To: sgrossklass
Thanks very much for your reply! The article on star grounding was quite good.
"ZTX718 might be a better candidate."
I looked at the datasheet for the ZTX718 and was somewhat dismayed by the "absolute maximum rating" of collector-emitter voltage at 20 volts. Seems rather dangerous in a circuit with +/- 17 volt rails, don't you think?
"reducing R5/7 to 7k5 like they are in later models."
How much will this affect the gain?
"replacing the opamp with a faster one can backfire badly, so be warned."
I have seen this warning as well. I was planning on putting 0.1 caps across the power rails at the opamp pins 4 & 8 to avoid problems. Do you think this would be sufficient?
THANKS for your help!
Thanks very much for your reply! The article on star grounding was quite good.
"ZTX718 might be a better candidate."
I looked at the datasheet for the ZTX718 and was somewhat dismayed by the "absolute maximum rating" of collector-emitter voltage at 20 volts. Seems rather dangerous in a circuit with +/- 17 volt rails, don't you think?
"reducing R5/7 to 7k5 like they are in later models."
How much will this affect the gain?
"replacing the opamp with a faster one can backfire badly, so be warned."
I have seen this warning as well. I was planning on putting 0.1 caps across the power rails at the opamp pins 4 & 8 to avoid problems. Do you think this would be sufficient?
THANKS for your help!
Err, oops. It would be marginal at least.
What do you think of 2SB1260 or ZTX550 then?
Slightly.
About +2 dB on the bottom end of the control range (due to 10k ||(7k5+7k5) vs. 10k||(15k+15k)), about +3 dB on the top end (due to higher Ic reducing 1/gm from ~26 ohms to ~13 ohms, so between emitters you're effectively seeing ~26R + 33R = 59R instead of ~52R + 33R = 85R). Not much in between.
This applies to both open loop and closed loop.
BTW, the schematic doesn't say anything about whether the line-in has some series resistors or a pad to bring levels down. Simulation reckons that you'd need 8k2 in series to handle +22 dBu (modified - the original circuit will handle about 2 dB more due to the difference in gain).
Simulated input noise @ max gain, 2N4403, 200 ohm source, 1 kHz:
2.39 nV/√(Hz), original
2.35 nV/√(Hz), modified
2.34 nV/√(Hz), modified w/ OPA213x
0 ohm source:
1.34 nV/√(Hz), original
1.24 nV/√(Hz), modified
1.23 nV/√(Hz), modified w/ OPA213x
That's (a) depressingly little difference, and (b) about -127 dBu with 200 ohms or -132 dBu shorted even "original" (well, with 2N4403s, but presumably the PN4355s weren't much worse). That would still be in a league with good modern mic inputs. I have a little Behringer Q1002USB mixer from 2014-ish that only gets to about -126 dBu shorted! (Many USB audio interfaces are not any better or even worse.)
I think you'd be best off leaving the actual mic pre alone (aside from trying some different input transistors in one guinea pig channel) and tackling any factors that might degrade real-life performance vs. these simulated results. I wouldn't be one bit surprised if cleaning board to board connectors, implementing low-impedance star grounding and beefing up the ground traces on the input boards makes a difference. (I would look at the return from the rail RC filters in particular.) Remember that the 200 had pretty cheesy (consumer grade) PCB material, so copper may not have been the thickest ever. It can't hurt to check resistors for tolerance either, even though at least the critical ones may well be metal film. And as stated earlier, recapping if it hasn't happened already.
I have my doubts. It would help (though a 100n between rails may be preferable in this noise-critical application), but you're talking a part with not only close to twice the slew rate but also over 17 times the GBW, and a layout from 1983 on a single layer pressed paper board. I'd try something a little less extreme like OPA2132/4 or OPA1642 first.
Last edited:
>> 20 volts.
> It would be marginal at least.
The peak transistor voltage is the idle voltage plus half the peak mike signal. The stock plan idles at 10V. The minimum gain is about 5, with <15V peak out, so 3V peak input before clipping, or 1.5V each transistor. Looks like 12V rating is ample as long as you do not clip it. If it has been *severely* overdriven, the broke-down junctions have permanent excess hiss.
> depressingly little difference,
Yes. The real hiss limit is the 22k NFB resistors around the opamp. For mike-amp gain less than 100 the VR1 value will be greater than the mike impedance and NF goes away. Even at higher gain there's just a lot of dumb resistance. "Better" mike amps make R8 R9 10k or 5k, even 2k or 680r. This requires VR1 scaled down in proportion (a hopeless part-mechanics problem on a low-price mixer).
