• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

DSD is it of any value

mocenigo

Major Contributor
Joined
Dec 8, 2018
Messages
1,274
Likes
1,034
The top performers from AKM, Cirrus, ESS, and TI are all multi-level designs. Achieving comparable performance with a single-bit design would be very difficult if not impossible.

This is true.
 

Tom C

Major Contributor
Joined
Jun 16, 2019
Messages
1,501
Likes
1,370
Location
Wisconsin, USA
It is not just about the bitrate: DSD is very inefficient at information storage, despite the compression. So at the same bitrate, PCM is going to hold more information.

Where DSD wins is if you build a true single bit “chipless” DAC. Just a flip flop, an output filter, a buffer. Then you potentially get all the advantages in terms of sound quality — if subjective reports are to be believed.
I’m surprised no one has produced a vacuum tube multivibrator that does this. Two small tubes and a power supply. Cheap to build, and you could charge a fortune.
 

ZolaIII

Major Contributor
Joined
Jul 28, 2019
Messages
4,069
Likes
2,409
I’m surprised no one has produced a vacuum tube multivibrator that does this. Two small tubes and a power supply. Cheap to build, and you could charge a fortune.
You think so?
https://patents.google.com/patent/US2927962A/en
This is before it got PDM name at the same time same year first silicon based transistor whose patented by TI, Bell Labs bipolar Germanium one's from 1947 ware pretty much tube alike.
 

Tom C

Major Contributor
Joined
Jun 16, 2019
Messages
1,501
Likes
1,370
Location
Wisconsin, USA
Of course, there was a time when most or all digital processing was done by vacuum tubes one bit at a time, and low power computers filled a room. But I was thinking of something in the present day, specifically for audio. But you make a good point: just because I’ve never heard of it, doesn’t mean it doesn’t exist. Maybe there is one out there, somewhere.
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
2,638
Likes
3,599
Location
Sweden, Västerås
Yes, you would not normally store data in that encoding, if you even have access to it. I'm aware of only a few chips that expose these signals externally.

.... then also DSD input to these DAC chips will eventually be turned into these internal formats for further processing I must assume , but I would not know I don't design these ? Would not a well designed modern DAC chip be input agnostic as everything ends up the same way just before it turns to analog ?
 

mocenigo

Major Contributor
Joined
Dec 8, 2018
Messages
1,274
Likes
1,034
Is it not sometimes 5 level sigma delta some internal format that does not at all look like any consumer or pro format just something suitable for the inner workings of the chip ?

Some chips have way more than 5 levels. I think Sabre chips use 6 bits, so they have 64 levels, for instance. AKM has now also 7-bit solutions.
 

mocenigo

Major Contributor
Joined
Dec 8, 2018
Messages
1,274
Likes
1,034
Yes, you would not normally store data in that encoding, if you even have access to it. I'm aware of only a few chips that expose these signals externally.

How many are there? I know only of the AK4191, which is in fact a pure modulator.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
.... then also DSD input to these DAC chips will eventually be turned into these internal formats for further processing I must assume , but I would not know I don't design these ? Would not a well designed modern DAC chip be input agnostic as everything ends up the same way just before it turns to analog ?
Many, if not most, DAC chips do indeed process DSD input digitally. I still consider DSD an inferior format to PCM since creating it in the first place is difficult to do without adding distortion. Even three-level sigma-delta is much easier to deal with.

At the A/D and D/A conversion interfaces, the best results are achieved using a fairly small number of levels at a high sample rate along with sigma-delta noise shaping. On the digital side, 24-bit PCM at normal sample rates is far more convenient, and conversion from the raw ADC encoding to 24-bit involves nothing more than a digital low-pass filter. Moreover, since DAC chips can have widely varying numbers of levels and sample rates internally (chosen by the designers to meet particular goals), using normal PCM at the primary interface is again the most sensible choice. To produce two-level (DSD) output, an ADC chip has to run the raw format through a second digital sigma-delta stage. That is obviously not doing anyone any good. Here's a block diagram from a TI data sheet:
1619258724587.png
 
Last edited:

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
Some chips have way more than 5 levels. I think Sabre chips use 6 bits, so they have 64 levels, for instance. AKM has now also 7-bit solutions.
Be careful about interpreting such numbers. Those bits may not represent a normal two's complement encoding. 6 bits could also mean 7 levels represented as the number of ones in each value.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
How many are there? I know only of the AK4191, which is in fact a pure modulator.
That chip does digital oversampling and sigma-delta modulation. It is intended to be used with the AK4498EQ DAC chip.

