Capacitance multiplier: which Darlington to choose?

[thor.zmt]

I have to admit, this project is not about a budget. Therefore, the PSU has to sound quite good at first. And it has to have sufficient good measurements. I'll try the MOSFET solution, the BJT and they have to compete with a fully tube regulated supply. So they better sound good, or they will never be applied.

I found, that if the effects of the different Transconductance are compensated by a series resistor, all open loop and closed loop supplies sound very similar (important though "all else equal"), the biggest difference are between series and parallel regulation.

Devices, if the objective differences are tared out, make miniscule differences.

Personally I found the same true for rectifiers. It is fairly trivial to match a tube rectifier (minus slow turn on) using soft switching diodes with added low voltage Schottky rectifiers and series resistors between transformer primary and diodes that simulate the tube rectifiers impedance and include a snubber cap on the diode input.

There is no Magik to electronics. If two circuits that have the same function and perform in basic ways similar, but have different sound quality, this is caused by objective differences that can be replicated in the subjectively "inferior" circuit to tare out the difference.

This is my angle, yes, I built and tested lots of circuits from others, but I seek to understand the root causes of the differences as this allows more direct and targeted design of my own circuits. And for me budget is an issue.

If I can deliver the same sound quality using solid state rectifiers and solid state regulators as others can only achieve with tubes, i have a competitive advantage.

I noticed for example that in the Mosfet regulator circuit you omitted the output resistor. I suggest you put it back and select the value so that it emulates the tube regulators output impedance.

If the tube regulator has a capacitor on the output, you need to place the same capacitor after the output resistor, failing to do so leads to incorrect conclusions.

You can, later, experiment with removing these elements and see what that dies, but for an initial comparison, the AC impedance of the regulator circuit tends to be dominant, which, if different disallows reliable conclusions about devices and topologies.

Thor

 

[Walter66]

Thor, thank you for providing hints for the implementation of the circuit.

 

[thor.zmt]

This analysis appears correct, yet you say it's not correct, so I'm curious where the difference lies.

I located your source. It is actually not a textbook:

The Capacitance Multiplier

This analysis is quite exhaustive and includes transconductance etc.

You abstracted a limited equation for a specific subset of analysis.

No reduction of the math yields Beta * C.

In fact, Beta is applied to Rload and in effect Rload * beta is placed in parallel with the filter's series resistor to give R in 1/(R*C) as parallel connection or Rload * Beta & R.

For Zout (which would be in effect Zc // R for the passive RC) the Article shows:


Note, Beta does not show up in this equation.

So the capacitance value is not multiplied by Beta.
For the analysis shown in the article to be applied to Mosfet's, simply remove any references RLoad * Beta (should be Zload * Beta to account for the fact that load is commonly not resistive) and the rest will work fine.

Bottom line R & C (plus parasitics like the lowish input impedance of BJT) set the corner frequency of an RC filter that is buffered by an active circuit.

The Output impedance is determined by the active circuit's behaviour, primarily the transconductance.

Thor

 

[Matador]

I located your source. It is actually not a textbook:

True, but that equation shows up in several textbooks, and it's easier to quote a web page than scan a book.

Note, Beta does not show up in this equation.

To be fair, r(pi) is directly related to beta via rpi=beta/gm (at least in the hybrid-pi model).

If I can paraphrase: the advantage of the active element (BJT or MOSFET) is that the corner frequency of the equivalent passive RC filter network is reduced (more ripple rejection, all other things being equal), however the output impedance is greatly reduced, especially at high frequencies. Stated another way, with a given RC, you get better filtering than just R and C would predict by themselves, and you don't lose I(load)*R across the series filter resistance for 'low' frequency load transients.

The advantage of the MOSFET is that since it has (essentially) infinite gate impedance, then modulation of the load current has no effect on the 'buffered' RC network, whereas for the BJT, the RC network will be modulated by the base current (which is the load current divided by beta). And practically, for tube supplies, finding high voltage MOSFET's with high gm is vastly easier than BJT's, provided precautions are taken with the gate->source junction.

Does that sound reasonable?

