Active Ribbon Preamp/Impedance Converter Idea - opinions?

[bcarso]

[quote author="clintrubber"]FWIW this T.H.E.-circuit is quite hefty with the nanoFarads as well ...

I guess a manufacturer has to go a bit more for 'sturdy under as many conditions as possible' - at the cost of a bit higher distortion.

Regards,

Peter[/quote]

The "phase" splitter approach is tricky, and Samuel is correct in warning of potential distortion effects (although these will likely be fairly small compared to the basic distortion of the FET). At first glance, knowing that collector impedance is much higher than emitter under normal bias, one would expect the high frequency response of + out and - out due to capacitive loading to be asymmetrical. But as mentioned above, because the current that the emitter supplies comes mostly from the collector, if the loading capacitors are equal there is compensation and the measured responses for each line are about the same. But the transistor is working harder.

If there is asymmetrical loading things get a lot different. For example, for the original circuit (the second schematic at the beginning of this thread), if you take away C13, the frequency response at the collector output loaded by C12 at 22nF is down 3 dB at only about 5.1kHz! Restore C13 and the response goes out to over 300kHz. This assumes light loading at the destination.

The double emitter follower approach of the T.H.E. circuit is intrinsically symmetrical, without much interaction due to loading, and the rolloff will be due to the series R plus emitter impedance breaking with the C's.

EDIT: from sim the loading effects on the phase splitter are surprisingly small, and also may induce distortion that tends to cancel some of that coming from the FET.

 

[clintrubber]

[quote author="bcarso"]EDIT: from sim the loading effects on the phase splitter are surprisingly small, and also may induce distortion that tends to cancel some of that coming from the FET.[/quote]
Sorry, how do you mean the second part of this ?

Thanks

 

[bcarso]

The predominant distortion is second harmonic, and cascading the stages tends to have the second subtracting somewhat from the first. The effect is small though.

Some numbers for the phase splitter itself: at 10kHz, 100mV p-p through 4.5k (roughly the impedance at the FET/4.7k junction of the original circuit), with 1nF loading caps the differential output THD is about 0.0037%; with 22nF it rises to about 0.0142%.

By comparison the overall circuit is running about 0.25% for a comparable output signal (EDIT: about 47mV rms differential). The only change made to the original circuit was to make the voltage divider at the base give a better spacing between collector and emitter.

These are big signals I suspect. As the distortion is strongly 2nd it will tend to diminish linearly with signal level. Third is just beginning to appear with the overall circuit.

 

[Samuel Groner]

Looking over the servoed version I notice I didn't take the 18 V zener tempco into account, and this actually dominates the new d.c. tempco since it runs about +15 mV/degree C.

As a quick fix one could synthesise the 18 V with 2x 6.2 V plus 2x 2.7 V--gives a much lower tempo and reduced noise at the cost of increased impedance (and parts count).

By comparison the overall circuit is running about 0.25% for a comparable output signal.

So that's for the common source configuration, right?

I know how noise is measured, but I don't own a "Rauschbombe" (whatever that is in English) for highest isolation. But then again I wasn't interested in absolute values so much as in perceived performance.

Measuring absolute values is not that easy, couldn't do it either. I'm more interest in the different power density spectrums.

Samuel

 

[PRR]

> condensers have a falling noise curve anyway due to the capsule capacitance.

Not that simple. Indeed low freq noise tends to be "huge". But at the top of the band the acoustic resistance of the diaphragm damping (the holes in the plate) can and probably should be larger than amplifier noise. You can reflect the acoustic resistance to the electric side (microphone/speaker dual) and work it as an electrical analysis, or observe that if the diaphragm gets a really good grip on the air (as it must for maximum sensitivity and S/N) then you hear individual air molecules randomly ping-ponging on the diaphragm.

You can "improve" this by adjusting stiffness or mass. You can get lower sensitivity or less HF frequency response. Or you can reduce the damping, giving a ringy high-end. Indeed I thot one fancy mike used low damping for low acoustic noise, and electrical EQ to take down the ring (and HF noise).

