The tweaking of the output filter
The D/A-converter in this player generates two rectangular, pulse-width modulated signals per channel. In this block diagram the internal construction can be seen. By subtracting and filtering these two signals, eventually the analog output signal is formed. The two opamps in the analog output section are responsible for this. But because the square-wave signal is only limited passively filtered before the first opamp, it is subjected to a considerable amount of high frequencies, and this can lead to distortion. Most of the times, the specifications of the applied opamp fall short (definitely true for the original NJM2114D) and it's input stage cannot 'keep up' with the signal. This way, intermodulation distortion is generated; a very unpleasant kind of distortion to the ears.
The amount of high frequency energy that enters the opamp can be attenuated by inserting an extra capacitor between the first 10k resistors after the DAC (RD21/23 and RD22/24). This way, an extra low-pass filter is created, of which the corner frequency depends on the size of the capacitor. The modification can be seen in the schematic of the filter, on the left. A value of 220pF seems like a good start, this results in a filter characteristic that lies somewhere between the original one and the Bessel curve that's described below.
This tweak is created by Mauro Penasa and was brought to my attention by Riccardo Romagnoli. He is a moderator on the Audiofaidate forum and he wrote an article about it: A Decisive Tweaking.
Another interesting possibility is to adapt the filter in such way that it gets a Bessel characteristic. This results in a constant group-delay in the passband and this is audible in the sound reproduction. In the filter, C605/606 are lowered to 430pF and R605/606 and R611/612 are adjusted to 14k. L601/602 must be raised to 470uH to bring the frequency of the notch filter back to 8*Fs.
The result for the sound is a more 'relaxed and slightly smoother' high-end. In this plot of the filter simulations the difference can be seen between the original and the modified filter. In this article by Pedja Rogic more information can be found regarding this subject.
In this simulation, the three different filter-slopes have been placed side-by-side in one plot: the one of the original filter and those of the above tweaks. The changes in the characteristic and the group-delay curves are clearly visible.
In theory, a combination of the above two tweaks is also possible.