Lions and Tigers and Grounding Problems, Oh My! DoZ Grounding...

[ilimzn]

Originally posted by EchoWars
[B]So ilimzn...you suggest the grounding of the PC board be done something like in this drawing?

Since I'm only showing one channel, I don't really show how the L & R inputs are both grounded, but the ground (signal return) from the input phono jack on both channels joins at the ground of the volume control (VR701), and return together (one wire) to the negative terminal of the cap.

OK, the separation of the grounds on the board is drawn correctly but the connection to the various ground returns is not. I'll get to this later.

Other suggestions I've gotten have implied that the input signal ground should also include the ground for Q1's emitter resistor, R3, and instead of returning the ground back to the power supply cap (as it is now, running from the ground of VR701 to the cap), to remove this wire and join all these input grounds to the PC board ground with a 10 to 100 ohm resistor.

The standard scheme of separating the power and signal ground by a 10 ohm resistor is not easy to directly apply here as it generally depends on the differential action of the input and FB connections in an amp. Normally, the 10 ohm resistor is really just there for safety as the ground return is provided anyway. But this is taking steps ahead od ourselves.

Let's first look at the schematic a bit. In your second pic the grounds connected in magenta can be considered signal grounds.

Why not the ground end of R3? Because Q1, in it's amplifying role, actually generates collector current proportional to base-emitter current (in fact, if we were strict, it would be base-emitter voltage), and this collector current then generates the bese-emitter voltage for Q4+Q5 combined. Therefore, the 'ground reference' for R3 must be the same as that for Q4+Q5, which is the emitter of Q5 and ground end of R5. This point can actually freely 'shift around' due to voltage drop in it's ground connection, without affecting output (in theory - in practise it depends on how good Q1 is as a controlled current source), because the voltage drop on R3 stays the same. If R3 was grounded somewhere else, the ground wire to Q4+Q5 would become an added emitter resistor, thus a form of emitter degeneration/feedback.

Why not the gorund end of Vr1? Because the DC level set by it remains the same either way, what we are concerned with here, is taking out the remaining AC from the bias voltage of Q1 - and this is what C6 does, so it's ground side must be firmly tied to the signal ground - otherwise any AC voltage difference between the signal ground and the ground of C6 gets directly injected into the base of Q1. What is worse, if there is a load current generated voltage drop between these points, it will be a form of positive feedback. The ground side of Vr1 can then be tied to 'another' ground point because C6 will filter out any injected AC from it anyway.

Why not the ground ends of D701 and C702? Similar reason as for the ground end of Vr1 - simpler and any remaining AC gets filtered by C6, while the DC remains largely the same - perhaps a few mv difference due to voltage drop in the ground line to Q4+Q5.

Why not C5 or the output zobel? Because the AC loops they close run directly around the output stage. With C5 one could argue that it does not enclose a part of the output current loop, because as drawn, that current loop has a separate ground onnection. This is valid to a point - but at some point one has to consider the added complication and diminishing returns. C5 is there to provide a short path for any currents generated internally in the amp, most of the load current decoupling will be done by the PSU filter banks (BTW there is a remaining problem there, I'll get to that later).

Finally, Why is C2 connected to the signal ground? Well, this is actually the ground reference for the amp's inverting input. The fact that it is quite low in impedance may confuse a bit, but never the less, if there is any difference in voltage between that point and the signal ground, it will be treated as a differential input voltage - not something we want. Obviously, if there is a long wire run from here to the rest of the signal ground, we risk amplification of any induced AC.

OK now let me get back to the connection of the power and signal grounds.

The ground rule is: theoretically, 'no current' should flow through the signal ground. In actuality, this means no load current, and especially no extra currents of any kind, be it induced or generated by rectification. The signal ground currents should only be those generated by the signal and feedback voltages into their respective load impedances, i.e. quite low! The reason for this is that typically any voltage difference between various signal ground connections become the amp input signal, and get amplified (in some cases by the open loop gain!). Even though it is sometimes possible to provide near zero current in the signal ground (with amps that have input and feedback impedances equalised), this is not the case with the DOZ - but the idea behind the ground rule still applies. It just means that any currents flowing through the signal ground connections, which of course need to have as low impedance as possible, will generate negligible voltage differences. In practise this means:
a) Load current must not flow along the ground path between two or more signal ground connections
b) Power supply current must not flow along the ground path between two or more signal ground connections
c) And especially, Rectification current must not flow along the ground path between two or more signal ground connections

Now have a look at the grond connections in the schematic you provided - and, BTW you whould really draw it for both channels, in order to see where possible loops can happen if the left and righ input grounds are connected (as they will be once a source is connected to the amp).

