DSP Robotics Support

I have just noticed the problem. I've found the original paper from where the graph. The graph shows a transfer function of several such stages chained:

In this simulation, the first gain stage employs the equation sin(x) + (abs(sin(x)))^8 – z. This
equation produces the most linear transfer between the input and the output. The second and third
stages employ sin(x) + (abs(sin(x)))^4 – z, which is slightly less linear and in the fourth stage,
sin(x) + (abs(sin(x)))^2 – z, this produces the least linear of the three equations, clipping the
negative portion of the waveform much more dramatically at lower settings.

I even think that the simulation also might have contained the filtering stages.

here is the paper:
http://ses.library.usyd.edu.au//bitstre ... rt%202.pdf
and I just found out, that in the matlab are negative values used for z. So e.g. -0.2 in the given formula and the function sin(x)+(abs(sins(x)))+0.2 with the given Sine-Wave.wav file, which is half as loud as a sinewave, the plot can be achieved.
But the complete signal chain with filters (I use bidir IIR, not FIR), sounds not that good für higher amounts of distortion
I get a lot of high frequency "smearing" and the low end just sounds not right...