Jump to content
Sign in to follow this  
blownsi

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

Rate this topic

Recommended Posts

Hi Miska,

 

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

 

I suppose you wanted to say "constant noise energy", not "full band energy", isn’t it?

 

Full band energy I consider as signal + noise.

 

Full band energy limited by dynamical range of analog part.

 

If noise energy is constant, any trying of transmiting of useful signal with band 80 or 1000 kHz lead to overload. Of course we must consider energy transmitted in band. Noise energy must be moved out new useful band via noise shaping.

 

 

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

 

You are right. It is not counter that you quote. Unfiltered (straight from DSF/DFF) DSD has oscillations about zero (-1/+1). Yes. It's modulation. After DSD demodulator's filter it become zero. Due these oscillations is high frequency noise :)

 

P.S. I like that the discussion order in brains some things. Ideas for next researchings.

 

Best regards,

Yuri

Edited by audiventory

AuI ConverteR 48x44 - HD audio converter/optimizer for DAC of high resolution files

ISO, DSF, DFF (1-bit/D64/128/256/512/1024), wav, flac, aiff, alac,  safe CD ripper to PCM/DSF,

Seamless Album Conversion, AIFF, WAV, FLAC, DSF metadata editor, Mac & Windows
Offline conversion save energy and nature

Share this post


Link to post
Share on other sites
I suppose you wanted to say "constant noise energy", not "full band energy", isn’t it?

 

Yes, to be clear I meant to say that the energy within fs/2 band is constant...

 

If noise energy is constant, any trying of transmiting of useful signal with band 80 or 1000 kHz lead to overload.

 

Certainly you can encode 80 kHz signal with DSD, no problem. Based on certain rules you'll need to give up on the peak level though.

 

Here is Schiit Loki DAC playing back 250 kHz sine using DSD64. SNR at that point is not so great anymore, but the output is still clearly measurable:

Loki-250k-sine.png


Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Share this post


Link to post
Share on other sites
Of course all DACs must have a reconstruction filter. For PCM there are the images above fs/2.

 

By the way, Miska, you made this measurement of one pcm r2r Burr-Brown dac some time ago and brickwall DF chip.

 

PCM1700-sweep-wide.png

 

Do I understand it correctly that what we are seeing here are the residual ultrasonic images from PCM coding that passed the brickwall?

Share this post


Link to post
Share on other sites
By the way, Miska, you made this measurement of one pcm r2r Burr-Brown dac some time ago and brickwall DF chip.

 

Yes, this is their DF1700 digital filter plus PCM1700 DAC chip.

 

Do I understand it correctly that what we are seeing here are the residual ultrasonic images from PCM coding that passed the brickwall?

 

Since the brickwall is digital filter, it has maximum output sampling rate, which in this case as in most DAC chips these days is 8x, meaning 352.8/384 kHz. Due to this, the signal spectrum is replicated twice, around each multiple of the DF output sampling rate (every 352.8 kHz in this case, up to around 4 MHz). Idea of oversampling is to move these images high enough in frequency so that analog filter after the DAC chip can successfully remove those. In this case the result is better (lower image level) than with many modern DACs because this DAC supports input sampling rates only up to 48 kHz. This allows the analog filter fc (cutoff frequency) to be placed low at 25 kHz.

 

Cutoff frequency of analog filters in modern DACs are quite a bit higher, closer to 100 kHz to support wide pass-band of hires content and to have better phase response in audio band (less phase shift in top octave). But still the oversampling digital filter output sampling rate is the same. Higher source rates are supported by just dropping necessary number of 2x filter stages from the beginning. Rest of the rate increase is commonly done using sample-and-hold (copying same sample value N times), which produces same result as a ladder DAC running at DF output rate would - because it's same operation.

 

This may result in discrete, correlated intermodulation tones in audio band if the analog filter cannot remove all the images. For example difference tone of 1 kHz due to images is 2 kHz, that you can see for example from this zoom-in of 1 kHz tone being played:

iDSD-PCM-image.png

Left is the negative frequency, 352.8 - 1 and the right one is the positive frequency, 352.8 + 1. More there are images, more the intermodulation products add up.

 

There always needs to be an analog reconstruction filter. For perfect filtering purely in digital domain, the digital filter output sampling rate would need to be infinitely high. There's a word for "infinitely high sampling rate", that's called "analog"... ;)


Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Share this post


Link to post
Share on other sites

 

There always needs to be an analog reconstruction filter. For perfect filtering purely in digital domain, the digital filter output sampling rate would need to be infinitely high. There's a word for "infinitely high sampling rate", that's called "analog"... ;)

 

So Miska, now you are turning to vinyl???

;-)

Share this post


Link to post
Share on other sites

Miska only said that analog low pass (reconstruction) filter should follow digital oversampling filters in DAC.

