Hotrodding the X-Fi: A Layman's Guide (No 56k) - Page 87

Yep, perfect, Much Thanks!
- finding similiar anomolies between simple RMAA tests, Bruel-Kjaer tests and theoretical simulations.
- of interest, at the moment, is harmonic/reasonant-coupling of all the PC switching supplies.
- including the local switcher supplying the SB0460 CA20K1.
- appears its "buck-noise" is coupling to the SB0460 ground plane and causing crap to appear in analog sections, above 32khz.
- OSCON's seem to have notched most of the parasitics down, by about -5 to -8db, depending on harmonic frequency.
- LM385 was to set "worst case" model reference and behavior template.

back to FFT's, Bode's and sims....
- hope you manage to get your backup card and sockets installed.
- would be interesting if I could send you "mystery" op amps and give you an additional headache. LOL
- and rock on with your "long leaded" by-pass stuff... planning to do a "long-lead" experiment to see interactions/coupling dynamics in a few days...

Picture: (new change: C72, OSCON 15uf @ 25vdc (-5vdc section)
http://www.esnips.com/doc/385fcb4e-9.../X-FI-MOD-063b

Tried lifting the caps off the board some but no difference. Interesting though is most of my D.C. offset disappeared with these caps. I have no explanation for this as it does not seem reasonable but it dropped from 100mV to less than 5mV. Distortion pattern remains the same though with high second order THD relatively speaking compared to stock, all others lower.

finally got around to doing the mod, but just stuck with the L/R opamp for now

lm4556 swap, bridged the coupling caps, replaced the power cap with a 1000uf muse KZ (its what I had spare from my millet MAX build)

noise floor seems to be darker, sounds more crisp then before, cleaner, just a sharper, punchier sound with better seperation then before the mod, overall i'm really pleased with the mod for the ~1 hour and $1.50 it cost me

after 2 hours of use, seems to be a bit more detail, hearing some things i didn't hear before, as well as some vocals that used to get lost in the song coming through a bit more

Congratulations. Sounds like it worked for you.

Did some verification fun on TI TPS54352 switching section, C177 cap change from low-ESR (Jamicon WL) to ultra-low ESR (OSCON).
- both TI and OSCON discuss design differences between ultra-low ESR, as opposed to low-ESR. (see ref below)
- did some simple validation with TI's design calcs, using Creative's 100uh inductor and guessing at the rest. (no public CA20K1 specs)
- compared theoretical with practical measurements below:

+1.2vdc waveforms, SB0460, C177, stock Jamicon WL 220uf @ 16vdc vs., OSCON 330uf @ 6.3vdc:
- taken while WAV playing looped, output max'ed into 50ohm resistor load.
- OSCON might be causing a bit of "ringing," compared to standard low-ESR.
- will have measure with better scope and possibly change 100uh inductor to smaller value.
- OSCON remains steady, while stock Jamicon "wanders" as load changes. (very small deltas, in reality)
- OSCON does attenuate noise coupling on analog sections, small deltas...
- did not find any "catastrophic" negative changes.
- recommend staying with standard low-ESR caps for C177 for now.....

Comparison Waveforms:
http://www.esnips.com/doc/23843031-4.../X-FI-MOD-064b

REFS:
TI TPS54352 Datasheet, pages 22~23
http://www.ti.com/lit/gpn/tps54352

OSCON TechBook v14, pages 70~86
http://www.edc.sanyo.com/english/pdf/oscon/E70.pdf

Suspicions confirmed, not the best idea to replace C177 with ultra low-ESR type capacitor, equiv. to OSCON.
- noticable effect from switching power supply, feeding CA20K1 and impact on analog sections.
- issue is TI TPS54352 PWM and Creative designing for low-ESR and not ultra-low ESR capacitors.
- requires different PWM inductor value, corner frequency, to match ultra-low ESR types.
- sonically, both have excellent qualities and is tough to hear any difference...

Other power caps, on linear regulators and power-in from PC bus, remain candidates for ultra-low ESR cap tweaks...

RMAA Test Results comparing C177 - OSCON 330uf @ 6.3vdc vs., Panasonic FK, 1500uf @ 16vdc:
http://www.esnips.com/doc/800ccc44-3.../X-FI-MOD-065b

Mystery Op Amp
- just a bit of evening fun
- Germanium, care to take a crack at it?
- will post op amp part number, after analysis...

Picture:
http://www.esnips.com/doc/d4060c65-6.../X-FI-MOD-066b

RMAA Test Results:
Summary 48/24/48: http://www.esnips.com/doc/75b77315-8.../X-FI-MOD-067b
Detail 48/24/48: http://www.esnips.com/doc/83411bbd-3.../X-FI-MOD-068b
Detail 96/24/96: http://www.esnips.com/doc/6fe645db-d.../X-FI-MOD-069b

Why shield the top of the chip, when it has such excessively long, unprotected leads? Worst thing you could do if your worried about interference.

Leads are short for reasons.

Have you tried other methods to get the IMD spikes less eminent? Like shielding paper/cover? Replace to the 4 6.3v1000uf?

