I have DSD2048 file that is compressed in high mode worse than in normal mode with Wavpack 5.1. File was made with low quality filter (32 bit version of this converter (http://pcmdsd.com/Software/PCM-DSD_Converter.html) with IIR Filter that is included with program), so there are much aliasing. But shouldn't high mode perform better than normal mode anyway? Maybe @bryant
will be interested in sample. Sample - https://www.dropbox.com/s/b0lx3bqx4qkndl4/DSD2048-n.wv?dl=1
Interesting file, thanks! I don't think I've seen a file that actually compresses better with the default mode except special cases like silence, which this isn't. Meow... :)
This file has so much aliasing (or is just strange in some other way) that the high mode does significantly worse than the normal mode: about 22% compared to the normal mode at 31%.
However you don't see that much difference because the high mode takes so much CPU that it actually makes sense to compress every block with both the normal mode and the high mode and choose the smaller. On this file every block uses the normal mode even when you select high (which is why it decodes as fast as the normal mode).
The reason you still see a slight degradation in the high mode is because that mode uses smaller blocks by default and smaller blocks compress slightly worse because of the overhead, but this would only be a couple percent or so in the worse case. You can add --blocksize=44100 and the difference goes away.
Is there an application in for a 90 megahertz
sampling rate at 1 bit? RF, video? This number is beyond the usual margins of hi-res. Seems strange to attempt to apply compression to an audio signal that has been inflated on purpose. One probably has enough spare bandwidth if he decides to use such a special format in the first place.
Is there an application in for a 90 megahertz sampling rate at 1 bit?
It is better to ask about this people who believe that conversion to DSD improves quality. I just tested software out of curiosity.
Ugh what a braindead file ...several hundred times higher bandwidth than CDDA. No audio codec and no audio commodity should be tuned for such signals.
What this format needs: http://www.davidstonedecor.com/Double-Monuments/DSD-2048
For 1-bit you'd have to double the sample rate 15 times to get the approximate equivalent of 16-bit depth, and 2.8MHz only doubles the CD rate 6 times. You don't even get the equivalent of 8 bits SNR across the spectrum. There have been advancements, including DSD-wide (8 bit depth at 2.8MHz), but the requirements from the beginning for such high sample rates are expensive to maintain without distortion because the hardware timing has to be exact, with state of the art voltage regulation and all.
Ultimately, DSD is a legacy format meant for recordings not up-converting PCM. DSD players these days will probably downsample to PCM. Rates above 352,800 is overkill as it is... even in bone conduction, human hearing can't register frequencies above 150KHz according to tests; and in air, I doubt anyone of any age can hear 48KHz.
Did you check the above file? It's not typical DSD.
And ultimately DSD is one big dithering. Doing audio PWM way. Even the most absurd DSD file ends up in your audio chain in a D/A converter characterised by some sane figures of merit. Even if it wasn't, diaphragm in your speaker doesn't move the binary way, and that fast.
Someone who said that DSD is better than PCM, could as well say that bits are better than bytes ::)
Most sensible way to compare them is in terms of bandwidth. And I wonder if there was an ABX test comparing DSD to PCM at the same bandwidth...
ultimately DSD is one big dithering. Not that there's anything wrong with that
For 1-bit you'd have to double the sample rate 15 times to get the approximate equivalent of 16-bit depth, and 2.8MHz only doubles the CD rate 6 times. You don't even get the equivalent of 8 bits SNR across the spectrum.
I don't understand. Do you mean that SACD (1 bit/2.8 MHz) has a SNR <8 bit? It would be very surprising and SACD would then be audibly inferior to CD—especially with classical music. Thus, ABXing SACD should be very easy according to your claims.
BTW, 2.8224 MHz / 44.1 Khz = 64, not 6. Sample rate is 64× higher than CD :)
44,1 kHz * 16 bits * stereo = 1411,2 kbps
2,8224 MHz * 1 bit * stereo = 5644,8 kbps
5644,8 / 1411,2 = 4
SACD has 4 times the CDDA bandwidth (data rate).
Bandwidth-wise one could compare SACD to 88,2 kHz 32 bit stereo PCM. Or 176,4 kHz 16 bit stereo PCM. Or 44,1 kHz 64 bit stereo PCM. etc... Feel free to dither and noise shape them if necessary :))
5644,8 / 1411,2 = 4
SACD has 4 times the CDDA bandwidth (data rate)
Indeed. Bitrate is 4 time higher ; sample rate is 64 time higher and bit per sample is 16 time lower. Comparison becomes clearly difficult between PCM and DSD (technically I mean ; because sonically they both sound identical). But DSD at 2.8224 MHz can't have such low SNR as ajp9 described: it's technically better than CD and certainly not <8 bit quality :)
I would rather say than SACD is closer to 24/88.2 than 32/88.2 performances. Why would you say 32 bit?
Because then the bandwidth match. I wanted to show that it's hard to directly compare these 2 encodings, and that at such high bandwidth it doesn't really matter because if you take 5644,8 kbps stereo PCM you can make it 8/352,8 with dithering noise shaped to ultrasonics and still it will be indistinguishable from SACD. The same as 64/44,1 PCM - we are above human perception.
If I were to ABX these formats I would go with CDDA vs SACD downsampled to 705,6 kHz because then the bandwidth match (and I wonder how it is with DSD resamplers...).
In reality I suspect that SACD SNR is D/A converter limited.
Then there surely is math to calculate dynamic range/SNR for dithered/PWM signals...
edit: as expected, from what I read, DSD dynamic range and SNR is high at low frequencies and low at high frequencies where all the quantization noise is moved. So probably at low frequencies it is equivalent to very high bit depths, and at ultrasonic frequencies it is ~1 bit - in contrast to PCM it's non-uniform across frequency range.
I think that the DSD is actually a delta-sigma modulation as its math is the closest that I know to the one of DSD, but you are nearly spot one and more in the edit.
I hate a lot of confusion about SNR and DR (dynamic range) especially with digital signals, those are related but not are always in a simple way and more when trying to compare different systems of audio encoding.
The resolution for bit depth is on an exponential scale not a linear scale; 64 times the rate is only the equivalent of +6 bits with dithering (64=2⁶). And yes, this dithering results in a non-uniform signal to noise ratio, with resolution decreasing the higher you go in frequency. The 22050 band will have the equivalent of 7 bits SNR; 11025 will have 8; 5512.5, 9; etc. So in terms of preserving the entire output spectrum, it's better to go PCM.
I would go a step further in how ridiculous very high sample rates are for acoustics: most microphones don't cover frequencies above 44KHz. Even the Earthworks M50 only goes up to 50KHz. So even 176KHz is overkill for recording via mic at an adequate bit depth.
Synthesized sounds can contain any frequencies, but yeah, on D/A or A/D frontier useless frequencies get filtered out anyway.
Are you sure your calculations are good? It seems weird that such a praised format has such a low resolution at ~5 kHz which yet is a well heard frequency...
Exponential resolution scaling is a nice feature, but if it is that low on audible frequencies then DSD is even more wasteful format than I thought (lots of bandwidth needed for decent resolution in audible range)...
Fortunately, high rate dithering naturally acts like a form of noise shaping that most of the quantization error is in the less audible (ultrasonic) bands. Because of this, the 5KHz is still rendered pretty well. It's just when the overall signal is soft, the quantization noise will be more noticeable.
BTW I have some PWM music generated by ZX Spectrum "beeper", and some "chiptunes" generated by AY-3-8910 chip - ZIP/RAR/7z often compresses their WAVs to half the size of no matter if FLAC, WavPack, or OptimFROG...