Active ribbon-mic

[Rossi]

Wow, that was fast, PRR. Thanks a lot. I'm gonna try this circuit next weekend. I rarely have time to do stuff like that during the week. I'll try to get some fat transistors locally to make experiment before the Reichelt stuff arrives. There aren't that many electronics shops left in Germany, either. There used to be a really good one just one block from where I live. The company still exists, but now they sell designer lampshades & stuff...

Reichelt is as good as it gets for DIY in Germany. It's a good company with a decent selection. But there's always parts they don't carry for some reason (or maybe no reason at all).

Companies, of course, have other sources available. They probably can get any part they desire. I don't think there's much of a problem with import limitations etc. If there was serious interest for certain parts, I'm sure they could be had. But serious interest does not include DIY. There seems to be an ever greater difference between what companies use in their products and what DIYers can obtain. Think of NJM4580 opamps (not that they're that great). They're the most frequently used opamps in "prosumer" equipment. Yet you won't find them in electronics stores.

Transistors that I saw used in mic preamps? The 2SB737 was very popular until it was discontinued. It was used for instance by Mindprint and Behringer. The common replacement now is 2SA1084. I think I saw BC546B used as the PNP counterpart. In the Behringer MX8000 (an early model) I found 2N4401 in conjunction with 2SB737. In old German broadcast preamps such as the said Siemens V272 you often find BCY58, BCY66 and BSX45. Neumann tended to use regular BC-transistors such as BC107.

Quite a lot of recent preamps are based on INA163 or INA217 chips. While I do like discrete designs, I found that those chips often perform better than most of those simple semi-discrete designs built from two complementary transistor pairs + opamp stage as found in Mackie, Behringer, and many many others. I don't know if that's because of poor design or budget compromises.

 

[bcarso]

Rossi: "Companies, of course, have other sources available. They probably can get any part they desire. " Well, sorta. If you have someone in the far east to buy them. Harman once paid Mouser a buck and a half apiece for 2SA1015GR's, before they could persuade a subcon in China to send bags of the things to Northridge.

Chips Technology Circuits in New Jersey has a highly variable stock of at least Japanese semis, and there are others. But they probably don't ship overseas.

The NJM parts from JRC historically have been very poorly supported in the US and elsewhere. They have had their quality problems too, but evudently both support and consistency (consistently good ) have been improving. And as you say they are not all that hot, but at around ten cents a dual in quantity for the 4560 I've populated many a mid-fi product with them to no egregious fault. Actually I think the Rohm parts are a bit better (BA4560 etc.) and they are not too badly supported.

BTW, there tends to be a slight advantage arising from the "N" base material to using PNP's for low voltage noise apps, all other things equal.

There is no dispute that the better ICs are fantastically better than they used to be. There is less fun to be had, clearly, in using them, and you're probably not going to learn as much. One of the areas that is always a potential advantage using at least some discretes is the reduction of thermal effects of a changing-dissipation output stage affecting other stages, especially the input. What one would like is a good supply of matched multiple transistors for places where the monolithic match will always be better than hand-matched single discretes, and then optimized singles for other places. Wouldn't hurt to have some extras so one could implement a temp-servoed substrate as well (like National did in their LM299 voltage ref part).

BTW On Semi has announced some Toshirola-Motoshiba big power biploars with on-chip bias components (I guess diodes). They need to do some for the driver parts too but this is a step in the right direction for really fast thermal feedback for bias control in power outputs.

Brad

 

[tmbg]

If you guys like this design and have it pretty well stabilized sometime soon, I think I'm gonna go way out on a limb and try my hand at a SMD layout for it. I bet I could make a nice tiny little board that would go well inside the mic....

I might be talking entirely out my posterior, however

 

[Rossi]

I'm pretty sure this circuit will fit into the mic without resorting to SMD layout. The mic is large, and, if I remember correctly, there's plenty of chinese air at the bottom opposite to the transformer.

