1970s Design Indulgence

Half the voltage drop appears where the output transistors emitter resistors join. If the output is offset it won't quite be half, but close enough for this explanation. It can be seen as a step voltage, but not a steep step due to the HT capacitor charging rate.

The positive end of the output capacitor will see this voltage step and its other end will temporarily stay at the same voltage until charged via the LS load.

At the end of the cymbal crash, the HT recovers, and then the output capacitor will charge to its original voltage, before the step. This might be seen as cone wobble, but it's doubtful it can cause damage.

The real problem is the effect on the quiescent voltage of the output stage, as the upper transistor supplies charge to the output capacitor. During the finite time the voltages will not match, and distortion results.

The workings are such that it cannot be solved, but like most things, we might be able to mitigate the problem.

The ear might ignore a short interval of distortion, and one way of shortening it is to reduce the output capacitor size. It might not be necessary to reduce it so much that it cuts the bass prematurely. 2000uF cuts at 10Hz, 1000uF cuts at 20Hz (assuming an 8 ohm LS impedance).

Large smoothing capacitors will minimise ripple voltage and better regulation of the transformer helps too.

If the output capacitor were large by comparison to the smoothers, the step period should follow closely to that of the smoothers. A smaller output capacitor will start to correct the step earlier - tortoise vs hare.

Perhaps 1x2200uF is better than 2x2200uf?

What triggered this light-bulb moment was that the sound was better with over bias, but it shouldn't be. The greater emitter current of the output stage makes the "DC on the signal" increase in quiescent current less obvious.

I've heard the argument that DC doesn't have "a sound" because it's DC, and yes, DC, if it is DC, doesn't have a sound, but come on! An unregulated power amp power supply is DC in name only.

In a push-pull amplifier you have double the music's frequencies on the wires from the smoothed supply to the amplifier's power terminals for a start. And so, the smoothers are zero impedance at the highest reproducible frequency. That would be twice 24kHz on 48kHz sampled digital files. Really?

We've covered Kirchhoff's circuit laws until we're sick of it. I shall just say that everything has impedance and there is no such thing as true DC except for what comes from a battery. A battery isn't going to be much use powering a power amp (although I've seen the equivalent of an EV battery under the lid of one power amp).

"DC" is the positive and negative going cycles of an AC voltage source where both halves of the cycle are put above zero by diodes. The diodes switch on but never cleanly off, as some datasheets sometimes show.

If we examine current waveforms the transformer primary current is not a sine wave, and the voltage waveform is clipped at its very peaks. Magnetising current is the "admission fee", and the "music current" has an effect on that.

At the same time as the magnetising current produces its primary spikes, the smoothing capacitor ripple current is spiking somewhere in the vicinity. It can be in sync, or earlier or later depending on the ever changing current demand of the music - the amplifier output delivered to the loudspeaker, and the loudspeaker's changing demands dependant on frequency and amplitude - which changes extremely quickly because music contains transients.

The load on the transformer is constantly changing and so the transformer design does have a bearing on how the music sounds!

Having pointed out the above, I can now say the RS Components Nuvotem Talema 160VA transformer is working better, to my ears, than all the other transformers in my transformer graveyard (at least ten).

I now await Terry's design from TransTronic.

I believe, or at least hope, this is the final leg of development. I also hope it isn't the beginning of another long journey.

The amplifier circuit, taken in isolation, is as good as you'll ever get from seven transistors (my opinion of course).

For you weary folk who have followed this ramble - perhaps wondering if it will ever end - you will have noticed we keep coming back to the power supply. And here we are again, but with a little new understanding of its mechanism.

To me, it now seems that the expert explanations of why output capacitors are bad, were never thoroughly thought out. I therefore guess it's going to be left to me to try and explain the real reason.

If all you ever do is measure distortion between cap coupled and DC coupled amplifiers, I guess the distortion you can hear, is easily explained as being the capacitor's fault.

Maybe it never occurred that it could be signal asymmetry? As I explained till I was blue in the face, if you want one watt, you need 25 watts to do the transients.

At one watt, the power supply voltage can be called "off-load" and at 25 watts, the power supply can be called "on-load". Off-load to on-load, the HT voltage is going to be different. A sharp or "single" transient (if such a thing exists) might not result in that voltage step, because of the smoothing capacitor energy storage.

However, when we follow what real music does, some of the instrumentation often follows the singer in reaching the high notes. The foundations of the song, the bass and rhythm, stay where they were.

We then have multiple high frequency transients for the period the singer is up there in pitch. The power supply voltage drops because the ripple voltage increases as the smoothers lose their stored energy.

The DC operating point of the junction of the push-pull must obviously move. Therefore, any symmetry set off-load, becomes asymmetrical. This indicates an increase in even harmonic distortion as the musical event hits the busy highs.

But, hang on a minute, isn't the transient intermodulation distortion test (or even the dynamic intermodulation distortion test), supposed to reveal this?

OK, so we set the analyser to do these tests, and it takes a few seconds for its reading to settle, during which time the DC operating point has settled, and the reading isn't bad at all!

These tests are therefore useless. The analyser hasn't a hope of keeping up with the fast fluctuations of the HT.

Correction; these tests are not useless if the DC operating point never shifts.

In a DC coupled amplifier, the DC operating point is constant. In a capacitor coupled amplifier (single supply, that is), the DC operating point shifts with the music.

As it shifts the distortion changes, and so we blame the capacitor - falsely!

If it were possible to maintain the DC operating point rigidly, the distortion would not change. When we have the distortion changing as it does, then the highs become so distorted that the ear cannot hear the foundation notes of the bass and rhythm - it goes bright.

We then add the other channel, seeing it is stereo. Not only do we have the difference transients from one channel distorting both the sum of both channels, and the difference signals of the other, we have any two of three assaulting any other of the three.

Shall we look at frequency bands now? OK, best not, I think you have got my point.

So, for DC coupled, it is OK, provided the power rails collapse equally. The capacitor coupled amplifier has one HT rail, which collapses to throw the output DC operating point asymmetrical. It is no fault of the capacitor!

Is there any way to mitigate this effect? It would be good if we could set the DC operating point such that is was the same on and off load, but for the fact it would still be asymmetrical, but at least constantly so.

The other way could be to separate the supplies. Sorry, we've been there. All that did is prevent the one channel corrupting the other, but the asymmetry was still there.

What if we regulated the power supply so none of this happens? Yes, and also separate each channel? A great advert for the Proprius!?

The work continues.

The pre-sweep will have the effect of letting the power supply "pre-adjust" to the power level - it lasts 600mS - and is immediately followed by the 1 second sweep 20Hz to 20kHz. With no pre-sweep the power supply has no chance of "pre-adjusting".

With "pre-adjustment" the THD measurements were better than without, and so the observations of my last post are true.

Also, the effect of two channels is worse than just one channel, as mentioned.

This is about the closest possible measurement available with current analyser technology. As for quantifying the distortions, the time periods don't lend themselves to establishing conclusive distortion levels. They simply show a worsening due to the change in the output's DC operating point.