Think about the plots, their scaling and the granularity of the data that is being used to generate the curve.
(The word "curve" has a specific usage when data is being plotted. A perfectly straight line can correctly be called a "curve".)
Your Hlog plots are generated by using points in a two-dimensional rectangular
Cartesian coordinate system. The
abscissae occur at a regular 4.3125 kHz spacing. Nothing we can do will allow the data to be "harvested" with a smaller spacing. (The precision of every abscissa is very good, it is 0.5 Hz.) Think about that 4.3125 kHz spacing . . . that width is greater than the audio band-width of a standard telephone! The "harvested" data give us values for the ordinates to four decimal places on a logarithmic scale. The ordinates spacing and precision are both good. Even if you used a different utility, a different method, to plot the Hlog curves
you will not be able to escape the granularity of the abscissae. Just look at the Hlog plot you have attached . . . It is possible to see the granularity by the stepwise appearance of the plot. You keep mentioning sub-carrier number 55. I look and
see nothing other than experimental error and a limitation of the method by which the data points are generated and are subsequently plotted.
I have nothing else to say on this subject. Please do not bring it up again.Finally, for good measure, I attach a copy of the Hlog plot for my circuit which was generated earlier today. To the limits of the data precision and the mode used to generate the curve of the plot I see nothing other than a perfectly normal representation of the transfer function, loosely the attenuation versus frequency (with 4.3125 kHz spaced abscissae), for an xDSL circuit operating in G.992.3 mode.