In a class B or AB amplifier, the rail currents are half wave rectified replicas of the output load current, with attendant harmonics. At any instant, only one of the rails is supplying current into the load. Therefore, twisting the supply cables together offers no benefit in terms of reducing the radiating area. In fact, you are likely to couple noise into the non conducting rail and any sensitive circuitry attached to it - filtering ( and even better, cap multiplier for conducted noise) and careful layout is the solution here. You can also use a sound card while dressing the cabling to find the null point (you will have to drive the amp hard while you do this). I have concluded that the best engineering solution is to co-locate the filter caps on the same PCB as the amplifier. This way you can keep the tracks short and the radiating area as small as practicable.
Chassis earth? Socket earth? Power ground? Is this unique to a specific grounding scheme, or will it apply to the usual scheme?
PS induced distortion & earths
His error is that he doesn't show the speaker return lead which is what completes the evil current loop. In fact he entwines his 'clean earth' with the PSU leads in Fig 6.9a which almost ensures it is within the evil speaker current loop. No wonder '6.7b is usually more effective' as this gets his 'clean earth' further away from these evil currents.
At the risk of repeating myself to this august audience ..
Ideally, the speaker return should follow the 'live' speaker lead back to the output devices and then the PS leads back to the PS caps .. all twisted of course
______________________
Also worth reading is John Roberts on the groupDIY and proaudiodesign forums. Us ex-mixing desk people are paranoid about such stuff.
Self doesn't have clear ideas about earths. Bob Cordell is better. see his Chapter 16.9
The relevant pic is Fig 6.9 'Countermeasures against the induction of distortion from the supply rails. 6.7b is usually more effective' in Self 4th ed.
His error is that he doesn't show the speaker return lead which is what completes the evil current loop. In fact he entwines his 'clean earth' with the PSU leads in Fig 6.9a which almost ensures it is within the evil speaker current loop. No wonder '6.7b is usually more effective' as this gets his 'clean earth' further away from these evil currents.
At the risk of repeating myself to this august audience ..
Ideally, the speaker return should follow the 'live' speaker lead back to the output devices and then the PS leads back to the PS caps .. all twisted of course
______________________
I always define 3 earths; Clean, Dirty & Chassis. They are ONLY connected at one point which will be the star point on a power amplifier.
- Clean is used ONLY for signal and feedback
- Dirty is used for decoupling and other yucky stuff. At times this might have to run in close proximity or parallel to 'Clean'
- Chassis is chassis
- Speaker returns are a special case and returned separately to the star point. In a complex device like a mixer, you might have to do this with several stages with large 'signal' current ... or adopt some form of 'heirarchical' earthing in addition to this star point stuff.
Also worth reading is John Roberts on the groupDIY and proaudiodesign forums. Us ex-mixing desk people are paranoid about such stuff.
Self doesn't have clear ideas about earths. Bob Cordell is better. see his Chapter 16.9
Last edited:
I am in broad with you here kgrlee. I went to a lot of effort on my e-Amp to address these issues. My next amp will take some of the learning s from that effort and refine them further, hence my comment about the PSU bring co- located on the same board as the amplifier where you can really control these aspects
If you're going to put the caps right next to the OPS, why not make the heatsink a ground plane? Cover it with heat-resistant dielectric tape and cut out the areas for the outputs. Or, even undersize the cut-outs for the outputs and just fill the gap with a kind of thermal putty that doesn't run. I read about someone doing this with transparent laser printer paper; something thinner would be better. You have to choose the right thermal paste, watching the breakdown voltage.
For power amps (and loadsa other audio), ground planes don't actually help the segregation of clean & dirty currents. In fact they usually encourage poor earthing practice. Clean audio is very different from evil digital in this.
CCIR IM & HD
More on chapter 22 which I really like.
Bob, in 22.4, you touch on the equivalence of the products of CCIR IM test with the order of the non-linearity.
In Jurassic times, I worked out the exact equivalent numbers for these vs 2nd, 3rd .. up to 7th Harmonic Distortion products and scribbled all this in our B&K 1902 manual which was used to measure CCIR IM with our B&K 2010 (This really dates me!) Scaling was as adjusted to exactly how this gear did the measurement.
My few remaining brain cells won't allow me to repeat the exercise But you might be able to persuade a tame Mathematician to do so for inclusion in your 2nd ed.
The exact relation exists for wide bandwidth devices .. ie if Closed Loop response is flat to 200kHz, there is a simple factor to convert each 20kHz Harmonic Distortion up to 10th order .. to the corresponding CCIR IM product(s).
___________
A useful point on doing CCIR (& other) IM is that you don't really need a supa dupa mixer. Just mix passively directly into the PA input. Two 10k metal film resistors have rather less distortion than even the best uber OPAs at the levels required to drive any PA to clipping.
