Active ribbon-mic

[Rossi]

Welcome back, Brad

I'm not sure, butI think I do have two 2N4403s somewhere. Unfortunately many of those 2N, 2SA, 2SB, 2SB etc transistors are hard to get in Germany (and I suppose the rest Europe, too), and when you do get them, they're often expensive. I think I paid about 1 buck a piece. Right now, I wouldn't even know a source. The 2SA 1015 is available at Reichelt, though, and pretty cheap too. But that's an online store, which means waiting. Do you know a BC or BD transistor that would be suitable? Those are the ones that the local dealers stock.

 

[bcarso]

It looks like, based on an ancient National Semiconductor transistor databook, that the PNP BC214LC part would be a good bet for relatively low source impedances. The caveat is that the Moto 4403's were often special compared to other parts such as the National ones.

For a comparable NPN part, I come up with no equivalent to the 2N4401 per se, but this databook is very old. I'll keep an eye out as sometimes the latest manufacturers show all of the part numbers that are similar on a given datasheet.

Brad

PS: remember that you have to run some current through these to get the benefits---the other noise term is essentially half the thermal noise power associated with the reciprocal of the transconductance, which is around 26 ohms for 1mA and 300K.

 

[bcarso]

I still haven't found a BC series equivalent to the 2N4401, but one part that looks like it might work, although lowish beta, is the BC635. It's characterized for switching up to an amp so probably is large enough to have reasonably low rbb'. There are PNP complements to the series among them the BC636.

 

[PRR]

> the other noise term is essentially half the thermal noise power associated with the reciprocal of the transconductance

Why "half"? I think you are right, or at least other sources agree, but I can't see the "why".

 

[bcarso]

Hi PRR. It's what falls out of the equations when you take the thermal voltage and collector-current-determined equivalent emitter resistance (or reciprocal tranconductance) and figure out what the noise voltage developed across it is by the shot noise in the collector current (presuming beta is reasonably high). So, you have the shot noise formula 2 times q sub e (electron charge) times collector current term for <i> squared, and then the usual resistance term, the thermal voltage kT over q sub e, divided by the current. The product of those is the equivalent noise voltage spectral density. So that e sub n referred to the base is kT times root (2 over [q sub e times Ic]), which can be expressed as root 2kTRe. If the noise were thermal per se associated with the equivalent emitter resistance, it would be given by root 4kTRe. So it is as if the Re were half as large for noise purposes.

There's a good discussion of the deeper reasons as to how this works, beyond just a fortuitous cancellation of terms, by the late Bernie Oliver of HP, "Thermal and Quantum Noise," from May 1965 IEEE Proceedings. It's been reprinted a couple of times. I hope he didn't dash it off overnight---I'm enough in awe of the man as it is. He points out the effect for P-N junctions, that is, that the noise power is half that to be expected from an equivalent resistance for a given forward-biased diode, and goes on to say that lots of things can display a less-than-thermal noise, like the input resistance of some amplifiers with shunt feedback.

This idea has been exploited to make circuit elements that behave like resistors but with less than thermal noise.

Brad

 

[clintrubber]

from PRR:

> They're 6H (@50Hz) i.s.o. 10 H prim inductance

Ignore those H values. Simulated Henries are free so I use big numbers. The key thing is that it be designed as a good-response microphone (or 150Ω line) input and 1:3-1:5 ratio.

Thanks !

 

[PRR]

> ...which can be expressed as root 2kTRe. If the noise were thermal..., it would be given by root 4kTRe. So it is as if the Re were half as large for noise purposes.

OK.

Now, for a space-charge vacuum triode, theoretical noise is essentially the cathode resistance (taken at cathode temperature, so voltage is about 2.5-3 times higher than room temp). And a chart in a 1950s book gives measured noise resistances of this size or higher.

So why is tube noise not "half"?

Is it because (√4kTRe) is the ultimate Gm of an amplifier? And while BJTs can come very close to this, tubes act like the sum of this Gm plus a "dead resistance" which acts as simple thermal resistance? i.e. BJT vs 12AX7 at 1mA, BJT has Gm of about 38,000uMhos or 1/26Ω, while 12AX7 has Gm of 1,600uMhos or 625Ω composed of 26Ω actual electron/field action and 600Ω of dead resistance? So then the thermal noise is "only" 613Ω instead of 625Ω, insignificant difference??? The 1% difference in noise voltage would vanish in 10% variations of electrode spacing and cathode temperature.

