Reducing Noise on Small Signal Generator

[helix1]

I am attempting to build a 1mV square wave signal generator (up to 50Hz) to use as a test input for home made ECG and EEG machines. I am using a 555 chip to generate the square wave, and then putting it through a voltage divider to get the required small voltage.

The problem so far is that there is a lot of noise on the small signal. I have tried putting the output from the 555 through a low pass filter and that takes out some of the noise but when the final output is 20mV, the noise superposed is also about 20mV peak to peak. When I try and get an output much below 10mV, it is almost impossible to distinguish the signal from noise.

I would appreciate any suggestions you might have. Is there a better way to design a 1mV square wave signal without getting all the noise?

[Andy Watson]

What are you using to measure the noise?

[barry14]

If the signal out of the 555 appears good, then any noise on it will be reduced by the same amount with your voltage divider.  Thus, if your divider output is noisy, you must be picking it up elsewhere.  Can you provide a schematic and possibly scope photos? What is the frequency and wave shape of the noise? Is it power line related? How is your circuit layed out? Were you careful with grounding? How are you measuring the signal?  What size resistors are in your divider?  If they are too large, they can pick up external inference and even generate their own noise. Are you using a shielded scope probe? I am trying to make the point that you have to pay attention to a number of factors when trying to generate very small signals. For a 1 mv signal to appear quiet, the noise level has to be in the microvolt level.

[helix1]

I've attached a few scope pics, one of the output from the 555, and ones of the output at the divider without and with low pass filtering (3dB point at 40Hz). I'm only testing at the moment, so my 555 unit is soldered on to stripboard, with the output leading to a breadboard where I have put the divider (100Ohm and 150kOhm). The filter unit is also on its own bit of stripboard, wired into the breadboard as needed.

[Andy Watson]

Remove the scope probe tip from the output of your attenuator and connect it to the ground point - leave everything else connected - including the probe ground. I think you will find that this noise is mostly "artefact" - i.e. it is not really coming from your circuit but is being introduced into your measurement circuit.

Also, are you using the bandwidth limit on your oscilloscope?

[helix1]

Ok this seems to be helping... when I touch the ground point, I can see the same amount of noise. I tried noise rejection on the trigger but that didn't do much, but when I limited the bandwidth on the vertical scale a lot of the noise went away, but not all of it.

Is there anything else I can do to reduce it further, or do I just have to accept a certain amount of scope noise as part of working on small signals?

[w2aew]

Ok this seems to be helping... when I touch the ground point, I can see the same amount of noise. I tried noise rejection on the trigger but that didn't do much, but when I limited the bandwidth on the vertical scale a lot of the noise went away, but not all of it.

Is there anything else I can do to reduce it further, or do I just have to accept a certain amount of scope noise as part of working on small signals?

Averaging will help to reduce the noise further, now that you've confirmed that the bulk of it is from the scope.

[tggzzz]

Is there anything else I can do to reduce it further, or do I just have to accept a certain amount of scope noise as part of working on small signals?

It is unlikely this is solely "scope noise". It is much more likely an artefact of your design, the details of its implementation, and the connection to the scope.

In order to have low noise, you need to get many grounding and shielding details in correct.

[Zero999]

How much noise do you get with the probe shorted to the ground clip?

Are you using a x10 probe? If so, try switching it to x1 and adjusting the oscilloscope settings to take account for it.

[helix1]

Averaging is improving things even more! Th 10mV no average scope pic below shows the result of just bandwidth limiting. The 10mV trace with averaging looks pretty good now. The 1mV trace with averaging still looks quite noisy, so I guess this is more due to design rather than scope issues now.

I am using a x1 scope probe.

So now it seems I need to learn about effective grounding and shielding issues. Does anyone know of a good reference for this?

[TSL]

What sort of PSU is powering your project ?

You might have a ground loop issue.

[helix1]

6V battery pack put through regulator to 5V, then converted to +/-15V with a MAX743 chip. Ground is negative terminal of battery pack, and +15V going to 555 chip.

[alsetalokin4017]

6V battery pack put through regulator to 5V, then converted to +/-15V with a MAX743 chip. Ground is negative terminal of battery pack, and +15V going to 555 chip.

Well no wonder. Why don't you just make a square wave oscillator with a couple of gates in a CMOS chip, that will run directly off of the 6V battery supply? Then you will probably have a much cleaner signal before your voltage divider. Something like this:


http://www.seekic.com/uploadfile/ic-circuit/s200972222553882.gif

or this:

http://www.piclist.com/images/www/hobby_elec/e_ckt13.htm

[helix1]

The short answer is that I was just grabbing things that I had to hand to try and make something that worked, and didn't realise I would run into problems! 

That circuit looks great and I will certainly give it a try. But my question is why would that circuit give a cleaner signal than the output of a 555? This is the kind of thing I don't really understand at the moment...

[ThomasDK]

The noise is most likely coming from the switching converter

[TSL]

The noise is most likely coming from the switching converter

+1 that !

you have to seriously decouple the PSU from your oscillator if you don't want all those artifacts ending up in your output signal.

[nuno]

Recently I friend came here with a circuit that generated a small audio signal (~10 - 20mV) that he wasn't able to read with his Rigol 1000Z, complaining of lots and lots of noise. I did some measurements, short GND on the scope probe, averaging, bw reduction, but the single most effective thing I did was to shield the whole circuit (put it inside a metal box). (We powered the circuit from a linear lab supply)

[codeboy2k]

That CMOS circuit won't work if there is any switching noise on the DC rails (including ground noise). Any noise on the power rails will get coupled through.

As for the op amp circuit example, yes, that CAN work if you use low offset (<100uV), precision op-amps with bi-polar supplies so you can get down to 0 and 1 mV with your square wave. The op-amp has good PSRR and will have good noise rejection on the power supply.

I also gave it some thought, and I thought that a good way would be a switched 1 mA current source over a 1 ohm resistance to get 1 mV.

So I spiced it up here: (click for a bigger image)

The TL431,  NPN transistor Q1 and R2+R3 act as a precision current source, sourcing 1mA over the 1 ohm R3,
giving you a precise 1 mV square wave that is minimally affected by noise coupled into the source.  I've demonstrated
this by including a random noise source at the input which is over 100mV peak-to-peak.

The low pass filter RfCf at the input  slows down the rising edge of the 555 output to remove some high frequency components, as well as offers a first line of defense at reducing some of the coupled input noise. You don't need the RC filter, and without it you will get some noise at the 1 mV square wave, but that noise is still very, very low at about 5 uV peak-to-peak due to the PSRR of the TL431.

The spice source is attached in the zip file.