But we are getting far away from the type of article I would expect to find in a Musician's Magazine. I have trouble putting two transistors, 6 pins, together right (which is why I have some uncommon debugging insight). Here I'm seeing 66 part-pins to wire-up. Is that typical for this magazine?

--------------

> That circuit requires exotic super low Rbb transistors, though. And by super low, I mean super low.

It isn't clear to me why that should be. Maybe some other parameter is the villan.

Did you explore the Neutrik small transformers for about 10 Eurobucks ahead of a simple buffer?

 

[bcarso]

[quote author="Samuel Groner"]

Looking over the servoed version I notice I didn't take the 18 V zener tempco into account, and this actually dominates the new d.c. tempco since it runs about +15 mV/degree C.

As a quick fix one could synthesise the 18 V with 2x 6.2 V plus 2x 2.7 V--gives a much lower tempo and reduced noise at the cost of increased impedance (and parts count).

By comparison the overall circuit is running about 0.25% for a comparable output signal.

So that's for the common source configuration, right?

Samuel[/quote]

For the servo circuit version, near-perfect tempco seems to be gotten with ~50k R's in the emitters of the servo transistors---the lower loop gain makes the 18V tempco and NPN Vbe's work optimally with the PNP "Vbe-multiplier" effect. I haven't checked yet to see over what temp range it will reliably correct though. I also like that approach because it keeps the servo more "out-of-the-way" as far as potentially affecting audio.

The distortion stated from simulation is with the Rossi original circuit, signal input to the gate, differential output from the splitter, with just a more appropriate divider ratio for base bias. The distortion is very slightly higher at the FET drain than at the output IIRC.

 

[Rossi]

[quote author="PRR"]But we are getting far away from the type of article I would expect to find in a Musician's Magazine. I have trouble putting two transistors, 6 pins, together right (which is why I have some uncommon debugging insight). Here I'm seeing 66 part-pins to wire-up. Is that typical for this magazine? [/quote]

I think your Ribbon Booster circuit and the reduced Schoeps/Dorsey circuit are doable. You can easily build them on Veroboard. I'm not sure about my own circuit (with the improvements Brad suggested). Parts count is acceptable, but it is a bit more complicated when you acutally put it together. Brad's circuits are certainly above muician's magazine level. But of course they'd be very interesting for anyone attempting to build a state of the art active ribbon.

My article is not all about these circuits, it is also easier mods, some of them are of a purely mechanical nature, others are about different transformers (Lundahl, Edcor), so I guess everyone can do whatever they feel they can handle.

--------------

> That circuit requires exotic super low Rbb transistors, though. And by super low, I mean super low.

It isn't clear to me why that should be. Maybe some other parameter is the villan.

No, I think it's just the fact that the output level is very low. If you put the mic in front of a guitar amp, there's no problem, of course. But if you want to use the mic for vocal recording (it sounds pretty nice), output is weak. Even though it has about the same sensitivity as a SM58 on paper, you don't use it in the same way. Mic distance is likely to be 1 foot or more, due to the huge proximity effect. So in that application the mic is noisy with anything but a very low EIN preamp. Using the not-so-bad 2SC1815 (Rbb = 50 ohms, IIRC) transistors, EIN is about -124 dB (150 ohm source), 2SC3329 (Rbb = 2 ohms) give about -129 dB. I tried two 2SC1815 in parallel (unselected) and got about 126 dB. In that situation an improvement of 2 or 2.5 dB is very noticeable.

Did you explore the Neutrik small transformers for about 10 Eurobucks ahead of a simple buffer?

I tried beyerdynamic transformers (1:7), but I didn't like the results. Noise was relatively high. Those small transformers tend to have quite high DC resistences. I didn't like the sound either, to be honest. There was quite some LF roll off. I suppose better results could be attained with more attention to proper temination. But I dropped the idea quickly because your circuit sounds so much better and isn't much more difficult to build than a buffer. I guess you'd have to use a really expensive transformer to get comparable performance.

 

[Rossi]

Hmm, so my circuit (with Brad's improvements) isn't so bad?