IF you could get a 2.5k reverse-taper pot to fit VR1's spot, made R8 R9 4.7k, resistor hiss voltage would approach half, a real improvement. C7 has to be much larger (to 6,800u in some desks). However max gain will be a little lower and THD at maximum gain significantly higher.
This is a $1 mike input. There is a "$10 mike preamp" around a TI chip which is much better, using multiple opamps and lower impedance. But not a drop-in for a budget console.
"Compensation" is not an issue. There's no overall NFB around both transistors and opamp. The opamp works at gain of 5 and is surely stable. The transistors work at variable gain (VR1) and are too simple to be unstable. (The flip side is THD.)
If you are close-miccing singers and amplifiers, this class of input serves well at a very popular price. I fear Dotneck's sessions are more subtle, and this is not the ultimate gear for such work.
> It would be marginal at least.
The peak transistor voltage is the idle voltage plus half the peak mike signal. The stock plan idles at 10V. The minimum gain is about 5, with <15V peak out, so 3V peak input before clipping, or 1.5V each transistor. Looks like 12V rating is ample as long as you do not clip it. If it has been *severely* overdriven, the broke-down junctions have permanent excess hiss.
> depressingly little difference,
Yes. The real hiss limit is the 22k NFB resistors around the opamp. For mike-amp gain less than 100 the VR1 value will be greater than the mike impedance and NF goes away. Even at higher gain there's just a lot of dumb resistance. "Better" mike amps make R8 R9 10k or 5k, even 2k or 680r. This requires VR1 scaled down in proportion (a hopeless part-mechanics problem on a low-price mixer).
IF you could get a 2.5k reverse-taper pot to fit VR1's spot, made R8 R9 4.7k, resistor hiss voltage would approach half, a real improvement. C7 has to be much larger (to 6,800u in some desks). However max gain will be a little lower and THD at maximum gain significantly higher.
This is a $1 mike input. There is a "$10 mike preamp" around a TI chip which is much better, using multiple opamps and lower impedance. But not a drop-in for a budget console.
"Compensation" is not an issue. There's no overall NFB around both transistors and opamp. The opamp works at gain of 5 and is surely stable. The transistors work at variable gain (VR1) and are too simple to be unstable. (The flip side is THD.)
If you are close-miccing singers and amplifiers, this class of input serves well at a very popular price. I fear Dotneck's sessions are more subtle, and this is not the ultimate gear for such work.
Yes, both of those are much more robust than the ZTX718; But the 2SB1260 is surface-mount only, though its output capacitance is only 25pF. The ZTX550 doesn't specify output capacitance. I am guessing that the ZTX951's output capacitance of 74 pF is why you don't recommend it---eh?
You have a good point there, and I really don't think the faster speed of the OPA1656 will offer much REAL benefit. I would LOVE to use OPA2134s (as they are available in PDIP), but they've gotten SO expensive (almost $5 apiece!) that they are outta my league. One or two, no big deal. But, 28 of them---yeah, that's a big deal (to my cheapskate budget). So the OPA1642 @ $2.08 each looks like the best bet. Also, I just noticed that its quiescent current (1.8mA) is almost the same as the stock TL072 (1.4mA), and WAY less than the 1656's 3.9mA; therefore avoiding any current-draw problems with the power supply.
Last edited:
Quote from another diyer about mic transistors:
Seems this may be the way to go after all. 1-2 db per channel can add up.
It's not so much a limit to ultimate noise floor but rather dynamic range. By the time output noise is dominated by input transistor noise, you are near the upper end of the gain setting range (>46 dB, so maybe the top 8-10 dB). So don't be shy with the gain control! I would aim for maybe 16 dBu peak output level for the mic pre. Maybe you'll have to rethink your gain staging a bit.
I would love to plot effective input noise vs. small signal gain in LTspice, and I bet there is some way of accomplishing it... probably a job for a .MEAS statement.
It is difficult to do much about this here, given that the 4k7 resistors are part of a pack of 4 matched ones. Much less than 22k would also have been unwise for the TL072.
It's basically just a transistor V/I into an opamp I/V stage, isn't it?
Mind you, BC560Cs are not exactly known for their ultra-low Rbb' to begin with. Bob Cordell's model shows 170 ohms (vs. 38 for 2N4403, Horowitz/Hill even list that with 17 ohms), and with cheapies you may apparently get many hundreds of ohms (800-900?). Their main strength is high beta even down in the 10s of µA. That's another definition of "low noise" (low current noise when dealing with high-impedance sources). For a mic pre, even a 2N3906 (33/40 ohms) is a substantially better choice.
