PWM signal problem

So the high pass time constant is in a 100 us Range. When I asked for the waveform scaling, I also referred to the voltage (level of wide noisy line and peak-to-peak value.)

If I understand the trigger level display correct, the base line is around zero and the negative peak is actually below zero. A level, that can never be driven actively by a FPGA I/O that's really connected to this signal. You could pick it up instead from an unconnected trace running directly beneath the driven signal.