All about Tubes, Tube Circuits, Tube Gear

Thursday, June 14, 2012

The Aikido Comedy (pt 5): How tests go wrong

From Aikido User Guide:

PSRR: "The Aikido circuit sidesteps power supply noise by incorporating the noise into its normal operation. The improved PSRR advantage is important, for it greatly unburdens the power-supply. With no tweaking or tube selecting, you should easily be able to get a -30dB PSRR figure (a conventional grounded-cathode amplifier with the same tubes and current draw yields only a -6dB PSRR); and with some tweaking of resistor R1’s value, -60dB—or more—is possible. Additionally, unless regulated power supplies are used for the plate and heater, these critical voltages will vary as the power line’s voltage falls and climbs with your house’s and neighbors’ house’s use, usually throwing the supposedly fixed wall-voltage askew. Nevertheless, the Aikido amplifier will still function flawlessly, as it tracks these voltage changes symmetrically."

 "PSRR is almost invariably measured in the absence of signal." - SY

"I am an expert, and I can assure you that PS ripples depend on signal, both their form and amplitude, since the load discharges filter capacitors. Without load there is no ripples theoretically. That means, ripples are not superimposed on the signal. They are modulated by signal. Especially when the amp is loaded such a way that peak output current almost equals to an idle current of output cathode follower (like in case of power amp)." - Wavebourn


"...both tubes have the same plate current and so the same AC plate resistance:
Rp = 2u/3G[Vg +Vp/u]^.5 or Rp = 2u/3G[Ip]^.333
So it would seem to provide a 50% B+ noise divider at all signal levels..."

No, this is incorrect: The fundamental flaw in this analysis here is simple:

The algebraic equations used for AC analysis of tube circuits
are not 'real', in that they express AVERAGES, not absolute impedances.

Thus any equation for Plate Resistance Rp) is not a factual statement of a fixed value, but rather an average, based on a range of instantaneous values and impedance slopes exhibited by the tube during a cycle, or a number of cycles
of an example sine wave input.

That is, the number you get when you calculate Rp
is an average value based on the size of input and voltage swing
at the plate against a load resistance with a fixed B+ voltage.

This is exactly the kind of thinking that caused Broskie to
fundamentally misunderstand the action of the first stage,
and subsequently model it as a voltage divider.

See further down for more details on real circuit behavior.

Certainly some noise reduction is achieved at times by the Broskie Aikido technique here.  And as noted, one very effective trick is to remove noise in the time-span between  musical signals.

But if that were all the Aikido circuit did, it would really be just a cheap trick,
and not a serious noise-reduction technique. Ideally we would want a design or method that removes noise consistently at ALL times.

That is, or ought to be the real definition of PSRR.

Certainly non-linearity of the tube amplification also is a significant source
of intermodulation, sidebands, noise and non-musicality, but its not the only one.

But operating a tube in its most linear region is no guarantee
that there will be no IM distortion, or unwanted products.

The power supply too is well-recognized as another source of modulation
and distortion components, which brings us back to Broskie's claim.

This is the very crux of the claim concerning Broskie's technique.

I would re-formulate the question as a fact of life this way:

Its very difficult to remove PS sourced distortion during a music signal,
while using ordinary resistive loads.

Only high impedance CSS based isolation from the power-supply noise
can keep it out of the signal path, along with its horrendous and undesirable side-effects.

"...That means, ripples are not [merely] superimposed on the signal.
They are modulated by signal."  - Wavebourn

This nails the scientific facts of the case on the head.

I am continually surprised by the penetrating insight Wavebourn shows.

He alone has seen the light, and probably guess very accurately
what I will be posting in the way of analysis.

 Let me recap what this thread is NOT about:

(1) theory vs. experiment in science (they are both essential: one doesn't trump the other).

(2) criticizing Broskie or his amps. They are as reasonably well designed and function as well as most other designs.

Now lets recap what the thread IS about:

(1) Why Broskie's 'Aikido' B+ noise cancelling circuit is a failure.

(2) Why the circuit doesn't and cannot do what is claimed for it.

I intend to demonstrate my thesis on theoretical grounds, not experimental.

The reason this is more than adequate,
is that the theoretical grounds I am using are well-understood,
and non-controversial.

I'm not claiming any secret insights or magical effects.
I'll be instead going over familiar ground,
in a way that shows exactly why the Broskie technique is ineffective.

I will also repeat what I said previously, in case it was missed:
There is no indication from the responses so far,
that anyone except Wavebourn has a clue what is wrong
with both the circuit and the historical analysis/claims for it.

To make my point clearer, and in fact simple enough for all,
lets suppose Broskie's first premise is actually true:

(1) The first stage acts like a voltage-divider.