 

That's common for R2R and delta-sigma DACs, but they use different D/A conversion stage between digital oversampling and the analog reconstruction filters.

 

In the case of delta-sigma DAC: Between oversampling filters and and the analog reconstruction filter yet delta-sigma modulator and the D/A conversion stage is placed. Delta-sigma modulator generates pulses from n-bit PCM input, it has one bit or more bit 2 level or multilevel output. This pulse delta-sigma modulator output is then somehow summed into analog signal and low pass filtered to become "more contiguous" analog signal. Depending on the quality of this low pass filter the DAC analog output may yet contain those high frequencies, which came from oversampling and therefore subsequent analog filter may be yet needed. It depends on how much analog filtering is performed in the DAC chip itself, that gives requirements for analog stage, following the DAC chip. In the case of Sabre chips that analog stage is very important. Some DAC chips do complete analog filtering on chip - for example my old iPod 5.5G with Wolfson WM87588G - that's not only DAC, it is special chip for digital audio players.

 

Miska, is my understanding correct, at least on popular scientific level? :D

Edited by bogi

Share this post


Link to post
Share on other sites
Yes, this is their DF1700 digital filter plus PCM1700 DAC chip.

 

 

 

Since the brickwall is digital filter, it has maximum output sampling rate, which in this case as in most DAC chips these days is 8x, meaning 352.8/384 kHz. Due to this, the signal spectrum is replicated twice, around each multiple of the DF output sampling rate (every 352.8 kHz in this case, up to around 4 MHz). Idea of oversampling is to move these images high enough in frequency so that analog filter after the DAC chip can successfully remove those. In this case the result is better (lower image level) than with many modern DACs because this DAC supports input sampling rates only up to 48 kHz. This allows the analog filter fc (cutoff frequency) to be placed low at 25 kHz.

 

Cutoff frequency of analog filters in modern DACs are quite a bit higher, closer to 100 kHz to support wide pass-band of hires content and to have better phase response in audio band (less phase shift in top octave). But still the oversampling digital filter output sampling rate is the same. Higher source rates are supported by just dropping necessary number of 2x filter stages from the beginning. Rest of the rate increase is commonly done using sample-and-hold (copying same sample value N times), which produces same result as a ladder DAC running at DF output rate would - because it's same operation.

 

This may result in discrete, correlated intermodulation tones in audio band if the analog filter cannot remove all the images. For example difference tone of 1 kHz due to images is 2 kHz, that you can see for example from this zoom-in of 1 kHz tone being played:

[ATTACH=CONFIG]18211[/ATTACH]

Left is the negative frequency, 352.8 - 1 and the right one is the positive frequency, 352.8 + 1. More there are images, more the intermodulation products add up.

 

There always needs to be an analog reconstruction filter. For perfect filtering purely in digital domain, the digital filter output sampling rate would need to be infinitely high. There's a word for "infinitely high sampling rate", that's called "analog"... ;)

 

Thanks Miska for explaining in such a detailed way the inherent problems related to PCM coding and filtering. Very informative read.

Share this post


Link to post
Share on other sites
Miska, is my understanding correct, at least on popular scientific level? :D

 

Yes, that's the case.

 

There are four main differences between PCM and SDM DACs.

1) Sampling rate.

2) Word length.

3) Word encoding.

4) Type of ultrasonic noise in the DAC output we are discussing here. PCM DAC has ultrasonic image frequencies which are fully correlated with the base-band signal. SDM DAC has ultrasonic noise which is on purpose decorrelated from the base-band signal.

 

It is also possible to make a hybrid DAC like TI has been doing for quite a while for PCM inputs. Sort of running two different paths in parallel with a certain job split.


Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Share this post


Link to post
Share on other sites
It is also possible to make a hybrid DAC like TI has been doing for quite a while for PCM inputs. Sort of running two different paths in parallel with a certain job split.

 

It's interesting to note that in such hybrid DAC the ultrasonic noise from the correlated PCM images is much higher than from the delta-sigma modulation, as can be seen in the measurement below.

 

dac_output.png

Share this post


Link to post
Share on other sites
I've always liked vinyl, and tape! :)

 

It would be great if the recent vinyl revival spurred some new developments in tape recording technology, maybe even a new analog medium for studio purposes offering 120dB of dynamic range, and top notch transparency. Would transfer great direct to DSD 11.2 ;)

 

Unfortunately we haven't seen anything like it so far, as far as I'm aware. And many companies are going the easy route, simply putting their pcm recordings on vinyl...

Edited by Hiro

Share this post


Link to post
Share on other sites
It's interesting to note that in such hybrid DAC the ultrasonic noise from the correlated PCM images is much higher than from the delta-sigma modulation, as can be seen in the measurement below.

 

It is the case for two reasons.