 

[thor.zmt]

If I can paraphrase: the advantage of the active element (BJT or MOSFET) is that the corner frequency of the equivalent passive RC filter network is reduced (more ripple rejection, all other things being equal)

No. That is incorrect.

Let's say I make an RC filter with 150mOhm resiststance and 150F/5mOhm ESR capacitance (something like that feeds my headphone amplifier BTW, at 5V and 5A max., so I'm not making up numbers).

I now have 0.007 Hz turnover 30dB maximum noise attenuation and generally levels of impedance few active circuits can match.

If I make the 150mOhm resistance a pair of 10uH inductors, I get a 2nd order filter at 3Hz, with -40dB at 30Hz with our original 7mHz first order filter superimposed, for huge noise suppression.

So if we make a true "equivalent" RC Network (which we can) the active circuit is not necessarily better.

What the active circuit does offer is lower cost, size and weight however.

The advantage of the MOSFET is that since it has (essentially) infinite gate impedance, then modulation of the load current has no effect on the 'buffered' RC network, whereas for the BJT, the RC network will be modulated by the base current (which is the load current divided by beta).

Yup, plus we can use smaller value, higher quality capacitors in the filter.

And practically, for tube supplies, finding high voltage MOSFET's with high gm is vastly easier than BJT's, provided precautions are taken with the gate->source junction.

Does that sound reasonable?

Yup, spot on.

I'd add that with a Mosfet, we could add a zenner diode (string) fed from a CCS ahead of RC and get a rather precise, non looped feed back regulator. With BJT'S looped feedback is commonly needed.

Using this with (say) my 100R resistor and a 22uF MKP capacitor on the output can make an excellent subjective sound quality powersupply for small signal tube stages. I use this a lot.

I also tend to use a "tube rectifier emulator" with solid state diodes and resistors ahead of this circuit.

I prefer fully passive filters (CLCLC for output stage of power Amplifiers) and real tube rectifiers combined with mains side AC voltage stabilisation (yes, I also did such circuitry in commercial high end hifi gear) strictly in terms of sonics, but such a set-up is not realistic for most situations, never mind most commercially made equipment.

Thor

 

[Matador]

To answer your other question, I do indeed use a variation of this circuit, using an IRF830 as the pass element, however the gate is controlled by an LR8, which allows for a precise setting of the output voltage. This is mainly used for screen supplies when I want a fixed voltage, but also for a few 'external' high voltage adjustable lab supplies (that are used for tinkering with new circuits). It also uses a current limiter to clamp the output current to less than a few hundred milliamps, to protect against shorted screens (and dropped voltage probes). With the relatively slow time-constant of the R and C it also gives a slow ramp-up.

I also have a 'negative' version that uses a Zener diode on the gate, fed by a 10M90S high-voltage constant current source, which is used to generate a low-impedance negative bias supply for output tubes that may be driven to class AB2 operation (around -80V fixed, but can be made different by changing the Zener).

In both cases, output ripple is almost un-measurable under standard operating conditions, and the entire mess is much less board real estate than a large bank of 100uF/450V caps and resistors.

 

[ruffrecords]

I am surprised nobody has thought to use Scroggie's method of improved RC filtering by cascading RC stages prior to the base/gate. For those who do not know suppose you have an RC filter consisting of a 30K resistor and a 30uF capacitor; this will attenuate 100Hz by 55dB. However if you cascade three stages of 10K and 10uF (so the same total R and C) you get a 100Hz attenuation of 108dB. Yes I know it is three times the components but low wattage resistors cost very little and low value film caps are cheaper than higher value ones.

Cheers

Ian

 

[thor.zmt]

I am surprised nobody has thought to use Scroggie's method of improved RC filtering by cascading RC stages prior to the base/gate.

It is so obvious, I don't mention it.

A third way, by using RCRC and retiring the second C to the output and not ground, we make an AC current source, a true "electronic choke" that lets the capacitor on the output dominate the impedance and linearity.



This board shown is a "Universal" powersupply that uses only non electrochemical capacitors, includes a regulated -120V bias/Aux supply and a selectable second order "capacitance multiplier" or "electronic choke" jumper selectable operation.

Thor

 

[ruffrecords]

It is so obvious, I don't mention it.

