Capacitance multiplier: which Darlington to choose?

[JohnRoberts]

Good question. I am asking myself that as well.

I used to have long thoughtful conversations with a fellow console designer (RIP) about what was that etherial hard to quantify difference that motivates customers to release their Benjamins. We never pinned down any single objective metric but suspected multiple ergonomic human factors design aspects.

I agree. I also have a fundamental understanding that not everything that is measured is audible AND that not everything that is audible is currently coblvered by standard measurements.

Thor

Indeed "standard" measurements?

I use null testing to help isolate hard to find differences between audio paths. The null test doesn't tell us what is right or wrong, but helps us quantify differences between similar audio paths.

JR

 

[abbey road d enfer]

That said I had to roll some of my own semi-custom test equipment to measure two-tone IMD before it was widely available.

IMD testers of the past were made cheaply, with a single oscillator, using mains voltage for the LF "modulator" and a very basic AM demodulator.
It has its significance, but not as useful as DFD.

 

[thor.zmt]

Out of curiosity I made a simulation of this circuit and I have quite different results, with HF regulation tending to zero. The MOSFET seems to behave as a negative impedance at about 200kHz.
Also I made a comparison with a similar circuit using a Darlington.
The Darlington behaves as expected.
The output impedance of the circuit is entirely dominated by the 100r series resistor and the 100uF shunt capacitor. No benefit from the active circuit in this area.

Interesting. I suspect a model in your sim is not 100% correct.

Mosfet's do not have a mechanism for a negative impedance, unlike BJT Darlington transistors, where negative impedance in followers is present.

Mosfet's have parasitic capacitances however in this circuit they have very limited impact and they are not that much larger than those in BJT's.

With the 160pF Coss, 100 Ohm series resistance and 100uF with bypass (say 100nF/0.1Ohm ESR any effects from parasitic capacitances should be pushed into the MHz region.

And at these frequencies all the capacitors are inductors anyway...

Thor

 

[JohnRoberts]

IMD testers of the past were made cheaply, with a single oscillator, using mains voltage for the LF "modulator" and a very basic AM demodulator.
It has its significance, but not as useful as DFD.

the Cheap Heathkit SMPTE IMD analyzer that I modified used internal 60 Hz and 7 kHz sine wave oscillators. I modified both oscillators to deliver 19 kHz and 20 kHz 1:1. Not precision bench test equipment but very useful for parsing out hard to measure nonlinearity in my phono preamp designs back in the day.

In my judgement the fault in SMPTE IMD for modern test benches, is the use of too easy to reproduce frequencies. In the way back machine 7kHz used to be HF and 60 Hz was a common LF cross modulation from projection lamps.

JR

 

[thor.zmt]

the Cheap Heathkit SMPTE IMD analyzer that I modified used internal 60 Hz and 7 kHz sine wave oscillators. I modified both oscillators to deliver 19 kHz and 20 kHz 1:1. Not precision bench test equipment but very useful for parsing out hard to measure nonlinearity in my phono preamp designs back in the day.

In my judgement the fault in SMPTE IMD for modern test benches, is the use of too easy to reproduce frequencies. In the way back machine 7kHz used to be HF and 60 Hz was a common LF cross modulation from projection lamps.

Modern Test gear, even fairly inexpensive, handles CCIF IMD etc with ease.

On the other hand, most measurements, such as HD, IMD etc. are actually proxies for fidelity impairments of the actual system, they are not a direct illustration of the fidelity impairment.

Thor

 

[abbey road d enfer]

Interesting. I suspect a model in your sim is not 100% correct.

I tried first with IRF240 because it's in the standard library; when I saw the strange behaviour I downloaded a "proper" IRF830 model, which does the same.
But it wouldn't be the first time a whole bunch of spice models are crappy.

 

[abbey road d enfer]

Modern Test gear, even fairly inexpensive, handles CCIF IMD etc with ease.

Yes, but SMPTE IMD is still part of the arsenal.

On the other hand, most measurements, such as HD, IMD etc. are actually proxies for fidelity impairments of the actual system, they are not a direct illustration of the fidelity impairment.

And that's the big isuue...

 

[JohnRoberts]

I do not find fault with most modern gear. I remember trying to design stuff before it was so easy.

The marketing pukes have to work a lot harder to gin up merchantable differences to sell their all too similar SKUs with.

JR

 

[ccaudle]

I modified both oscillators to deliver 19 kHz and 20 kHz 1:1.

That seems pretty common for published measurements of devices these days. All the measurements published in Stereophile have that.
I checked back to the PCM-1630 review published in 1987 and it had a result for 19kHz + 20kHz IMD.

 

[JohnRoberts]

That seems pretty common for published measurements of devices these days. All the measurements published in Stereophile have that.
I checked back to the PCM-1630 review published in 1987 and it had a result for 19kHz + 20kHz IMD.

