> I love yahoo. The link works for like 10 minutes, and then dies.
Yahoo does not like you linking from other pages (bypassing their ads), and is liable to suspend the account if traffic is high (and a post on this forum can make a lot of traffic in an hour).
bc> I don't know how good the 990 is in terms of initial offset and drift.
It is an LM394, with a current-balance load. Offset is quite low. Not zero in the older forms: the few-Vbe difference in collector voltages induces offset, so recent models from some builders even trim this out. The leading-edge 990 comes darn close to the LM394 datasheet.
> this design should yield 70 dB of gain, (including 10 dB in step-up
Why? What are you recording, with what, that can sensibly use 70dB of gain?
Your output noise level will be up around -58dBu, FAR-FAR above the noise of any recorder. You can record with less gain and fix it in the mix. You don't have to pin the meters.
And 60dB of amplifier gain, without a DC block cap, assuming 1mV DC offset, means one VOLT of DC at your output transformer, which will probably saturate it. (990 should be less than 1mV, but still "a lot" after 60dB gain. The trim pot can make it lower, but it will drift away from zero.)
To see what your plan is missing: take off the DC offset trim (frill), replace the $$$ 990 with a $0.15 LM741, and look in any basic op-amp text for "non-inverting amplifier". You need a resistor from output to in- to give feedback, and a resistor from in- to ground to set gain.
(Actually, as shown, IF you could get the output anywhere near zero DC volts, the gain would tend to be about 3dB because of those 220pFd+120pFd caps. Which make no sense except in context of resistances you have not specified yet.)
Using 741 values: the feedback resistor generally has to be more than 2K, because the op-amp has to drive it and a 741 can't drive lower loads. The resistor from in- to ground "should", for low noise, be less than the source resistance, about 2K for a 200Ω mike and a 1:3 iron. However if we will want high AC gain, we probably will need a DC block cap, and the resistor from in- to ground should not be so low that we need an impossible cap value to get 20Hz response.
Take 2K feedback. For 60dB or 1:1,000 gain, then the resistor from in- to ground is 2Ω. This will work, until you calculate over 5,000uFd DC-block cap in series with the 2Ω. This usually has to be non-polar, which rules out common electrolytic caps. Now, you can get a 6,800uFd NP cap from Panasonic and DigiKey, but it is quite a few dollars and VERY big for a low-noise low-impedance circuit. Unhappy answer.
If you only want the 60dB gain, you set the resistor from in- to ground to about 500Ω, the cap can be a few-dozen uFd NP. Now the feedback resistor is 500K. This will almost work. The R and C values are workable. The LM741 DC bias current in 500K will be a problem, but a TL071 will work with much lower bias current.
I'm not serious about LM741 (anyway +/-24V will kill it soon). If you look up the bandwidth of a 741, it is 1MHz at gain of 1, 100KHz at gain of 10, and 1KHz at gain of 1,000. Or gain will be 60dB from 20Hz to 100Hz, -1dB at 500Hz, -3dB at 1KHz, -7dB at 2KHz, -20dB at 10KHz. Sounds like the Treble control is stuck on zero. The TL071 is about 6 times better: -3dB at 6KHz, still kinda muffled.
And even before the response droops, the feedback factor has fallen to "almost none" and all the 741's/071's distortion appears without correction. Not nice.
The 990 IS better. But not a whole heap better. It can be flat to around 20KHz at gain of 1,000, and indeed it was sometimes used this way in the old days. But it is starting to dropp, and the THD is starting to rise. It is usable, but frankly a couple halves of 5532 in cascade will gibe 30dB+30dB of gain with a lot less strain (and 1/50th the cost).
OK. I've used things a lot worse than 990 running 60dB gain and cashed the check for the gig. And I don't sense that you want to start on a dual-amp project today. So let's go with your plan.
Making the shunt resistor 500Ω as before, the feedback resistor is again 500K. Problem: the 990 has huge bias currents and you will get output offset around 0.5V, even with a DC-block cap. Ignore that a moment. You can change gain by reducing the feedback resistor. In fact with 990 you can reduce the feedback to zero (almost) because a 990 will happily drive the 500Ω shunt resistor. The gain-resistor should be a normal audio-taper pot, wired as a rheostat. One problem with this is that, if the pot wiper loses contact, gain rises to infinity and blatts your speaker cones across the room. (Yet I have used such systems without much problem.)
The alternative is to fix the feedback resistor and vary the shunt resistor. If the feedback R is 500K (which has problems), then the shunt resistor is 500Ω for Gv=60dB, 5K for 40dB, 50K for ~20dB (30dB with iron-gain). This gives low noise at high gain, but at 40dB the shunt resistor's thermal noise is twice as high as the mike self-noise, and horrible at 20dB. If we make the shunt resistor 5Ω to 500Ω, with 5K feedback, the shunt resistor never has large impact on noise, but we do need a cap that is about 5Ω at 20Hz, or 2,000+uFd NP. In this case, the shunt resistor is a rheostat-wired REV audio-taper.
So when all is considered, you did order a good pot. 10K Rev Audio is 1K at mid-turn and can be set about 50Ω-100Ω at the high end. Make the feedback 5K or 10K. Put 100Ω in series with the pot so the gain does not go infinite at the end. The capacitor has to be <100Ω at 20Hz or about 100uFd. That gives -3dB at 20Hz at maximum gain, but as gain is reduced the bass cut-off decreases, to 0.02Hz at 20dB gain. And there is almost no place where it makes sense to have 70dB total gain all the way to 20Hz: the world is full of rumble and 1/f noise. 20Hz at max gain is plenty. (And who records 20Hz? I have some true 32Hz organ tones on disk, using modest-output mikes, and
only needed around 48dB gain working at the far end of the church.)
While the 100uFd NP cap is a tolerable size, we might like to eliminate it. I will state without proof that the output transformer does not want more than around 20mv of DC on its primary. Without the cap, at 60dB gain, that says 20 MICROvolt input DC error. Now, we can do a lot of math. The offset of LM394 is 50uV typical, 200uV max, if everything else is test-lab perfect. Thermal drift and long-term drift are actually quite low: if you cook it for a week, trim to zero, and keep it in a studio you might never see more than 5uV of offset. The bias-current error might also be about that size. But in real life, not test-bench, you get microvolts of thermocouple effect in joints and resistors. I've messed with a few precision amps, and I would not expect to hold the output offset to 20mV in a real-world project, not without some major add-on (more than a trim-pot).
I note that that offset trim seems very sensitive to DC supply voltage un-balance. Was it designed for a super-good supply regulator?
I'd be real inclined to put a DC-block cap in front of the output iron. 100uFd NP there would for-sure keep the iron from being drenched in DC. Then a few-Volt amplifier DC error would just be a trivial reduction in maximum output (which is already too high IMHO). Add 50Ω series resistor to kill the Q, and 1uFd-5uFd film-cap across the NP electrolytic to take some of the curse off of it.