For THD, why there are spikes at 2k, 3k, 5k, etc? Any ways to reduce these?

For crosstalk, why raised dramatically after 1kHz till 44kHz (-84dB)? Any way to get better results?

It shows the left channel line and right channel line are not superimposed in graphs of your testing, and very clear in your crosstalk graph? Do you think it's caused by shorting the caps? Any suggestions?

Thanks!

I'm going to replace the clock crystal, with one with better tolerance and stability. Since the stock one probably isn't that great.

FYI, if you want to do the same, the clock has a CL of 30pf. and needs to be 24.5760mhz.

I'm getting a Fox one with 10ppm +/- tolerance, and a 30ppm +/- stability.

Comparing test 60b old to 68b newest.

slightly higher even order THD in 68b compared to 60b but fewer high order spurie in 68b.

According to graph slightly lower IMD swept in 68b.

Slightly lower IMD in spectrum analysis.

Noise & dynamics too close to call in graphs.

Slightly better crosstalk at midband but worse at high frequencies.

According to test analysis provided by RMAA they appear very evenly matched accross the board but graphs tell a slighly different story especially in IMD distortion where 68b definatly does better. I think the test interprets the even order THD increase as part of the IMD score but IMD swept & spectrum analysis are definately better in 68b according to the graphs.

Wish you had done tests at 96KHz so I could see the higher order spurie but overall looks pretty good. Slightly higher THD but lower IMD distortion according to graphs.

The THD spikes at 2K, 3K & 5K while they can be reduced especially the 2K spike the overall spectrum always seems to stay about the same. Capacitor coupling reduces the 2K spike somewhat dramatically but increases the others slightly. Due to the increase in second order harmonics (the 2K spike) the overall score for THD is slightly worse but second order THD is really nothing to worry about. If you do a lot of reading about THD you will find that It is preferable to have higher second harmonic distortion & lower high order products. In my experiments the 3K & 5K distortion is actually lowered slightly by direct coupling. Only the second harmonic is higher.

Crosstalk will almost always tilt up at high frequencies due to capacitive coupling. My crosstalk is actually substantially better across the board than creatives own tests for my card.

The difference between channels for crosstalk is present in almost every graph I have seen & not just from my card. There is a slight difference in the frequence response of the left & right channels which developed after moving the capacitors around to different locations on the board that didn't originally exist & moving them back to the original locations did not fix it. originally after direct coupling both the DACs & the ADCs the specs were even & were +.01/-.05 but the left channel went back to +.01/-.07 & the right channel stayed at +.01/-.05. Don't ask me why because I don't know & all attempts to fix it failed but it still sounds very good none the less. bare in mind though these are really tiny differences.

For some reason the big 12uf caps seems to have fixed the D.C. offset to where there is almost none where there was 100mV before, Again don't ask me why as this is very strange behavior but again welcome at the same time as there is no more pop when booting or shutting down.

Your wish is my command
- added RMAA 96/24/96 detailed results to "Mystery OpAmp," #1297

Specs don't look so good on IMD at 96KHz. Substantially worse compared to the IMD taken at 48KHz sample rate. Even at this level it is really meaningless as it is not likely to be heard by mere mortals like us.

High order distortion spurie decreased @ 96KHz compared to 60b but IMD distortion increased @ 96KHz compared to 48KHz

Did you change anything else between the 48KHz pass & the 96KHz pass??

The 96KHz pass didn't seem to do too much better, very little improvement anywhere except above the audio band & that is only slight as far as THD is concerned.

Cross talk better at midband than 60b test but like the 48KHz test worse at the high end.

Nothing special to report on Noise or Dynamics. Both look very good & are too hard to tell which is better.

By the way the MDR-D777LPs arent really worth the effort to modify. They are way too much midrange oriented. My modifications reduced the excessive warmth caused by the lower midband & upper bass but the area center around 1KHz still too emphasized. The high frquencies sound better modified & the bass is still very strong yet pretty well defined but that midband hump is just way over the top. It tries to drown everything else though not really successfull at it. In a word even at thier best very colored presentation. Looking forward to trying the Denon AH-D1001 or the Sony MDR-SA1000 earphones.

Thanks for the analysis, Germanium.
- mystery opamp is the original NJM 4556A.
- no, no hardware changes between 48k and 96k.
- was cleaning home-lab and couldn't just throw out clipped opamps.
- decided to try soldering leads to SOIC stubs and test new lead-free solder-paste, just for grins.
- would also serve as "stock" reference."

Sony MDR-D777LP, Denon AH-D1001, MDR-SA1000, closed earphones, eh?
- closed get too hot for me.
- considering building simple buffer amp with Burr-Brown/TI BUF634 to use with old Sennheiser HD-414SL open headphones.

Just a few more experiments left with this SB0460...
- might try a trick I used when designing LNA GaAs MMIC parts, ten years ago, based on "close-field" electrostatic theory.
- when I finish calcs, will send modification to you for trial and analysis.
- considering playing with either Auzentech X-Fi Prelude or E-MU 1212M (balanced-out).