I am pretty sure they knew what they were doing when picking the PN4355, which does not appear anywhere else in the mixer (even though e.g. BC560Cs do!). Isn't it supposed to be a Douglas Self design anyway?
As PRR already indicated, you don't have to turn down input gain very far for input transistor noise to become less and less relevant. 10 dB below max, and ZTX951 vs. 2N4403 vs. 2N3906 becomes essentially a non-issue. 16-20 dB below max, and (better) BC560Cs would no longer distinguish themselves.
Do you mean the input-referred noise?
In this case, right-click on the graph, and select add trace and then V(inoise) (provided you have selected the correct source as input):
Attachments
Last edited:
Oh, I never noticed that you could plot more than just V(onoise)! Cool. Gotta remember that. So far I've always divided V(onoise) by small-signal gain at a defined frequency by hand... All doable but cumbersome - V(inoise) seems decidedly more convenient. Thanks, Elvee.
What I meant was:
* .STEP gain pot
* determine set of AC gains
* determine set of V(inoise)
* plot V(inoise) over AC gain for a particular frequency
Which I'm sure is doable.
And then ideally all of that for several sets of fixed resistor values (via .STEP + .PARAM TABLE, hopefully).
I'll have to look into it later.
What I meant was:
* .STEP gain pot
* determine set of AC gains
* determine set of V(inoise)
* plot V(inoise) over AC gain for a particular frequency
Which I'm sure is doable.
And then ideally all of that for several sets of fixed resistor values (via .STEP + .PARAM TABLE, hopefully).
I'll have to look into it later.
Well, it appears that the best solution for me is the "Complementary Feedback Pair" documented here:http://www.thatcorp.com/datashts/AES129_Designing_Mic_Preamps.pdf
They show an example (page 27) of a "simple mic preamp" which is almost EXACTLY what is in my Soundcraft 200 console. Then they show an improvement---the CFP, which offers ~5db less noise and >20 db less distortion; it also will fit nicely in the existing space and is inexpensive to do. The only question remains---what transistors to use? A lot of the older low-noise transistors are no longer available (2SA1085, 2SA1316); I have seen recommendation of the KSA992FB/KSC1854FTA. Suggestions?
They show an example (page 27) of a "simple mic preamp" which is almost EXACTLY what is in my Soundcraft 200 console. Then they show an improvement---the CFP, which offers ~5db less noise and >20 db less distortion; it also will fit nicely in the existing space and is inexpensive to do. The only question remains---what transistors to use? A lot of the older low-noise transistors are no longer available (2SA1085, 2SA1316); I have seen recommendation of the KSA992FB/KSC1854FTA. Suggestions?
I had considered suggesting a CFP input (à la ESP P66), but simulated distortion performance turned out exactly the same. Mind you, I did leave tail current as-is (as the 4k7s are not easily replaced), while in the linked presentation it is increased 6-fold. That is the main advantage of a CFP, being able to run more input stage current without a corresponding input current noise penalty.
ZTX790A inevitably is quite a chunky part still - Cobo = 24 pF, Cibo = 225 pF. Rbb' of 5 ohms still is well into diminishing returns territory for a 150+ ohm microphone application, too. No doubt, it certainly makes a far better 2SB737 substitution candidate than the ZTX951.
If CFP, I might go with 2N4403 + 2N3904 or BC550C or something, maybe 2SC1815 at >5 mA.
ZTX790A inevitably is quite a chunky part still - Cobo = 24 pF, Cibo = 225 pF. Rbb' of 5 ohms still is well into diminishing returns territory for a 150+ ohm microphone application, too. No doubt, it certainly makes a far better 2SB737 substitution candidate than the ZTX951.
If CFP, I might go with 2N4403 + 2N3904 or BC550C or something, maybe 2SC1815 at >5 mA.
Still not understanding why you prefer the 2SC1815 over the ZTX790A. Datasheet for the 2SC1815 does not show Cobo, Cibo, or Rbb'. The '-BL' version of the 1815 does indicate a high hFE of 350-700 @ 2mA Ic. ZTX790A shows hFE of 300-800@ 10mA. 2N4403 shows hFE of 100 @ 10mA. 2SB737 is, of course, unobtanium. Self wrote "Nobody now would consider devices like the 2N4403' Other low-noise transistors mentioned that are unobtanium are 2SA1084 and 2SA1316.
