Time advanced error feedback correction

[smoking-amp]

Here is an idea for those who don't like the fact that normal error correction or negative feedback always ends up correcting the error after the fact. Thus creating higher harmonics, as the error in the error signal propagates endlessly around the loop. The basic idea here is to generate a slightly time advanced error feedback signal for "just in time" addition to the input. Admittedly, one can just use a small capacitor in the feedback to phase advance the feedback signal, but this does not take into account the dynamic distortion curve of the amplifier to prevent generation of higher harmonics. So, here is the plan:

To the input signal of the amplifier, a small, high frequency, sawtooth or triangle wave dither signal is added before application to the amplifier. Since the actual input signal will be either increasing, decreasing or stationary, the dither signal will produce the future signal moments later (u seconds) somewhere in its - to + ramp waveform. By synchronously demodulating the error feedback, (repeatedly sampling at a specific phase point in the repeating dither waveform) we can select the future error correction needed for that later signal level if we know at what point to sample it. Hence we can have the error feedback already available when the signal is just coming in at that level.

A differentiator circuit (RC circuit) monitors the input signal for polarity and magnitude or rate of change and selects the phase angle for the synchronous demodulation so as to select the appropriate future error feedback from the dither ramp levels. The output of the sync. demodulator is then low pass filtered to remove the dither frequency and then summed with the input, as a normal negative feedback signal would be. In this case however, the error feedback is up to date with the input signal rather than the usually delayed form. (Since we have predicted the future signal and the amplifier and sync. demod. have conveniently computed the error correction for us already.) So no more of the endless around the loop stuff trying to fix the error in the feedback because it's too late.

The output of the amplifier will also need a low pass filter to remove the dither signal. Think this will work?

Don

[tubetvr]

Admittedly, one can just use a small capacitor in the feedback to phase advance the feedback signal,

It doesn't change the the time delay of the original signal, you cant change anything after the fact as you cant go back in time.

I haven't looked into your scheme into details but similar things has been tested using analog and digital pilot signals in RF linear amplifiers but this method is not commonly used today.

If you are concerned about feedback adding new distorsion products there are many solutions of which I can name a few:

1 Minimise distorsion before feedback is applied, (this is one of the main reason why tube amplifiers often have less of this problem then SS)

2 Minimise time delay trough the amplifier and feedback circuit

3 Increase feedback, if feedback is high enough then also higher order distorsion products will be soo low that they are not any problem, (Holcro?)

4 Use feedforward correction or digital or analogue predistorsion, this is a form of distorsion cancellation and can produce good results.

All of these methods have their own problems and personally I prefer 1 and 2 as I like to correct problems at the source, using high feedback to reduce distorsion is I believe possible but it adds a lot of other problems.

Regards Hans

[Layberinthius]

It doesn't change the the time delay of the original signal, you cant change anything after the fact as you cant go back in time.


No, but you can delay the final signal while the original signal modifies it.

Step 1: Half-wave comes from Splitter stage at plate voltage of that stage and is placed to the proper grid voltage by Res 1.

Step 2: At the same time the capacitor is showing a signal on the cathode, which is also the same wave, but is very faint.

Step 3: The grid acts as a feedback to the cathode /internally/
and externally.

Step 4: When the plate is saturated by the cathode and grid is not supressing electrons, This time at which it is acting as a Tetrode, when the grid is supressing electrons it acts as a feedback path to cathode and the plate is as if it wasn't even there.

Step 5: the variable capacitor acts as a fine tune to the time at which it takes for the signal to pass back to the cathode and re-create the loop, essentially cancelling out a small amount of it's own signal.

I think that's about it, tell me what you think...

I think the only way I can describe it is that it is a threshold diode which turns into a Tetrode while the screen is energized.

Maybe this will supress noise, maybe it'll fry.

I have apsoloutley no idea how it would work nor made calculations or simulated it.

[tubetvr]

No, but you can delay the final signal while the original signal modifies it.

Yes, that is called feedforward and is well known, also the problems with it.

What I originally commented is that as an amplifier always will have delay however small and feedback can not be corrected in any way to counter act that.

Regards Hans

[SY]

I'm losing you at your premise.

Here is an idea for those who don't like the fact that normal error correction or negative feedback always ends up correcting the error after the fact. Thus creating higher harmonics, as the error in the error signal propagates endlessly around the loop.

I'm not an engineer, but in the derivations of basic feedback equations I've seen, the delay is assumed to be negligible. The conversion of high levels of low order harmonics to low levels of high order harmonics is a consequence of the feedback itself, and this conversion is independent of any delay and independent of the frequency. Now where delay DOES enter into things as a practical matter is in loop stability at very high frequencies, but that's not really relevant to the question of production of higher harmonics in the audio band. Even with zero delay, you'll still have those higher order harmonics.

[Layberinthius]

/me looks stunned at SY as usual


WHAAAAAA!!!!!???

All hail Holy and Wise SY!

