Hi,

on which scale do the adhesion forces differ? If they are "the same" on a scale of a few hundred nanometer, you can do e.g. 10 force curves and make some averages. Just sum all curves and divide by the number of curves. That's only possible, if the noise has a real noise character. If the forces are different on the atomic scale, then, you have to use atom tracking technique (there is a paper for that ;))... don't apply some lowpass filters or something, which manipulate also the useful data!

Till some 5 years ago I was active in separating signal from noise in AFM data, typically in images of scanning probe AFM. Different flattening procedures were collected in a Matlab toolbox. I hope my colleagues from Linz, Johannes Kepler Univ. Biophysics are still using these progs. In general, if you know the properties of the noise produced by your device, the extraction of the signal from the record is well studied, e.g. ML toolbox System Identification. AFM is very sensitive, both to electrical and mechanical noise components. Therefore, some time even averaging is not sufficient, because in long term studies the noise appears nonsteady-state.

You're right, but it could give a first order approximation to average over some curves. At least to smoothen out some of the noise. Do you have any further information about how to do these signal processings? They sound quite interesting!

BTW: The best noise reduction is done before the measurement starts - so, buy a RF modulated laser for example and enhance your detection system. We have a paper about noise sources and how to avoid them (DOI:10.3762/bjnano.4.4 for example)...

I agree with Matthias' last point, but I am sure this is a practical matter and you wish to analyze the data you have. If the noise is relatively regular and if you are simply looking for adhesion values, you can consistently measure valley to valley (or peak to peak) at the bottom of your adhesion peak and "out of contact" region of the curve. It is a problem though if there is only noise in the contact region and not the other portion as adhesion measurements must span both parts of the curve. For accurate statistics, we measure 10 curves at each point, for at least three different but similar regions of the sample, on at least three different samples. This will give you an average with standard error, with a reasonable n - the noise usually comes out in the wash as part of the error. However, if the noise is really large compared to your adhesion values, then I would strongly recommend Vassili's post-processing option, or as Matthias suggests, getting rid of it from your measurements.

Hi All,

Thanks for the suggestions. But I think I should elaborate little bit more on noise. I am collecting force maps (128x128 force curves) in aqueous media, so there is lot of random waviness (few nm to 1 micron) in the non-contact region. In air I don't have any of these issues. This waviness is making it difficult to find the point of contact between the tip and the surface. For now I am calculating adhesion energy difference between approach and retracting curves but for further analysis I would like to remove these random waves. I think there is some viscous drag which might be causing this. They don't look like any electrical or acoustic noise.

But I'll still look into suggestions that you gave me...

Your curves may have what is known as virtual deflection, which is a virtual deflection is caused by a systematic artifact in the beam path of the diode used for optical lever detection. The tell-tale sign is if the baselines are straight but no flat or have a slight quadratic arc to them (especially for soft samples for some reason). In the operating software for some AFMs, virtual deflection can be corrected prior to measurements. The correction is to simply subtract the baseline from the entire curve by fitting it with either a straigh line, or in the case of soft samples, sometimes a quadratic fit is a better approximation. In either case you can make these corrections after the fact on your data. Just determine the contact point (be conservative, so that you are sure only to get the baseline region), then do a fit on the baseline, and then subtract this from the entire curve. You will now see that the baseline is perfectly flat, and the contact region of your curve will be slightly tilted with respect to the original data. The caveat is that it should only be used in instances where your If the baselines are crooked or wobbly for some other reasons, then the data should not be used.

If you are getting big wobbly baselines, you may be pulling tethers out of your sample that are sticking to the tip, and the wobble is caused by the z-piezo feedback. If you fit through the wobble, and it is flat, you may be able to get away with using the data, but I would be careful. You may be better off doing the measurements again.

Actually, baseline (zero force level) inbclination should not affect much adhesion. This type of defect I would say is much more destructive for stiffness measurements, where one needs to find the correct slope of the contact part of the curve, to calculate optical lever sensitivity and then the true force during the indentation (BTW, this often makes the apparent indentation be negative, which is a nonsense). For ahdesion, yes, some averaging after smoothing, perhaps after removal of optical interference frequency in a FFT and transforming back, could probably work decently

Jack, I actually do virtual deflection correction before starting the experiment but still i get slopes in the baseline. I have tried using polyfit on matlab to subtract slope and it works quite well. Most of my curves do have slight quadratic arc. I'll try what you have suggested. I think this is what i was looking for...thanks.

Marco, i understand your concerns. Once was trying to convert raw data into f-d curves. I think its was park systems AFM. I had to include certain conversion factor along with optical lever sensitivity. Initially I didn't had that and indentations were coming out to be negative. Using FFT to remove optical interference sounds cool, I'll give it a shot.

Good luck!

to Varun Vyas

Typically, AFM produces a number of force curves (an example mentioned Tanya Dahms, 10 curves at each point) containing the same number of data points, and all the fitted polynoms have the same degree. For such situations my POLYFIC (vectorized POLYFIt, for data Columns, cf FEX in MATLAB) is the tool of choice, saving your time.

·

In case you are using a Bruker AFM to collect your force curves, you can use the Nanoscope Analysis software (v1.4) to do a 'baseline correction' to your force curves. The correction is based upon a 1st order (line) fit of the non-contact part of your curve, and will not correct for any oscillations you might have. Contact me offline in case you do not yet have a copy.

Actually i would check the setup you use again for noise sources. I also do a lot of force distance curves in water/organic solvents and i rarely see any noise in the non contact area. I used to see a lot of noise when measuring in plastic petri dishes, obviously all acoustic noise was coupled into them and transferred through the liquid to the cantilever. Using thicker glass petri dishes as well as "home made" acoustic isolation using styrofoam material solved the problem finally.

You may try the SPIP software, which has options for baseline corrections. When having a lot of curves you may also take advantage of its batch processing facilities

https://www.imagemet.com/products/spip/modules/force-curve-analysis/

Also be mindful of whether the tip may in fact still be in contact with the sample after retraction, e.g. after pulling a long tether out of a cell membrane. This can cause the effect you describe. 

Flattening the baseline can be done with a linear fit in post-processing e.g. in MATLAB. The issue then lies in whether or not to rotate the contact region of the curve, and of course identifying the contact point.