How do I decide the optimum smoothing method for my noisy data?

I have a raw signal from a calibration process, in which I applied a precisely known excitation signal in several steps. As it can be seen, the noise in the response signal increases with the excitation signal. I want to smooth the signal and remove the noise, so I can characterize the calibration points for each excitation value.

I have been playing around with the smooth function, and I got a result like the one in picture two, using the sgolay and loess methods respectively. However, I don't know what criteria I should consider when choosing a smoothing method. I have read on other answers that the optimum smoothing method depends on the data set and source of noise. I would like further information, if someone could develop this, please.

Here I leave the sources of documentation I checked before asking, and none satisfied me.

http://de.mathworks.com/help/curvefit/smoothing-data.html

http://de.mathworks.com/matlabcentral/answers/221628-smooth-a-curve-using-sgolayfilt

http://de.mathworks.com/matlabcentral/answers/48298-best-smoothing-function-in-matlab

Thank you.