More on chapter 22 which I really like.
Bob, in 22.4, you touch on the equivalence of the products of CCIR IM test with the order of the non-linearity.
In Jurassic times, I worked out the exact equivalent numbers for these vs 2nd, 3rd .. up to 7th Harmonic Distortion products and scribbled all this in our B&K 1902 manual which was used to measure CCIR IM with our B&K 2010 (This really dates me!) Scaling was as adjusted to exactly how this gear did the measurement.
My few remaining brain cells won't allow me to repeat the exercise
But you might be able to persuade a tame Mathematician to do so for inclusion in your 2nd ed.The exact relation exists for wide bandwidth devices .. ie if Closed Loop response is flat to 200kHz, there is a simple factor to convert each 20kHz Harmonic Distortion up to 10th order .. to the corresponding CCIR IM product(s).
___________
A useful point on doing CCIR (& other) IM is that you don't really need a supa dupa mixer. Just mix passively directly into the PA input. Two 10k metal film resistors have rather less distortion than even the best uber OPAs at the levels required to drive any PA to clipping.
I agree with everything in your reply except
In my mind there is no disagreement. The PSU supply must be a twisted triplet. Your suggestion of using the shield as the return may be even better, I don't have any means to confirm that.
I have stated on this Forum every point that you have confirmed as worth consideration, even your "here is an interesting idea.
You and I are certainly not in disagreement.
I am currently reading H.Ott. He says much the same message, although I am learning that at VHF the practicalities are slightly more complicated.
Exactly.
It's the Flow and Return currents that are balanced and it's these that must have minimum loop area.
I definitely agree. I don't propose a ground plane as a way of cutting corners. The reason I'm interested in ground planes is that they shield the sensitive nodes and control parasitics. They make the parasitics more predictable and controllable. The relaxed restrictions result in greater performance in capable hands. High-performance circuits need shielding at their sensitive nodes.
Andrew, what are you reading? Any links? I would at least like to have some idea of how VHF really behaves.
I would not expect any links to a currently available book.
The Forum would rightly delete a link as infringement of copyright.
And they have announced that they will also remove any link to a site that had or currently has any infringing info.
They even said they would impose penalties on Members who broke those anti-infringement rules.
Last edited:
Hi Bonsai,
First of all, I completely agree about having substantial filter caps right on the board close to the output transistors. These do not take the place of the reservoir caps, but should be probably between 1000 and 4700 uF. Clearly, they should be well-bypassed with smaller low-ESR caps. We want to provide a tight local circulating loop for those half-wave rectified currents.
The reason for twisting the rail leads going back to the power supply is not to reduce area, per se, but rather to have the two half-wave-rectified magnetic fields created by the two rail lines to sum to a linear representation of the signal, to the extent possible.
Cheers,
Bob
Hi Bob,
Is the twisting of the power leads for the same reason for staggering the npn/pnp output transistors. I do recall a earlier discussion on this thread and there was much confusion of which signals sum or don't sum.
Cheers,
Yes, to first order it is the same reason, but not quite as accurate in summing the fields. If each NPN-PNP pair is right next to each other you keep the fields as close as possible to summing to a linear field. If all of the PNPs are on one side of the heat sink and all of the NPNs are on the other side of the heat sink there is minimal chance of mitigating the radiation of the half-wave-rectified fields. If each PNP-NPN pair has its own (smaller) bypass capacitor to keep the currents circulating very locally, that is even better.
Cheers,
Bob
This would appear to be another advantage of a dual-bridge PS. The supply and return currents for each bridge would have to be the same, so twisting the supply and return to each bridge would provide cancelation. With star ground on the board, and twisted pair for speaker leads, each individual loop is easily "recognizable" for minimization. Also, as stated in a previous post, keeping interfering wire(s) at 90 degree angles wrt one another is a no brainer.
The Driven Cascode
Back to the driven cascode, where we discussed the reduction in distortion that can happen when the cascode bases are driven with a version of the common mode signal that appears at the tail of the LTP.
Recall that there were two issues discussed (at least). The first was Self's assertion that the effect of the distortion from a non-driven cascode could not be seen in SPICE simulations.
The second was how to derive the signal to drive the cascode bases. One was to take it directly from the tail of the LTP, level-shifting it by a DC amount suitable for driving the cascode bases. In this approach, sometimes the tail voltage is buffered before being level-shifted and applied to the cascode.
The other approach, described in my book, is to create a replica of the feedback signal with another feedback network and use that signal to drive the cascode bases. This signal is essentially equal to the inverting signal applied to the LTP, but in the feedback amplifier this is essentially the same as the common mode signal at the tail up to fairly high frequencies. This approach has the advantage of not messing with the sensitive LTP tail node.