Just a question about the resistors on the output. Do they look about the correct value? As I'm low on math, I "calculated" them by reducing the load impedance to something like 200 ohms and tweaking the output resistor values until both output graphs (XLR-2 and XLR-3) matched closely.

 

[Rossi]

[quote author="Samuel Groner"]
Measuring absolute values is not that easy, couldn't do it either. I'm more interest in the different power density spectrums.

Samuel[/quote]

You could just build a reduced Schoeps/Dorsey circuit with a dummy load and put in different FETs. As FETs aren't too consistent, I suppose you'd have to measure noise and/or analyze the noise spectrum for a number of each type. You could even try different loads and thus find the best FET type for a given load impedance.

 

[Gus]

Couple things

did M.J. use 170s before?

Next why "Schoeps Dorsey"? it should be just Schoeps IMO

 

[bcarso]

[quote author="Rossi"]Hmm, so my circuit (with Brad's improvements) isn't so bad?

Just a question about the resistors on the output. Do they look about the correct value? As I'm low on math, I "calculated" them by reducing the load impedance to something like 200 ohms and tweaking the output resistor values until both output graphs (XLR-2 and XLR-3) matched closely.[/quote]

Remember that balanced does not mean equal and opposite polarity levels on each line, but rather equal impedance on each line. As discussed above, the phase splitter output stage is inherently unbalanced, although indeed very close to equal and opposite levels. For short runs and low interference in the local environment you should be o.k.

 

[Rossi]

I don't know what Michael Joly used before, in fact I don't even know exactly what he's using right now (he doesn't specify the type). He probably compared to the stock FET (which I think is fine, at least the ones in my Oktavas).

I write Schoeps/Dorsey because I use a reduced version that is closer to Dorsey's stripped down schematic. The Schoeps original has different bias voltages, too. The Zener is a 6.2V type, and the FET drain is at 4.5V.

So, Gus, what's your favorite FET? I remember you prefering Process 50 FETs for a while. Is the love still strong? :wink:

 

[Rossi]

[quote author="bcarso"]
Remember that balanced does not mean equal and opposite polarity levels on each line, but rather equal impedance on each line. As discussed above, the phase splitter output stage is inherently unbalanced, although indeed very close to equal and opposite levels. For short runs and low interference in the local environment you should be o.k.[/quote]

I see, but if I use a very low load impedance (perhaps I should go lower than 200 ohms), different output impedance ought to produce different gain loss. So if (after tweaking the output resistors) I get congruent curves at such heavy loads, output impedance must be (nearly) equal for both lines, no?

 

[bcarso]

[quote author="Rossi"][quote author="bcarso"]
Remember that balanced does not mean equal and opposite polarity levels on each line, but rather equal impedance on each line. As discussed above, the phase splitter output stage is inherently unbalanced, although indeed very close to equal and opposite levels. For short runs and low interference in the local environment you should be o.k.[/quote]

I see, but if I use a very low load impedance (perhaps I should go lower than 200 ohms), different output impedance ought to produce different gain loss. So if (after tweaking the output resistors) I get congruent curves at such heavy loads, output impedance must be (nearly) equal for both lines, no?[/quote]

No---see a previous post about that. The nature of the beast is that loading on the emitter produces a compensatory level increase on the collector. However, modeling a line disturbance with two identical current generators reveals that those signals produce a large differential signal---i.e., bad news for susceptibility.

 

[Samuel Groner]

You could just build a reduced Schoeps/Dorsey circuit with a dummy load and put in different FETs.

I mean measuring equivalent noise level of the entire microphone.

Samuel

 

[Rossi]

There are one or two articles about microphone noise on the Neumann website. Stephan Peus of Neumann told me that there is another lecture by a younger Neumann Engineer about more recent research on dynamic microphones, which also includes observations about noise spectra. But that one wasn't online the last time I looked.

As a rule of thumb (it is a bit more complicated, as PRR explained), dynamic mics produce noise that resembles white noise whereas condenser mic noise looks more like pink noise. So, at the same measured noise level, dynamic mics tend to appear noisier, because there is more noise in the upper frequencies where you hear it more distincly.

