) the current is coming in in the positive terminal and out at the negative, therefore absorbing power. In a source the current is coming out of the positive terminal therefore a negative resistance generating power.

At high frequency, circuits are analyzed with twoports. It can be shown that if a feedback network is replaced by its twoport equivalent, an equivalent stability condition to the loop gain analysis is : a two port is unstable if the input/output admitance is real and negative when the output/input is loaded with a passive load. Intuitively this means that the circuits generates power to the input and because the output is passive, it means that the circuit oscillates to generate this power.

If we load a buffer with a capacitance // with the load, this capacitance resonates with the output inductance and the admittance is real at resonance. If the remaining resistance of the circuit is smaller (in absolute value) than the negative resistance of the buffer output impedance, oscillations will occure.
.
The output coil of an amplifier is a cure. At high frequency the coil is high impedance in // with resistance. This resistance will add with the negative output resistance and keep the admittance positive. At low freq. the resistance is bypassed and no power is lost. A test would be to increase the resistance.

Another view is peaking. The output of the buffer is a // LRC tank. To decrease the Q so the peaking at resonnance, a zobel or snubber may be introduced. If this network is a resistance at resonance of the LRC thank, the resistive load will be decreased ( load//Rzobel) and so the Q so the peaking.
A test would be: use a low zobel resistor but make the C high enough to have the brreak frequency of the Zobel zero way below the resonance of the load but not too high to influence the audio range
 
A trace over a ground plane can have less inductance. More than a direct connection to a ground plane, but less than a plain trace. One option is to run star/bus ground schemes normally, and simply add a ground plane which is only connected to one ground point and nothing else. This way no currents can flow in the plane except those generated by mutual inductance with nearby traces.
 
Harry, I don't necessarily disagree. But this pretty sophisticated stuff and not always adhered to by diy-ers.
It's much easier to avoid all that and terminate the zobel for instance at the same point that you terminate the speaker return (assuming you do that sensibly).
After all, you wouldn't terminate the speaker return at the ground plane either, right?

Jan

Hi Jan,

Not sure I agree.

First of all, grounding choices and topology appears to be somewhat of an art, in all fields of EE. So there is honest disagreement and different experiences.

Also, grounds in SPICE are "perfect" and introducing appropriate real-world parasitics into simulations is almost impossible.

The Zobel is there to make sure that the EF output stage has a reliable, resistive, load at HF. For that reason, it really should be on the amplifier side of the L-R network and should be tightly associated with the HF return path of the output stage rails.

Although star grounding is the Holy Grail for some, I do not believe in taking it to extremes, especially when to do so introduces a lot of extra inductance. I have never built an amplifier with a ground plane, although it has been tempting. If so, it would be a dirty ground plane. Signal ground should be treated somewhat as if it was a signal. Also, as I have mentioned, I always have the power supply-proper as a separate group of components, and connect it to the amplifier circuit board through a twisted triple of +, - and ground. Not everyone agrees with this.

I believe that currents, especially at high frequencies, should be "resolved" locally. Ideally, only fairly low-distortion sinewave currents should be flowing in the triple of power supply lines that connect to the board. I also do not believe that the power supply lines need to be huge and bulky, and have even once or twice wondered if a little resistance in those lines might be a good thing, promoting more local resolution of the currents and a tiny bit more filtering against the 1000uF or so located on the amp board. 10 or even 20 milliohms might not be bad.

Indeed, I think once or twice I have used microphone cable to connect the power supply to the amplifier board. Whenever I prototype an amplifier, I have a triple of lines at least 18 inches long going back to a power supply module, for convenience of testing, and have never had a problem. This is a strong example of NO star ground back to the power supply - but a local quasi-star ground on the amp board. This is what I refer to as the star-on-star topology in my book. I don't think rigid aderance to a single star ground architecture is necessarily good.

The key is always that we watch where the currents go and manage their path knowing that it will be the one of least impedance.

Some of these issues may also be the reason that dual-mono power supplies are favored by some.

Just some thoughts and personal prejudices.

Cheers,
Bob
 
But the loop gain of the regulator ( if based on an opamp) includes the closed loop gain of the opamp itself acting as an error amplifier. This opamp is cabled in the loop as a positive voltage amplifier (closed loop) with large bandwith.

