[tangent]

I dunno how many people this will interest, but here's the new measurement preamp I'm working on. (Schematic.) It's primarily intended for testing linear-regulated AC power supply noise, which on a decent good linear supply is way below the noise floor and/or resolution range of typical hobbyist measurement equipment. In general, it's a useful addition to a digital oscilloscope or a PC spectrum analysis setup.

This preamp amplifies the AC part of its input by 40 or 60dB (switchable) while adding minimal noise. There are two single-pole passive high-pass filters and one two-pole active filter to give very steep DC and low-frequency suppression. The filters have fcs in the 2-10Hz range, as I recall.

There are three amplification stages. The first stage is a simple 10x inverting stage. The second is an adjustable 8-12x inverting stage; that range is necessary to be able to trim out the amplifier to precisely 100x or 1000x gain even with worst-case values of 1% resistors. The last stage is a switchable 10x/1x noninverting gain stage as well as an active filter; this stage uses a big, fast CFB op-amp made for cable driving.

All the resistor values are low for low noise.

The amp's power source is 8xAAA NiMH cells, which sit in two back-to-back 4xAAA holders in the cutout in the corner of the board. (See the layout image.) There is a built-in compacted version of the PPA battery board charger. The power switching setup is made so the amp is charging the batteries from the wall in one switch setting, and running the amp from the batteries in the other, since this amp should only be run on batteries to avoid ground loops disturbing the test. Using NiMH cells lets us pull a fair bit of current (even 0.25A is easy to maintain) while keeping good circuit performance and not making the circuit expensive to run.

There is a simple TLE2426-based virtual ground circuit. The current should be quite sufficient despite the low resistor values, since the signals will be so small. The low supply voltage ensures this: between that and the headroom limits of the op-amps, this amp won't produce more than about 2Vrms. That runs about 20mA to ground through the output resistor. I've half-considered center-tapping the battery pack instead, but that complicates the power switching setup; you end up needing a DPDT as well as an SPDT to get the charging-or-amping switching.

This whole thing fits into a Hammond 1455L1201 case -- the shorter version of the 1455N12. This one is harder to get than the 1455N12, but I don't think it's worth using that one since the extra space would only be useful for more batteries and a higher supply voltage would only allow for a higher output signal, and that isn't helpful or useful in this application.

Thoughts, comments, complaints?

 

[morsel]

I understand the appeal of capacitor coupled input in this application, since we are interested only in the noise, not the DC level of the power supply, but why the choice of high value electrolytic coupling capacitors, why so many of them, and why couple the output?

 

[aos]

Typical noise measurements are done within a determined frequency range, so that you can get meaningful, comparable numbers. Typically it is 10Hz - 100kHz. While noise is typically larger at low frequencies (as some types are dominant there), since you are using wide bandwidth opamps, you might be adding substantial amount from over 100kHz. You should consider having a low pass filter set at 100kHz, even if just a single pole.

 

[tangent]

Quote:

why the choice of high value electrolytic coupling capacitors

Because the fc must be quite low, and the following R is 220 ohms. If I lowered the C, I'd have to raise the R to keep the same fc, and that would raise the amp noise. That's really a bad idea in a 1000x gain amp, especially in the first stage. I'd make the input impedance of the amp 50 ohms if I thought I could get away with it.

You didn't ask, but I'll explain why the output cap is large, too. The LT1206 is a high-speed CFB op-amp, and it gives the flattest gain vs. bandwidth curve when the feedback resistance is a specific value based on a particular load on the op-amp. (See the table on page 4 of the datasheet.) I chose the 150 ohm output load since that is the highest value given in the table, and it lets me use the lowest gain resistor values. Just like on the input, since R is low, C must be high to keep fc low.

Quote:

why so many of them

There are only three. One across the rails, and two in RC filters. All the other polarized caps are tantalums, since they're doing bypass on high-speed op-amps.

Quote:

why couple the output?

a) it gives one more pole for a sharper LF falloff (and it's almost free)

b) all of the op-amps are bipolar input, but I haven't tried to balance input impedances, and I don't have DC trimming on any of the stages. This final filter strips any DC offsets picked up through the amp stages.

Quote:

You should consider having a low pass filter set at 100kHz

I considered that, but I left it out for a number of reasons:

a) single pole isn't sufficient*, so I'd have to use an active high-order filter and I don't have the room for that or the desire to go with a bigger board or a smaller battery pack to make room

b) besides my DMM, my primary measurement tool will be a sound card feeding spectrum analysis software, which is limited to 96kHz already. (48kHz, really...Nyquist)

c) I wanted a general-purpose preamp for use with an oscilloscope.

d) the typical low HF content of linear regulator noise

If I decide I need a low-pass filter, I can add it after the fact, in a separate box.