 

[bcarso]

You've got me reaching for Cherry and Hooper 1968 (Amplifying Devices and Low-Pass Amplifier Design) on that conjecture. It may be a while...

Brad

 

[Rossi]

Thanks, Brad. The BC214LC has been discontinued, I think. But the BC635 and BC636 are easily available and cheap, too.

 

[bcarso]

Rossi, I'll be interested to know what your results are. Some parts of this sort have other things wrong with them that don't seem to be captured in the nominal specs.

The BC214LC still comes up on the Faichild website at least, just now. I don't know how good their European distribution is.

Philips also has a BC327-40 (the -40 is a high beta sorted part) that looks like it might be pretty close to the BC214LC.

Brad

 

[Rossi]

Well, I haven't gotten a better transformer so far. Looks like I'll have to order the neutrik stuff online. The place that I saw them (Conrad, a chain like RS) has two catalogues: one for normal mortals and another one for business customers. The Neutrik transformers are in the business catalogue, and, unfortunately, I only qualify as a normal mortal. They don't like to sell the business stuff to mortals in the local Conrad stores, and of course they do not stock the business items. Yes, I love Conrad, I really do.

Apart from that, I built PRR's impedance converter (the balaced one). It performed a little better than my previous one. My crappy transformer didn't sound quite as crappy but still wasn't really usable. There's serious loss in bass response, and the treble sounds gritty. Gain was okay, though, and noise wasn't bad either, considering that nothing was shielded.

I didn't have 2N4403s after all; I misremembered the number. I tried 2N5401 and BC560C; there was no significant difference between those two in noise or sound.

Even the better online dealers like Reichelt don't list the BC214LC. In fact, I don't see any BC2xx. BC327-40 is available everywhere and pretty cheap. I'll try those.

 

[PRR]

> My crappy transformer didn't sound quite as crappy

{sigh...}

OK, spend some power (more than Phantom can easily give) and try transformerless (using active devices to avoid buy good passive devices; amplifying yourself to your goal):

This eats 6V of battery. If you use four D-size cells it can run weeks without a power switch, with AA-cells you want to turn it off after the gig and a set of AAs should last a year. I might be able to do this from Phantom but not easily, not tonight. Let's try the "simple" battery power first: it may have too much noise in which case no point in developing the idea.

Note that everything sits Negative of the Ground. This is a mind-twist but it ensures the output caps are polarized whether the load is grounded or at Phantom voltage.

Gain is 11 or 12 dB, THD is around 0.02% at 20mV input, mostly 3rd.

Transistors should be big for low noise. It will "work" with little transistors like 2N3904 but if it is noisy try 2N4401/4403/etc big fat-Base switches before you condemn it. LM394 or the MAT or THAT similar parts are also good bets.

 

[clintrubber]

Transistors should be big for low noise. It will "work" with little transistors like 2N3904 but if it is noisy try 2N4401/4403/etc big fat-Base switches before you condemn it. LM394 or the MAT or THAT similar parts are also good bets.

Stupid me - I asked why better use big switching types i.s.o. low-noise types but got misguided by 'low noise'. I've never compared, but then we have the situation that the larger types which don't have the label 'low-noise' are in fact lower noise than those which are advertised as such (BC550/560 etc).

Thanks,

Peter

 

[Rossi]

Thanks a lot PRR! I'll try that circuit next weekend, I won't have time before that. But I do wanna get those Neurik transformers. Even if they don't work well with the first circuit, it'll be nice to have them for further experiments. The Neutrik stuff usually isn't total crap.

2N4401/4403 are impossible to get around here. My Siemens V272 uses big NPN switching transistors in the output stage. They're labeled SST117, but I think they're identical to BSX45. I'm sure I could get those, though perhaps not in local stores. Would small power transistors such as BD138/BD139 be suitable? At least they're big... :roll:

 

[PRR]

> why better use big switching types i.s.o. low-noise types

If you have a free choice of source impedance, wind to about 1K-10K impedance and use a small high-gain transistor. It is easier to get high Beta on a small die, the odds of a defective spot are lower, and small transistors are so cheap you can sort-out the noisy ones.

