The My_Ref is particular project with a lot not so obvious design choices.
The entire amp is in fact 'floating' so that noise referenced to ground is simply 'ignored' by the circuit; the only one that matters, in this context, is common mode (ie between rails) noise that is 'handled' by C4 and C7.
Any bypass to ground demonstrated to be detrimental, soundwise.
With Rev C there is an additional reason illustrated in a whitepaper I've linked some posts ago since the LM318 and LM3886 have been feed-forward compensated.
Hi Andrew,
probably 'floating' it's not a proper term.
I'll use Mauro words to try to explain it better (we already discussed about it some years ago on Madisonears thread)
Una limitata, ma utile, funzione di contenimento dei disturbi provenienti dalla sezione di alimentazione verso gli amplificatori operazionali (sia LM3886 che LM318) è garantita dai condensatori da 100nF inseriti tra le linee+Vcc e –Vcc . Di norma, nei circuiti audio, si vedono delle batterie di condensatori sempre inserite in modo “duale” tra le rispettive linee di alimentazione e la massa. Queste tecniche servono per aumentare l’ efficienza della “massa virtuale”, che rappresenta il riferimento di tensione del circuito. Dal punto di vista del “immunità al disturbo di alimentazione” tipica di un OPAMP queste reti non servono a nulla. Gli OPAMP infatti usano dei riferimenti interni legati esclusivamente a +Vcc e –Vcc, per cui sono sensibili ai disturbi “non comuni” tra questi collegamenti. Dei condensatori inseriti direttamente (e localmente) su queste linee garantiscono una maggiore immunità a questo problema.
In English:
A limited but useful function of containing noise from the power supply to the operational amplifier (for both LM318 and LM3886) is guaranteed by 100nF capacitors placed between + Vcc and-Vcc lines. Usually, in audio, you can see bunch of capacitors inserted in a "dual" manner between their supply lines and ground. These techniques are used to increase the efficiency of the "virtual ground", which is the voltage reference for the circuit. From the point of view of "immunity to supply noise" of a typical OPAMP these networks are useless. The use of internal references OPAMP is in fact related only to + Vcc and-Vcc, so they are sensitive to noise"uncommon" among these links. Capacitors connected directly (and locally) on these lines provide greater immunity to this problem.
On an old post Mauro elaborated the concept affirming that the virtual ground is in fact 'floating', following noise present on both rails so the circuit, basically, ignores it.
Last edited:
A pushpull output stage (all chipamps as far as I know) sends out current from the output device to the speaker and brings it back (return signal) to the audio ground and from the audio ground the circuit must be completed by PSU and decoupling.
LF passes all the way back into the PSU and around to come back out of one or other supply rail.
Or the medium frequency signal passes from the audio ground to the electrolytic decoupling into one or other supply rail.
or the HF frequency signal passes from the audio ground to the HF decoupling capacitor and thence to one or other supply rail.
In every case, the PUSH PULL topology demands that the speaker return currents go from speaker return lead via a capacitor (or group of parallel capacitors) to ONE supply rail. Both supply rails come into play their part on alternate halfwaves.
The rail to rail cap does not help with decoupling the supply rails from load currents that return via the "centralised" return route.
LF passes all the way back into the PSU and around to come back out of one or other supply rail.
Or the medium frequency signal passes from the audio ground to the electrolytic decoupling into one or other supply rail.
or the HF frequency signal passes from the audio ground to the HF decoupling capacitor and thence to one or other supply rail.
In every case, the PUSH PULL topology demands that the speaker return currents go from speaker return lead via a capacitor (or group of parallel capacitors) to ONE supply rail. Both supply rails come into play their part on alternate halfwaves.
The rail to rail cap does not help with decoupling the supply rails from load currents that return via the "centralised" return route.
Thanks Andrew for the interesting post.
I've a question, though.
From what I've read, even from NatSemi ANs on LM3886, the small value bypass caps we're talking about should shunt HF noise to ground to improve stability.
From what you wrote that noise (originated from current returns) affects 'Both supply rails come into play their part on alternate halfwaves'.
So, in a given moment, that noise would be present only on one of the two rails.
Shouldn't a rail to rail decoupling cap filter it too?
Last edited:
True, but it does ensure that any HF signals on either rail gets coupled on the other rail also. The trace inductance between the rail-to-rail bypass cap (C4) and C1, C2 ensures decoupling of the HF path from C1, C2. My interpretation is that C4 is primarily for HF bypass and Class-AB cross-over conduction sonics, i.e., when both transistors in the output stage are conducting, causing a rail-to-rail conduction path within the chipamp.
It should not hurt to have a larger C4, up to say 0.47uF, maybe bypassed with a very linear film/foil cap like a 10 nF or smaller, FKP2, FKC2 or styroflex. I'll experiment with this on my next build.
C7 is rail-to-rail for a different reason altogether - as Dario stated, it's the recommended bypass when there's internal feed-forward compensation inside the opamp, which is the case in the LM318. Here, small film/foil types work best - FKP2, FKC2, FKS2, PFE225, since the opamp does not have to source or sink much current into the Howland.
I am not talking about noise.
I am talking about the load current and the route that load current follows
Noise is a random uncorrelated phenomenon - it has nothing to do with the load current. Load current may cause interference, or distortion, or other correlated behaviour, which has nothing to do with noise.
So from my layman's view, It is sounding like some of the magic/ingenuity designed into the MyRef is that there are virtually two circuit layers - one focused on what actually makes music and a second that minimizes/isolates a particular kind of electrical noise.
I'm thinking double layer chocolate cake with caramel frosting in between.
I'm thinking double layer chocolate cake with caramel frosting in between.
I have seen measurements where noise levels increase with playing of sound, as well as change in noise pattern with playing of sound. Additionally, EMI/EMC testing require something being played during the test. This is not theoretical.
I don't know if Farnell has them, but Goodsky RW-SH-124D is a direct substitute for the Omron and is confirmed to work fine with no failures to date (in 80+ monoblocks supplied to date).