On the ADC side, the TI PCM4222 can provide the raw 6-bit modulator output. There may be others with similar capabilities.
 

Pluto

Addicted to Fun and Learning
Forum Donor
Joined
Sep 2, 2018
Messages
990
Likes
1,631
Location
Harrow, UK
If anyone could explain this last part with DR slight increase after normalisation for DSD track's I would love to hear/learn about it.
Simple rounding error, in all likelihood.

To what level were you normalizing?
 

charleski

Major Contributor
Joined
Dec 15, 2019
Messages
1,098
Likes
2,239
Location
Manchester UK
Note that every single DSD release out there (unless it's some kind of single-take affair like direct-to-disc vinyl recordings were once hyped) will have massive noise-shaped ultrasonic noise due to the fact that the bitstream needs to be converted to PCM for editing, mixing and mastering, and then reconverted to DSD.
The noise-shaping is inherent to the process and will be present even if you record straight to DSD. It's got nothing to do with format conversions, though these just add to the problem. This is because oversampling is less efficient in terms of increasing SNR than adding bits: each bit gives you another 6dB of SNR, but you need to increase the sample rate 4 times to achieve the same result. For raw DSD to have the same SNR as redbook CD it would need a sampling rate over 16000 times higher. Noise-shaping is absolutely vital for DSD and the only reason the format works at all.

In general, SNR = 6.02N + 1.76dB +10log (fs/(2*BW))
where N = no of bits, fs = sampling rate and BW = signal bandwidth

The biggest flaw in the whole DSD enterprise is that you can only do DSP with a PCM stream, so it needs to be converted to a sensible format anyway. Running your hifi without DSP simply means you're depriving yourself of one of the most effective ways of improving sound quality.
 

mocenigo

Major Contributor
Joined
Dec 8, 2018
Messages
1,274
Likes
1,034
Be careful about interpreting such numbers. Those bits may not represent a normal two's complement encoding. 6 bits could also mean 7 levels represented as the number of ones in each value.

True. I have seen designs where the bits are just in parallel instead of in a ladder, and to reduce noise and distortion the allocation is randomised once the circuitry has determined the weight to use.
 

xerxesro

Member
Joined
Apr 7, 2021
Messages
10
Likes
2
I really don't follow people talking nonsense around hire. Bit is bit no matter if it's written in word or as a stream. DSD 64 bit stream is equal regarding data amount as PCM 176400 Hz 16 bit. DSD as a format is only relevant for music labels because it provides them a good protection. DSD can be compressed as any other data. However PDM stream is great for transport protocol to DAC or from ADC in DSD or any other form because simplicity and efficiency. DSD is not editable without converting and it's stiff (regarding it's layout) and proprietary format. I think another transport protocol based on PDM is needed which will be more flexible and open source but I don't mean repackaged DSD like DoP but fundamentally different regarding unfolding/processing.

Most people around here talk nonsense and throw "technical" terms around without understanding them. They probably feel good if they talk about "noise shaping" and the like. I bet almost no one knows what that means.

Anyhow, in terms of the theoretical amount amount of information that can be stored, DSD64 is equivalent to 24/88.2kHz. But that doesn't really mean much, a few of those bits are actually noise, not information.
DSD vs PCM is nothing but one encoding format vs another. There are advantages and disadvantages to both but the only real difference is how your DAC converts these into analogue. Some turn all input to PCM or to DSD, some upsample PCM signal but not DSD signal, the digital filters may not have similar quality etc. It's not the format itself but rather how it is used. Like you said, PCM is simpler to work with, but DSD is simpler to transport as electrical signal.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
True. I have seen designs where the bits are just in parallel instead of in a ladder, and to reduce noise and distortion the allocation is randomised once the circuitry has determined the weight to use.
It's called dynamic element matching, and most high-performance DACs use it.
 
Top Bottom