SMD transistors may not be the best choice anyway. As Brad and PRR pointed out, size matters in this case. Even SMD loving parts scrooges such as the big B use non-SMD transistors on their otherwise SMD preamp boards.

 

[bcarso]

Rossi: "if I remember correctly, there's plenty of chinese air..."

And what about Chinese pace, rhythm, and slam? ;-)

Some of the better chips are only available in through-hole, but there are more coming along in SMD, including the ones with a decent heatsink tab if you are running some serious current or otherwise high dissipation. In some cases (not this particular one, probably) the lower lead inductance can help performance.

It's usually not a matter of the actual chip area though: take a crack (no pun intended) at removing the case from a ~1A rated TO-92 case switching transistor and look at how big the chip itself really is.

Linear Integrated Systems is making their Toshiba 2SK170 JFET substitute, their LSK170, available in SMD, although the published first-draft datasheet shows a too-high max. power rating imo for that package. They have plans to make a dual, a replacement for the discontinued Tosh 2SK389, also available in SMD.

Brad

 

[tmbg]

I believe the 2N4401 is a 500mA part, the SMD version of the 4401 is 200mA max, possibly still enough die area to provide low noise.

The biggest hurdle in the SMD layout I'm working on is the fact that the input caps are 100uF 50V parts, and they're 10mm square. That eats half my board space right there. I was thinking about moving to miniature radial lytics to save space, but they're just as big...

What practical voltage do those need to be? I can save a LOT of space by sizing those down somewhat.

Anyway, regarding the SMD vs non-SMD issue, I'll probably do one of each... this is more learning experience for me than anything, and if it works out, I'll certainly make the layouts and gerbers available to anyone that wants to make one.

I'm gonna order it through sparkfun.com, I actually have an order in to them right now, $10 including shipping for one 3 square inch board. I've talked to others that have used them, and they have raved about the quality.

 

[bcarso]

The die area is probably the same if it is called a 4401. The difference in current rating will be due to power dissipation considerations, most likely.

There's hardly any d.c. across the 100u caps if the transformer floats. If it has a center tap referred to circuit ground then the caps see about 3V. There is a slight preference for higher voltage 'lytics when you have room for them for various reasons, but in your case you might want to get some 6.3V parts and test them for leakage and dissipation factor, and weed out any super-leaky/lossy ones. Leakage will generate noise. The stated leakage characteristics for most 'lytics are very pessimistic btw. If the foil is good and the electrolyte proper they can get to be quite low leakage after sitting at the operating potential for a while. Again, use the highest voltage you have room for.

According to some recent and controversial results from Cyril Bateman, published in an Electronics World (UK) issue some time back, it pays to used nonpolar 'lytics in audio circuits even if there is a bias voltage across them. They tend to be about twice as large though so maybe not what you want here.

Brad

 

[tmbg]

Good point on the 4401 die... and since we're not actually concerned about dissipation issues, sounds like a win win.

Good to know about the caps, I'll reevaluate what I can get and do a new layout...

Without those enormous caps, I may be able to fit this circuit into a half a square inch, which would be very very nice!

 

[bcarso]

Another possibility for the bold and the brave (and possibly the foolish): Substitute matched LSK170 JFET's in SMD for the biploars, and make the coupling C's SMD films of say 1uF. Now the bias R's can be raised from 4.7k to 470k, say. Take out the 22 ohm R's (or down to a couple ohms selected to balance the residual output offset voltage) and you'll get about the same, maybe a bit lower, voltage gain. The better the match between transistors the lower the even-order THD. One caveat: the SK170 is rated for 40V. Under severe overload you can exceed that, although it is probably non-destructive. Also, gate leakage is beginning to climb at the power dissipation and drain-gate voltage, but at these impedances shouldn't be a serious problem, although it's just beginning to get worrisome. It will be worse in SMD because of the higher thermal resistance from junction to case. Noise should be comparable to the 4401's.

