Inside High res music: Jazz at the Pawnshop NAXOS (DSD)

[Frgirard]

It does. for 192Khz its approx 1/(192000*2^16).

Note by the way that this for a FS signals. The number of quantisation steps that can capture the phase of a signal is less at lower ampiltudes.
Maybe we should do an experiment to show this

Please read the Redbook and others books like this

http://www.dspguide.com/ch14/2.htm

 

[amirm]

@amirm
In the dsd (SACD) standard they add a 50KHz LP filter after the 1 bit dac to filter out 'HF' noise.
I assume your Adobe sw does not emulate that when showing these spectrum images?

Today people use DACs not SACD players for these since we are talking about downloaded music and at multiples of DSD rate used in SACD. What analog filtering each DAC has, is unknow without measurement. But yes, the analysis is purely digital and indicative of the bits you pay for and download (in the form of file size). To wit, this *single track* is whopping 869 *megabytes* or almost one Gigabyte!!!

 

[Harry1973]

Interesting video. Some years ago I bought Soundgarden Superunknown (remastered) album (PCM 192/24).
The spectrum has always looked funny. Wondering if this is from DSD source? -I have not listened to this a lot. -The original 1994 CD has better mastering, more dynamic and over all much better in sound quality (IMO).

 

[thyristor]

DSD capable dacs are supposed to filter out that ultrasonic noise so that it doesn't end up on your speakers. It's not meant to be played.

Converting DSD to PCM with software depends on the software how it handles the ultrasonic noise. For example Saracon has a low pass filter at around 25 kHz for DSD to PCM conversion.

 

[amirm]

I don't understand the relevance or usefulness of this.
It's no secret that it's a conversion from tape. No one is being bamboozled.

If I was going to buy it, I'd buy the DXD, because it's one less generation of conversion before making the release version.

But in any case people who are buying the DSD are filtering out the ultrasonic noise on playback as part of the DSD>analog process.

See my answer above. The tape did not have the severe noise shaping. Nor do you know what your DAC will do when playing DSD256 as far as analog filtering. Don't confuse SACD players and today's DxD digital downloads. The former had some governance as far as recommended analog filtering but the latter does not. If all ultrasonics were filtered, no one would buy these tracks as they would not be considered "high-res" anymore.

Just looked up AKM4493 and this is what they say for DSD filtering:

They have two filter settings. For DSD64 (SACD) you easily have bandwidth to 50 kHz in setting 1 and 100 kHz in Setting 2. If you use DSD 128, it doubles these. The track I analyzed is DSD256 which means filter 1 gives you 200 kHz bandwidth and filter 2, whopping 400 kHz bandwidth!

So no way you are filtering the ultrasonics for such content. The whole point of the format is to produce ultrasonics or it wouldn't be "high-res."

 

[Hayabusa]

Please read the Redbook and others books like this

http://www.dspguide.com/ch14/2.htm

Ok, it seems I have to make an example how amplitude quantization has influence on temporal resolution...
Note that I think Redbook is perfect enough to capture temporal resolution for human ears
I only want to state that if you really go down to last pico seconds there is more temporal resolution with higher sampling rates when using the same number of (amplitude) quantization steps.

 

[amirm]

Converting DSD to PCM with software depends on the software how it handles the ultrasonic noise. For example Saracon has a low pass filter at around 25 kHz for DSD to PCM conversion.

For what DSD rate? I just looked up its specs and it doesn't even handle DSD256: https://www.weiss.ch/products/saracon

Sampling frequencies

• Sampling frequencies supported: 44.1 kHz, 48 kHz, 88.2 kHz, 96 kHz, 176.4 kHz, 192 kHz, 352.8 kHz, 384 kHz, DXD, DSD64 (2.8224 MHz), DSD128 (5.6448 MHz)

 

[Frgirard]

Resolution is 2exp( nbre of bit)
Hi res is only high resolution. Nothing to do with sample rate and any frequencial content.
The resolution is used to calc the amplitude 20*log(resolution).

 

[Frgirard]

Ok, it seems I have to make an example how amplitude quantization has influence on temporal resolution...
Note that I think Redbook is perfect enough to capture temporal resolution for human ears
I only want to state that if you really go down to last pico seconds there is more temporal resolution with higher sampling rates when using the same number of (amplitude) quantization steps.

The sample rate and the resolution have no link.
Résolution =2exp(number of bit)

The sample rate: read the Shannon theorem.

 

[DSJR]

Apologies if this is drifting a bit, but it makes me laugh with these old analogue recordings being 'hi res' mastered.

When Decca were Decca and based in Belsize Road London, the masters were placed on the (Studer A80?) playback machines and wound back to the beginning (stored tails out). My mastering engineer pal told me that basically, the four test tones were used to equalise the tapes (Decca had their own eq slopes, so lord knows how 'flat' third party Decca masterings are today) and the 15khz or so high tone was wound up and up as the machines heads wore until finally, there was no eq room left. Only then were the heads replaced (not sure if they were 'lapped' in the interim). No idea what the 15 - 20khz level was on part or well worn playback heads.

I have some FLACS of the Kraftwerk catalogue and a crude wave editor shows output up to 17khz on the older recordings but next to nothing higher. the 'Minimum Maximum' live recordings do seem to go right up to 20kHz though.

The Pawnshop albums did 'sound' very nice, the best systems allowing you to forget the playback system and enjoy the atmosphere and 'vibe' in the venue. After what, forty years, just how 'far up' do these now elderly tapes go? I bet it's nowhere near 30khz let alone higher... Anyone interested in hearing this recording must be getting on a bit too, so what's the average hf limit for a mid fifty year old these days, one who hasn't had the benefit of 70's rock groups blasting out 120db or more at gigs (I'll never forgive led Zeppelin at earls Court in 1975, as that experience started off the Tinnitus I now severely suffer from

 

[restorer-john]

The whole point of the format is to produce ultrasonics or it wouldn't be "high-res."

Resolution and bandwidth are two different things (and I know you know that).

My concern with all things noise shaping, is the completely unnecessary and undesirable HF noise content which wide bandwidth amplification doesn't take kindly to.

My rules are simple. Analogue sources and amplification are not bandwidth limited (within reason). Digital sources should have correctly designed LPFs with no chance of aliases at significant level anywhere up the spectrum.

Your analysis is certainly helping to pour water on the Hi-Res bandwagon.

 

[voodooless]

I only want to state that if you really go down to last pico seconds there is more temporal resolution with higher sampling rates when using the same number of (amplitude) quantization steps.

Let's have some proof then

 

[Ata]

IMO, a "suspicion" becomes "spot on" only when it is confirmed. I don't see anything being confirmed in your post.

If we are to go into such deep semantic analysis, I'd say "confirmed" and "it seems" do not belong in the same sentence. Since the latter phrase is in my response, clearly nothing is confirmed...

 

[Hayabusa]

Let's have some proof then

Ok, I have been taking a closer look how to proof this
And while thinking further basically I think you are right!

The only way I could construct a different phase difference of a sampled sine wave at two different sample frequencies is a very specific one:
The sine wave has to be an integer divide of the FS
It has to be aligned with sampling and quantitation grid in such a way that all quantization errors point in one direction and in such a way that a 2 times up sampled version of the same signal has quantization errors in the other direction for the interpolated samples.

 

[voodooless]

And while thinking further basically I think you are right!

Actually, I could argue the opposite: adding more samples might yield a greater error and thus less time resolution. imagine two low sample rate samples: they are 1 LSB in value apart. When you reconstruct this you would get a nicely reconstructed smooth line between the points. Now let's add more samples in between the two points. Since there is only a 1LSB difference, somewhere you'll get a flip from one to the other value. Now when you reconstruct this, you will not have a smooth line, but rather a more sudden change in value. The error is therefore bigger.

Obviously, this is all academic. The noise in a 24-bit system is obviously far bigger than 1 LSB. In the 16 bit world, however, we can add dither to mitigate part of this issue. Noise shaping can make this frequency dependant. So you could argue that in a 16-bit world, a higher sample rate would let you do noise-shaped dither better. It's far easier to add a few bits though.

 

[Herbert]

I discovered "Jazz at the Pawnshop" at the beginning of the CD-era,
of course because of the stereo journals. Maybe this recording does
not differ fro the "established" High-End brands that get debunked here.
Always though the source was at least a decent 1" tape, but a Nagra IV?
Worked with one on location during a 35mm film shoot, we chose it over DAT back then
because we were "in the Wild" in a rather dusty and hot environment
and did not want to relate on rotating heads
I rremember mine was rather noisy.
And a portable Nagra IV indicates that there was no mobile studio to monitor the "audiophility."
of "Jazz at the Pawnshop"
So this recording seems more myth than fact, and interstingly,
besides a volume two, nothing else was released using this quite ordinary setup...

 

[thyristor]

For what DSD rate? I just looked up its specs and it doesn't even handle DSD256: https://www.weiss.ch/products/saracon

Sampling frequencies

• Sampling frequencies supported: 44.1 kHz, 48 kHz, 88.2 kHz, 96 kHz, 176.4 kHz, 192 kHz, 352.8 kHz, 384 kHz, DXD, DSD64 (2.8224 MHz), DSD128 (5.6448 MHz)

Even if you convert DSD to 24-bit 384 kHz PCM it has low pass filter at around 25 kHz. And yes it doesn't support DSD256.

It's an old program. Latest version is from 2010.

 

[dfuller]

Actually, I could argue the opposite: adding more samples might yield a greater error and thus less time resolution. imagine two low sample rate samples: they are 1 LSB in value apart. When you reconstruct this you would get a nicely reconstructed smooth line between the points. Now let's add more samples in between the two points. Since there is only a 1LSB difference, somewhere you'll get a flip from one to the other value. Now when you reconstruct this, you will not have a smooth line, but rather a more sudden change in value. The error is therefore bigger.

Obviously, this is all academic. The noise in a 24-bit system is obviously far bigger than 1 LSB. In the 16 bit world, however, we can add jitter to mitigate part of this issue. Noise shaping can make this frequency dependant. So you could argue that in a 16-bit world, a higher sample rate would let you do noise-shaped jitter better. It's far easier to add a few bits though.

You mean dither.

 

[voodooless]

You mean dither.

Indeed I doo

 

[sarumbear]

A 1966 Teac 4-Track for example:

Frequency Response @ 0 VU
@15ips = 30 to 18kHz (that's chewing through a lot of tape!)
@3 3/4 ips = 50-7500Hz

Signal To Noise @ 0 VU
1-track = 55db
4-Track = 50db

Total Harmonic Distortion
1% @ 0 VU (so it's a lot higher than that in real world use)

Allow me to correct the record.

1- The recording was done a decade later in 1976

2- I personally heard from the recording engineer Gert Palmcrantz that he used a pair of Nagra IV-S recorders through a pair Dolby A 361 noise reduction units (I was working at Abbey Road then).

3- Nagra IV-S has a 20-20,000Hz +/-2dB response at 15ips, which was the speed used. Specs…