Don't forget, most members here are not electronic engineers.

Cheers

Iann

 

[thor.zmt]

Don't forget, most members here are not electronic engineers.

Ok, then, instead of a simple RC circuit on the Grid/Base/Gate of the circuit, a wide range of circuits may be used.

They may include:

zenner diode voltage regulators followed by RC (stabilises the output voltage regardless of input voltage, reduces noise dramatically over simple RC)

multistage series RC circuits to increase the slope of noise suppression and or use smaller values

an active Salen Key lowpass wrapped around the device where a 3rd order roll-off is possible

a multistage RC filter where the final capacitor is not returned to dround, but to the Cathode/Emitter/Source of the active device, literally creating a virtual choke, with rising impedance as frequency rises

One might even use a zenner chain as level shifter driven at the bottom by an Op-Amp to make a highly precise, super low noise and output impedance "high voltage Super-Regulator"

Plus a variety of other options that do not immediately jump to my mind...

============

Active devices too can be compounded for better performance.

For example a Darlington or Sziklai Pair for BJT's (especially the latter)

or a mix of MOSFET (or even tube) with a high voltage, low beta BJT as Fetlington or Fetsziklai (that is called IGBT if made as single device)

are all possible and have different benefits and drawbacks.

Thor

 

[Walter66]

Hello,
would it be good enough for channel separation of the circuit, if both channels are feed by the same rectifier tube (like the schemo drawn) or could the PSU be improved by using two separate rectifier tubes, feeding two cap multiplier circuits, completely channel separated?

I'm just asking because Shindo San has done something that looks like channel separation, maybe the way I painted it?
He used one transformer AC voltage, one rectifier tube.
After the rectifier, the circuit seems to be two independent channels for PSU and amplification, because the left IC chamber looks very channel independent build.
Would this be the best solution or could it be improved? Thank you.

 

[Walter66]

I am surprised nobody has thought to use Scroggie's method of improved RC filtering by cascading RC stages prior to the base/gate. For those who do not know suppose you have an RC filter consisting of a 30K resistor and a 30uF capacitor; this will attenuate 100Hz by 55dB. However if you cascade three stages of 10K and 10uF (so the same total R and C) you get a 100Hz attenuation of 108dB. Yes I know it is three times the components but low wattage resistors cost very little and low value film caps are cheaper than higher value ones.

Cheers

Ian

Ian, still don't have decided about the place where to put the PSU chokes best.
Shindo San seems to have just put a resistor and one capacitor in front of the cap multiplier IC, but four, channel separated, choke coils behind it.
Which variant will give the best signal to noise ratio, heavy filtering in front or behind the cap multiplier?
Think I have to test this on the bench within the circuit.

 

[Walter66]

Another thing I'm a little worried about. The chinese seem to copy everything these days and instead to produce good quality, they can degrade quality, too. We all know this, but with IC circuits, it can become dangerous.

In a video, one person had constructed a constand current source using LM 317. By changing the IC, the circuit completely lost its ability to deliver a constant current, instead the current drifted steadily. It was a cheap chinese fake copy of the original IC, low quality and not original circuit inside.
So how to spot those bad copies and how to find reliable suppliers?

With electron tubes, I do know a big supplier from UK, who sometimes sell rebranded fake tubes. Many think, he is a trusted seller. And not long ago, I ordered electron tubes from a german pro- seller, but received Korean tubes. The picture on the net showed Philps tubes, and thats what I ordered, but necer reveived.

How to protect being supplied with bad parts that are not able to deliver the original performance?

 

[thor.zmt]

Hello,
would it be good enough for channel separation of the circuit, if both channels are feed by the same rectifier tube (like the schemo drawn) or could the PSU be improved by using two separate rectifier tubes, feeding two cap multiplier circuits, completely channel separated?

Using dual mains transformers and everything else separate, or at least separate windings for LR and everything separate is certainly an option. I have used this myself.


In this picture T-03 is the Heater and HT transformer with separate windings per channel.

The PSU circuit is Tube Rectified HT with LC/RC/RC filtering (chokes top Left/Right) and Mosfet follower regulator.

