Yes I am well aware of what it is, a dying format that brings no value. Topic for another thread.
It's certainly a nice format when you want to DIY your own DAC, see the noDAC and Signalyst DSC1 threads, but it has some drawbacks: unsuitable for most kinds of digital signal processing, like those used during editing, too few quantization levels to allow normal dithering, incompatibility with the highest-performance DAC technologies unless it is first converted to PCM. This is very off-topic of course.
Last edited:
What about it the digital audio started out as 16/44 (or 24/192) and was converted to DSD512? Is it still incompatible with the highest performance dac technologies?
No, high performance DACs are not single bit anymore. Thus it must be converted. Might as well be PCM to begin with. Surely, you see how weird it is to record it, convert to 352.8 PCM (DXD) to do all processing, convert it back to 1-bit DSD, transport it to a DAC, which then converts it back to some internal 6-bit representation.
In a world where we can just use high res PCM to begin with, I don't see the point. It might have made sense if 44.1 is the only alternative or the world stuck with 1-bit converters.
Last edited:
Not exactly. Don't know about all dacs, but do know about Sabre dacs. They handle DSD somewhat differently from PCM, and the data sheet shows a very distinct difference (unfortunately under NDA). If conversion to DSD512 is done very carefully, subjective SQ can better vs high sample rate PCM. However, dacs with high quality external PCM interpolation filters may compete very well with high sample rate DSD. For most dacs in diy use, it is easier to obtain good sounding DSD using computer software vs producing good sounding PCM using an FPGA interpolation filter.
No offense, I don't think you understand what's going on. I have the datasheet, just took a look at it. This is off topic so best to just leave it.
If you start another thread or PM me you can maybe explain how the DAC does not convert it to the internally required 6-bit format as suggested by the block diagram on the second page in the datasheet.
Yes Scott, as you probably know by now (google), died piloting a plane.
I never met him but have to say, he did an incredible amount in the time he was here, from electronics design / manufacturing to FOHS for Elvis, Bruce, Streisand...
Pretty amazing life by any standards.
quantran,
If you go over to computer audiophile, search for threads about HQplayer software and specifically input by the designer, name = 'Miska' , (I forget his actual name).
Incredibly talented software engineer that does share some insight.
Generally speaking. If real-time streaming (i.e., streaming audio) performance isn't required, then a software filter can easily be made to any arbitrarily high performance. Just add additional math instructions, or increase the filter coefficient precision bit-depth, it's only code. If real-time throughput is required from a software filter, then the underlying hardware processing engine (CPU or DSP) must also be of sufficient performance.
Hardware filters can easily support real-time streaming data throughput, but are not as easily made to an arbitrarily high filter performance.
Best way to research it is trying it yourself. It's free to try it if you already have a computer + audio interface. You may not hear any meaningful difference, and you can sleep better.
IIRC BruteFIR was capable of 3.4 million floating point taps on a 1G processor real time, years ago.
From: BruteFIR
You remember correctly.
I'll need to double check (if I can find a decent benchmark), and it will depend on FP vs INT calculations, but I'm betting on that Athlon being quite a bit slower than a single A53 (big) core of a RPI3 chip clocked similarly.Edit: here we go! https://www.element14.com/community...ensive-raspberry-pi-comparison-benchmark-ever
Quite a bit slower was incorrect, but trading blows fairly closely!
Last edited:
Currently I do not know, why no one uses the spline as WADIA did.
I still have the WADI 27 and have still more music (as I am able to here) as AKM 4490 DAC or ESS9038Pro DAC. Always listening to acoustical music content as boom boom..
Things gets even worst (listening wise as pre/post echos who simple do not exists) not using the fast filter mode.
Hp
That is correct, there is a huge body of literature on polynomial spline interpolation. The mathematical trick is to keep the derivatives continuous so there are no discontinuities which are physically unreal. IMO frankly the whole idea was based on the "doesn't this look nicer" principle and it has the usual problem with aliasing.
BTW has anyone ever tried a brute force FFT based filter at the length of an entire song. A 3 minute piece at 96K would take a 17,280,000 point transform, a modern PC should be able to handle that at double precision floating point.
Last edited:
WADIA used the R2R BB PCM1702/04 chips with high SR as x64. Today's AKM/ESS DAC's are not able to feed this high SR. Also with the WADIA spline some early roll of at 22kHz was the trade off.
BTW: Did a 16M FFT ver. 82 times avg 256k FFT, looks close as equal. IMHO a spectral or TFD (time freq distribution as used to analyze birds) would show much more the real beef.
Picture Note: Lindberg 2L-087_stereo-96kHz_06 source (nice to see some spurious
)Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.