* If I went with a single pole low-pass filter, that would only drop HF by 9dB at 1MHz, assuming the -3dB point was 100kHz. That isn't helpful for a number of reasons. First, my meter seems to have a similar drop-off in AC mV performance already, so this curve doesn't swamp the meter's behavior (this is implied in the specs, not given). Second, the bandwidth of the preamp isn't much higher than 1MHz already, so only being 9dB down at 1MHz isn't a great help. Third, the idea of such a filter would be to be as close to a brick wall to HF as is practical, and this clearly is not.

 

[aos]

you don't really need a high order filter. Single pole will do fine, actually.
All your opamps are high bandwith except the last one which I don't know about in detail (what is 10nF cap for, if it is compensation then you're in effect already limiting bandwidth) so claiming only 1MHz bw seems very unlikely unless you layout sucks. Claiming that your measurement equipment does that already is fine, but if you want universal use you should add it instead of reying on chance. It may be as simple as an RC filter at the output (that cfb should have no trouble driving that).

Actually single pole filter is 20dB/decade so you should get a solid 20dB at 1MHz, not 9dB.

 

[tangent]

Quote:

All your opamps are high bandwith except the last one which I don't know about in detail

With a 500pF load and a gain of 2, I should be able to get about 25MHz out of it. I'm assuming a 470pF filter cap plus 30pF of cable capacitance. That requires a 3.3K resistor to get -3dB at 100kHz. The g=2 comes from a datasheet graph; I assume the bandwidth will be more like 5MHz at g=10.

Quote:

what is 10nF cap for, if it is compensation then you're in effect already limiting bandwidth

It's a compensation cap to limit a big HF peak when running the op-amp into a capacitive load. I had added it because I wasn't sure about the capacitance of the cable to the test equipment. This cap does reduce bandwidth, but not drastically. Total bandwidth, all things considered, should still be 2MHz or more.

Quote:

unless you layout sucks

I do welcome comments on it...

Quote:

if you want universal use you should add it instead of reying on chance.

That limits the device's usefulness to noise measurements. What if I want to amplify a 1MHz signal some day?

I'm not hard against this, just trying to decide whether I want generality or utility specific to the purpose I'm making the device for.

Quote:

single pole filter is 20dB/decade so you should get a solid 20dB at 1MHz, not 9dB.

Ooops, I confused that with 6dB/octave.

What do you think of making the -3dB point 100kHz? Would it be better to push it farther out, so that perhaps the filter is only 1dB or less at 100kHz? If it's at 100kHz, then it's 23dB down at 1MHz.

 

[tangent]

My 30pF assumption above bugged me, so I looked into it, and it's wrong but not important in practical use. To drop bandwidth by 2x over my estimate above, the cable capacitance would have to be more like 400pF, and even if that happens, I should still have over 1MHz of total amplifier bandwidth. The two cables I'm most likely to be using are 50 and 120pF.

 

[aos]

If your output amp is limited to 2MHz, then you probably will never look into amplifying 1MHz. Any amp is only really useful in the range where it's gain is flat. The thing is was the 2MHz point of unity gain (in which case your actual useful bandwidth is (far) less than that depending on the gain) or -3dB point, in which case you could potentially still use it for 1MHz amplification. However, in reality I'd think a simple output cap is probably just one part of the solution if one wants to limit noise to 100kHz - one would probably want to limit the bandwidth of all the preceding stages, by compensating the opamps or using lower bandwidth ones in the first place (the usual rule of the thumb is to never use more bandwidth than necessary, which some people deny in audio because it limits the slew rate, but in this case limiting bandwith might be a good idea). I wouldn't worry about amplifying 1MHz in the future though (what would it be anyway, video signal? ADSL?), you'd probably want to redesign at that point using some modern high speed opamp that has low distortion into high MHz, perhaps CFB all the way.

 

[aos]

Actually it looks like -3dB bandwidth of your LT1206 is about 40MHz from the datasheet for the values you selected (10 gain, etc.).

 

[tangent]

Quote:

one would probably want to limit the bandwidth of all the preceding stages

This is a really good suggestion. I've added a phase-lead cap on the first stage. I like this idea because with an output RC, this gives a 2-pole filter, which makes me a lot happier than a 1-pole filter.

EDIT: Schematic and layout have been updated, so reload to see the changes.

I can't do this on the second stage due to the adjustable feedback R; it would give an adjustable filter as well, which isn't a great idea. I can't do this on the output stage, since you can't use a feedback cap on a CFB op-amp. Anyway, the first stage is the important one.

I did consider putting more RC filters in between the op-amp stages, but either the C would be too high for the recommended op-amps (LT1028 and AD797) or the R would be too high for noise reasons.

The -3dB point is now at 100kHz. I decided that would be better than making 100kHz a -1dB or under point, since it's widely understood that "100kHz bandwidth" usually refers to the -3dB point.