Here you have 2 problems:

1) All grounds end up on the filter cap
In theory, this would be fine, except the actual connections do have non-zero impedance, and the cap charging current from the rectifier is VERY high. With all the terminals stacked and a screw running through them, are you sre you know where the actual rectification current is going, and if there are any differences between the various ground attachments? Remember, we are talking sub-mV stuff here - and 1A over 1m ohm will already be 1mV.
On the other hand, you don't have such a problem with the + terminal, as you only have the rectifier attachment. The RC filter resistors take care of limiting any remains of the rectification current into the secondary filter caps. In order to solve this problem, the same approach can be used for the ground side.
Here is how you do it:
- Attach ONLY the rectifier ground to the ground side of the main filter cap.
- Connect the two ground lugs of the secondary filter caps by a thick wire. If needed, re-orient them to make the wire as short as possible, however this is less important than re-arranging the ground connection as I am about to explain, so an optional if it's not practical. You can use thick bare wire for this, in any case you will need a place to connect two more wires to the middle of this bridge connection.
- Connect a single wire between the ground of tha main filter cap and the midpoint of the ground 'bridge' between the secondary filter caps that I described above. The purpose of this mod is to keep the largest possible amount of rectifier ripple current flowing through the main filter cap, and keep it as much as possible out of the secondary filter caps. The midpoint of the bridge wire where the ground from the main filter cap attaches is now your clean power ground - clean because it is as much as possible free from rectification current, i.e. it is 'well filtered'.

2) Now that a clean power ground has been created, you still have the problem that you actually need two - because you have two physically dislocated amp boards. Because it is impractical to solder many ground returns to the clean power ground (and you would need at least 5!), although it's not the 'perfect' solution, the best would be to use a thick wire and connect this clean power ground point to the copper ground bar.
At this point you need to separate the clean power ground into 3 different ground wires per channel, plus the chasis ground.
Remembering the rule that (theoretically) no currents should flow between ground connection points on the signal ground, and keeping in mind you have 5 attachment points on the ground bar, you can use the ground bar as follows, attachments from RIGHT to LEFT:
- Amplifier board power grounds (attach both channels to the same point)
- Clean power ground from the filter caps
- Load return (attach both channels to the same point)
- Signal ground (attach both channels to the same point, but read further)
- Chasis ground

The chasis ground connection, being 'blind' - i.e. no current flowing - could in theory be attached anywehre, and if it is more practical, you can move it around as it suits you. However NEVER attach signal grounds between two other attachments that have load or power supply current flowing between them!
The clean power ground attachment should be as short as possible as it is a direct addition to the filter cap ESR/ESL, however, this is econdary as fars a hum and noise is concerned, high impedance here 'just' lowers channel separation.

[ilimzn]

Final part:

One more possible problem:
Because the signel grounds are the amp input ground reference, it ispossible to introuduce foreign signals by forming loops around the wires making up the signal ground connections (which eventually end up at the input connector grounds) and the grounding in the input interconnect cable to the source.
What you get is a sort of diamond-like structure, with the source grounds of both channels connected on the source, the signal grounds of both channels connected at the ground bar, and the actual signal ground points on the left and right channel PCBs. If this loop has any AC magnetic field going through it, there will be induction. The loop sections themselves will for a sort of bridge-like divider, and the voltages along the sections will get superimposed with the input signal, and amplified.
One way to sort this out is to minimize the area of this loop - by carefully routing the signal ground wires to the boards and the input connectors.
The other way is to reduce the current that can flow in this loop should any induction happen, so that the voltages developed along the ground wire connections are minimized. This is where the famous 10 to 100 ohm ground loop isolation resistors come in.
In the DOZ the feedback path impedance is quite low, and it is not the same as the input impedance, so something like 10 to 100 ohms would be far too much, because there would indeed be signal dependent voltage generated on these resistors. Something like 1 ohm would be more appropriate - and even that will reduce any currents induced in the loop by a factor of 100 or so, because the impedance of the groung wire or shield in the input interconnect cables is much lower than 1 ohm. That being said, these resistors whould NOT be necessary if it is possible to take care of the problem 'mechanically', i.e. by proper ground wire routing.

