Thread: Groundbreaking DXD recording for SACD

Posts: 36
Page: prev 1 2 3 4 next

Post by Beagle December 20, 2006 (11 of 36)
I echo Julien: "your post is making some light in my brain!".

Your comment that DXD is "able to capture the ambience around the instruments" due to a clearer impulse response, explains the spatial resolution heard in the 2L Mozart violin concertos.

The high frequencies are the range where fast sampling is needed to prevent noise, right? Is it true to say that the human brain gets its best spatial clues from the high frequencies, and maybe even the very high frequencies?

Post by Karlosak December 20, 2006 (12 of 36)
Beagle said:
Is it true to say that the human brain gets its best spatial clues from the high frequencies, and maybe even the very high frequencies?

A very good question Beagle. From what I've read, the human ear (brain) isn't very good at hearing very high frequencies let alone their localization.
But what it's good at, is the left-to-right localization of common sounds. As you know, we are able to localize nearby sounds with an accuracy of several inches. To do so, we must be able to distinguish phase differences in the order of microseconds(!) between the sounds coming to the left and right ear. This technique is only useful in the left-right direction though.

As the sound enters the ear, it is filtered by the auricle and ear canal in a specific way dependent on the sound direction. From the day we were born, with everyday practice our brain continually relates the altered frequency spectrum to the source of the sound. That's why we can precisely localize sounds even from front/back/top/bottom.

Getting back to the recording formats, we don't need the high sampling rate to capture high frequencies per se (we can't hear high frequency tones, although studies show that there is still some human brain activity involved) but primarily to faithfully preserve the phase differences between channels. That's why a very good impulse response is desirable.

Regarding the bit depth, 24 bits is certainly enough. The input waveform is sampled with the precision of 16,777,216 different possible steps, which translates to more than 144dB dynamic range(!), certainly more than you can actually capture with todays instruments (either you blow up the microphone with some jumbo jet engine or you get the inherent noise).

Post by Beagle December 20, 2006 (13 of 36)
Karlosak said:
... we must be able to distinguish phase differences in the order of microseconds(!) between the sounds coming to the left and right ear...

From what I've read, phase-difference is significant for spatial cues from lower frequencies, and amplitude-difference at higher frequencies (the phase-differences are easier to detect in 'slow' waves). And a pitch-distortion curve, similar to doppler effect, enables us to locate sound sources above and behind, to a limited extent.

I have a suspicion, though, that the brain makes use of high frequencies, of which we are barely conscious. When we can hear the rosin on the strings, we have a very precise image of where the violin is.

One of the fascinating aspects of all this, is the implication that the brain has a very fast, very accurate 'clock'!

Post by Karlosak December 20, 2006 (14 of 36)
Yes, you are right, the higher frequencies are less prone to sound refraction, so it's easier for us to simply compare which ear gets the louder sound :-)

The funny thing about this is, that the actual amount of information traveling down the aural nerve is quite small, something comparable to MP3 data rate (I don't remember the exact numbers). The brain must be using some very efficient coding then...

Post by Beagle December 20, 2006 (15 of 36)
Karlosak said:
The brain must be using some very efficient coding then...

Maybe it's dynamic coding, aka DSD ;-)

Post by DAD-Digital Audio De December 22, 2006 (16 of 36)
Yes, DXD and PCM is the same multi-bit principle, however DXD do not have an anti aliasing filter, and DXD are using 32 bit floating point in the editing system.

The 32 bit floating point in the DAW is important since audio is dynamic and fx’s and sum mixing then can calculate at the weak signals as well.

The bit resolution determines the possible dynamic range. One bit gives roughly (not precisely) 6 dB of dynamic range so I agree with Karlosak, 24 bit is more than enough for the AD conversion since the format itself has more than 144 dB of dynamic range.

In our new AD converter design we have been able to obtain a dynamic range at >122 dB and a THD+N < -118 dBfs @ 48 kHz sampling rate. Consequently we are only using 21 of the possible 24 bits when delivering the signal from our AD converter.

Regarding the discussion about multi bit/single bit it is a fact that multi bit is much more efficient that single bit, at the same bit rate.
Internally in our A/D converter we are using a 5 bit signal at 5.6 MHz when sampling for DXD or DSD.
This format has a resolution at 32 x the resolution of DSD, however the bit rate is only 10 x the bit rate of DSD.
When this is said we should keep in mind that a PCM sample always starts its life as a single bit stream.

I think the discussion about pure DSD has some of it’s foundation in the quality of the DSD modulators in the DAW systems. Here there is still room for significant improvement within the existing DSD format.
Another important issue is of course the trans-coding between DSD and a multi bit (PCM) format in the editing phase.
If a DSD signal is recorded to disk, every small changes except from cutting and pasting has to be trans-coded in order to perform the editing. I think the introduction of DXD has eliminated this problem.

Most SACD’s are up-sampled from an original PCM recording at 44.1 to192 kHz and have therefore lost the possible impulse response of the DSD format in the recording/editing phase.

For SACD lovers this is good new indeed, because better audio quality can be expected in the future with existing technology.

As Karlosak mentioned it is very true that we can't hear high frequency tones.
I do not know a lot about the ear, the aural nerve system or the brain, but I think Beagle is right, the brain must have a very fast and very accurate clock.

When we in electronic are detecting a pulse we are actually detecting a threshold. When the voltage exceeds a specified level we can conclude that we have a pulse.
If the pulse presented to our level detector only contains the lower frequency spectrum it will be detected a little slower that if is containing the higher frequency spectrum as well.
If the pulse only contains the higher frequency spectrum it will most likely not be detected at all, since the amplitude do not exceeds the threshold.

Regarding the amount of information’s in the aural nerve these pulses are analog. I think it will be more accurate to compare them wit bytes instead of bits.

Marry Christmas, Happy New Year, and best regards to everybody,

Peter

Post by Karlosak December 22, 2006 (17 of 36)
Peter, thank you for your input. I wish you and your team all the best and many splendid recordings in the forthcoming year!

Post by dvda-sacd December 22, 2006 (18 of 36)
Thank you very much, Peter.

May I send you all our very best wishes for 2007.
Merry Christmas!

Post by Windsurfer December 22, 2006 (19 of 36)
Is the 2L recording of Mozart Violin Concertos the only current DXD recording extant? For some strange reason I had believed that Channel's new Mahler 2nd was also one but I see that was an error on my part. (That Mahler is simply superb!)

Post by Julien December 22, 2006 (20 of 36)
Windsurfer said:

Is the 2L recording of Mozart Violin Concertos the only current DXD recording extant?

Apparently this one isn't DXD either:

Immortal Nystedt - Ensemble96

Page: prev 1 2 3 4 next

Closed