One could also brute-force-down the drain voltages with a stiffer bias divider, but this would make C6 want to be larger. Raising the drain currents by reducing R5 or raising the bias voltage could also be done. It looks like taking R6 to 100k helps a lot; FET current goes up but dissipation goes down and drain-gate voltage goes down a lot.

A more elaborate all FET circuit would add two more higher-V FETs in cascode and this would fix the gate leakage problem, if any.

My sim says the distortion with a 10mV peak differential signal is about double-ought four, but again this is for perfectly matched Q's. With the suggested change in R6 it drops to < double-ought three. The differential gain at this point is 12.2dB with one ohm R's for R3 and R4---those being just placeholders for an offest voltage tweak. The sims say the voltage gain is 12.2dB, compared to the biploar 12.8. FETs have much larger parameter variations in a given type than biplors, so these numbers are approximate to say the least.

For the even braver or more foolish the caps and bias R's could be omitted and the transformer center tap tied to the bias voltage divider. Hazards include d.c. through the tranny winding if the power doesn't come up symmetrically, so maybe it's not such a good idea.

Brad

 

[tmbg]

Other thing is, I'm simulating this circuit, and I get 12.757dB of gain, and it's only 1dB down at 1Hz. Do we REALLY need that much extended frequency response? If we can afford to knock those caps down an order of magnitude, that will also make this project smaller and cheaper.

With 10uF caps, I'm getting 1dB down at ~11Hz. Phase shift begins at a higher frequency with the smaller caps also, I don't know how much of a concern that is.

 

[Rossi]

The transformer has no center tap to ground (would have been dangerous for the ribbon in case phantom power was accidentally turned on).

FET version sounds tasty, but for now I'm gonna stick with the bipolar one. Quality FETs are mailorder only. I tried to get 2SK170s before; my local dealer offered to order some for me (at about 1.5 Euro a piece) but never called back. BF245 are about the only FETs you can buy cheap around here. I don't suppose they're appropriate substitutes.

Brad, you say there's a preference for higher voltage electrolytics. Do you mean for sound or for safety reasons? Or both?

 

[tmbg]

Hm, I wonder if it's the way I drew this circuit, or if it's because I'm not familiar with the software yet, but my sim is showing 0.831% THD on the sinusoidal source, and 6.198% THD on the output. That seems a tad excessive :/

 

[bcarso]

One reason to have that low a coupling tau is to minimize noise and distortion. As far as distortion, the less you slosh around that 'lytic stuff the better. OTOH, the bigger 'lytics have more leakage so if there were much d.c. across them that could also be a problem.

Also, 'lytics have very poor tolerances on the actual C value. So you will often see a.c. coupled diff amps using high values of coupling C's to assure decent common-mode rejection at much higher frequencies than the cutoff.

Also, 11 Hz sounds low but will have associated with it a large phase shift at fairly much higher frequencies---one more variable to be wary of in a multi-miked situation. To that end, I demoed a buffer stage for a guy who heard (without knowing what he was listening for) the difference between a electronics lf cutoff at 140 milliHz vs. the 100 milliHz he was used to in his system. I was impressed.

I like the idea of d.c. coupling with the FETs but it calls for some contrivance to assure d.c. is kept out of the transformer, maybe. They could be run roughly floating gate, or with large value bias R's located strategically, but this asks a lot from the matching, board leakage, and common-mode rejection.

 

[PRR]

> the input caps are 100uF 50V parts

Oh, no. 6V is ample.

> it's only 1dB down at 1Hz. Do we REALLY need that much extended frequency response?

Transistor Bases spit-out noise current. This flows through the base circuit and makes noise voltage. We want the base circuit to be very low impedance, basically no excess impedance other than the "good" impedance of the source (microphone).

With 5uFd (10uFd+10uFd) in series with 260Ω (200Ω mike, 2X30Ω base resistances), noise will start to rise below 120Hz. If we are talking hundred-buck mikes in home-studio work, that's probably perfectly acceptable: non-massive studio walls let in all the truck rumble in town, and if you really cared you'd be using better mikes in better studios. So yeah: try 10uFd 6V.

