DSD: Explain it to me because I'm not getting it

[AudioDoctor]

SO just for the sake of saying I tried it. Can I upsample to DSD with JRiver on a Mac?

[audiventory]

My point was that practical limitations of analog implementation need to guide what is done in the digital domain. And the bandwidth is not as simple as having a single digit. Your transition band has substantial impact on your step response...

 

There are many aspects in analog reconstruction and the digital domain approach has huge impact on the net result.

 

I prefer to optimize both frequency and time domain, to get as good step response as possible too. As you can see here:

Squarewaves from DACs - Blogs - Computer Audiophile

Optimizing for both simultaneously is a nice challenge.

 

 

About DSD, audio signal band and dynamical range

After all our discussion me seems I understand, how illustrate it.

 

 

1. Maximal energy available in frame dynamical range of analog part is E

 

M=S+N

N - energy of quantization noise

 

S - maximal energy of useful signal

 

 

2. Energy of quantization noise N correlate with maximal energy of useful signal S

N=f(S)

 

3. As example, we extend bans of useful signal 2 times

Additional energy of signal birth additional energy of noise.

 

Energy of new signal will

 

E=2*S+2*N=2*(S+N)=2*M

 

I.e. we get 2 times more energy what allow dynamic range of our analog part.

Thus the analog part is overload. It appear for any band extending.

 

More band extending of useful signal - more overload.

 

If we extend useful signal band - we must limit its dynamic range.

It fairly as for DSD as for PCM.

 

However it don't cancel advantages of high sample rates - better adapting to real analog filters.

 

 

 

 

About ultrasound and square perfection

 

I agree - more band more square at output.

 

Other side, even 100 kHz or 1 MHz band can't ideally restore square. Only infinite band. Where edge of perfection?

 

Currently no reliable proofs that ultrasound impact to perception of music.

Thus if step response a bit worse due band limitation 20 … 25 kHz - according current theory, no problem for listening.

 

There are theories about wavelet analyze in brain.

 

I remember, we was discussed it some time ago.

 

However for more full checking all details of new theory need time.

 

Currently high sample rates, including DSD, very suitable for building simpler analog filters, in my opinion.

 

Possibly in future theory will changed. As usually

[bogi]

1. Maximal energy available in frame dynamical range of analog part is E

 

M=S+N

N - energy of quantization noise

 

S - maximal energy of useful signal

 

 

2. Energy of quantization noise N correlate with maximal energy of useful signal S

N=f(S)

 

3. As example, we extend bans of useful signal 2 times

Additional energy of signal birth additional energy of noise.

 

Energy of new signal will

 

E=2*S+2*N=2*(S+N)=2*M

 

I don't consider myself an expert, but I'm really unsure if this is correct. The result of oversampling is lowered quantization noise floor and wider distribution of quantization noise over frequencies.

 

From Demystifying Delta-Sigma ADCs - Tutorial - Maxim

see Figure 1,2,3

An FFT analysis shows that the noise floor has dropped. SNR is the same as before, but the noise energy has been spread over a wider frequency range.

[bmoura]

The best mastered versions I've heard of classic albums by artists like the Stones, Who, Dylan, Pink Floyd, Dire Straits, Beach Boys, The Band, Grateful Dead, Rickie Lee Jones, Tom Waits and others are in SACD or DSD format. This seems likely to continue through outlets like Mobile Fidelity, Acoustic Sounds, and CD Japan.

 

And don't forget Audio Fidelity when it comes to SACD remasters.

https://audiofidelity.net/category/sacd

[audiventory]

I don't consider myself an expert, but I'm really unsure if this is correct. The result of oversampling is lowered quantization noise floor and wider distribution of quantization noise over frequencies.

 

From Demystifying Delta-Sigma ADCs - Tutorial - Maxim

see Figure 1,2,3

 

Yes. If be exact (I must some correct previous post):

 

1. Energy of quantization noise N correlate with energy of useful signal S

N=f(S)

 

S - energy of useful signal

 

 

 

 

2. Noise quantization distributed between signal band and non-signal band

N=Ns+Nns

 

Ns - noise energy in signal band

 

Nns - noise energy out of signal band

 

If band is wide but signal band is narrow (in comparing first one)

Nns significantly greater than Ns.

For DSD and DXD we displace quantization noise Ns to Nns (out signal band). It called as "noise shaping".

 

For DSD and DXD (that like to DSD in PCM frame) noise in signal band Ns in ideal go to zero.

 

 

 

3. Thus for simplification we can accept

N=Nns

Therefore we can use formula:

 

Maximal energy available in frame of dynamical range of analog part

M=S+N

N - energy of quantization noise

S - maximal energy of useful signal

[Hiro]

My point was that practical limitations of analog implementation need to guide what is done in the digital domain. And the bandwidth is not as simple as having a single digit. Your transition band has substantial impact on your step response...