The other day I was investigating a simple input stage for another project, and revisited the driven cascode matter. The input stage included an LSK389 JFET cascoded and then loaded with a current mirror. All very conventional. I simulated the amplifier less the output stage. The whole thing was a very conventional amp architecture with a single-ended VAS run at 10mA and a driver circuit run at 15mA. The output was buffered by a unity VCVS before being applied to the feedback network, so no distortion would result from any loading of the driver.
Here are the results at 1kHz and at 20kHz for the IPS-VAS with and without driving the cascode with the replica signal. These results are at an output signal level of 20V rms, representing 50 watts into 8 ohms.
At 1kHz, distortion went from 0.000042% down to 0.000013% when driven, a 3.2:1 reduction.
At 20kHz, distortion went down from 0.0008% down to 0.00008% when driven, a 10:1 reduction.
Cheers,
Bob
Back to the driven cascode, where we discussed the reduction in distortion that can happen when the cascode bases are driven with a version of the common mode signal that appears at the tail of the LTP.
Recall that there were two issues discussed (at least). The first was Self's assertion that the effect of the distortion from a non-driven cascode could not be seen in SPICE simulations.
The second was how to derive the signal to drive the cascode bases. One was to take it directly from the tail of the LTP, level-shifting it by a DC amount suitable for driving the cascode bases. In this approach, sometimes the tail voltage is buffered before being level-shifted and applied to the cascode.
The other approach, described in my book, is to create a replica of the feedback signal with another feedback network and use that signal to drive the cascode bases. This signal is essentially equal to the inverting signal applied to the LTP, but in the feedback amplifier this is essentially the same as the common mode signal at the tail up to fairly high frequencies. This approach has the advantage of not messing with the sensitive LTP tail node.
The other day I was investigating a simple input stage for another project, and revisited the driven cascode matter. The input stage included an LSK389 JFET cascoded and then loaded with a current mirror. All very conventional. I simulated the amplifier less the output stage. The whole thing was a very conventional amp architecture with a single-ended VAS run at 10mA and a driver circuit run at 15mA. The output was buffered by a unity VCVS before being applied to the feedback network, so no distortion would result from any loading of the driver.
Here are the results at 1kHz and at 20kHz for the IPS-VAS with and without driving the cascode with the replica signal. These results are at an output signal level of 20V rms, representing 50 watts into 8 ohms.
At 1kHz, distortion went from 0.000042% down to 0.000013% when driven, a 3.2:1 reduction.
At 20kHz, distortion went down from 0.0008% down to 0.00008% when driven, a 10:1 reduction.
Cheers,
Bob
The dual bridge PSU has been discussed here:
http://www.diyaudio.com/forums/power-supplies/161865-dual-bridge-ps-does-phase-secondary-matter.html
The difference between a single and dual bridge PSU in this case AFAIK would just be that between twisting 3 wires together and twisting 4 wires together. Not much of a difference in my understanding. Unless there is an advantage to forcing the loops to stay separate, which I don't see for this case. Actually, triple twisted wires conform more naturally so should cancel more consistently. Almost all of my alligator clips are twisted in pairs and I have a few triple twisted sets, and I find this much improves prototype quality and workbench tidyness.
Tidyness aside, I would just twist two wires together, one pair to each bridge. The ground current(s) would not be a factor in this. In fact, this is one of the main advantages of the dual bridge.
Perhaps, but if the idea is to keep the pairs separate, then you are spreading them out in the chassis and filling more space with stray fields that need to be kept away from sensitive components. If you don't twist all 4 wires together but route them next to each other anyways, then what is the point?
Could you explain the benefits of a dual bridge PSU? I can't think of any that weren't shot down in that thread I linked to.
I have heard there are advantages to using parallel bridges to separate channels, but I know a guy that does that just using diodes at the supply input to the board. That way it's the music that modulates the rail isolation instead of the 60Hz.
Could you explain the benefits of a dual bridge PSU? I can't think of any that weren't shot down in that thread I linked to.
I have heard there are advantages to using parallel bridges to separate channels, but I know a guy that does that just using diodes at the supply input to the board. That way it's the music that modulates the rail isolation instead of the 60Hz.
Each PAIR is a separate loop. There is no loop where both pairs are involved to "filling more space with stray loops".
The best explanation I've seen for the advantage of dual bridge is in post #38 here.
I don't think the idea was very well shot down in your link. Most of the shooting was done by unsupported opinions and by those that find the idea of losing an astronomical .7v as abhorent. While the cost may add up with expensive style rectifiers, there are cheap bridges. I'll leave the value judgement up to the individual constructor. You could always use cheap bridges and upgrade later.
Last edited: