Power Amp PCB design, grounding

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I started a Leach Low-TIM amp PCB design... as I'm finding so many errors in existing power-amp pcb designs, mainly with grounding.
I remember the old ETI 150W mosfet amp (1982) where they got 1% THD and simply re-did the board traces a few times to get it down to 0.05% !
All that DIY work getting top quality parts, spice simulations, machining aluminum, soldering, wiring etc. and it frustrates me to see otherwise good designs do poorly.

I went back (to the future) and reviewed Leach's Double-Barreled/Super-amp (1980) and the Low-TIM amp (1976).
I found changes when both the Low-TIM and Super-amp went from double-sided (non-plated thru) to single-sided pcb. Low-TIM I think it happened from V3 (1976) to V4 (1986), and the Double-Barreled (2000). The input cascode zener bypass caps got connected to the dirty ground, whereas before they went to the signal ground. I don't think it's good- any AC on the dirty ground gets coupled into the cascode base, a very sensitive spot. How would you guys do it.

I know Jen Rusmussen put a lot of energy into updating the Low-TIM pcb and it looks like a great layout. But I don't agree with the pcb grounding scheme, being a mix of techniques.
 
www.hifisonix.com
Joined 2003
Paid Member
Whats important in the layout in order to get low noise is the order of the grounds and knowing when to 'T' off. Starting from the signal input ground of the amplifier:-

1. signal ground (note: RCA input socket must not be connected to chassis ground)
2. lower feedback resistor ground

You can separate (2) and (3) with a 5.6 Ohm resistor for added insurance that there are not ground loops.

3. local front end decoupling capacitors
4. output stage decoupling capacitors
5. speaker ground return
6. Zobel ground return - this is a separate ground that runs off th e PCB and goes to the main 0V 'T'
7. 0V Main 'T' - this is a 'T' off the filter capacitors 0V connection point
8. 0V Filter Capacitor connection


If you change the order of any of the above connections, you will get noise and or give rise to distortion.

One other point: on the feedback connection from the amp output, the connection must be 'T'd off just before the the output filter - not simply connected where all the emitter degen resistors are connected.

If you use this approach, it s very easy to get completely hum and 'wiring distortion' free designs.
 
Last edited:
Okay, here is a first draft pic, let me know if it's what you have in mind. I included the NFB takeoff point (Kelvin connected) and a safety ground (9) for noobs.

I'm not sure about the cascode-bypass cap gnd (marked ???). It must go to the "quiet gnd" (1) but then there's ~10mA of zener current which adds 56mV offset across the 5.6 ohm resistor :(
I consider point (3) a "dirty ground, although the input front-end decoupling caps don't have as high ripple currents.
 

Attachments

  • power-amp-grounding-rev0.png
    power-amp-grounding-rev0.png
    11.2 KB · Views: 2,147
Hi,
I too have doubts about where the cascode reference should go.
Taken to the clean ground ensures that the AC through the reference must use the clean ground as the return.
Taken to the dirty ground ensures that the cascode voltage varies with load and decoupling currents.
There is a third alternative. Decouple the cascode supply to dirty ground and take the cascode to clean ground.
And a fourth. Take the cascode between the supply rails then you cannot contaminate the ground.

I think that 3, 4 & 6 can be combined in the amplifier to simplify the wiring. In fact the Zobel is a low inductance load specifically intended for high frequency signals. I think it must take the very shortest route from output device back to decoupling ground, through the decoupling and back into the output devices. This figure of 8 loop must be very small to allow the HF to pass with least reactance.

The loops formed by any circuit must be small (or self canceling- see Dr Cherry). This comes down to trace and component layout. The Input RCA to input base around the filters and signal ground must be short and very low loop area. The circuit from -IN & feedback components and signal ground must be small area. BUT! Where is the return route for the current that flows from output through Rf & Ri and signal ground?

Point 1 shows the link to the main Audio Ground. I would say the link can go from the input RCA socket barrel to Audio Ground.

Options, alternatives and choices.
How do we decide?
 
Last edited:
www.hifisonix.com
Joined 2003
Paid Member
I have taken a quick look - nice diagram BTW.

However, I would recommend the following refinements:-

You have shown the decouplers star connected to the 0V point of the filter caps. This is dangerous because it only takes a small connection error to introduce noise and or distortion problems. They connections must be 'Td' off in the order described in my post above. Starting at the noisiest point, the 0V connection point between the filter caps and the transformer CT. A bit back off this, the Zobel, a bit back from this the Speaker return, a bit back from this, the output stage decouplers and so on.

The cascode decoupling cap must not go to the clean front end ground for the reasons you state and because cascode connection will not in itself inject any noise into the collector current of the VAS amplifier. It should be connected to the front end decoupling capacitor ground point.

For the feedback point, I'd take this off just before the Zobel network.

The reason the Zobel has a separate return to GND and not just joined to the PCB ground is that at high frequencies, you don't want the HF currents from the Zobel network being injected into the PCB ground. Think about the inductance in the main ground return and what happens with fast transients if you just connect to the PCB ground return. An ugly situation.

The point made above about keeping the loop areas as small as possible is very important as well.

For the main supply wiring, you can wire the +ve and -ve supply lines together (i.e.next to each other) but do not bundle them together with the amplifer PCB ground return. The reason is that there are heavy half wave current pulses flowing in the supply lines (full of harmonics) and these can couple into the ground return line effectively causing it to 'bounce around' and then couple into the sensitive circuit nodes. If you can arrange it, the speaker return line should be located adjacent to the power supply lines as this will reduce the loop area. Practical layout considerations make this quite difficult however.

For safety wiring, Rod Elliot has an excellent artice on how to wire up a ground lifter for further assurance against noise, but taking into account safety issues.
 
Last edited:
Thanks guys for your input :cheerful:! Bonsai, I drew the grounds in the sequence you list, and got puzzled. Are you saying a star point doesn't work well? I'm not sure about thinking of it as a series run- is this on a microscopic level? I am aware of voltage gradients on high current pcb pads (and even a bolt with stacked lugs), in practice all nodes can never be equipotential. Maybe this is the (lug or wire) sequence at the star point? That I can understand.
Overall, I find two techniques are popular:

1. The star point is on the amp chassis as a single bolt, with the amp PCB having two ground wires (quiet gnd and noisy gnd) tied to it. With 5-15 wires it gets crazy and I dislike the extra wiring inductance. I've 'scoped a lot fuzz, despite a using a big bolt or copper bus wire. The Leach amps use this technique.

2. The star ground point is on the amp pcb, with a single ground wire to the PSU. The loudspeaker and zobel ground also connect to this pcb star point. Silicon Chip magazine has been using this guessing 10 years, in the Studio 350 and ULD Mk2 etc. I like this scheme a lot and would only add RF bypass to it.

On the odd-ball ground point, my reckoning is AC on the cascode decoupling caps/zener ends up modulating the transistor's base, so I'm not seeing how the diff. amp doesn't react. It's not common-mode noise?
 

Attachments

  • series-gnd.png
    series-gnd.png
    4.3 KB · Views: 1,781
the diff pair collector cascode reference is a input signal gnd referenced circuit, doubly so as it is also the diff pair tail bias too

the Zener bias currents should cancel given the complementary input topology

better regulation can be had by splitting Leach' R13,14 and bypassing the midpoints to quiet gnd with the 100uF C4,5
the RC time constant is much lower with the split R rather than trying to bypass the very low Zener dynamic R

I wouldn't use the 5.6 Ohm in the gnd - a "star" at the the speaker gnd such that input, feedback and speaker return all the same gnd "point" is better

"dual mono" with separate ps for each channel is the best approach but with a single shared supply I would make a heavy "technical gnd" out of 16 oz Cu roof flashing if necessary to short the 2 board's signal, feedback and speaker gnds

otherwise you really need differential sensing across your "gnd isolation" resistor to tell where the the "dirty gnd" of the speaker is - and it is the precise control of the differential V across the speaker terminals that is the whole point
 
www.hifisonix.com
Joined 2003
Paid Member
Prairymystic,

yes, new drawing is how I described it.

The star ground is a great way to wire up an amp and I am not proposing that anyone scraps this if this is what they are using and getting good results. However, it is quite easy in a PCB layout to route a track to the wrong point in a grounding system and to then end up with a problem. If you use the 'T' approach discussed here, its easy to visualize and to get a great result. Note, the order of the connections only has to be separated by a few mm - we are not talking great distances here.

In terms of actual layout, all the ground connections can be done on the PCB except for the Zobel ground connection and the speaker ground connection. The speaker gound connection should come back from the output ground connector back to the central ground (so don't take it back to the PCB) and the Zobel ground needs to be run separately from the PCB, observing the order of connections, to the central ground as in your diagram.

jcx,
to clarify, my comments about the cascode refer to a standard cascode connection and not the Leach connection as shown above. I've done some simulations on the standard cascode connection and don't find reference noise, within reason, is a problem - in a practical amp as well. That said, the cascode reference ground should in any event go to the front end decoupling ground (#3 in Prairiemystic's diagram).

Re the 5.6 Ohm ground isolation resistor for th e front end - sure, you can drop it off if you find that when you short it out on the actual amplifier you have a perfectly noise free design. Its a good technique however for preventing 'accidental' ground loops. Only a small wiring error can cause a problem here.
 
Last edited:
Disabled Account
Joined 2006
Ref. the scheme in post #3; I've never been a fan of input / output separation via the resistor, and would echo the view of jcx on this point. Better (in my humbly grumbly opinion) to consider placing it between the 'star point' and the chassis ground.

G.
 
Better (in my humbly grumbly opinion) to consider placing it between the 'star point' and the chassis ground.
NO, absolutely not.

The fault current that can flow from mains to audio and to Safety Earth can be kA.
Pass that through a resistor and it will go open circuit so quickly that the mains fuse may not rupture and the audio side will now be at mains potential and the amp will sound dead.
Touch a cable or terminal and you could be dead.
 
Disabled Account
Joined 2006

Hello Andrew,

Perhaps you did not understand what I said? I did not say place the resistor in the mains ground line. I said place it between the chassis main earth point and the audio 'star' ground point.

A small resistance between mains ground and audio ground is fine and works nicely. It is normally bypassed by a small capacitor and often with large anti-parallel diodes.

This has been common practice for many years. A quick scout around finds:
> A Bryston amp using a 10R resistor, and they are a long-established company.
> The Pass Labs Aleph Ono service information shows a CL60 thermistor with a cold resistance of 10 Ohms.
> The engineer behind ESP also recommends a 10R resistor...click: Power Supply for Power Amplifiers .

It's OK to disagree, however there seems to be a weight of opinion to suggest that it works well and works safely.

Kind regards,
G.
 
I remember the old ETI 150W mosfet amp (1982) where they got 1% THD and simply re-did the board traces a few times to get it down to 0.05% !

Everything has to go wrong to achieve that. In practice grounding isn't that critical when it comes to THD, it more strongly affects noise. THD comes more from general layout (Self discusses this in his book).

All the best, Hannes
 
Gordy.
I mean NO

A mains fault that puts mains potential on any non chassis part inside the amplifier can only blow the fuse it the mains can escape to Safety Earth.

If you say it works with a resistance in there then you have not tested for that fault condition.

If you require something other than a direct link from Audio Ground to Safety Earth then you must using the Disconnecting Network. A resistor will not do.
 
Cherry's paper (A new distortion mechanism in class B amplifiers - Edward M. Cherry, AES Engineering Report, May 1981) has this explanation (as 10 ohms instead of 5.6 ohms):
"The 10ohm resistor provides a high-frequency connection between the quiet ground track and the ground ends of the bypass capacitors; 10 ohms is less than the reactance of a 100nF bypass at frequencies up to 1MHz, but is large enough to reduce hum current in the quiet ground track to the order of a milliampere". It's easy enough to experiment with this, so I'm too worried about it.

An amplifier's Class-B currents, which Douglas Self calls "induction distortion" are worth going into, when I get a chance...
 
Gordy.
I mean NO

A mains fault that puts mains potential on any non chassis part inside the amplifier can only blow the fuse it the mains can escape to Safety Earth.

If you say it works with a resistance in there then you have not tested for that fault condition.

If you require something other than a direct link from Audio Ground to Safety Earth then you must using the Disconnecting Network. A resistor will not do.

AndrewT and Gordy,

Just to mention, Douglas Self advocates putting the AC mains safety ground at (1), on the input RCA jack. This looks good from a hum-loop point of view, integrating with other gear, but IMHO is totally impractical. One, the 10 ohm 1/4W resistor cannot tolerate any fault currents and would just smoke :flame: Two, with no 10ohm resistor at all, input pcb traces and wiring are never meant to take ground fault currents, a trace would just melt.
It's hard enough to come up with decent (system) grounding then put the constraints of electrical safety on it.
 
Disabled Account
Joined 2006

I have provided factual information yet you continue to shout no! You make me smile. To any reader that would like to see fact instead of listening to Andrew shout…

Go here (click) >
http://bryston.com/BrystonSite05/pdfs/SSTAmplifiers/3B-SST-SCH-1C(Oct02).pdf

Observe that the mains input ground is top middle of page 2. It joins the chassis ground and then leads out at the top left of the page and on to page 1. On page 1 it leads to the top left where it passes through a 100R resistor in parallel with a capacitor and anti-parallel diodes. It then leads back to page 2 where it forms an ‘audio ground’ point at the junction of the transformer secondary and the capacitors.
 
I'd think you guys on the other side of the pond would be up on CE requirements! I think if you read CE regs (or UL regs here) and consider every part that can be contacted with the official finger probe, you'll see that almost all schemes using low value resistors would be considered unsafe under certain failure conditions. I won't shout, but I think "no" is still the right answer.

One of the Ott books on grounding and noise in instrumentation systems covers this where-to-ground-it problem very well, but the book has gotten expensive like so many technical books these days.

CH
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.