Last edited:
For a differential stage you need to match Vbe for low offset, and gains for symmetrical swing (which only affects headroom in a true differential stage). Only the actual input transistors matter for Vbe in the CFP config, but the second transistors also matter for gain.
Gain varies with magnitude of collector current so you need to measure/match at the right current levels if possible. Vbe matching is more consistent across the decades of current.
Gain varies with magnitude of collector current so you need to measure/match at the right current levels if possible. Vbe matching is more consistent across the decades of current.
I meant 2N4403 + 2SC1815 in case that wasn't obvious. The second transistor in the CFP isn't too critical, you just need to make sure that one is considerably faster than the other under the given operating conditions to avoid ending up with a double pole and associated response peaking. If in doubt you can always give the second one a few pF worth of Miller cap.
BTW, 2SC1815 (at least the Toshiba original) does actually have an Rbb' spec - 50 ohms typ. For the 2SA1015 pnp it's 30 ohms typ - no worse than a 2N3906 and still adequate for a mic input. These high-voltage, medium beta parts tend not to have that much Rbb' in general. (I still wonder how Behringer managed to bugger up the Q1002USB mic input to the point where it's at roughly -126 dBu shorted. Either their super special mic input transistors have pretty high Rbb' or it's actually power supply noise thanks to some rather dried-out dodgy Chinese caps and/or thin copper, much like what I'm suspecting for the Soundcraft 200. Resistor matching may not be of the highest standard either.)
Matching BJTs usually isn't a major priority since (within the same batch) their Vbe tends to be matched fairly well to begin with. If you're picky about this you are probably better off making sure they are bonded thermally. (-2 mV/K!) Otherwise output DC offset has a tendency to drift with the slightest bit of air movement over the circuit. Other than that it's important to keep your currents matched (resistor matching).
If you do want to make a CFP you are probably best off putting it all together outside after taking measurements. I assume this would make epoxying the input transistors a bit easier. Or perhaps solder up each side, epoxy input transistors and fit the whole shebang while the epoxy is still malleable.
I have now revisited my simulations, and lo and behold, the CFP configuration plus increased current does in fact yield a substantial reduction in distortion, in line with the THAT Corp. presentation. With 680 ohm resistors in the CFP and 7k5 instead of the 15ks it's about 7 dB down at highest gain, dropping to about 28 dB down at moderate gains - quite significant. You could still try pushing you luck by going a bit higher with the current, but 6k2 seems to be about the limit before performance at high input starts degrading. I'd say hand-match some 6k8 metal films and be done with it.
BTW, 2SC1815 (at least the Toshiba original) does actually have an Rbb' spec - 50 ohms typ. For the 2SA1015 pnp it's 30 ohms typ - no worse than a 2N3906 and still adequate for a mic input. These high-voltage, medium beta parts tend not to have that much Rbb' in general. (I still wonder how Behringer managed to bugger up the Q1002USB mic input to the point where it's at roughly -126 dBu shorted. Either their super special mic input transistors have pretty high Rbb' or it's actually power supply noise thanks to some rather dried-out dodgy Chinese caps and/or thin copper, much like what I'm suspecting for the Soundcraft 200. Resistor matching may not be of the highest standard either.)
Matching BJTs usually isn't a major priority since (within the same batch) their Vbe tends to be matched fairly well to begin with. If you're picky about this you are probably better off making sure they are bonded thermally. (-2 mV/K!) Otherwise output DC offset has a tendency to drift with the slightest bit of air movement over the circuit. Other than that it's important to keep your currents matched (resistor matching).
If you do want to make a CFP you are probably best off putting it all together outside after taking measurements. I assume this would make epoxying the input transistors a bit easier. Or perhaps solder up each side, epoxy input transistors and fit the whole shebang while the epoxy is still malleable.
I have now revisited my simulations, and lo and behold, the CFP configuration plus increased current does in fact yield a substantial reduction in distortion, in line with the THAT Corp. presentation. With 680 ohm resistors in the CFP and 7k5 instead of the 15ks it's about 7 dB down at highest gain, dropping to about 28 dB down at moderate gains - quite significant. You could still try pushing you luck by going a bit higher with the current, but 6k2 seems to be about the limit before performance at high input starts degrading. I'd say hand-match some 6k8 metal films and be done with it.
Last edited:
A study in resistor matching and shared grounds. Looks severe but it seems noise level would be constant irrespective of gain, and ultimately just reducing maximum dynamic range somewhat. We're still talking a few µV of output noise even with a relatively noisy supply, nothing super significant.