Measuring the equivalent noise level of the entire mic would of course require a dead quiet environment. For me, what's important is not absolute values but the perceived performance. Most users don't care about numbers very much. What the readers really want to know: does the mic appear noisy in real world use? That's why I do comparative listening tests (levels closely matched, of course). As a reviewer, the main part of my job is to say all that is not in the datasheets. Still, I'm thinking about building a "Rauschbombe" for more intricate work such as this.

 

[Rossi]

[quote author="bcarso"]
No---see a previous post about that. The nature of the beast is that loading on the emitter produces a compensatory level increase on the collector. However, modeling a line disturbance with two identical current generators reveals that those signals produce a large differential signal---i.e., bad news for susceptibility.[/quote]

Hmm, so if I understand you correctly, the beast just looks balanced, but doesn't behave like a balanced circuit, because interfering signals cause it to become imbalanced.

What about this:

This is (supposed to be) an impedance balanced version with just one driven line. Would this behave better?

 

[PRR]

> I'm thinking about building a "Rauschbombe"

"Rauschbombe" == "Intoxication bomb" ???

"Intoxication" could be noise.

"Bomb" could be the explosion, or the heavy metal shell around the explosive you want to throw at someone else.

There is "bomb calorimeter". You burn fuel inside a heavy metal shell and measure the total energy released from the temperature rise of the shell and water jacket. You could be testing an alcohol-air mix. But it is a sin to burn good intoxicants.

So I'm thinking: concrete box, sealed lid, foam pad, sealed XLR connector. In a room of 25dB SPL noise, the inside of the box will be near or below 0dB SPL noise.

4" heavy iron pipe with pipe-caps would work. However I would not buy one short pipe and two caps all in the same store on the same day. Concrete flower-pot inverted on dense foam (carpet underlayment) might give 10dB-15dB isolation, if all gaps are very small. A US Army ammunition box has a quick-open good seal, though the walls are prone to resonance. A pile of blankets will give 5dB-10dB isolation... you really want mass, stiffness, and seal.

I'm not sure there is a standard english word for this. "mike noise test chamber"?

As a relative test, just put good, less-good, and unknown mikes in the chamber, and read the output. A-Weighted filter is entirely appropriate. Flat weighting may be dominated by LF or supersonics: test for anomalies, but not for perceived hiss. Listen (and spectra): a narrow resonance may be annoying all out of proportion to the reading on the wide-band. (Someone just came in with a recording like that: +20dB at 3KHz and less than an octave wide.)

The chamber should be much-much bigger than the microphone, so it does not cramp the mike's near-field and change acoustic loading on diaphragm. If you wanted to calibrate with a sound source, it must be even bigger and anechoic.

An absolute reading will require a reference sound source inside the chamber. This is a job for a Calibration Geek, not you or me. It would be less insane to just measure the sensitivity in a dead room and adjust. Of course un-flat mike and noise spectra will confuse things.

 

[PRR]

> an impedance balanced version with just one driven line.

In small studio work, single-ended often works fine. The main problem is that we still can't be 100% sure that the amp behind an XLR jack will handle unbalanced sources nicely.

For CMRR, you would like zero source impedance on both lines.

Long lines, and some transformers, may be flatter with some series impedance. ~100 ohms (50 per side) may be best for MHz response. Some fancy input windings expect 50 - 100 ohms out there to damp their resonance (however such inputs will usually be lower noise than my two-BJT plan).

In very-simple designs, there is a BIG reason to always throw a few dozen ohms in series. In your latest plan, change R2 to zero. Apply a square wave to the FET. For infinitely fast rise-time and zero ohms, the current through Q2 and C12 (and line capacitance) is infinite. But not for very long. Even with non-infinite/zero practical test-bench situations. I killed two fairly beefy transistors before I figured it out. 22 or 47 ohms has been working for a couple decades, no failure.

Actually, that was intended for very long lines, and I was using 0.01uFd and 0.05uFd caps as dummy lines, and high frequency signals. I bet just 10 ohms would protect your plan for any possible transient, and you probably want 22 or 33 ohms to terminate some board inputs.

 

[burdij]

I think its:

das Raushen - noise
die Bombe - shell