I did this back in the late 70s early 80s. I didn't invent the concept but did I mention I am cheap (my test bench was populated with Heathkit and some old 2nd hand test equipment.)

Now they do three tone IMD testing (maybe more). IIRC Jon Risch my resident "golden ear" at Peavey published an AES paper on the subject of three-tone testing ('90s?).

JR

 

[thor.zmt]

IIRC Jon Risch my resident "golden ear" at Peavey published an AES paper on the subject of three-tone testing ('90s?).

Haven't talked to Jon in ages. Tell him hello if you talk.

My personal test gimmick is noise loading. It's the inverse of multitone testing.

Take white noise with 0dBFS, apply 1/2 octave wide notches to -144dB (24 Bit) in octave distances.

Thor

 

[JohnRoberts]

Haven't talked to Jon in ages. Tell him hello if you talk.

My personal test gimmick is noise loading. It's the inverse of multitone testing.

Take white noise with 0dBFS, apply 1/2 octave wide notches to -144dB (24 Bit) in octave distances.

Thor

I haven't seen Jon in over a decade but while working at Peavey I used him on several projects (he was my secret golden ear).... He designed one simple generic (accessory plug in) crossover for me that IMO was very good. Some staggered pole whoo-hah with a nice only -6 dB per octave response in the critical mid range crossover region. I am not a speaker guy, but it sounded really good.

Last I heard he was still working at Peavey, but I escaped over 20 years ago. He was involved with some nice custom driver designs. We used have some strange debates about microscopic effects inside capacitors, or wire, or whatever. It was fun.
==
noise loading sounds a little like noise dither for quantization errors.

JR

 

[Walter66]

Out of curiosity I made a simulation of this circuit and I have quite different results, with HF regulation tending to zero. The MOSFET seems to behave as a negative impedance at about 200kHz.
Also I made a comparison with a similar circuit using a Darlington.
The Darlington behaves as expected.
The output impedance of the circuit is entirely dominated by the 100r series resistor and the 100uF shunt capacitor. No benefit from the active circuit in this area.

Do I understood this correctly, a cap multiplier using a Darlington (which type?) makes no great difference to a passive filter, but the MOSFET does work with a great benefit on ripple rejection in a cap multiplier circuit? A transistor was used by the IRT V81 PSU to enhance the filter qualities of a passive filter. I think it should be good for a choke emulator circuit at least.

 

[Walter66]

I agree. I also have a fundamental understanding that not everything that is measured is audible AND that not everything that is audible is currently covered by standard measurements.

Thor

Maybe everything is measurable, which should be the case. We can measure much better than hear differences in tone. But we have problems to transfer acoustical effects into a physical (means mathematical) model to build a measurement unit for them. What our brain analyzes in the data stream of acoustic waves is much more complex than simple physical models transferred to measurements.

 

[metalockpick]

In this circuit IMD & HD are indeed rising in proportion, however IMD is lower than HD. More to the point, the lower HD circuit used aggressive HD cancellation in the output stage and increased loop feedback.



Are we suggesting that the IMD cause by nonlinearities that are H2 dominant SMD have < -110dB HD are audible at normal listening levels?

I consider this a bit unlikely.

If we can agree that HD and IMD at -110dB H2 dominant and -130dB H3 dominant should be inaudible, then it cannot be the difference in HD & IMD that are audible.

Instead, the difference in HD/IMD are indicative of different and audible differences between the two largely identical circuits, differences not captured directly using traditional tests.

Thor

I would add that if ears can hear it but it "cannot be measured", You need to get somebody on the job who can measure.
Furthermore, when old systems are proven suspect or disproven, it is the duty of science to come up with a better way to quantify, and to quit parroting "specs" and "measurements* which were poor criteria or blatant marketing claims from half a century ago. Now, if somebody wants to take the old gear and measure it on a current FFT, I would love, love, love to see how it fairs against the current state of the art, measured on the same machine. That might be telling.
And while it's easy to argue that "nobody can hear beyond 22kHz", One could easily construct a double stop of 25, 050 Hz and 25,000 Hertz, where the 50 Hertz beat frequency would be coming through clearly through the low frequency transducer (unless the program were low pass filtered, and steeply to remove both of those carrier frequencies). So, there's a case where average ears could easily hear the difference between having HF in excess of human hearing and rolling it off.
I think looking at the transient response, the pulse response over the time domain is far more crucial to getting true to life sound than making sure the "harmonic distortion" is low.

 

[abbey road d enfer]

Do I understood this correctly, a cap multiplier using a Darlington (which type?) makes no great difference to a passive filter, but the MOSFET does work with a great benefit on ripple rejection in a cap multiplier circuit?

No. All I said is some of the benefits of the capacitance multiplier are lost because of the protection resistor, which is admittedly a very simple solution.
The MOSFET or Darlington circuits are arguably better than the vintage circuit with a Ge transistor.