The Transformer T-02 is primarily an Autoformer to allow the system to adjust the system to differing mains voltage and supplies the AD Converter and solid state first stage.

The circuitry to switch the autoformer taps is on the PCB below the mains transformers with the blue plastic Transformer. The logic and display is supplied from this transformer.

It is this unit:

Abbingdon Music Research PH-77 Phono Equaliser

I'm just asking because Shindo San has done something that looks like channel separation, maybe the way I painted it?
He used one transformer AC voltage, one rectifier tube.
After the rectifier, the circuit seems to be two independent channels for PSU and amplification, because the left IC chamber looks very channel independent build.
Would this be the best solution or could it be improved?

I suspect it would be adequate. I suspect my solution is at least conceptually superior, if much more involved.

Which variant will give the best signal to noise ratio, heavy filtering in front or behind the cap multiplier?

I used:

EZ80 -> 2.2uF (Film) -> 20H 2-section C-Core Choke -> 100uF ->

(noise 56mV Peak-Peak, almost pure sine)

Zenner Reference & IRF830 Regulator -> 100R -> 220uF -> 1K -> 220uF ->

Regulator has a 4.7M/4.7uF Filter on the gate after a chain of 3pcs 91V zenner diodes to fix the voltage.

At this point noise a miniscule fraction of feck all, we are talking nanovolt levels. What is there will be johnson noise, chemical noise from the electrolyte of the capacitors etc.

split into decoupling per stage

1k -> 22uF MKP -> +B3 (Output Stage)

1k -> 220uF -> 5.1k -> 22uF MKP -> +B2 (middle stage) -> 39k -> 22uF MKP -> +B1 (first stage)

Of course, channels are completely separate.

All capacitors that directly decouple the +B of a given stage are custom made MKP, with respectively 7Hz (output stage), 1.4Hz (middle stage) and 0.2Hz (first stage) corner frequencies of the RC Filters.

Personally, I think splitting channels at the output of the capacitor multiplier to L/R would be more than sufficient to get excellent SNR and channel separation.

So how to spot those bad copies and how to find reliable suppliers?

For a commercial entity that needs to purchase on the open market, you need stringent incoming goods control and you need to weed out fakes.

I implemented that with a "sample book" set having "golden samples" and a set of test fixtures and a requirement to test the important parameters needed to make sure that incoming stock matches the "golden sample".

The only reliable suppliers are the likes of Mouser & Digikey.

How to protect being supplied with bad parts that are not able to deliver the original performance?

Only buy from reliable large size suppliers that have a strong anti-fake policy.

Of course, you will not find NOS tubes there. Or any NOS and obsolete/discontinued parts.

So adjust your circuits to use currently available parts, which means no "toobz".

Thor

 

[user 37518]

GM= ID/Vgs

JR

gm = 2*ID/(Vgs -Vth), Vth being the threshold voltage.

 

[Walter66]

Thor, you made my day. Thanks so much For giving advise on this topic.
Have to try this Out after Holidays. Will Report later.

 

[thor.zmt]

gm = 2*ID/(Vgs -Vth), Vth being the threshold voltage.

Really? That is not even a good first order approximation.

Best look at the curves in the datasheet.

Here Gm / Id for a MOSFET.



Gm is highest at low currents and quite linear with Id at low currents. That might explain my preference for Mosfet VAS.

Thor

 

[user 37518]

Really? That is not even a good first order approximation.

Best look at the curves in the datasheet.

Here Gm / Id for a MOSFET.

View attachment 110322

Gm is highest at low currents and quite linear with Id at low currents. That might explain my preference for Mosfet VAS.

Thor

Seems like you have some basic math deficiencies, because that expression I gave and you dismissed as "not even a good first order approximation" is exactly the same as the one you posted:



The expression which uses the exponential is in weak inversion, that is in sub-threshold region when the channel is not yet fully formed, namely, when the transistor is not working in the saturation region. Also note that that is the so-called small signal approximation, the linearity happens only with very, very small signals, such that it can be considered a constant gm; in fact, for larger signals, the gm may be expressed as a sum which includes modified Bessel functions, namely, not linear at all. Even worse is if the MOSFET is allowed to transition between sub-threshold (weak inversion) and saturation (strong inversion) since the non-linearity will be stellar.

