Matching Transistors for Valley People MicPre

I am building a Valley People Trans Amp.

It uses paralleled PNP transistors, diff circuit, with a constant current source.

I am trying to figure out the best way to match transistors.
The VP uses closely matched transistors that are thermally coupled, one transistor from the + side of the signal input gets thermally coupled with one from the minus side.

I have used Emitter Base voltage to match them, Collector voltage, etc.
(dc matching)

I am now using a more dynamic matching method, injecting a sawtooth wave form and comparing the two transistors.

Bias and collector resistors have been matched to the nearest ohm.

I am looking at the output voltage on the collectors to match the transistors up.

My problem is that I can get a match at 100 hz, but at 10,000 hz, they no longer match. So what frequency should I use as a standard to match with?

Thanks,

cj

Let me know if youm need a schematic.

A schematic would be helpful, but if they are directly in parallel the most important parameter is Vbe. Be aware that this is highly temperature sensitive. What I do is tape down a whole strip of candidate Q's and use a DVM with a diode test mode on the ohms scale to measure Vbe at what is usually 1mA. Don't touch the transistors for a while and avoid drafts; avoid cycling the HVAC system in the vicinity. Write down the reading on the paper below each device. It's not going to test everything but it will get you close.

Ideally you would also match for Vbe at the operating current, ditto beta; parasitic Rc, Re; rbb'; capacitances etc etc. But again for direct or low-R ballasted parallel op the Vbe is the most important.

If you match to a millivolt things should be fine.

PS: and it goes without saying that the transistors should all be from the same date code.

How would I ramp Ic up from lets say 4 ua to 4 ma with a sawtooth generator?
Right now I am simply injecting the sawtooth ito the bases via a couple of tant caps and looking at the waveform on the collectors of the transistors in question.

I have heard of parasitic Re befor. What the heck is it?

thanks!

I don't know what your test circuit looks like, but for generating a current ramp just make a current source out of the appropriate polarity transistor, i.e. a fairly good-sized resistor in the emitter, and attach your sawtooth gen across base to supply. Your function gen probably has a d.c. offset so by playing with that and the amplitude you should be able to dial in the two endpoints of current. Take the current output from the collector. This won't be perfect but who cares. Float the generator if it makes it easier.

Re would be the "dead" resistance that appears in the emitter even at high collector currents. The regular r sub e (the reciprocal of the transconductance) is about 26 ohms at a milliamp of emitter current and is inversely proprortional to that current (at about normal room temp). If you increase the current the r sub e keeps going down but limits out somewhere---that is, the transconductance only rises to a point with increasing collector current. When it levels out you have a measure of the "dead" emitter resistance. Usually by this time the device is getting hot so it is a tricky measurement--rbb' is also rising since the tempco of bulk silicon is positive.

This is often why you will see the spec of "pulsed measurement" in transistor datasheets, usually with a 2% duty cycle.

This parameter is usually small for most apps but gets important for apps where the exponential/logarithmic conformance is critical---for example, in log-antilog based VCA's.

Thanks!

I had some goofy scope leads. In times 10 mode, they were giving different readings at higher freqs. Go figure.

OK, sig gen goes from base to supply ground, or supply positive?
These are PNP's, so emmitter goes to plus. Bias resistor on collector base.

OK, I think I got it. I raised the dc offset to lower the current down next to zero on the minus cycle. You can also play with the supply voltage to further tweak it.

I assume you want a slow freq, like 1/2 cycle per second so you can watch the ammeter.

Nice to have an analog meter in this situation!

So this is kind of like a poor man's curve tracer.

OK, back to do some matching tests.

Yep yep. You have got your mojo workin'.

I could see no difference between curves on the sawtooth test. So I tried your diode checker. No wonder. These guys came out of a box of qty 2500. They all measure within a millivolt of each other. 659, 659, 660, 659,,.

So time to shave them down to the substrate and piggy back these guys
with some heatsink pads.

Thanks!

cj

In this circuit....

Match Vbe at 1mA to better than 5mV.

That used to be tough. As you see, today you just buy all-at-once and you probably get matching much better than you need.

Just to be sure: check Vbe at 10mA and weed-out any oddballs. They have parasitic resistance. (You might not find any.)

Note that 1 degree C gives 2mV change: you need to be VERY isothermal when checking Vbe to a few mV.