 

There are many aspects in analog reconstruction and the digital domain approach has huge impact on the net result.

 

I prefer to optimize both frequency and time domain, to get as good step response as possible too. As you can see here:

Squarewaves from DACs - Blogs - Computer Audiophile

Optimizing for both simultaneously is a nice challenge.

 

I'm surprised that anyone would want to totally ignore one of these two basic aspects of audio performance. But lo and behold, it turns out that some do just that!

[bogi]

For DSD and DXD we displace quantization noise Ns to Nns (out signal band). It called as "noise shaping".

 

Oversampling without noise shaping also lowers noise floor by changing it's distribution within wider frequency band. For more than 2x oversampling the majority of quantization noise appears in inaudible range even without noise shaping.

 

Noise shaping only moves yet more quantization noise from audible to inaudible band.

[audiventory]

Oversampling without noise shaping also lowers noise floor by changing it's distribution within wider frequency band. For more than 2x oversampling the majority of quantization noise appears in inaudible range even without noise shaping.

 

Noise shaping only moves yet more quantization noise from audible to inaudible band.

 

Yes. Noise shaping is one of ways that allow decrease level of audible noise. It is base of DSD.

 

DXD, as DSD's derived, at high frequencies has more noise energy than usual "native" PCM after applying of noise shaping.

 

Here need remember that here we talk not about noise floor (level) but about energy (summ of spectral components).

 

I.e. we have two alternatives avoid overload:

 

1. Use 1 sine signal with level 0 dB, or

 

2. Use 2 sine signal with level -6dB (2 times less) each.

 

If we use 1 sine -6 dB, 2 sine -5 dB we will get overload.

[Hiro]

DXD, as DSD's derived, at high frequencies has more noise energy than usual "native" PCM after applying of noise shaping.

 

DXD (played as PCM) has more high frequency noise energy than DSD128 played via native DSD DAC. As you can see below, DSD128 has still 100dB of dynamic range at 100kHz. 2L records DXD recordings I've seen measured exhibited higher levels of noise.

 

[audiventory]

Dynamic range is constant value of energy M. More exactly Power (energy per second).

 

Here we have only possibility balance between noise energy N and signal energy S.

 

M=S+N

 

Or more noise, or more signal.

 

If we want save level signal and get more wide band we must decrease energy (not level) of noise.

 

Decreasing of noise energy solved via filtration.

 

If I don't mistaked, DXD compatible devices must filter high frequency noise.

[Hiro]

If I don't mistaked, DXD compatible devices must filter high frequency noise.

 

I would think so. Though I'm not sure if they do.

 

If one were to play them on a PCM DAC, you would end up having higher noise than with DSD128 played on a DSD DAC (noise floor still at -100dB at 100kHz).

[Miska]

If I don't mistaked, DXD compatible devices must filter high frequency noise.

 

Of course all DACs must have a reconstruction filter. For PCM there are the images above fs/2. Usually DACs that support DXD have a second or third order analog LPF with fc somewhere lower, typically around 100 kHz.

 

DXD itself is just ordinary PCM. Since for example in 2L case it is down-conversion from the original ADC's SDM there's leftover ultrasonic noise.

 

There's similar leftover noise for example in 192k PCM output of my Focusrite Forte ADC:

...which is of course also sigma-delta ADC (just like all audio ADC chips currently in production).

[Miska]

DXD, as DSD's derived, at high frequencies has more noise energy than usual "native" PCM after applying of noise shaping.

 

Not necessarily, if you look both at the same bandwidth. You need to remember that PCM replicates it's spectra twice at every multiple of fs...

 

The bigger problem is usually that the ultrasonic images in PCM are fully correlated with the base band signal, while the ultrasonic noise shaping noise of DSD (when correctly done) is uncorrelated with the baseband signal.

[Hiro]

Of course all DACs must have a reconstruction filter. For PCM there are the images above fs/2.

 

It's funny that some people measure DSD without LPF / reconstruction filter, and they think they are making a point.

[audiventory]

while the ultrasonic noise shaping noise of DSD (when correctly done) is uncorrelated with the baseband signal.

 

Yes. Here possibly add what uncorrelated by phase (if said rough - wave form), but correlated by energy.

 

If input signal of sigma delta modulator is zero:

unfiltered DSD has oscillations about zero, but filtered signal at DAC output has zero level (in ideal).

 

If at input the modulator appear ordinary signal:

in unfiltered DSD at spectrum rise "hump" of shaped noise.

[Hiro]

Not necessarily, if you look both at the same bandwidth. You need to remember that PCM replicates it's spectra twice at every multiple of fs...