Finally, a bit of advice:
It is not necessary, and indeed may be counter-productive, to use very thick wiring between the transformer, rectifier and main filter cap.
One way of looking at rectification and ripple, is to remember that the transformer + rectifier + wiring impedance vs filter cap impedance (including ESR/ESL) forms a sort of divider. If the left side of this 'equation' results in a low impedance, the net weefect is an increase of ripple current, and HF portions of ripple voltage (because the vawevorm edges are steeper, ESL becmes more signifficant + diode reverse recovery becomes more of a problem). Besides, there is a second stage of RC filtering right after the main filter cap. Using thinner wires (or even chokes, low value resistors, RFI supress rings...) actually results in better filtering, and less stress on the transformer, rectifier and cap. Keep in mind that the PSU internal impedance is actually that of the filter cap, NOT the rectifier and transformer - in fact, in the short time these are 'visible' by virtue of the diodes conducting, thu get ripple - a necessary evil, not something you want to make bigger. So, no need to double up the wiring there, quite the oposite.
Also, no need to go overboard with braiding - the purpose is to reduce the cross section of the current loop - braiding insures that once so reduced, the wires stay put. But, so will cable ties on parallel run wires, or co-axial cable. You chose what is most practical, everything else quickly runs into diminishing returns.

[nigelwright7557]

Originally posted by ilimzn
Final part:

One more possible problem:
Because the signel grounds are the amp input ground reference, it ispossible to introuduce foreign signals by forming loops around the wires making up the signal ground connections (which eventually end up at the input connector grounds) and the grounding in the input interconnect cable to the source.
What you get is a sort of diamond-like structure, with the source grounds of both channels connected on the source, the signal grounds of both channels connected at the ground bar, and the actual signal ground points on the left and right channel PCBs. If this loop has any AC magnetic field going through it, there will be induction. The loop sections themselves will for a sort of bridge-like divider, and the voltages along the sections will get superimposed with the input signal, and amplified.
One way to sort this out is to minimize the area of this loop - by carefully routing the signal ground wires to the boards and the input connectors.
The other way is to reduce the current that can flow in this loop should any induction happen, so that the voltages developed along the ground wire connections are minimized. This is where the famous 10 to 100 ohm ground loop isolation resistors come in.
In the DOZ the feedback path impedance is quite low, and it is not the same as the input impedance, so something like 10 to 100 ohms would be far too much, because there would indeed be signal dependent voltage generated on these resistors. Something like 1 ohm would be more appropriate - and even that will reduce any currents induced in the loop by a factor of 100 or so, because the impedance of the groung wire or shield in the input interconnect cables is much lower than 1 ohm. That being said, these resistors whould NOT be necessary if it is possible to take care of the problem 'mechanically', i.e. by proper ground wire routing.

Finally, a bit of advice:
It is not necessary, and indeed may be counter-productive, to use very thick wiring between the transformer, rectifier and main filter cap.
One way of looking at rectification and ripple, is to remember that the transformer + rectifier + wiring impedance vs filter cap impedance (including ESR/ESL) forms a sort of divider. If the left side of this 'equation' results in a low impedance, the net weefect is an increase of ripple current, and HF portions of ripple voltage (because the vawevorm edges are steeper, ESL becmes more signifficant + diode reverse recovery becomes more of a problem). Besides, there is a second stage of RC filtering right after the main filter cap. Using thinner wires (or even chokes, low value resistors, RFI supress rings...) actually results in better filtering, and less stress on the transformer, rectifier and cap. Keep in mind that the PSU internal impedance is actually that of the filter cap, NOT the rectifier and transformer - in fact, in the short time these are 'visible' by virtue of the diodes conducting, thu get ripple - a necessary evil, not something you want to make bigger. So, no need to double up the wiring there, quite the oposite.
Also, no need to go overboard with braiding - the purpose is to reduce the cross section of the current loop - braiding insures that once so reduced, the wires stay put. But, so will cable ties on parallel run wires, or co-axial cable. You chose what is most practical, everything else quickly runs into diminishing returns.

I think this thread has simply gone way over the top now.

I have built numerous amps with their driver boards on veroboard and just connected all the ground along one track on it and they all worked great. No hum at all.

I am pretty sure messing around and seperating ground will cause problems rather than fix them.

Another cause of hum is poorly bpassed supply lines.

[EchoWars]

I had hoped that that would be the way it would work for me...no such luck.

I'll get the amp back on the bench soon and try something else following your suggestions ilimzn. But honestly, I'm so tired and frustrated trying to solve the problem that I nearly get ill just looking at the thing.

I'll see if I can't get back to it by the end of the week.