 

Primary reason is that the digital filter, as typical goes to 352.8/384k which is also output rate of the "PCM side", but also because further oversampling is done using sample-and-hold which means that there's no filter and thus the output spectrum is not affected by the sampling rate increase.

 

So you can see similar output from SDM DAC chips like the AD1955 or AK4399 (8x DF):

AK4399-sweep441-wide.png

 

Here is E-MU 0404 USB:

EMU0404-sweep441-wide.png

 

...this is why I like to do proper digital filtering and modulation before passing the data to DAC... :)

 

Second reason is that top 6 bits of the input are handled by the "PCM side" and the rest 18 bits are handled by the 5-level "SDM side" because it improves low level linearity drastically. They also need to do it this way, because the modulator is only 3rd order and it wouldn't be good enough alone.

 

Essentially this means that the DAC becomes pure SDM DAC below -36 dBFS input levels. Note! That also means that sample values on any level waveform that don't fall into range +- 8388608 - 8388576 are handled only by the SDM. So it is just really the waveform peak values where the "PCM side" kicks in.


Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Share this post


Link to post
Share on other sites
So you can see similar output from SDM DAC chips like the AD1955 or AK4399 (8x DF):

[ATTACH=CONFIG]18299[/ATTACH]

 

Here is E-MU 0404 USB:

[ATTACH=CONFIG]18300[/ATTACH]

 

...this is why I like to do proper digital filtering and modulation before passing the data to DAC... :)

 

 

Makes sense, as these PCM noise components still reach -65 to -45dB.

 

I have long argued that all modern delta-sigma DACs should have direct sigma delta inputs.

Edited by Hiro

Share this post


Link to post
Share on other sites

Hi Miska,

 

Yes, to be clear I meant to say that the energy within fs/2 band is constant...

 

Here we say about DSD signal? You want say «energy within [25-30 kHz … fs]»?

 

 

 

 

Certainly you can encode 80 kHz signal with DSD, no problem. Based on certain rules you'll need to give up on the peak level though.

 

Here is Schiit Loki DAC playing back 250 kHz sine using DSD64. SNR at that point is not so great anymore, but the output is still clearly measurable:

[ATTACH=CONFIG]18208[/ATTACH]

 

Interesting to see full spectrum from zero to fs.

 

 

...this is why I like to do proper digital filtering and modulation before passing the data to DAC... :)

 

Yes. Desktop PC usually have more CPU performance and memory than single chip decisions. It allow simpler apply and debug more sophisticated math.

 

Single chip decisions has such advantage as fast start.

 

At FPGA we can realize more fast decisions than PC. However we meet restrictions like ratio [elementary components number] / price for floating point math.

 

Best regards,

Yuri

Edited by audiventory

AuI ConverteR 48x44 - HD audio converter/optimizer for DAC of high resolution files

ISO, DSF, DFF (1-bit/D64/128/256/512/1024), wav, flac, aiff, alac,  safe CD ripper to PCM/DSF,

Seamless Album Conversion, AIFF, WAV, FLAC, DSF metadata editor, Mac & Windows
Offline conversion save energy and nature

Share this post


Link to post
Share on other sites
Here we say about DSD signal? You want say «energy within [25-30 kHz … fs]»?

 

Since there's no clear boundary between audio and noise bands, I don't like to draw any specific number there. The SNR just decreases as function of frequency depending on the used modulator.

 

At FPGA we can realize more fast decisions than PC. However we meet restrictions like ratio [elementary components number] / price for floating point math.

 

There are many severe restrictions on configurability, FLOPS/Watt, complexity and price/performance with FPGA. If you want to have high performance for something simple and fixed like AES encryption, then FPGA is nice. If you'd like to have something flexible and adaptive, then it is not good. For example HQPlayer can deal with any input sampling rate, not just the usual ones. Same for output sampling rates too. So something like 88207 Hz sampling rate would be just fine.

 

Latest Intel CPUs are manufactured in 14 nm process, giving pretty good FLOPS-Watt-flexibility ratio...


Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Share this post


Link to post
Share on other sites
Since there's no clear boundary between audio and noise bands, I don't like to draw any specific number there. The SNR just decreases as function of frequency depending on the used modulator.

 

Yes. Noise management is one of main part of sigma delta modulator / demodulator. Right combinations modulator / demodulator able provide proper quality.


AuI ConverteR 48x44 - HD audio converter/optimizer for DAC of high resolution files

ISO, DSF, DFF (1-bit/D64/128/256/512/1024), wav, flac, aiff, alac,  safe CD ripper to PCM/DSF,

Seamless Album Conversion, AIFF, WAV, FLAC, DSF metadata editor, Mac & Windows
Offline conversion save energy and nature

Share this post


Link to post
Share on other sites
It would be great if the recent vinyl revival spurred some new developments in tape recording technology, maybe even a new analog medium for studio purposes offering 120dB of dynamic range, and top notch transparency. Would transfer great direct to DSD 11.2 ;)

 

Unfortunately we haven't seen anything like it so far, as far as I'm aware. And many companies are going the easy route, simply putting their pcm recordings on vinyl...

