|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
74.74.162.231
In Reply to: RE: I'm not convinced ASRC is the culprit though posted by Dave_K on July 26, 2015 at 04:27:05
ASRC (upsampling)CS8422: 0.05
SRC4392: 0.008
DAC chip (oversampling)PCM1796: 0.0002 (for sharp roll off)
CS4396: 0.0001 (for 32, 44.1, 48)
WM8742: 0.000058
Edits: 07/26/15 07/26/15Follow Ups:
Are any of these used by the Cambridge DacMagic and Benchmark DAC1 HDR?
Yes, of course.
DacMagic 100 has WM8742: 0.000058.
Benchmark DAC1 HDR has SRC4392: 0.008.
By the way, DacMagic and DacMagic 100 are very different. The old DacMagic and the current DacMagic Plus use upsampling, but the current DacMAgic 100 does not use upsampling.
Benchmark DAC2 has SRC4392 and Sabre DAC. Benchmark DAC1 used AD1896 SRC and AD1853 DAC.
I was less sure about the Cambridge, but a little Googling revealed that the original DacMagic and DacMagic Plus have an Anagram Technologies sample rate converter (ATF in the original and ATF2 in the Plus).
Pass-band ripple spec on the AD1896 is +/- 0.016 dB, but the ripple pattern looks pretty conventional, typical of an elliptic filter, not like the one you posted earlier. See the pass-band ripple plot (TPC35) on page 13 of the data sheet. Link below.
In the graph you mentioned, the horizontal axis is linear scale. The measurement graph I linked is octave scale (logarithmic).
The location and spacing of the peaks in the ripple pattern aren't the same even accounting for scale. More importantly, the roll-off in the Ken Rockwell plots starts a lot earlier and with a shallow slope. The AD1896 data sheet shows the classic "brick wall" filter response whereas the Ken Rockwell plot shows a response down by 0.05 dB at 10 KHz and 0.2 dB down at 20 KHz. The only thing that really matches is the magnitude of the ripple. There might be a connection there, but it's no smoking gun.
Besides all that, I don't think we've connected this to the use of ASRC. There's no reason to think that ASRC lends itself to using a particular filter design function. If you want no pass-band ripple and can tolerate a wider transition band, you use a Cheby II function (or even Butterworth). If you want the narrowest transition band, you use an Elliptic function. The different filter performance metrics can be traded off easily with most filter design software. I can't think of any reason why these basic filter design tradeoffs would be dependent on the output sample rate of an interpolation filter.
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: