With a signal to noise ratio of 3 to 6 I'd have difficulties in drawing any conclusions from the data. Analytical chemistry uses S/N=3 for the detection limit and S/N=10 as a minimum for quantification. With a S/N of 3 your maxima/minima have an error of 33%. I'd focus on reducing (electrical?) noise before any data is acquired (Optimal A/D range?), but that is most likely not your job.

The next best thing would be oversampling. 16 times oversampling (averaging 16 measurements) should give a 4-fold increase in S/N to a workable 12 to 24.

With the current data set you have to face the uncertanty principle in x (correct frequencies, impeded by averaging) and y (correct amplitudes, improving via averaging). Classical approaches like derivatives produce too much noise. The only recognisable feature is the steep rise for each pulse. I'd declare a new pulse if the signal had a monotone increase for a certain number of samples and the combined increase has a S/N of at least 3.