Short of this paragraph which, if I understand correctly, is wrong (and so are the eventual inferences from it), I can't find any difference to what I said. Only lots of extra details.

It is as simple as that: if you do a loop gain analysis of the amp, the output impedance of the amplifier, and the load cap are an RC cell that introduces a pole in the forward gain. If the load cap is large, this pole can become dominant, and the amp is perfectly stable.
 
Hi Jan,

Not sure I agree.[snip]

Just some thoughts and personal prejudices.

Cheers,
Bob

Bob,

I think with all the subtleties and stuff we really don't disagree very much. The initial remark came from the fact that the OP returned the zobel to his ground plane without further details. My view is that this is a dangerous practise, nothing more.
It may work fine, or it may be the cause for some low-level hf oscillations.
It depends on many details but I just wanted to hint at another thing he could check on.

A page further on I get slapped with zero-length wires. Go figure.

jan
 
...............I believe there are ways to achieve low impedance at HF and VHF using physical traces and/or wires, rather than planes.

.............with discrete wires you are always up to the unaviodable L. How can you circumvent that then?............

If the wire is zero mm long then the SMD component is attached to the node with zero trace impedance.
All that is left is the impedance of the smd termination.

Ahhh yes thanks for reminding me; I'd have to reorder some of those, plain ran out just today...........
You asked a specific question and I gave an answer.
Then you say:
A page further on I get slapped with zero-length wires. Go figure.
If you interpret "omitting a wire" as a slap then ask a different question.
I was responding to the thought that planes solve everything and an alleged claim they are the only good solution.
 
I was responding to the thought that planes solve everything and an alleged claim they are the only good solution.

OK, yes, with a star ground you could achieve low impedance at HF, but only if you run all your star ground traces (“return” current) next to or underneath the traces that are carrying the corresponding “out” current. Let’s remember that current flows in loops and you cannot consider the length of the “ground wire” by itself. What matters is the area of the loop enclosed by the path of the current in question. Ground planes are a straightforward way of ensuring that these loop areas are as small as possible, as current will always flow in the loop of lowest impedance.
 
It seems that it is easier to mess things up than it is to get it right and that is why this is such a contentious issue. As Bob said grounding by itself is an art that should be appreciated. How many times do you look in a consumer audio amplifier and all the wires are routed so nice and pretty and every wire is run in parallel, no twist anywhere to be seen, oh my that looks nice, so professional! Then again just because you use a ground plane does not mean there are not multiple attachment points for various return paths and here we go again, so it seems messing things up on this side of things is mostly the standard.

Personally not having the education with regards to board layout and all the rules of running traces near each other it just seems that the twisted pair and twisted triples are the easier to us and implement. I would think using ground planes would take a higher level of understanding to get correct and not create a nice flat antennae.
 
I believe there are ways to achieve low impedance at HF and VHF using physical traces and/or wires, rather than planes.

Yes, in ground planes at HF the return current flows directly under the trace if it can, so there is similar impedance in the loop
if an uninterrupted ground return trace directly under the trace is used, instead of a plane.
 
Last edited:
Originally Posted by AndrewT If the wire is zero mm long



He meant that an smt component can be directly connected to a ground plane located on the same layer,
with only the shortest (say, 5 mil) trace length for thermal relief.

I understand that, but what does this have to do with where to ground a zobel?
Posts like this pull the tread into oblivion even before you have any chance to develop a meaningful exchange.
 
it just seems that the twisted pair and twisted triples are the easier to us and implement. I would think using ground planes would take
a higher level of understanding to get correct and not create a nice flat antennae.

The HF ground return currents automatically follow the traces above the ground plane if allowed to, with no breaks in their paths.
This is much better in general than routing traces. You do need to take special care with high impedance junctions like op amp inputs,
sometimes removing the planes in those areas.
 
Ground planes are a straightforward way of ensuring that these loop areas are as small as possible, as current will always flow in the loop of lowest impedance.

You should say " most of current will flow in the loop of lowest impedance" otherwise someone my think that in parallel resistors all current will flow trough the resistor of the lowest impedance. As this is not true, a ground plan is not an easy way, specially not for us hobbyist.