Quote:

using lower bandwidth ones in the first place

If I make some of these boards for others, I'll mention the option of using an OPA228 or NE5534 in the first two stages as options. These are cheaper and have about half the bandwidth of the recommended chips, but still sufficient. If you have any other low-noise chip suggestions in this bandwidth range, let me know.

While I'm thinking of it, if anyone wants one of these boards, let me know. I'll be doing this for the cost of the boards only. I expect a very small run, so each one will probably be about $30-40.

Quote:

Actually it looks like -3dB bandwidth of your LT1206 is about 40MHz

I guess you're getting that from the table on page 4, but I think the graph in the top right of page 5 overrides this. With a load of 300-1000pF, bandwidth is in the 10-20MHz range at g=2, so divide by 5 for g=10.

 

[aos]

Actually, yes, using it on the first stage makes the most sense as you're limiting the input instead of amplifying it. Second stage can live without it. I'm not that sure anymore on that output filter that I originally mentioned - it should be there somewhere though the better place would be the on the other side of the cable. If that was in the same enclosure then it would be fine, but if you're driving a cable then it's no good - that's not a way to drive the cable for sure. It was a bad suggestion to make. So perhaps just stay with limiting the first and third stage and don't use that output filter. The right solution is probably to have bandwidth limited buffer on the receiving side, or R/C filter after that buffer.

As for the third stage, you would have to adjust the feedback resistor as well if you were to have bw drop to 10-20MHz. What that one tells you is how to get 0.5dB peaking with capacitive load. If you don't adjust the resistor, you'll get higher peaking instead, but the bandwidth will remain. With CFB you control the bandwidth with the feedback resistor (and do not ever use a feedback capacitor unless you wanna make an oscillator - a high power high frequency one too, not a good thing to have). If you increase resistor you'll cut the bandwidth. What you can also do is to introduce the SD resistor, which lets you cut the supply current by a rather large factor, also cutting the bandwidth. A win-win situation in this case, as well as if using it as a voltage dividier (I think I'm using it for that purpose myself).

 

[tangent]

Quote:

it should be there somewhere

I don't really see where a simple RC filter could fit. It can't be between the amp stages, because with C=100pF (the max the first two stages are happy driving), R would have to be 16K, way too high for good noise performance.

It can't go on the input, before the high-pass filter, either. If you put it as the first element, you make a voltage divider with the 100 ohm input impedance of the first stage. I don't see how to put it after the high-pass filter without screwing up the feedback loop.

And as you say, it's no good on the output.

Quote:

don't use that output filter.

I agree, the R in series makes it no good.

Quote:

If you increase resistor you'll cut the bandwidth

If I made Rf 2K, this should put the bandwidth in the high single-digit MHz range. But I'm inclined to leave Rf alone. Raising the resistor to 2K adds something like 60nV/rt.Hz of noise. (~6nV/rt. Hz for a 2K resistor, x10 since it's in the feedback loop.) That's about -100dB relative to full scale over the measurement bandwidth, but that's just one stage; noise adds up. And, measurements on the quieter power supplies will probably be -20 to -30dB below full scale, even on the 60dB setting.

How about this for a plan: I could turn that first stage into a 2nd order Sallen-Key filter, the inverse of the third stage.

Quote:

introduce the SD resistor

I have no objection to that.

 

[tangent]

Layout and schematic updated.

First stage hasn't changed yet. That will require actual design work to even see if it's practical.

EDIT: Oh, and the per-board cost will only be $12 now. Again, this is my cost; just letting others jump on the order if they want some.

 

[aos]

I meant, the low-pass filter should not be in this device but in the one down the line. And maybe not as simple as RC due to even a few k generating too much noise for your purpose - you're right about that. I've been dealing with some A/D recently and they all use very low value resistors (less than 100 Ohm) in the series with the opamp for exactly that reason. An active filter is more likely, but I prefer cutting the bandwidth of all stages where possible.

 

[tangent]

I've done the design work on the 2-pole active 100kHz low-pass filter stage, and I'm thinking it's not such a good idea to add it. There are a bunch of reasons:

1. It can't be the first stage because you can't have an input cap before this filter. The complex reactances that result completely wreck the filter response.

2. If you make it the second stage (10x gain plus low-pass) you have to make the first stage the adjustable one. That would work, but it puts higher resistances in the first stage, increasng total noise.

3. A 2-pole Sallen-Key filter is more dependent on the tolerance of its passive parts than the current phase-lead filter. If you adjust the cap +/-2% in the current circuit, the fc only changes by maybe 3-4% either way. If you adjust both caps in a Sallen-Key filter by +/-2%, the worst combinations move the fc -20% to +40%. The combinations also vary in the "peakiness" of the filter, so it would be impossible to compare results between two of these preamps, since the area under the amplification curve would be quite different.

I think I'm going to run this board as you see it now, unless someone has an objection to this reasoning.