This didn’t exactly work, but it was fun to play with this BJT 2N3904. Ideally one should have 2 BJT’s back to back to switch a load with the same polarity that you trigger it with.
For the first picture, I noticed that the peak to peak voltage was pretty low (about 1V), so I decreased the resistor connected to the collector. Which is the second figure.
The third figure showed a pretty high, but noisy, PWM. But once I connected the load (a servo), it went down to about 2.5 peak to peak (with noise gone), which is the forth picture. Last two pictures show my setup, with my 100 MHz oscilloscope (an unlocked DS1052E hehe).
I do have some optoisolators that I may use for this, although I will try to find the “proper” solution to the 3.3V-> 5V control logic.