Supply Bypass Caps recomended values

[Little Ricky]

I had installed a tda2040 in place of a tda2030 in a Vox Pathfinder. It clipped horribly and I'm currently upgrading the transformer amd chip to an lm3886. But I'm still not suer why it was clipping. I thought there was not enough juice from the transformer. J M Fahey  mentioned it might be ocsillating, and I only heard the clipping.

I've been going over the spec sheets for the tda20XXs and the schematic for the Pathfinder. I noticed the the supply bypass caps on the PF are under recommended values for a 2030, which would make them way under spec for a 2040. Notes on the spec sheet warn that lower values may cause oscillation. JM was probably right all along.

My main question is why a manufacturer would choose such values for the bypass caps, I'm sure it was done for a reason.

[J M Fahey]

Well, I *had* them oscillate
Not only continuous oscillation, but harder to detect instability which makes it oscillate briefly at certain points of the wave, and only with speakers, not on resistive loads.
100 or 220uF is basically the same, but don't skimp and use, say, 10uF or less.
And thge ceramic .1 is very important, and must be *real* close to IC legs.
Even tack soldering to +V, Gnd and -V pads .
There is a successful "Gainclone" Hi Fi construction which does exactly that, for impressive bandwidth and stability:
Although this is too extreme, it conveys the idea of getting the bypass caps real close to the chip:

You can always tack solder a couple .1 ceramics to proper chip legs.

[Roly]

I have to say I don't think this is "extreme" at all JM; nice and close, VHF-style.

Why a particular value?  It needs to be big enough to do the job, but as electrolytic caps get bigger they also tend to get more expensive, so the lower and upper limits are set by effectiveness (accounting for the spread in component values and chip specs), and cost (size, weight, &c).

The kind of "sometimes" oscillation JM refers to is also called "parasitic" oscillation because it only occurs when there is a signal present, and then typically only over part of the signal wave, most commonly near the peak, but could be anywhere such as crossover, and can be devilishly hard to observe unless the signal conditions are right, even if you know what you are looking for (e.g. can occur at frequencies up to VHF, above the bandwidth of your CRO or meters so you just don't see it).  In some cases it may only occur at intermediate drive levels.

{In Yo Olde Days it was common to have an AM radio at the bench for a bit of music, and I've heard guitar signals or other crud suddenly burst forth while an amp is under test - a sure sign of parasitic oscillations.}

The simple solution is normally to provide "more than enough" bypassing.

So if there is already hundreds of microfarads there, what are the small 0.1uF's doing in parallel with them?

The bigger an electrolytic cap gets the more self inductance it has.  Because instability may be RF at HF or even VHF frequencies this self inductance make the big electro useless as a bypass at those frequencies, so we connect small caps with very small self inductance in parallel to provide HF and VHF bypassing, hence the need to place them very close to the chip.

You will also notice a resistor and cap in series, R4 and C7 in your diagramme, across the output.  This is called a Zobel network and its often misunderstood function (according to Doug Self) is to prevent HF and VHF instability, so it too needs to go close to the output and ground connections of the chip.

Another stability component, missing from these circuits, is a small inductor in series with the output, after the Zobel network (which is very commonly misplaced after the inductor, even in commercial amplifiers).  Here is a typical example, in parallel with a damping resistor;



Its value isn't critical and my favorite method is to take a 47 ohm 1 watt resistor and wind one layer of as much enameled copper wire (of a gauge suitable for the output current) as will fit, using the resistor as a coil former, then solder the ends to the resistor leads.  An equally effective alternative is to take a 5 watt wirewound resistor (the square ceramic ones) of around half to one ohm and simply use that as-is {I had a "Realistic" PA that was suffering terrible break-through from a local radio station because it had no such inductor, and this was a complete cure}.  This produces slightly more power loss, but below 50 watts with an 8 ohm load it's insignificant.

HTH

[Little Ricky]

Thanks. Amazing what you guys can explain clearly in 2 paragraphs that "get me" in the books.

I understand commercial amps, especially budget commercial amps are speced by the bean counters. Is there any reason other than saving a penny that Vox would for example use these "under recommended" values. Is there a sonic difference?

The recommended values for a 2030 are:

• c5-100uf
• c5-100uf

Yet Vox uses:

• c26 (c5)-1uf (actually on board is a 10uf)
• c23 (c5)-10uf

And for the resistor in the Zobel network they use what looks to be a 7ohm

[Roly]

Books are sold by the pound.   

Oooh, those caps are way under value aren't they?

Sonic difference?  Yeah, well parasitic oscillations sound a bit like "sand in the eyes"; I don't really know how to put it, but they certainly don't make an amp sound any better.  Too much bypassing won't make any difference, but too little almost certainly will.

Seven ohms eh?  Meh; maybe they were trying to shut up some instability that way, but values around 10 ohms and 0.1uF are pretty common.  Looks like that might be a 0.22uF next to it; again, in the normal range.

When in doubt try following the datasheet first - the example circuits might not be the last word, but at least you can normally be sure they work.