(2) The Powersupply hum (and noise also) can be viewed as an (unwanted) signal voltage.

(3) This voltage (Vps) appears across the load-resistor and tube (in series).

A percentage of this noise-signal appears at the output terminal of the 1st stage.
(4) We can ignore the issue of the non-linearity of the first tube stage entirely.
We can concede for argument's sake that additional components (harmonics, sidebands, IM distortion etc.) added by stage 1 directly during idle are even smaller relative to Vps than Vps. Thus for instance, if Vps were 10 mV, the additional internally generated noise components might be only 1 mV. This effect will not be significant, because Broskie's circuit would presumably still cancel out 90% of the B+ noise.

(5) However, the tube stage is still a VARIABLE voltage-divider, not a static one.

The tube internal resistance fluctuates according to the AMPLIFIED input signal.  ...
At each peak in the sine wave input signal, the noise signal is being divided across a different voltage-divider:
The Power Supply noise is being Amplitude-Modulated
by the input signal, as only Wavebourn perceived and noted.
Since in a typical musical signal the frequency (frequencies)
vary widely over the entire music spectrum,
what we have is a Ring Modulator, with no musical relation to the input signal.

The simple resistor-divider network of Broskie is is incapable of cancelling out this AM signal, and only (by clumsy design and adjustment) cancels out the PS hum and noise when the circuit is in idle (not amplifying an input signal).
(6) The fact that the Plate Current in either or both stages only fluctuates mildly or even in unison or in a complimentary fashion has no relevance to the AM voltage signal generated in each tube by power-supply noise reacting with an input signal.
Even virtual constant-current Class-A circuits will AM modulate power supply noise and result in a deteriorated music signal. Each stage will AM modulate the same noise-signal independently and in a random fashion in relation to the noise signal, and they cannot cancel out. Remember, it is not the current that is being amplified in these input stages but the voltage that appears at the output of the previous stage. The criss-crossing of the voltages of each stage (signal inversion) guarantees there can be no alignment or canceling except at two instantaneous time-points when the signals cross their idle-voltage lines.
This is why it is so important to isolate amplifying stages from power-supply noise and other noise sources.

(7) Broskie's Voltage-divider/sampler of the B+ noise-signal is a STATIC divider.

No experiments are necessary, contra SY and others.
This is simply Ohm's Law in operation.

Broskie's circuit as first described and presented achieves no reliable signal-cancellation except during idle.
If the overall circuit performs well,
it is for reasons of good tube choice/general circuit design,
all in spite of Broskie's innovation.

This analysis is based on Broskie's own claim that
the first stage acts like a voltage-divider to the PS noise-signal. (hence the irony)

Some however are still making the same analytical mistakes:

(1) The Rp published in a datasheet is an approximation and an average. Its not hard data, but the centerline of a range of values presented by tubes coming off the assembly line.

(2) Also, the Rp has little to do with the actual resistance presented to the power-supply connection, which here is functioning as a secondary "input" for AC hum and noise.

(3) The Rp value is given by the maker for the purpose of calculating algebraic approximations for AC impedances AS SEEN BY Grid Input and Anode Output circuits, its not meant for calculating impedances for signals presented across the power-supply connections.

(4) The Rp value is based on Class-A amplification in which a true (full cycle) AC signal is superimposed upon the voltage at the plate below the load from inside the tube, through the grid.

(5) The powersupply "input" (from ground to top of load connection)
is neither a true AC impedance, nor is it a DC resistance:
It can only be properly modeled in an equivalent circuit as a fixed resistor, a diode, and a variable resistor in series. This is critically important for understanding how the 'diode' characteristic of the tube chops PS A.C. signals in half, effectively multiplying their harmonic content by an order of magnitude.
When people reach a certain level of expertise at circuit analysis,
typically by taking electronics technology 'degrees' or studying 1st year electromagnetism, they think they understand what they are doing,
because they have gained a certain skill in manipulating some algebraic equations.*

Electronics courses however, very rarely cover the real physics behind the algebraic equations and their 'variables'.
Nor do they sufficiently impress upon the students that the algebraic equations themselves are just convenient approximations, and cannot be relied upon as if they were Newtonian Laws of physics.

As indicated above, you can't just use the published "Rp" value
as if it were the actual impedance seen by signals originating in the powersupply and appearing across the cathode-tube-load circuit-path.


 *(Few people know for instance that virtually all of modern electronics came from Heavyside's brilliant breakdown of Hamilton's Quaternions into 3 and 2 dimensional components, pretty much inventing all of modern 'vector calculus' and circuit analysis. )

No comments:

Post a Comment