How to attenuate signal to test LNMP?

[psilosome]

So I made a square-wave generator from a CMOS 555 timer chip and it makes nice waves, but the output voltage is too high for my purposes. I want to get a millivolt-level signal for testing my almost-completed LNMP, but the circuit puts out about 5 volts. I can probably reduce this by powering it with 3 volts instead of 9, but I imagine it will still be much higher than I want.

I tried a 50k pot in-line, but it had almost no effect. If I understand correctly, the input impedence of both the LNMP and my multimeter are very high, explaining the ineffectiveness of the pot.

Is there an elegant way to attenuate my 1 kHz square wave down to millivolts?

 

[NeilR]

I ran into the same issue when I built my LNMP. In my case I wanted to attenuate a sine wave from my commercial signal generator.

I just soldered two resistors together, a 100K and a 1K resistor, I think, that I had in my bin. That should have given me an approximately 100:1 voltage diviider (actually it is about 101:1 and the right values would be 99K and 1K in this case)). Connect signal source to the ends and tap the LNMP input between the two resistors. That will give you your attenuated signal. Adjust resistor value ratios to suit.

This worked, but the input to the LNMP was not even close to the expected value, presumably because of the way this voltage divider loads the LNMP input. Possibly other resistance values may work better. I'm a little fuzzy on the details but I recall that I was trying to calibrate the LNMP and the voltage across the two resistors (the signal generator source) should have been equal to the LNMP output with the 3rd stage disabled, or 10x the input with the 3rd stage enabled, but it didn't work out that way.

The biggest hurdle with calibrating the LNMP is accurately measuring a 1-5mV AC signal at the input. I get different results from different meters. I ended up calibrating it as best I could based on a direct reading of the voltage divider's output to the LNMP with the meter I trusted most.

I would like to find a more accurate way to do this, by getting a precise known voltage division but I haven't gone back to this because for the time being I'm reasonably satisfied with my calibration and other DIY projects are in the way (isn't that the way it always is?).

 

[psilosome]

Cool, the voltage divider is a great idea ... thanks!

 

[NeilR]

When I revisit this, I'm going to try different voltage diviiders with higher and lower values, maybe 10K/100R and 1M/10K just to see how the combined circuit behaves and to see if I can resolve my calibration problem, or at least take notes so I can discuss it. My 100K/1K is not necessarily the best values but it will do what you want to do, which is to simply attentuate the signal without burying it in the noise floor.

 

[tangent]

Try using a CD player with a test tone disc.

First, make it put out a full-scale tone. Measure it: it should be somewhere between 1 and 2 V. 1 mV, then, would be a -60 to -66 dB tone, which should still have reasonable dynamic range.

EDIT: If you don't know how to make such tones, I added that to my troubleshooting guide about a week ago.

 

[NeilR]

Tangent,

Any thoughts on how to accurately do a voltage divider off of a singal generator output to get, say exactly 1mV into the LNMP without relying on a meter?

 

[tangent]

Quote:

Originally Posted by NeilR Any thoughts on how to accurately do a voltage divider off of a singal generator output to get, say exactly 1mV into the LNMP without relying on a meter?

It depends on why you want to do it.

If you're only looking at audio frequency, I'd still go with the CD player, even with that other toy on your workbench. The CD player is just a better tool for the job.

If for some reason you wish to test the LNMP at higher frequency, then you do have to use the signal generator. But, you'll probably find that it has a 50 ohm output impedance. The LNMP's input impedance approaches 100 ohms as frequency goes up, so you can see why you had problems earlier. There are a couple of ways you can go to cope:

1. Cut all the gain stage resistors in the LNMP in half, then just factor in the extra divide by 2 stage, that being the interaction between the generator's output and the LNMP's input. Theoretically you could leave the values the same and calculate the division factor of a 50/100 ohm divider, but there are other advantages to doing it this way. Be careful doing it this way, because I probably did it the way I did for a reason; I haven't thought about it deeply, but just off the top of my head, I think you'll wreck the main filter's Q if you don't also change some cap values. There may be current driving concerns, too. If you don't want to modify the LNMP, I'd move on to the next option regardless:

2. Buffer the divider. Take a pair of LM6171s, configure them for unity gain, and put the divider between them. Don't use a 6172, because you will also want to use the offset nulling feature. I'd probably get a little more clever and make the second stage an active divider instead of a simple buffer. Design it like test equipment: you want an active, wideband, calibrated divide-by-100 stage here.