The quick, easy, and reasonably accurate way to check Vbe is with a plain old (yes old) VTVM, the kind with a C-cell for the ohm-meter function. Set it to the range that reads 1K at mid-scale: this is really a 1.5V battery plus a 1K resistor, and a voltmeter. "Read the ohms" of the Base-Emitter junction. It will read something less than half-scale. In fact if you ignore the fact that you are in "Ohms", and look at the 1.5V scale, that is the Vbeo. Assuming Vbe is about 0.6V, the current is (1.5V-0.6V)/1K or darn-close to 1mA, the current this design runs at. To do the weed-out at 10mA, use the "Ohms" range that reads "100" at mid-scale.

You have to squint to resolve 1mV on a VTVM. I used to be able to do it, on mirrored-scale, when I was younger. But 2 or 3mV is fine for this plan: that gets balance to 10% which is more than good enough.

Note that the VTVM drifts 10-100mV hour to hour, day to day. Let it cook overnight, then test all your transistors quickly.

This Vbeo is not the same as you will observe with current in the collector, but if all parts are from the same lot they will still match.

The large Base resistors make Hfe somewhat significant. Nominally the Hfe is 200. I doubt the exact value matters, but the match might. With a large quantity, sort out a dozen in the 150-250 (at 1mA) range, then find four that are within 10% of each other.

The ones that match Vbe but not Hfe can go in the current sources.

All that stuff about thermal bonding was overkill. Yes, if you put a "matched" pair in a draft, it won't match. Remember: 2mV per deg C, so even a part-degree difference gives 1mV "unmatch". My 1970 book says simply putting a cardboard shield around the input pair (or quad) makes a big difference. Bob Pease says similar. A copper "S"-clip used to be common to bond two transistors. I have also seen metal tape smooshed around a pair.

Shaving to the substrate is NOT needed. There is no large heat-flux. Get the transistors much closer to each other than to air-drafts and they will be fine. Mount them touching, then drip some epoxy between them. Wrap in a scrap of plastic to keep the air off. Stay away from heat sources and their cooling vents, but you do that anyway.

The only major effect of an unbalance is some DC at the output, and that's maybe more an annoyance to the designer than a real issue for the audio engineer. But unbalance also increases THD numbers (by adding even-order nonlinearity), so a prolonged unbalance can give high THD readings.

Note that this plan has a balance trimmer. Ideally you won't need it. If you can pull 1mV matches and be reasonably isothermal, you could short-out the trimmer. (Unless you believe that residual unbalance is part of "the sound".)

Got it. Thanks you guys!

Going with 8 ea. 3906's since they are matched so close. 4 on each side. Will this load down a dynaqmic mic too much?

Should I stick with 4 becuase of diminishing returns?

Going to tie wrap the pairs together with a piece of transistor heat sink pad in between, then put in a case with some type of goop.

I don't like that balance pot either. Going to trim it with resistors if needed. Pots can drift.

cj

Probably the circuit has significant feedback to the transistors, so it's not going to likely be an issue of loading. That is, the input impedance will still be high.

More germane is what is the combination for lowest noise. With the rather high rbb' of a 3906 you may be still improving matters with 4 per side---hard to say as it also depends on bias conditions. If good 4403's were used 4 might be overkill.

bcarso posted about 2n4403s first. I was going to post just use two of the 4403s

Cool ideas. I am doing a circuit board with 8 spots for transistors. I can always snip a few if I find 8 is overkill. Will be fun tto try different transistors in there. Original silicon was MPS4355.

[quote author="CJ"]Original silicon was MPS4355.[/quote]Well, don't you have any of those? :green:

Seriously, go for the 2N4403.

HTH!

39096 has hfe=100. 4403=60/ i will give 3906 a shot first since i have a box of 2500 to play with and the 4403 will cost me a whooping dollar for five.

Any recomendations for a dual opamp, something better than a 5532/4 or tl072?

i will socket the darn thing so i can plug and play.

CJ,

Do you want FET or BJT dual OA? I have some OPA2134 or OP275 I could send you, if you need them to try.

HTH!

You could always try That corp's 320P. 4 matched pnps in a DIP14 package. Pricier for certain, but very low noise, and it doesn't get much better in terms of thermal coupling.

Only disappointing thing about the THAT stuff is the betas aren't that high.

> something better than a 5532/4 or tl072?

What could be better than a 5532? Though as I recall, the opamps were TL072-like BiFETs. (and how can we advise you when you won't post the opamp end of the schematic {which I thought I'd seen already on some public site}?)

And... are you trying to "recreate a vintage module", or making something modern out of something old and over-complicated?

The Valley stuff was good, then and even today.

If you are going to fake it with 3906... Why? A handful of 3906 isn't as good (for THIS use) as one LM394. So it is NPN: turn the whole DC around.

Though I suppose if you are stuck with 2,497 of those 3906s, slathering them on mike-amps is one way to get the box out of your way.