c-multiplier, real life

I think the results line second to the bottom is supposed to be 170uV rather than 170mV. :-)

Right. Typo. The dB rejection is right.

The numbers here aren't far from the 2N3904 LT Spice sim I posted a week or two ago. The actual BCX70 is a bit worse than the 2N3904 sim.

There must be a way to convert these numbers into Early voltage, but my brain is used up for the day. I barely have the energy left to stagger home and stir up a rum+coke.

The Tek AM502 and a digital scope, signal averaging, has a nice signal-recovery capability. This would easily resolve a 1 nV signal.

John

Not exactly. The depletion width modulation from the Early effect acts as a conductance from collector to emitter. The base current and voltage are not altered, and the shielding provided by the base region has no effect.

This means the transistor collector-emitter can be modeled as a resistor in parallel with a capacitor.

In order to get substantial ripple reduction, hang a large electrolytic from emitter to ground. Phil uses a 10uF ceramic. I recommend using the

The scope probe will not be sufficient to measure the ripple. The ground lead has enough inductance to pick up all kinds of noise radiated from the equipment and coax cables.

This measurement will need a coax connector the same as the ones you are using, with a very short coax to the preamp.

The AM 502 has 25uV noise. If you are planning on measuring 25nV signals, it will require (25e-6/25e-9)^2 averages, or one million. Since you want to find ripple much lower than that, it will take correspondingly greater averaging.

The liklihood of drift during the averaging is very high, which will wipe out the results. So your equipment will limit you to a minimum detectable signal level, perhaps in the region of 250nV.

I find the leads in your layout are quite long. These will radiate signals and act as antenna. Also, soldering the coax connectors along the edge of the pcb means they will pick up the noise currents that are forced to flow along the edge of the pcb due to skin effect. This is surprisingly effective even at fairly low frequencies, say in the tens of KHz.

A better arrangement would be to solder the connectors directly to the copper near the signal. You might be able to bend the legs on the existing ones enough to tilt them up so the coax can be screwed on. Failing that, there are coax connectors with legs that can be soldered vertically to the copper. Or drill a hole and use a bulkhead connector.

When you start reaching decent ripple attenuation, radiation from the coax shields will start limiting the results. You will need better coax cables with 100% shielding. Or go to hardline.

Another problem is the reference voltage driving the base. When you finish making the ripple measurements, you need to find a way to supply the base with well-filtered voltage from the same supply as the collector. This will give an indication of the overall performance of the ripple filter.

The filter in the base circuit will require farily large series resistance, which will give additional voltage drop that is dependant on load current, beta, temperature, and the phase of the moon. This is probably why Phil went with a MPSA14 darlington.

What you are trying to do is not trivial. Most people end up with a shielded box, low noise preamplifiers, and battery operation.

Anyway, good luck.

Mike

Yay, you re-invented the capacitance multiplier :-)

What's a fun generator? I want one of those ...

Sorry, this is not very clear. The collector-emitter capacitance may be quite low, perhaps 400nF. This means the feedthrough will be small until you get up to 10KHz or so.

Please see what it took Jim Williams to do a 775 Nanovolt Noise Measurement in AN124:

Shielded box, low noise preamplifiers, and battery operation.

And that's only 775nV. I believe you wanted to see down to several nV.

I know that the c-mult works at high frequencies. What I wanted to measure is how it works at low frequencies. Re is pretty low, so the output pole is in the KHz range if you use a reasonable cap. Plus Re and the ESR make a divider. 3300 uF wouldn't fit inside my current product. I'm using a 120 uF polymer aluminum cap and some ceramics.

The scope probe is measuring the input ripple, 200 mv p-p. Collector lead. It works fine for that. And I'm signal averaging 64:1 on both scope channels anyhow.

The emitter output is via coax, to the AM502.

As you can see, the signals are pretty big. They don't even need averaging... it just makes them prettier.

Wild overkill at 400 Hz!

Not at 400 Hz!

That's calculable. It's the transistor I'm measuring here. I wanted to see if the LT SPice models were in the ballpark. Looks like they probably are. There was some conjecture in a previous thread that teit Early voltages were unrealistically low.

A volt or three of Vce seems to improve rejection. So a resistor from base to ground is good, if you can waste the voltage.