In this case we have a specified fixed source impedance: the 150-250Ω of a commercial "Lo-Z" microphone. If transformers were ideal and free, we would use a step-up transformer into a small transistor.

Transformers are not ideal or free. Transistors are very nearly free today, and you can find one with the several critical parameters very close to ideal.

However the "low noise" transistors were developed 30 years ago when there were a lot of non-ideal excess-noise transistors around. They have high Beta and selected processing and are still good choices for impedances like 1K-10K (though most any high-Beta transistor will work about the same).

These transistors, aimed for medium-Z sources, have base resistances (dead resistance in series with the Base) of 50-300Ω. This is "small" compared with 1K-10K sources. But it is "large" compared with a "lo-Z" microphone. So the noise resistance would be 200Ω unavoidable resistance in the mike, 200Ω dead resistance in the transistor. Total noise power is twice the noise of the mike alone. 3dB Noise Figure, where 0dB is perfection (no amplifier noise added to mike self-noise) and 1dB is possible with a transformer and good transistor or tube.

So we want a transistor with low base resistance and good Beta. At 100-200Ω the Beta is less important (it transforms current noise which tends to be a small problem); we want low Rb.

The 2N4401 is a "Switch". You would use it to turn-on lights, relays, other part-Amp loads. And designers often want a switch that is easy to drive, like a car with power steering, to lessen the load on whatever is driving it (may be a cheap CMOS gate or processor). So a good switch will have fairly high Beta, but also a low base resistance so that you can stuff excess current (for sure switching) into the Base without needing a high voltage.

I don't have the 2N4401 datasheet open. But I think it is rated for 500mA collector current with small Collector-Emitter voltage drop (for efficiency). And that Beta is over 50, so for solid switching you would force 50mA of base current to ensure solid 500mA load current, yet the 2N4401 will have less than 0.8V or 1V of Base voltage at this 50mA base current.

Hmmmm.... even a big-die device will have about 0.7V intrinsic Vbe at 50mA into the Base. If the specified Vbe is 0.8V at 50mA in the Base, then resistance is causing (0.8V-0.7V) 0.1V extra drop. 0.1V/0.050A= 2Ω of dead resistance in the Base. Actually the specs often say 1.0V Vbe at heavy drive, so it may be 5Ω-10Ω. Still this is much lower than the "low noise" parts, and lower than our 200Ω source, so the base resistance noise is low compared to mike self-noise.

{later...} Here's a snip of the 2N4401 data:

This is typical Vbe when you force Base current to be 10 times Collector current. I want to see Base current so I have re-labeled the bottom.

Note that from below 0.1mA to well past 1mA in the Base, the Vbe plotted on log-lin coordinates is a straight line, as we would expect from the exponential action of an ideal junction. At higher current it diverges, as we expect from the real-world fact that we can't make perfect contact to a junction. If you measured Vbe for base currents of 0.1 Amp to 1 Amp it would be curved on these coordinates but straight on lin-lin coordinates: very nearly a pure resistor. Except a real 2N4401 would melt very quickly at 1 Amp! So I've extended the "ideal junction" line from 1mA-10ma, where it does follow ideal action very nearly, up to 50mA where it is clearly non-idea. At 50mA we expect about 0.8V, we get about 1.0V. The difference is 0.2V. 0.2V/50mA= 4Ω. The transistor acts like an ideal device with 4Ω of dead resistance in the Base. (Some of this is really Emitter dead R transformed by Beta; we don't much care. Re is usually small, much smaller than Rb. And if it were large, this would be a lousy Switch transistor, so we know the maker kept Re as small as practical.)

This derivation is sloppy and unrealistic. It is based on data from soaking the junction with current. Noise resistance in the linear amplification range is higher, maybe much higher. For 2N5089 (classic "lo-noise" part) the forced-Beta curves suggest Rb is 15Ω but the measured noise curves say 800Ω in the audio band.

Still the idea is good. Pick a transistor that makes a good switch at currents 100 times higher than the proposed amplifier current (and be sure Beta holds up at your lower current), and non-ideal parasitic effects will be small. (This also assumes modern clean Silicon: 1967 Silicon and most Germanium has enough leakage current to complicate noise estimation.)

 

[bcarso]

Good discussion and reasoning PRR. I think the extrapolated numbers are a bit optimistic based on curves of e sub n in Motchenbacher and Fitchen (or the later book with Connelly), but the general idea is sound.