 

The bigger problem is usually that the ultrasonic images in PCM are fully correlated with the base band signal, while the ultrasonic noise shaping noise of DSD (when correctly done) is uncorrelated with the baseband signal.

 

People who think that PCM isn't generating ultrasonic noise should only take a look at this measurement.

 

[Jud]

SO just for the sake of saying I tried it. Can I upsample to DSD with JRiver on a Mac?

 

Yes. That does at least offline upsampling, don't know about inline.

 

I'd also suggest you might try HQPlayer for inline upsampling and/or Audiophile Inventory for offline upsampling if you have the time and inclination.

[Paul R]

Yes. That does at least offline upsampling, don't know about inline.

 

I'd also suggest you might try HQPlayer for inline upsampling and/or Audiophile Inventory for offline upsampling if you have the time and inclination.

 

JRMC does on the fly transcoding of PCM to DSD64 or DSD128 on the Mac, easily and without a lot of drama. Since it generates and keeps its own database, you can run it on the same media files as anything else without issue too. Try JRemote with it on an iPad.

[Daudio]

Benchmark is a bit of an odd duck there, as they already up-sample everything inbound to the DAC to some high rate, but last I looked that rate was less than 192K and way less than DSD data rates. That's probably changed a bit on the Benchmark 2, ...

 

Paul, FYI

 

The original DAC-1 Series with an Analog Devices AD1853 dac chip, Ultralock ASRC worked at 111khz, and only supported USB 1.1. Thus 96k was the highest true sample rate supported. The DAC-1's accepted 192k input streams, but then got Sample Rate Converted down to 110k.

 

The DAC-2 series with ESS Sabre32 ES9018 dac chip, Ultralock2 ASRC works at 211khz, and supports USB 2 (Async) and 192k sample rates. The Ultralock ASRC is supposedly out of the circuit for the USB input*, and Analog, of course.

 

* http://forum.benchmarkmedia.com/discuss/forum/other/benchmark-dac2-hgc-usb-and-asrc

[alfe]

JRMC does on the fly transcoding of PCM to DSD64 or DSD128 on the Mac, easily and without a lot of drama. Since it generates and keeps its own database, you can run it on the same media files as anything else without issue too. Try JRemote with it on an iPad.

 

Today is Sunday...

[Argopo]

JRMC does on the fly transcoding of PCM to DSD64 or DSD128 on the Mac, easily and without a lot of drama. Since it generates and keeps its own database, you can run it on the same media files as anything else without issue too. Try JRemote with it on an iPad.

 

When JRMC resamples hi-rez PCM files (24/96 or 24/192) to DSD128, I get static pops on playback. I assume this has to do with the maths that are being interpolated and the data being received by the DAC chip which is confusing it.

 

For now, when I want to playback DSD or resample PCM to DSD128, I use HQPlayer. (There are no static pops when resampling hi-rez PCM to DSD.)

 

JRMC is currently setup to resample to 352.8KHz or 384KHz depending on source PCM material.

[audiventory]

JRMC does on the fly transcoding of PCM to DSD64 or DSD128 on the Mac, easily and without a lot of drama. Since it generates and keeps its own database, you can run it on the same media files as anything else without issue too. Try JRemote with it on an iPad.

 

Here possibly look DSD converters comparison:

 

Archimago's Musings: ANALYSIS: A Comparison of DSD Encoders & Decoders (KORG AudioGate, JRiver MC, Weiss Saracon)

 

Archimago's Musings: ANALYSIS: DSD-to-PCM 2015 - foobar SACD Plug-In, AuI ConverteR, noise & impulse response...

 

Most good thing that these tests possibly repeat even at home

[AudioDoctor]

Yes. That does at least offline upsampling, don't know about inline.

 

I'd also suggest you might try HQPlayer for inline upsampling and/or Audiophile Inventory for offline upsampling if you have the time and inclination.

 

JRMC does on the fly transcoding of PCM to DSD64 or DSD128 on the Mac, easily and without a lot of drama. Since it generates and keeps its own database, you can run it on the same media files as anything else without issue too. Try JRemote with it on an iPad.

 

OK, I will give it a try and send upsampled DSD to my DAC and see what happens...

[Miska]

Yes. Here possibly add what uncorrelated by phase (if said rough - wave form), but correlated by energy.

 

No, there is constant full band energy with DSD...

 

If input signal of sigma delta modulator is zero:

unfiltered DSD has oscillations about zero, but filtered signal at DAC output has zero level (in ideal).

 

If at input the modulator appear ordinary signal:

in unfiltered DSD at spectrum rise "hump" of shaped noise.

 

Noise should be constant regardless of the input signal... You certainly want to avoid phenomenon called noise modulation.

[Jimmypowder]

What software are you using to do the conversion?

 

Audirvana+ and Onkyo player on apple devices.