Purpose of precision high-current series regulator's resistors and capacitor?

I'm sorry, which zip file is that? The old one?

Excuse me, 0.08nV is too noisy? You have a low Zout.

So, trusting your SPICE numbers, you haven't actually measured your real circuit on the bench?

You've argued that the high intrinsic noise of the 2n3904 is somehow overcome by placing it in a feedback loop with a noisier element. No fair adding a massive output cap.

OK, I have been using SPICE for 45 years, on many hundreds of circuits, and have found hundreds of faulty results. I mostly use SPICE for careful analysis of a small portion of a difficult analytically-evaluated circuit, using my own vetted models. Maybe because I'm evaluating a difficult aspect, often non-linear, it's easy to see the serious weaknesses that are there. Many models are very poor.

It's not SPICE that's at fault, per se, but the models. In these instances I've embarked on a call to urge folks to vet models on the bench, before trusting a SPICE result. I haven't seen you do that. Hey, I'm sorry, the bench rules.

Build your circuit, get out your spectrum analyzer, take the measurements, and report back. SPICE plots don't count.

Let's make the "teaching grandmothers to suck eggs" point extremely explicitly...

In a response to a post from *Win Hill*, Steve Wilson wrote: "After 40 years running SPICE, it is more likely the results are highly accurate, and offer performance you could never dream of. Hundreds of thousands of engineers have been taught SPICE in colleges and universities. How did you miss out?"

Mr Wilson might benefit from reading, amongst other things,

But, a warning: SPICE takes its models literally, and therefore goes seriously off the rails when the models are inaccurate. We saw this here, with the poor generic QN5088 model we tried first. Similar problems are found with MOSFET models, many of which fail completely in

Author? Win Hill, of course.

The ATF38143 pHEMT model I posted awhile back overestimates the 1/f noise b y _seventeen_orders_of_magnitude_.

Cheers

Phil Hobbs

Yes. The mission has changed and the file has been updated.

Re = 27 / Ie(Ma) = 2.7 Ohms

How do you measure between 20pV and 40pV/root Hz?

I switched to a 2N4401. The LT431 controls the output voltage. There is a 3-pole lowpass RC filter between the TL431 and the transistor. The frequency response of the filter is shown in the file.

I now revise that down to 35 years.

Most likely 35 years, same as me.

Intusoft was founded in March 1985:

It is interesting that you barely mention SPICE in AOE2, and waited for AOE3 to start using it.

The SPICE simulation is essential to tell you if a circuit is worth spending time on. It is invaluable to help troubleshooting when something goes wrong on the bench. It allows you to evaluate different component parameters and see which ones are critical to obtaining the desired results. It allows you to optimize the circuit and components for best performance. You can examine the open-loop response in a feedback circuit and optimize the loop gain, and gain and phase margin. This can be difficult or impossible to do on the bench.

It can be difficult and time-consuming to change components on the bench. You may have to fight noise, poor connections, unwanted oscillations, test equipment limitations, bad components, and a host of other problems that interfere with your evaluation. You may lack the equipment needed to make critical measurements that you can easily do in simulation.

In fact, in AOE2, you state:

"SPICE (originated by L. W. Nagel) models the operation of your trial circuit (using a library of sophisticated component models), predicting the gain, distortion, noise, frequency response, etc. You can ask it to show you the waveforms (voltage and current) at any point in the circuit - a computerized oscilloscope! Thus you can fiddle with your hypothetical circuit, trying speedup capacitors, exploring component and circuit trade-offs, and so on."

"In fact, good simulators even let you see the effects of component tolerance, via either conventional worst-case analysis or a more sophisticated "Monte Carlo" statistical treatment. Some modeling programs also perform a "sensitivity" analysis, telling you which components affect performance the most."

LTspice rules.

My HP 8566 spectrum analyzer is geared for high microwave frequencies up to 22GHz. It only has a dynamic range of ~100 dB, and is unlikely to be able to measure such low noise.

You state:

"SPICE (originated by L. W. Nagel) models the operation of your trial circuit (using a library of sophisticated component models), predicting the gain, distortion, NOISE, frequency response, etc."

The e_n noise in a BJT is not given by r_e, but instead by the base r'bb, which is much higher. Read pages 481 to 486, and study Figure 8.12.

We all recognize that your filter can present a low noise to the base, but your chosen noisy transistors will determine the output noise.

Better go to the bench. You'll get 1.35 nV/rt-Hz with a 2N3904, with an r'bb of about 110 ohms, and 0.84 nV with a 2N4401, both at 10mA, more or less, according to our measurements. If you chose some of the other parts in Table 8.1, you can improve on that by up to a factor of four to 0.2 nV. Our 80pV preamp can test all of them in your circuit. With G=50 to 100, you won't need a quiet analyzer.

I began using SPICE about 1972, when it was only a text program. Most engineers prefer to be able to calculate their circuit parameters, and know what's going on, and what's responsible for what, rather simply stab in some component values, and rely on an overall SPICE output.

AoE2 was completed in 1988, before manufacturers began routinely providing SPICE models.

AoE is all about circuit design, and SPICE was not then and is not now at the center of circuit design. SPICE was designed specifically for IC design, and fab houses work very hard to provide accurate models. As we learned from Jim Thompson, these are usually proprietary and tightly guarded. Good, but secret.

By contrast, discrete-component manufacturers provide horrible models. Yes you can often do the bench work, testing for specific parameters you care about, and create a good model, but short of that, watch out. Don't boast about your SPICE circuit's performance, until you measure it on the bench.

You asked about the output impedance. I was not using that to calculate noise.

As mentioned, I am limited by the transistors that are available in both LTspice IV and XVII. As mentioned, you are free to select better ones.

The output noise is determined by the 1,000 uF bypass capacitor from the output to ground, not by the emitter follower. TL431A and TL431B should make that clear.

IC manufacturers have much better programs for IC design.

LTspice is specifically designed for circuit analysis, and is the preferred method over pages of equations. How would you analyze my circuits by hand?

The 64,000 members of the LTspice group should make this clear.

I've lost count of how many times Spice has given me kilovolt results for little low voltage circuits with no inductor.

NT

And just now we have a SPICE results thread about a 2GHz op-amp buffer output stage using a 2N3904 and 2N3906, ha haw! And this from an article by Dr. Sergio Franco, who should know better! One can save themselves considerable embarrassment with a little common-sense analysis, and/or a bench test, before broadcasting their SPICE results.

Paul just returned from a grand Peru vacation, lucky guy, and has completed measurements on Steve's circuit. Steve said, "450 pV/root Hz at 10 Hz, and 15 pV/root Hz at 10 kHz and beyond." But as we expected, given a 2N3904's noise, e_n was 1.2 to 1.4 nV/rt-Hz. Perhaps Helmut's TL431 model lacked any noise. Thus his SPICE version of a noisy IC, was able to quiet a noisy BJT by a factor of nearly 100. Hah, in reality, an RC filter, plus the BJT follower, can slightly quiet the noisy TL431, not the other way around. Again, my advice is to not trust your SPICE model until it's been bench vetted.