I'm going to leave the whole thing as it is, TOO much for me!, glad however that I came up with "feedforward" in a matter of 2-3 hours...I bet any Phillips employee would have been proud to have me

[smoking-amp]

tubetvr said:

If you are concerned about feedback adding new distorsion products there are many solutions of which I can name a few:

1 Minimise distorsion before feedback is applied, (this is one of the main reason why tube amplifiers often have less of this problem then SS)

2 Minimise time delay trough the amplifier and feedback circuit

3 Increase feedback, if feedback is high enough then also higher order distorsion products will be soo low that they are not any problem, (Holcro?)
......

---------
Yes, I agree, these are more practical approaches.

Sy said:

The conversion of high levels of low order harmonics to low levels of high order harmonics is a consequence of the feedback itself, and this conversion is independent of any delay and independent of the frequency. Now where delay DOES enter into things as a practical matter is in loop stability at very high frequencies,

---------

I agree here also, the hope would be that one could apply higher gain stably to reduce the errors if the feedback delay were eliminated. On thinking more closely about my idea though I think this still can not be done. The dither signal is filtered out of the feedback, so is operating open loop for starters. The synchronous sampling also is a problem for noise since one can only get a single sample rather than averaging a bunch of them.

The "proto" design for all this was based on having two exactly identical amplifiers "A" and "B". The input signal to ampl. "B" would use the input signal plus a small linear predicted step from a differentiator. Amplifier "A" would have just the original input signal applied. Amplifier "B" would have a normal feedback loop for itself. Amplifier "A" would use the same feedback signal from "B" since it would be "just in time" rather than delayed for its input. The impossibility of building two exactly identical amplifiers led to the earlier described dither and sync. demod. idea for implimentation on a single amplifier. Obviously, however, the amplifier "B" in the proto design is still limited by stability considerations, so cannot have arbitrarily high gain to reduce errors to arbitrary smallness. So only the delay artifact can be canceled but not the errors from lack of open loop gain. So I guess we can scratch this idea. Hmmm... but maybe with feedforward...

[Sch3mat1c]

Here is an idea for those who don't like the fact that normal error correction or negative feedback always ends up correcting the error after the fact.

Which is why bandwidth is ALWAYS limited a distinct amount in any well-designed amplifier. Limiting it reduces the HF signals that are delayed, such that they have no effect and the amplifier remains stable.

Tim

[tubetvr]

The conversion of high levels of low order harmonics to low levels of high order harmonics is a consequence of the feedback itself, and this conversion is independent of any delay and independent of the frequency.

Yes that is true as long as the delay is neglible compared to the signal time period, it has however a lot of importance in linearised RF amplifiers but that is outside this discussion.

Now where delay DOES enter into things as a practical matter is in loop stability at very high frequencies,

I think it is more easy to grasp if you talk about phase characteristics instead of delay as a constant delay is equal to phase changing continously with increased frequency. Normally amplifiers doesn't have significant constant delays but phase can anyway change as frequency increases. Example: if the delay is 100ns it will give 0.036 degrees phase lag at 1kHz, 0.36 degrees at 10kHz. 3.6 degrees at 100kHz and so on, normally this is neglible compared to the phas shift introduced by the high frequency poles in the amplifier.

I agree here also, the hope would be that one could apply higher gain stably to reduce the errors if the feedback delay were eliminated.

As I tried to describe above the constant delay is not the problem but phase caracteristics are, an amplifier that never have phase changing close to 180 degrees, (preferably with some margin) will always be stable no matter what amount of feedback is applied, (it sounds more simple then it really is)

Which is why bandwidth is ALWAYS limited a distinct amount in any well-designed amplifier.

If you express it like that then most classic tube amplifiers are not well designed as they normally are bandwidth limited by the output stage and not the input stage.

I think it is better to say that "a well designed amplifier should have a bandwidth that is either; 1 wider then the applied signal OR if that is not possible 2 it should have its bandwidth limited by a low pass filter on the input in order for condition 1 to be fulfilled" This is more or less what Matti Otala and others after him came up with.

The reason why classic tube amplifiers have less problems then the early SS amps is that the many tube amps have wider bandwidth then the applied signal and therefore a bandwidth limitation is not necessary, for the early SS amplifiers it is different as they had very limited internal bandwidth and they where not bandwidth limited on the input and therefore produced different kind of distorsions, (TIM, DIM etc).

Regards Hans

[SY]

If you express it like that then most classic tube amplifiers are not well designed as they normally are bandwidth limited by the output stage and not the input stage.

Well, open-loop, most classic tube amps are bandwidth limited by their compensation. And I don't think in general that they slew-limit; all the ones I've tested have a constant rise-time, not slew rate. Crowhurst made a big point of this in some of his articles. he was a big proponent of not trying to make the amp ultra-high bandwidth, and let first stage compensation be applied strongly enough to greatly reduce or eliminate the need for the lag compensation cap in the feedback network ("faking" a good transient response was his way of putting it). His arguments in that direction make great sense to me.