[EchoWars]

Well if anyone was wondering what happened (doubtful...), the story goes like this:

Someone contacted Rod, and showed him this thread. Rod answered this person back with the opinion that the level of hum experienced seemed to be too low to be explained away with shared grounds between the input and output, or ground loops. His suggestion was that the hum was the result of power supply ripple, and that I should strongly consider building the capacitance multiplier for the power supply. Rod was adamant that there was nothing wrong with my wiring, and that no changes should be made to the PC board.

Indeed, the level of power supply ripple with the 20,000µf 50V cap in the supply was about 140-something millivolts (I can't remember the exact number). But all I had read, here, on Rod's site with design and building notes, and in Rod's forum, led me to believe that the design was not terribly finicky about supply ripple, and I doubted that this would make much difference. But, I had a 82,000µf 50V cap here for another project and decided to install it. If the hum 'got better', then perhaps there was something to Rod's suggestion.

So I installed the big cap. The hum dropped from about -67db to -77db. A 10db improvement, and measured ripple dropped from 140-something to about 66mV. So perhaps there was something to this.

It took a while, between other projects and a 3 week vacation, but last night I got the cap multiplier finished. The multiplier is similar to the positive side of the design shown in Figure 3 here, only using a 2SD669A for the small transistor, and a MJ15004 for the large pass transistor. The cap on the base of the 2SD669A is a 2700µf 50V Panasonic. The big TO-3 transistor is mounted on a new heatsink on the top side of the amp, and a small self-fabbed PC board containing the rest of the circuitry is mounted underneath.

Anyway, I crossed my fingers and fired it up. Available supply voltage dropped from about 41V to about 37.5V, but I measure about 1mV of ripple on the supply voltage. After resetting the center voltage and the bias current, I connected the Tek analyser and measured noise at about -89db under 1W. Not as good as I had hoped, but the analyser lumps all noise together, so perhaps the figure is misleading. At any rate, it is a 23db improvement over where I started.

Connecting speakers (JBL L36 Decade's), I can hear no noise or hum at all coming from the woofer. From the tweeter, I can hear a very faint buzz, close to inaudible even with my ear next to the driver.

I've fought this amp for long enough, and while PC board changes could 'possibly' result in a lower noise level, it seems apparent to me that the effort would probably not justify the gains.

I appreciate the help that was given to me by you all, but seems we may have been looking left when we should have been looking right.

[fotios]

Originally posted by EchoWars
Well, the AC wires running to the rear panel (that go to the remote transformer) have been removed and twisted, and the AC wires running from the barrier strip to the diode bridge replaced with twisted wires as well. The positive and negative wires from the bridge to the main 20,000µf power supply cap were also replaced with twisted wires. (see pic)



Made no difference in the hum. Or, if it did make any difference, it is too small to tell.

Perhaps it is time to tackle the grounding at the PC boards?

Why you are using these twisted pairs of cables going or arriving to/from the center capacitor to the bridge rectifier or to/from the copper bar? To increase you the thickness of conductors? Have you tried to replace these cables with thicker but single cables? To my eyes the twisted yellow, black, red, blue-brown pairs may cause some little loops. Usually big problems caused from small details. Can you try please to replace all these twisted pairs with single cables? And then to tell me if is there any reduction in the buzz - hum - hiss etc,etc, noise?

Fotios

[EchoWars]

Well, you see, it was suggested to me that they should be twisted, so I did. Didn't change a bloody thing, unfortunately.

The problem here is that everyone seems to have different (and conflicting) theories as to the root cause of the trouble. Honestly, if building an audio amp was so deep into 'black arts' to make something that sounded good and didn't hum, we'd all be listening to ipods.

I appreciate the attention and suggestion, but perhaps you didn't read my previous post?

[fotios]

Unfortunatelly you are right in that you reffer about the lot of different oppinions. You see that, everyone has a different experience in different amplifiers. This makes the audio amplifiers-preamplifiers a "black art" as you said.
From the other hand, and for a concrette architecture that i use in my projects (by making from time to time some experiments if i can to improve something) i am sure for this that i propossed to you, it might cause problems. Unfortunatelly the gnd tracks and straps are antennas. One time as i remember, i had placed double binding posts per channel for bi-amp connection of speakers. From the center tap of gnd i had connected seperate cables for each black binding post but during the first tests (i use a 15" proffesional woofer of Peavey which due to its high sensitivity, it reveals yet the smallest noise when connected in the output) the buzz noise it made me crazy for enough days. I had changed everything you can or you can't imagine. And finally the problem was in this stupid second gnd cable because it was not terminated in a load. When i removed it, every noise dissapeared.
Well my friend, for this reason from then, when i have noise problem, from the begining i check all my cables. This is mine expertise, someone else has a different expertise and so on.

Fotios