> for the bold...: Substitute matched LSK170 JFET's

Makes perfect sense to me. The R(noise) seems low enough. Bias can be simpler, maybe none. Distortion will be more gentle. And Gate current noise should be a non-issue even with quite small caps. Or no caps at all? Remember I drew a long-tail diff-pair for habit and bias stability: we don't actually need CMRR, the next stage can do that. And if we need industrial-strength CMRR and CM voltage range, we shouldn't be mucking with cheap mikes and cheaper boosters.

My sims say ~11dB gain. Are you guys simulating with reasonable source and load resistances, or zero/infinite Z?

 

[tmbg]

zero/infinite!

 

[bcarso]

Rossi: ' lytics at higher voltage will have typically lower leakage and hence less problems with noise and bias shifts etc. They also tend to have lower loss. That shold equate to better sound but I can't personally attest to hearing that myself.

Yeah, getting the FETs is a bitch, and yes, the easily obtained ones are usually not very good. FETs didn't used to make sense at all for low Z sources until the really good ones came out. When they are large enough to have low voltage noise they tend to be fairly high capacitance as well, although probably not an issue in this app. It is nice to have mostly negligible current noise though.

tmbg: your generator should be perfect if it behaves like mine, maybe just showing some numerical-noise-related THD and acting like a perfect voltage source. But you may be looking with a single-ended probe and seeing crap from the implied reference to ground. So be sure you are looking at the inputs and outputs differentially. I have a little 3-ideal-opamp instrumentation amp block that I paste in to new circuits when needed--two voltage followers driving the standard four-equal-R differential amp. The single-ended output(s) distortion of this stage is fairly large by itself and is almost all 2nd. Also, be careful not to overdrive too much, and do a reality check on d.c. levels to make sure everything is really hooked up right. The circuit should display perfectly matched values of voltage and current for the symmetrical components.

 

[bcarso]

PRR: "My sims say ~11dB gain. Are you guys simulating with reasonable source and load resistances, or zero/infinite Z?"

What's a good value for the load Z?

I'm sure that and the bit of source Z accounts for the slight reduction.

 

[PRR]

I assume about 2K input Z for a modern mike amp.

This makes a small difference here because I scaled the booster output Z up higher than we might like, to 300Ω. The increase in gain and output voltage outweighs the increase in current-noise of the next amp.

 

[bcarso]

PRR, yes, that accounts for the gain difference. With the 270 ohm source Wes Dooley says his mic looks like and the 2k load, I now get 11.2 dB with the bipolar version.

 

[PRR]

> I have a little 3-ideal-opamp instrumentation amp block that I paste in to new circuits when needed--

Some kind of lame simulator?

My antiquated one has ways to measure differentially. There is a "differential probe". If you are close enough to the SPICE to see punch-cards, in many places you can write V(R43) and get the voltage across a resistor. And there is the VCVS, 4-terminal Voltage Controlled Voltage Source, which has an ideal input and output, both fully floating. (Small problem with that: it is so "ideal" that you may need a dummy 1Gig resistor to keep SPICE from complaining.)

> my sim is showing 0.831% THD on the sinusoidal source, and 6.198% THD on the output. That seems a tad excessive :/

The basic "ideal" SPICE source runs 0.00? THD at mV-V levels due to rounding error. Even if you short it with a huge diode: it can supply infinite current. So something is odd.

Realistic signal levels are 20mV RMS (28mV peak). I figure 0.050mV at 74dB SPL (~10dB less than SM57/58), and music played 52dB higher. Rock-studios may have higher levels, but then you don't really need boost. And it won't suck at 90mV peak which is over 135dB SPL which is more than you get at the surface of a guitar or PA speaker cone. It starts to go over 1% THD, but so would your ear in the same location.

At levels more than a few mV, reading just one side of the output will show more THD than you'll get with a proper differential input. I didn't model that, and should have, since low-price gear might have non-diff inputs on XLR jacks.