Stepper motors are externally-commutated motors. So the speed is controlled by the driver which varies the speed operating on how quick the phases are exchanged.
Often they also run in open loop, i.e. there is not an encoder to tell if the motor is spinning or not.
All this makes it necessary to control the current very well: at every step change, if there is too little current the motor stalls, and if there is too much current the motor wastes energy and can even have resonance which leads, again, to stall. They are also driven with a much higher voltage than what their characteristics would imply. This is because at every phase change the coil inductance limits the slope of the rising of the current but, instead, current is needed quickly.
So, stepper motors are driven by controlling current in the phases. You energize phase A(+), then B(+), then A(-) and so on, each time with a PWM or a chopper (an automatic circuit which regulates PWM in order to maintain a fixed given current).
Then there is microstepping, which signifies that the single phases (steps) are modulated with a varying current in order to create sub-steps. In this case, instead of changing from A(+) to B(+) abruptly, the change is made gradually, for example A(+100%), A(+75%) / B(25%), A(+50%) / B(+50%) and so on. (In reality it is not so simple). In this case, a scope on a single phase will see different PWMs every step.
I can't tell what is you situation, the scope diagrams you posted are not so clear to me, but I see two options. One is microstepping. The other option could be some algorithm to optimize current/torque/speed/whatever.
The fact is that stepper motors are a different beast than normal DC (brushed) motors, but they have anyway, but more more strange, the same effects: back EMF, coil inductance and resistance etc. They are simple, robust, cheap, and very problematic, and every drive may implement more or less "hidden" optimizations on how to energize the coils.
You should monitor the current, together with PWM. That could reveal (or not) why the PWM is changing. A typical cycle on a phase (when not microstepping) would reveal a high duty PWM, while the current starts to increase; when the current reaches the correct value, the PWM duty reduces in order to mantain that current. I call it PWM, but often it is instead a waveform generated by a chopper, which simply gives full voltage if the current is too low, and shorts the coil, when the current is equal or bigger than the preset value, for a fixed short time. This chopper generates a quite messy signal, also depending on what the motor is actually doing.