 

[NeilR]

Tangent,

This has nothing to do with my signal generator per se, it was just more convenient at that moment and I have good shielded coax test leads at hand; regardless of my source- signal generator or some digital source, I still have to deal wtih the following statement in your assemblt guide:

"Measure the signal at the output of the amplifier: it should be about 1000× as high as at the input with S1 closed, and 100× as high with S1 open."

That simple statement opens up a HUGE can of worms, relating to accurately measuring any AC signal from any source at the 1mV level. That means, for example, to get my LNMP gain set to an accuracy of +/- 5%, which is not terribly accurate, I need to measure the ac signal to within +/- 50 uV! This is a classic catch 22. The purpose of the LNMP is to accurately measure small signals but in order to calibrate said LNMP, I have to accurately measure that same small signal without the aid of the device that I built to allow me to make that measurement!

To be clear, I am using low audio frequencies of 400-1000Hz, nothing fancy and nothing out of the audio band.

Your idea of a opamp based voltage divider is interesting. I don't want to tamper with the LNMP itself because this is basically a one time requirement. However, to satisfy the need, whatever I do has to be accurate without external measurements for the reasons I stated above.

This entire subject hinges on the accuracy of my meters at the 1mV level, of course, but I don't know of any easy way to verify that other than using a number of different meters to arrive at some sort of concensus, which is the approach I took for my initial LNMP calibration.

I assume from your response that given the higher output impedance of a CD analog out, my simple resistor based divider might be more accurate than using my singal generator, which, I believe does have a 50R output impendence?

 

[tangent]

Quote:

Originally Posted by NeilR That simple statement opens up a HUGE can of worms, relating to accurately measuring any AC signal from any source at the 1mV level. That means, for example, to get my LNMP gain set to an accuracy of +/- 5%, which is not terribly accurate, I need to measure the ac signal to within +/- 50 uV! This is a classic catch 22.

You can call it a Catch 22 if you like, but it's the way the world is. You must use a more accurate device to measure accuracy in a less accurate one. That's why $10,000 DMMs and $100,000 scopes are successful products.

Quote:

The purpose of the LNMP is to accurately measure small signals

You might have that purpose, but that is not why I designed it. I designed it to be able to make qualitative comparisons of power supplies. If one supply has twice the noise of another, I expect to get 2x the voltage difference at the output of the LNMP. It is not greatly important to me if the numbers are 50 uV and 100 uV, or 100 uV and 200 uV. The LNMP serves its purpose, for me, if it shows the correct ratio; if it gets the absolute value right, too, that's nice, but not necessary.

I am not saying that you are wrong to have this goal. I am just poining out that you may have mislead yourself into unwarranted expectations. The LNMP docs make no claims of accuracy. If you want some particular level of accuracy, that's your responsibility as the DIYer.

Quote:

To be clear, I am using low audio frequencies of 400-1000Hz, nothing fancy and nothing out of the audio band.

Okay, then use the right tool for the job. Unless your generator cost in the thousands of $$ new, it's going to be far less accurate in the audio band than any halfway decent digital audio source. In fact, from a distortion standpoint, even a few thousand dollars won't be enough to equal the performance of a decent digital audio source.

(I said "CD player" earlier on the assumption that a standalone CD player is your most accurate source. Maybe I'm wrong, and your most accurate source is a high end PC sound card. Whatever, the point is the same.)

Quote:

I assume from your response that given the higher output impedance of a CD analog out

Now, I didn't take that into account. I assumed, perhaps wrongly, that the average CD player's output is all but zero. If I'm wrong, then add a simple buffer stage between them. Again, the LM6171 would be fine here, but if you have a spare AD797 on hand, that would be fine, too. Indeed, it might even be better. I'm uncertain because the 797 really isn't made for unity gain.

 

[NeilR]

Fair enough, Tangent. I don't have a specific reason to make the LNMP as accurate as possible, but in a wold of meters measured in 0.something% accuracy, it was just a knee jerk reaction.

I'm not criticising your project or assembly instructions, just stating the world as it is, just as you reiterated. I do think you should put an asterisk next to that statement and at the bottom of the page, add:

* - Good luck!

Beyond that, my error could be 50% or more and I would not know it without independent confirmation. And no, I don't have 100K to throw at the problem

. I was just trying to do better with stuff out of my parts bin.

Interestingly, my 4 Wavetek meters all agreed within about 10% of each other but the DMM on my Fluke 97 scopemeter disagreed by about 35%. I assume it has something to do with the added noise level in a digital based meter and I went with the Wavetek readings, but I don't have a high degree of confidence in that decision. On the one hand I didn't want to throw out a multi-thousand dollar (in it's day) Fluke meter but on the other hand, the 4 meters all were in agreement. And no, none of the 5 meters are recently calibrated, and no, I probably won't spend several hundred dollars to do that.