 

Negative there ghost rider, check out United Home Audio. These guys have done things with tape that are absolutely amazing. When I talked to Greg about the details I asked about noise reduction techniques (like Dolby). What I learned is that through modern technology, and a better understanding of the magnetic transport of tape, he was able to develop his own pick-up circuitry and avoid the need for noise reduction entirely. The proof is in the sound, and it sounds like no other - NO OTHER! The best DSD can come close (native DSD), and it blows vinyl away (and I'm a vinyl guy).

 

Why don't I have one? Because the software just isn't there, and that's a shame. Yes there is The Tape Project and there are some other sources floating around but you're not going to take your precious favorites and find master copies of them - at least not easily. Of course Greg has all kinds of great music, but then he has been building his tape library for over 20 years. I was saving money to buy one of their tape units when I finally decided to punt (due to the lack of software) and used that money to enter into computer audio - and here I am in the world of CA.


Analog: Koetsu Rosewood > VPI Aries 3 w/SDS > EAR 834P > EAR 834L: Audiodesk cleaner

Digital Fun: DAS > CAPS v3 w/LPS (JRMC) SOtM USB > Lynx Hilo > EAR 834L

Digital Serious: DAS > CAPS v3 w/LPS (HQPlayer) Ethernet > SMS-100 NAA > Lampi DSD L4 G5 > EAR 834L

Digital Disc: Oppo BDP 95 > EAR 834L

Output: EAR 834L > Xilica XP4080 DSP > Odessey Stratos Mono Extreme > Legacy Aeris

Phones: EAR 834L > Little Dot Mk ii > Senheiser HD 800

Share this post


Link to post
Share on other sites
Negative there ghost rider, check out United Home Audio. These guys have done things with tape that are absolutely amazing. When I talked to Greg about the details I asked about noise reduction techniques (like Dolby). What I learned is that through modern technology, and a better understanding of the magnetic transport of tape, he was able to develop his own pick-up circuitry and avoid the need for noise reduction entirely. The proof is in the sound, and it sounds like no other - NO OTHER! The best DSD can come close (native DSD), and it blows vinyl away (and I'm a vinyl guy).

 

Why don't I have one? Because the software just isn't there, and that's a shame. Yes there is The Tape Project and there are some other sources floating around but you're not going to take your precious favorites and find master copies of them - at least not easily. Of course Greg has all kinds of great music, but then he has been building his tape library for over 20 years. I was saving money to buy one of their tape units when I finally decided to punt (due to the lack of software) and used that money to enter into computer audio - and here I am in the world of CA.

 

I didn't mean to be negative at all, and would actually love to see next-generation analog solutions like the one you wrote about get traction in the recording market. Both Analog and DSD lovers could benefit from it.

 

If record labels continue to put PCM DIGITAL files on Vinyl, the analog medium which enjoys a revival today will be doomed to become, to borrow Neil Young's phrase a fashion statement.

Share this post


Link to post
Share on other sites
...

If record labels continue to put PCM DIGITAL files on Vinyl, the analog medium which enjoys a revival today will be doomed to become, to borrow Neil Young's phrase a fashion statement.

 

Have you examined the reasons for the current "vinyl revival"? I think you'll find your fears have already been realised.


"People hear what they see." - Doris Day

The forum would be a much better place if everyone were less convinced of how right they were.

Share this post


Link to post
Share on other sites
I've always liked vinyl, and tape! :)

 

Got the girlfriend a Technics SL-1800 for her birthday. I have already built a list of 19 mods I want to do to it :P


Dedicated Line DSD/DXD | Audirvana+ | iFi iDSD Nano | SET Tube Amp | Totem Mites

Surround: VLC | M-Audio FastTrack Pro | Mac Opt | Panasonic SA-HE100 | Logitech Z623

DIY: SET Tube Amp | Low-Noise Linear Regulated Power Supply | USB, Power, Speaker Cables | Speaker Stands | Acoustic Panels

Share this post


Link to post
Share on other sites
I didn't mean to be negative at all, and would actually love to see next-generation analog solutions like the one you wrote about get traction in the recording market. Both Analog and DSD lovers could benefit from it.

 

If record labels continue to put PCM DIGITAL files on Vinyl, the analog medium which enjoys a revival today will be doomed to become, to borrow Neil Young's phrase a fashion statement.

 

That presumes that people would notice. Vinyl LP releases have been sourced from PCM and DSD tapes for some time now.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

×
×
  • Create New...