A transistor was used by the IRT V81 PSU to enhance the filter qualities of a passive filter. I think it should be good for a choke emulator circuit at least.

Agreed. However the low current gain (hFE) of the period Ge transistor certainly puts a limit on its performance, compared to what a MOSFET or a modern Darlington could achieve.
I don't have a valid model for simulating the vintage circuit, but I guesstimate the ripple rejection to be about at least one order of magnitude less.
OTOH, the absence of protection resistor probably makes it a stiffer voltage generator, which is a good point, since it reduces interaction between stages.

 

[thor.zmt]

Out of curiosity I made a simulation of this circuit and I have quite different results, with HF regulation tending to zero. The MOSFET seems to behave as a negative impedance at about 200kHz.

I did a Sim with Tina TI, using the build in IRF830 Model and a discrete TIP122 (BC109+BD140-16) for comparison, with the RC at the base scaled to 68k/10uF insted of 6.8M/100n.

The Mosfet Model is Spice Level 3. Using "ideal" capacitors the Mosfet has better HF noise rejection than BJT/Darlington:



Getting more real by adding ESL & ESR to our capacitors and adding 10 Ohm source impedance, we get:



Up to 80kHz the Mosfet is ahead (by a lot actually) but above 2kHz the parasitic capacitances of the Mosfet show up. The main issue is the gate stopper, if we make it into a ferrite bead (first order RLC approximation) with 100 Ohm @ 100MHz (which usually eliminates oscillation just as well as a resistor) we get this:



Conclusion, to suppress mainly the 100Hz/120Hz (and harmonics) power supply ripple noise the circuit originally presented vastly outperforms (~ 20dB) the darlington transistor version with identical RC filter cutoff.

If optimisation for high frequencies (> ~100kHz) is required, the simple gate resistor oscillation suppression must be evaluated.
That said, really high frequency noise is best dealt with using discrete LC networks.

Conclusion 2, one must make simulations realistic enough if one desires results that match reality closely.

Thor

Final sim file attached as zip for anyone wanting to play with this.

 

[thor.zmt]

Maybe everything is measurable, which should be the case.

Yes. Every real physical effect CAN be qualified and quantified. But just because something, in general principles CAN be done, does not mean it is done.

Therein lies one of the many fallacies of the "hardcore objectivist" position, namely that everything is measurable is extended to mean "everything is being measured". Another objectivist fallacy BTW is "Because it is measured it matters".

But we have problems to transfer acoustical effects into a physical (means mathematical) model to build a measurement unit for them. What our brain analyzes in the data stream of acoustic waves is much more complex than simple physical models transferred to measurements.

This is the ongoing dichotomy.

What is measured lacks any demonstrated reliable link to audibility of fidelity impairments.

Thor

 

[thor.zmt]

No. All I said is some of the benefits of the capacitance multiplier are lost because of the protection resistor

That is intentional. It sounds better with that 100R resistor, after testing.

Current limit could be applied differently. If we, for arguments sake, use a standard current limiter, the output impedance at 50mA becomes ~ 7 Ohm and the 10kHz PSRR with Ferrite bead instead of resistor is over 100dB.



We can also omit current limiting entierly, , in that case the DC output impedance is ~ 1.4 Ohm.

The circuit I presented was optimised for reliable ability to be replicated by "average DIY'ers" while offering good performance to kill main based linear PSU ripple and avoid the kind of "sound" simple tube circuits produce when fed by power supplies optimised for lowest internal impedance using active solid state circuits.

It was never meant to be optimised for HF behaviour or maximum ripple rejection or lowest output impedance, non of which, in this application, have reliable correlation with reduced subjective fidelity impairments.

OTOH, the absence of protection resistor probably makes it a stiffer voltage generator

And maybe easier to destroy with a slipped probe during service or if the turn on sequence is out.

It makes for a less reliable circuit, especially if taken out of it's original specific context and treated as "general" building block.

Again, a lot in design is a compromise informed by the actual design purpose. Showing a simple circuit that should be able to be reliably replicated and applied generally as building block by Amateurs with limited experience and instrumentation had different requirements to taking such a circuit and carefully designing it into an overall product ground up and carefully testing edge cases to assure reliability in production.

Thor

 

[JohnRoberts]

A bit of a veer but that reminds me of something I came up with years ago.

The cheap 78xx/79xx 3 terminal voltage regulators were inexpensive and ubiquitous. They were serviceable but based on modest semiconductor technology internally (similar gain/bandwidth to the 741 op amps). The effective result was a rising output impedance at higher frequencies as the internal circuitry lost gain. Prudent designers added capacitors across the PS rails to stiffen them up at HF. I determined that hanging a 1,000 uF electrolytic capacitor directly across the regulator output extended the low source impedance for another octave or two.

Caveat: check for stability issues trying this trick with modern regulators and modern electrolytic capacitors.

JR