Unfortunately as regards processing, I understand that for some discrete devices there was a golden age when the material got better and the processing was more or less optimal. Ex Oxner (retired, ex of Siliconix) says that when they went to ion implantation to make FETs things got noisier, and basically nobody cared. Once you've banged up the lattice there's only so much that annealing can repair. I wouldn't be surprised if something comparable happened with some bipolars.

The T*shiba folks seem to still use diffusion for a bunch of their parts, and I have found their bipolars to be the best noise performers. Their FETs reign supreme as well, although I have some ancient Teledyne Crystalonics C413 (a long-defunct company) parts that are comparable. I wonder if Customs and the Patriot Act etc. would allow me to ship small quantities of silicon into these poorly supported regions of the world?

Brad

 

[Rossi]

Thanks for considering, Brad, but I finally found a source here: http://www.fibra-zw.de/ They seem to have much of the rare stuff, and at good prices, too. But for some reason they don't carry a lot of the stuff that's available elsewhere. Why can't there be a place that has *everything*? Looks like I have to order stuff for this project at three different places.

 

[Marik]

If you guys are interested, here is a schemo for Oktava ML51 ribbon mic. It did not have a transformer. Once I had this mic, it sounded nice, but was noisy like hell:

 

[PRR]

> I think the extrapolated numbers are a bit optimistic

Perhaps more than "a bit".

More to show why a "fat" transistor may be better for low noise with very low and fixed source impedance.

> here is a schemo for Oktava ML51 ribbon mic.

Is that what Octavia really put in it? Very odd biasing. Not sure it could work at all in production. Maybe they had some low-Beta transistors sorted-out from another model.

Q1 probably works at about 1mA current. I'd expect the noise resistance to be OTOO 15Ω-30Ω. Even if the ribbon were stretched and thinned up to 1Ω, that's a terrible noise figure. Q1 apparently works at gain of ~2K/28= ~70, which is in the ballpark of voltage gain needed for a 0.2Ω ribbon but seems high for a 1Ω ribbon.

It does not make sense as drawn, but if it is anything like that plan I'm not amazed that it had high noise.

Parallel 25 to 100 low-Rbb devices, flow about 100mA through them. You might get in sight of a ribbon's self-noise, and you won't need an output buffer. I suspect this extrapolation is too simple, and certainly layout of 0.2Ω systems will be problematic.

 

[PRR]

> BD138/BD139 be suitable?

Possibly. Near as I can tell, they are somewhat like a double-size 2N4401 in a power package. Beta at low current is good, though if I understand the Euro Beta-number system then you probably prefer the -16 (hi-Beta) version.

Since they make no pretense of low noise, you can't complain if they hiss more than theory predicts. Clasically, excess noise suggests dirty Silicon which is a reliability problem too. And dirt-induced unreliability seems to have been solved in the last couple decades. (You can't make million-transistor many-Watt Pentiums and DRAMs if you have 1% dirt-specks or any excess leakage; even the mundane discrete foundries have inherited much cleaner techniques from microprocessor development). But as Brad says, there are new ways to skin cats that meet all specs yet don't work like the good old cats. Instead of slow-baking arsenic into the wafer, they can zap it with electrons to get a P or N layer. Probably a lot more repeatable (same "done-ness" every time) but does violence to the silicon crystal. Gain and breakdown and leakage may all meet specs, but the ragged crystal may not flow as smooth as the classic diffusions.

I would assume the LM394, MAT and THAT are still made the old-fashioned way. (Though I think one MAT part is now really a LM194 die in a MAT package and price.)

THAT300 is an interesting alternative to LM394. Profusion http://www.profusionplc.com/ will sell small quantities at GBP £ 6.684/2 to 5.348/5 ($13-$10). This is twice the price I can find LM394 for, but the THAT300 is a quad part and at these low impedances we might want to parallel things. Rbb is claimed to be 30Ω, so in a 200Ω differential input you could get a slightly lower Noise Figure with all four in differential-parallel than with just a pair. But to get near a 0.2Ω ribbon you want over 100 devices, 25 quad-packs: £83, $160, ? 120 ! And about a 10"x10" or 250mmX250mm board. Enough to make a transformer look cheap.