 

[tangent]

Quote:

Originally Posted by NeilR I do think you should put an asterisk next to that statement and at the bottom of the page, add: * - Good luck!

No, the instructions were indeed wanting. I've rewritten the test steps to include what we've hammered out above, plus some more things I thought of while rewriting. Step 10, in particular, is about 3x longer now.

Quote:

I was just trying to do better with stuff out of my parts bin.

As I said (both above, and better in the new docs), I think you can do a lot better with a digital audio source, if you set it up right. Please read the new material, particularly noting that I changed my advice to an inverter. We're talking about $10 in parts, and maybe an hour of construction time.

Quote:

I assume it has something to do with the added noise level in a digital based meter

No, it's more to do with the fact that to get speed, they have to trade noise floor. It's just the nature of analog to digital conversion. 16-bit audio ADCs are a dime a dozen, but just go and try to find a 100 MHz 16 bit ADC. If you do find such a thing, it'll be damn pricey.

Quote:

And no, none of the 5 meters are recently calibrated, and no, I probably won't spend several hundred dollars to do that.

In that case, I'd relegate the Fluke to looking at fast signals, then. You now know you can't trust it for anything else. Personally, I'd get it recal'd.

 

[NeilR]

I think the additions to the setup are great. I'll give the inverting opamp "buffer" a try.

I still don't understand why it is so important to use a high quality audio source. Even if my signal generator has a few percent distortion and a little more noise than a digital source, I would not think it would impact a simple ac measurement as long as I start with a relatively high level signal and use a divider. I'm not trying to get it 0.01% accurate, just within, say 3-5% for starters.

I guess I am suggesting that once the signal is divided down to the 1mV level, the general noise level from the LNMP, the test leads, and the general environment would probably swamp a little noise and distortion on the source signal. Not so much the LNMP, because I believe your 6uV noise floor estimate based on my tests, but it all adds up.

One of the reasons, by the way, that I would like to make it as accurate as possible is that I might compare my numbers against someone elses. For example, my Steps measured a bit higher than the levels you posted on your PSU comparison page; as I recall around 70uV or so, although it was jumping around on an intermittant basis and I didn't bother to "sterilize" my testing environment as I would do with an RMAA test. I just wanted a quick test to get a reasonable number to verify the instrument was working correctly, which I did. I can see from this discussion, though, that comparisons with different users/builds of this instrument will not be very meaningful unless the numbers are very out of line.

 

[Garbz]

Quote:

Originally Posted by psilosome Cool, the voltage divider is a great idea ... thanks!

This is how a attenuating pot should be hooked up anyway. H to output, L to GND, wiper to input makes a variable voltage divider.

 

[tangent]

Quote:

Originally Posted by NeilR I still don't understand why it is so important to use a high quality audio source. Even if my signal generator has a few percent distortion and a little more noise than a digital source, I would not think it would impact a simple ac measurement as long as I start with a relatively high level signal and use a divider.

If you tell a digital CD player to put out a -60 dB signal, I expect it to be well within 1% of 1/1000 the voltage of a 0 dB signal. If you run a signal through a divider made from a pair of 1% resistors, you have up to 2% error right from the start.

Quote:

my Steps measured a bit higher than the levels you posted on your PSU comparison page; as I recall around 70uV or so

There are so many things that can account for the 12 uV difference that I can't see how you can reasonably even define success.

 

[NeilR]

Quote:

Originally Posted by tangent If you tell a digital CD player to put out a -60 dB signal, I expect it to be well within 1% of 1/1000 the voltage of a 0 dB signal. If you run a signal through a divider made from a pair of 1% resistors, you have up to 2% error right from the start.

When you say in your instructions "Now you need to divide that down to 1 mV, or thereabouts." I thought you meant to use a resistive divider. I guess I am a little hung up over trying to accurately get a 1mV signal out of any active circuit. For a couple bucks I can use 0.1% resistors and that gets me to 0.2%, far beyond what I would hope to get after factoring noise and other issues. If my digital source is really that accurate and the noise floor is far enough below that (counting cables and everything else in the system), that is certainly easier than trying to buffer the analog generator's output to make the voltage divider work as accurately.

Quote:

Originally Posted by tangent There are so many things that can account for the 12 uV difference that I can't see how you can reasonably even define success.

I'm not sure what you meant by that. To me "Success" was getting a reasonable measurement. If I measured 5uV or 500 uV, that would not have been "success".