I think these numbers are good; 140 dB would be tricky, but 66 ain't. But if anybody wants to reproduce them, I'd be delighted.

If I have any Darlingtons around, I'll try one of them. Any predictions?

John

Mostly overkill for what's almost a microvolt! I wonder why he used differential jfets. That's just throwing away 3 dB of noise performance.

I don't need nV for the c-multiplier here. But nanovolts are easy if you do can do narrowband tuning or synchronous detection. JW was trying to measure noise.

This Rigol scope will do digital bandpass filtering, signal averaging, and FFTs. I wonder if you could turn them all on at the same time. I suppose ADC overload would limit how far you could push that.

At low source impedances, like this situation, there are opamps like the Lt1028 that have noise below 1 nv/rthz.

John

Sorry, your previous posts said you needed nanovolt-level noise levels for the new circuit, and that it would be easy to do averaging and get down to

But you did not mention your goals had changed for this measurement.

Mike

Nanovolts are never easy. Again, your goals seem to have changed, but you do not mention the new ones.

Narrowband tuning or synchronous detection constrains you to sine waves. Even then, you need good low-level preamplifiers. Your AM502 won't be much good down to nanovolt levels.

What is the input noise level? That pretty much determines everything else.

When you get into the millions or hundreds of millions of averages, the system will likely drift during the measurement. This will render the results unusable. So you will probably need a good low-noise preamp to boost the signal into the scope.

I posted this list earlier. "nV" stands for nV/sqrt(Hz). The maximum differential input voltage is shown, along with any minimum gain or maximum input current if applicable.

AD797 : 0.90nV +/-0.7V 25 mA AD8099 : 0.95nV +/-1.8V +/-10mA G >=2 ADA4898 : 0.90nV +/-1.5V LMH6624 : 0.92nV +/-1.2V LT1128 : 0.85nV +/-1.8V +/-25mA LT6200 : 0.95nV +/-0.7V +/-40mA OPA687 : 0.95nV +/-1.2V G >= 12 OPA847 : 0.85nV +/-1.2V G >= 12

Mike

He needs low dc drift.

When you get down to 1uV on your breadboard, do show us the scope photos.

Mike

I had to review the Analog Devices AD630 Balanced Modulator for verification, but synchronous detection will only give a DC output. This is of little use when you are trying to trace noise and need to view the actual waveforms.

Mike

On a sunny day (Wed, 02 Jun 2010 17:02:45 -0700) it happened John Larkin wrote in :

LM317 has > 60 Db ripple rejection? Why bother with all this?

When I tried a Darlington in a Cap-multiplier I found that it reduced the DC impedance of the filter. (As one would expect) but that there was more noise on the output. Something like 4nV/rtHz versus 1nV/rtHz with a 2N3904. I'm not sure what darlington I used... perhaps the MPSA14 or BC517. (Those are in my darlington parts drawer.)

George H.

My goal was to better characterize the c-multiplier. I did it.

If the post-detector is narrowband enough the AM502 if fine. If I heeded more selectivity than the scope's signal averaging can handle, I could drive a spectrum analyzer or a selective voltmeter.

But it's academic. I'm seeing signals at the emitter in the 100 uV and up range. I have tons of signal, not that tons of signal is what you want from a noise filtering circuit.

John

His entire circuit is AC coupled.

John

You keep saying that. What I said on the subject is exactly

"I don't need nV for the c-multiplier here"

My goal here was to characterize the low-frequency performance of a transistor used as a c-multiplier. And I did it, in real life, not simulated.

As far as the product we're designing goes, we do need nV front-end sensitivity. That will measure itself.

I am considering doing an AM502-like box, but much lower noise. 0.5 nV/rthz might be good.

John

It has gobs of output noise.

John

I'd expect the darlington to be worse, but not by that much!

Can you explain what you mean by "reduced the DC impedance of the filter"?

How did you measure the noise?

Mike

He is trying to measure very low amplitudes at low frequencies. A single- ended stage would have tens or hundreds of microvolts per degree C shift in the dc level and make it virtually impossible to keep the signal on screen. Differential inputs can give less than a microvolt without chopping.

All the low-noise op amps have differential inputs. They still can give less noise than a 50 ohm resistor.

Mike