Just give an update on my SB0550 mods.
Replaced all OSCON decouple caps to tantalum ones. It appears the ultra low ESR OSCON has some side effects when directly replaced the electrolytic caps, and could cause regulator instable. Normally when OSCON is used, the circuit need to be modified slightly to prevent its high ripple current due to low ESR. The tantalum ones have medium ESR (~0.5ohm at 100KHz) are ideal for direct replacement. But tant caps are not very tolerant to voltage overrating, general guidance is the voltage rate should be at least 2X, sometimes 3X of the actual voltage.
Also found the AUX power connector on my mobo (Abit IN9-32MAX) next the sound card introduced significant noise, since I don't run SLI, removed that connector improved both SNR and dynamic range by about 2db.
I also made similar DIP socket adapter like
bichi's for line-out opamp. Now using the DIP 8-pin LME49860, decoupled with 2 10uf/35v Sprague tant and the dynamic is noticeably improved. However, the RMAA score does not change much, as we are reaching the limit of the ADC SNR and dynamic range up ceiling. All three opamps for line-in have been changed to LME49860 SOIC.
Changed the DAC coupling caps from 66uF MLCC to 88uF MLCC, thanks
bichi for giving these caps to me to try. The bass is further improved, but I personally don't feel the high is too bright. Maybe because this time I only listened thru my LD MK III tube amp, but not directly connect the headphone to the line-out.
I played a little bit with the ADC coupling cap C27, C15 (stock 10uf NP) and C29, C14 (stock 47uf/16v), now all socketed. See the following result for comparisons:
I settled with the combination of 22uf MLCC for C27, C15, and 68uf tantalum for C29, C14. The frequency response is awesome with moderated IMD+Noise. I also tried 66uf MLCC for C29 and C14, but the IMD+Noise jumped to 0.0065, so definitely don't put too many MLCC in the audio path. The only thing troubles me is my SB0550 stereo crosstalk have some funny pattern, it starts climbing from 1KHz almost linearly in 24/48 and 24/96 testing. The stock elite pro had the same issue even before I start modding, and the
official RMAA X-Fi test report illustrated the same pattern.
stock X-Fi Elite Pro before modding
To identify the cause of this crosstalk issue, I ran a few tests. 1) The DSP internal loopback (select record input "Wave" in Audio Creation Mode) testing. 2) SB0550 digital out to my receiver to decode and the receiver preamp out is connected to SB0550 line-in. This bypasses the line-out section. 3) SB0550 line-out connects to a laptop's line-in (w/ realtek chip) to run RMAA record-only test on the laptop, this bypasses the line-in section. 4) Just for comparison, the laptop's external loopback crosstalk patern.
Can anyone who have X-Fi Elite Pro also check you stereo crosstalk pattern? My SB0550 is the original release that does not have heatsink (I added heatsink myself) and the X-RAM is only on the component side of the PCB and the back side PCB does not have soldering pad for X-RAM.
Bichi's SB0550 repair picture shows his one has X-RAM on both side of PCB, and it has stock heatsink, so looks like the SB0550 has two version of PCB and maybe the older version has some design issue, or firmware issue. Can anyone confirm?