All of them are tested with 32768 FFT size, it would look "cleaner" with 1/2 the FFT size, like this:
http://i.imgur.com/CffGQ4k.png
Under the same condition:
closed form 11.2m
http://i.imgur.com/FfoXO22.png
Last edited:
Given that DSC1 board is based on Amanero, I have some news about it and how it plays on Linux -
I have just finished talking on the phone with Domenico, the producer of Amanero USB boards.
We had a very friendly chat, also because he comes from a town 60Km away from me.
He explained that it already exists a firmware capable of DSD512 under Linux, but he just offered the tools to designers and at the moment it is up to each one of them to patch OS in order to work in DSD natively.
I suggested to contact ALSA maintainer to make such patch public as part of Linux next release.
He will do so and keep me informed.
So sooner or later Amanero board will be able to play DSD in bitstream (non DoP) on Linux.
I have just finished talking on the phone with Domenico, the producer of Amanero USB boards.
We had a very friendly chat, also because he comes from a town 60Km away from me.
He explained that it already exists a firmware capable of DSD512 under Linux, but he just offered the tools to designers and at the moment it is up to each one of them to patch OS in order to work in DSD natively.
I suggested to contact ALSA maintainer to make such patch public as part of Linux next release.
He will do so and keep me informed.
So sooner or later Amanero board will be able to play DSD in bitstream (non DoP) on Linux.
Just curious, which modulator was used for this? I recommend trying with HQPlayer 3.12 and ASDM7 (note that 3.12 fixes one bug in ASDM7 initialization).
What is resolution of the ADC used for the measurement?
Last edited:
It was ASDM7 under 3.11. I'll test again under 3.12 soon.
ADC is QA400 https://www.quantasylum.com/content/Products/QA400.aspx
For the reference:
AD-DA loop of QA400
http://i.imgur.com/Ydio4eY.png
1k sine soekris R-2R
http://i.imgur.com/1gN8HCh.png
Last edited:
I did some measurements with dScope III today too. Still without case, just boards on a table with open wire hassle.
dScope gives me -90 dBr noise level at DSD128 including PSU hum, which is still better than most power amps ref 0dBW. Background noise is pretty much white noise without any correlated or spurious tones/peaks. The background noise shape peaks at 7.5 kHz and I believe it is PSU related so that's one of the areas to be improved. I just regret the opamp choice and wish I would have stick to the ones I used before. But IMO, still not bad for rev.A board.
Good thing is also that Jtest24 spectra is quite clean, only visible side bands are PSU related +-100 Hz somewhat below -120 dB level.
1k wide spectrum:
IMD spectrum:
1k at -96 dB level:
PCM 0-input upsampled to DSD128:
R2R ladders tend to generate forest of correlated discrete tones throughout the spectrum, which I personally consider more audible than white noise.
For comparison, Metrum Musette discrete R2R:
1k wide spectrum:
IMD spectrum:
And 1k wide spectrum with upsampling to 352.8k:
dScope gives me -90 dBr noise level at DSD128 including PSU hum, which is still better than most power amps ref 0dBW. Background noise is pretty much white noise without any correlated or spurious tones/peaks. The background noise shape peaks at 7.5 kHz and I believe it is PSU related so that's one of the areas to be improved. I just regret the opamp choice and wish I would have stick to the ones I used before. But IMO, still not bad for rev.A board.
Good thing is also that Jtest24 spectra is quite clean, only visible side bands are PSU related +-100 Hz somewhat below -120 dB level.
1k wide spectrum:
An externally hosted image should be here but it was not working when we last tested it.
IMD spectrum:
An externally hosted image should be here but it was not working when we last tested it.
1k at -96 dB level:
An externally hosted image should be here but it was not working when we last tested it.
PCM 0-input upsampled to DSD128:
An externally hosted image should be here but it was not working when we last tested it.
R2R ladders tend to generate forest of correlated discrete tones throughout the spectrum, which I personally consider more audible than white noise.
For comparison, Metrum Musette discrete R2R:
1k wide spectrum:
An externally hosted image should be here but it was not working when we last tested it.
IMD spectrum:
An externally hosted image should be here but it was not working when we last tested it.
And 1k wide spectrum with upsampling to 352.8k:
An externally hosted image should be here but it was not working when we last tested it.
Jussi are those measurements taken with any USB conditioners?
I try different ways to get a better measurements apparently using a USB hub gives me better measurements around 1k (shorter "grass")
Also better PS definitely helps a lot, Reflector-D works wonders with DSC1
http://i.imgur.com/A2BIAL5.png
I'm still not able to get it below -120 like you did, however -110 is definitely doable.
I try different ways to get a better measurements apparently using a USB hub gives me better measurements around 1k (shorter "grass")
Also better PS definitely helps a lot, Reflector-D works wonders with DSC1
http://i.imgur.com/A2BIAL5.png
I'm still not able to get it below -120 like you did, however -110 is definitely doable.
You are aware that the height of the 'grass' in FFT plots is a function of the FFT bin width (i.e. the measurement bandwidth)? So far Jussi's not given us the effective bin width of his plots so comparisons with your own plots will be difficult. It may be that his FFT has more bins than yours.
When I talk about my experience I always try to compare apples to apples.
Same condition just different PS:
http://i.imgur.com/6u5cwXu.png
No conditioners, only HiSpeed certified USB cable with ferrite beads at both ends (original factory installed). Computer is 2011 MacMini. I also tried with iFi iUSB (with their bundled USB cable) which has given improvements before, but now it was just giving worse results.
The configuration I used for my measurements was 500 (499) R1/R2 and 1k R806/R816.
I've been thinking about this and the noise I have at low frequencies compared to 30 kHz is not PSU noise, the noise level at 30 kHz is the PSU noise (PSU is the normal DSC1-PSU), which could be lower but not the main issue. Electrolytics on the DAC board are Nichicon MUSE. One could also try OS-CON as C11/C12 (I wonder why I didn't spec it like that at the beginning!).
I'm now pretty sure the 10 dB extra noise I have below 30 kHz is intermodulation from the ultrasonic noise. Coming from early stages, I/V and filter. My main suspect is the AD844, I was thinking to try LM6181 as alternative. Does someone have other good alternatives in mind? I will also try to lower the slew rate pressure on it by trying 22p capacitor in parallel to R1/R2 again (I think I tried 220p before and got unwanted results).
Another thing that needs fixing is use of dual op-amp in the SK filter stage. I was trying to save some board space and components, but it wasn't such a great idea. Reason is that the channel separation between the opamps drop as function of frequency, so the HF noise can jump from one side to the other. Searching for an alternative dual op-amp that would have higher channel separation would be beneficial. So in DSC2 I will probably go back to my tried and tested setup of 2x AD797 although it needs pile of extra components per op-amp compared to ordinary ones to perform at it's best.
I also have two of my own "discrete op-amp" designs, but I need to run some SPICE simulations on how well those perform in HF applications like I/V and the SK filter. So far I've used those for normal audio buffer and gain stage applications only.
yeah
Too bad on that, but existing stocks should be enough for a run of DSC-2....
Too bad on that, but existing stocks should be enough for a run of DSC-2....
Yes, that looks good too! Not available as PDIP though, but the TO-99 should be fairly easy to put in place instead.
I thought that I had seen that before too, but now when I was doing searches it didn't show up.
Is it really? The SOIC model is marked LIFEBUY, but the TO-99 which I anyway like more is ACTIVE. (one of my very long time favorites OPA627/OPA637 is also TO-99)
Last edited: