What you have discovered is quite normal for acquiring data in the time domain. If you were only interested in the frequency domain, you could follow the Nyquist criteria and sample at a frequency at least twice your desired frequency (6 Hz). Since you wish to examine the data as a sine wave (to view and recognize), you need to sample the waveform at a rate of at least ten times the desired frequency. Hence, to plot and recognize a 3 Hz signal, you would need to sample at a rate of 30 Hz. This action allows ten actual data points per wave cycle.
As a rule, I use 16 times "oversampling" for my application where I desire to manipulate the data in the time domain. Twenty times oversampling would be even better. It really depends upon the resolution you would like to
see in your data.