 

[thor.zmt]

Seems like you have some basic math deficiencies,

You presented that large signal equation, which is valid mainly in the high current, switching region of the FET.

I point out that this is not correct to describe the FET at low currents which are used here.

In our specific example we have a 4.5A rated FET that operates in the transition or subthreshold region at drain currents below ~1A if I read the datasheet right.

The circuits discussed as load have current consumption in the single to tens of mA max.

Ergo your math is not correct, in context.

So yeah, clearly I'm math challenged and you are absolutely and completely right.

Thor

 

[thor.zmt]

the non-linearity will be stellar.

Analog Circuit Design · Samuel Groner · Resources · Power Amplifiers

Page 31 for "stellar nonlinearities"...

QED

 

[user 37518]

You presented that large signal equation, which is valid mainly in the high current, switching region of the FET.

I point out that this is not correct to describe the FET at low currents which are used here.

In our specific example we have a 4.5A rated FET that operates in the transition or subthreshold region at drain currents below ~1A if I read the datasheet right.

The circuits discussed as load have current consumption in the single to tens of mA max.

Ergo your math is not correct, in context.

So yeah, clearly I'm math challenged and you are absolutely and completely right.

Thor

Look, they are your equations that you provided, I just did the algebra for you. You can twist it all you want by quoting datasheet parameters, it doesn't change a bit that you were wrong. And no, its is not valid in high currents, it is in the saturation region, that is when the Vgs is higher or equal to Vt, which is were most MOSFETs are used in general. You are wrong, the FET operates in saturation as long as VGS is greater or equal than Vt, and as long as VDS is larger or equal than VGS - Vt. The use of the MOSFEt in the subthreshold region requires VGS lower than VT. You can still have a very low current by making VGs close to VT, and make the MOSFET operate in saturation, where my equation is perfectly valid. You seem to lack basic understanding of the physics of the MOSFET

Also, I advice you to read a couple of textbooks, that equation I provided is shown under small signal approximation, not large signal as you claim. Large vs small signal refers to the variation around the bias point, NOT the absolute value of the current, which is what you seem to suggest. Small signal gm, which is the one I provided, is given as the partial of Id with respect to VGS, that is a rate of change, if VGS variations are small, it can be considered constant, which is exactly what I did provide.

Ergo, math is math and you are the one who doesn't even know basic algebra, since you dismissed my equation (which is not even mine) as not even a good first order approximation,and then proceded to post an image with the same equation. Also, in my original post I was responding to JR, and he wasn't referring to the subthreshold equation.

 

[user 37518]

Analog Circuit Design · Samuel Groner · Resources · Power Amplifiers

Page 31 for "stellar nonlinearities"...

QED

QED what? That example is not even relevant to what I was saying, I said that if a Mosfet operates in subthreshold and it is allowed to swing between subthreshold and sat regions it is not linear, are you saying it is linear? What are you talking about, your mere equations betray what you say. That piecewise approximation used in your graph (which is an approximation) shows that the Mosfet transitions between a quadratic and an exponential, is that linear to you?

 

[abbey road d enfer]

Please, gentlemen, I feel the heat coming...
I believe there are some reasons to disagree in several parts of this thread, but let's stay courteous.

 

[user 37518]

Please, gentlemen, I feel the heat coming...
I believe there are some reasons to disagree in several parts of this thread, but let's stay courteous.

You are right. It is just that he constantly tries to boost his ego even when he is wrong. And I've had it

 

[Walter66]

The complete Tube preamp I designed as a phono LCR and Line unit will consume 50mA. Therefore, For that small load on TO-220 cap multipliers, May it be preferable to use only one instead of using two For each Channel? Will the Mosfet Work better with Higher current load?
As Seen with Shindo Giscours, His bipolar IC seems to be two of them, so I suppose this Part wont have Problems even with loads of less then 20mA. Thats what His preamp consumes per Channel.
Is there a difference in Low current Performance between bipolar and Mosfet?
P. S. Sorry, Im Just an Amateur, dont understand Higher Mathematics.