Multi Bit DACs vs. Delta Sigma DACs

[Audio_ELF]

I hope I get this right Paul ... but a transducer (speaker for reproduction, microphone for recording) has a natural rest place, the "sound" is recorded as a voltage which is measured either side of the rest position (0v).

 

Eloise

 

[idiot_savant]

Eloise is correct - sound is made of waves, which are represented by samples, centred around 0.

 

Now, as for this distortion, I've been thinking about this...

 

In one respect, a DAC that separates the decoding into a +ve and -ve DAC has to be careful about the crossover point - as I've said, the PCM1704 seems pretty good in this regard.

 

As for padding the LSBs, sign dependent padding is not a good idea - in this case you definitely add a step in the transfer function - imagine you had a DAC with 1 bit of extra resolution than the input: If we input a small ramp into it going down we get:

01000 : 8

00110 : 6

00100 : 4

00010 : 2

00000 : 0

11110 : -2

11100 : -4

11010 : -6

11000 : -8

 

If we pad according to the sign we get:

01000 : 8

00110 : 6

00100 : 4

00010 : 2

00000 : 0

11111 : -1

11101 : -3

11011 : -5

11001 : -7

 

or a guaranteed half-LSB DC offset for negative samples.... ( bad ) - this is relevant because the DAC will have (say) a 24 bit input, and the source may be only 16 bits, and how do we transport from the source to the DAC?

 

 

As for the sigma-delta thing, the point is that yes, the overall noise will be higher, but you can't remove the noise shaping and still call it sigma-delta.

And the quantisation noise in-band is not limited by the number of bits, providing you do everything properly ( i.e. have enough bandwidth to put the noise in ).

According to DSD, a 1-bit sigma delta has an overall SNR of 6dB, but better than 120dB in the area of interest ( DC-20k )

 

Sigma-delta processors do not have any intrinisc problem with noise, or transients, or many of the things proponents of mult-bit DACs like to think. They do have problems with idle tones, proper dither, and the removal of the out-of-band noise, and may be more susceptible to jitter.

 

There are good multi-bit and good sigma-delta designs. There are equally poor versions of each,

 

your friendly neighbourhood idiot

 

 

[manisandher]

"Sigma-delta processors do not have any intrinisc problem with noise, or transients, or many of the things proponents of mult-bit DACs like to think."

 

In my case, it has nothing to do with thinking one thing over another. It has everything to do with hearing what I hear.

 

I don't like the sound of delta-sigma chips. Have I heard them all? No. But the ones I have heard, I don't like (including the ones I've owned in the past and the one I still own).

 

There is a very good reason why the 1704 still exists, and the likes of Naim still use it...

 

Mani.

 

[idiot_savant]

As I have said before, everybody is entitled to have a preference. I was merely responding to:

The argument against goes along the lines of:

"all remaining DAC's with > 16 Bit resolution except the PCM1704 are delta-sigma (with multibit cores), so any such debate is completely futile anyway, as such DAC's have native resolutions (before noise-shaping and other digital tricks) of much less than 16 bits".

 

Notice how I said "intrinsic" problems with transients. We've been here before, and I have no great wish to go into it all again,

 

your friendly neighbourhood idiot

 

[manisandher]

"We've been here before, and I have no great wish to go into it all again..."

 

Ditto.

 

Mani.

 

[Paul R]

Physical sound is not discrete, but rather a continuous analog signal. Whatever frequency any particular sound signal is at, the amplitude may center around zero, but the energy (power) in the signal is always positive, as zero power means zero signal.

 

So to rephrase my initial question with more clarity, why use a negative voltage when it seems just as easy to put the "zero" mark at say, 5 volts and vary the amplitude of the signal from zero volts to 10 volts. Or clamped to any voltage you want as your lowest amplitude value. Any sample at all above zero will generate a positive change in amplitude.

 

Wouldn't this eliminate the need to digitally deal with negative sample values? The actual word size of the buffer in a DAC never changes, so the data is manipulated only in software or firmware. The "centerpoint" of any signal is then easily manipulated as digital information, and can be shifted as necessary without need to worry about sign bits - just over or under flow.

 

Okay, admittedly, speaker drivers are probably just as driven by negative voltages as by positive voltages (I do not know the answer to that) but since I am not at all well grounded in AC theory (pun intended!), I find myself wallowing in a sea of ignorance. There are probably a thousand things I am not considering, especially so since things just are not designed that way. I'm kind of asking for the reasons, other than "that is just the way it is done!" And I am asking only out of curiosity, not out of a desire to argue.

 

Would you consider doing a "Signal To Sound For Idiots" essay in the not too distant future. If I am missing huge gaps in my understanding of the path from bits on a computer disk to sound waves in the ear, probably other folks have minor gaps in it. I bet everyone would welcome such.

 

I understand how sound is transmitted in a medium, I understand software logic, and I understand basic electronics, including old style discrete A/D conversions. (I worked on flight simulators for years.

 

I just cannot put all the pieces together the same way you guys do.

 

 

[PeterSt]

Hi Paul,

 

Let me try to stir it up a bit (just because Bob Marley plays that here, right now :-)

 

If you just look at the speakers, they push forward air, and distract it (suck it back !). The first is the positive voltage, the latter negative. So, it *is* negative. Really ! You can try this with a battery on a speaker driver, and swap + and -.

 

Now you know why the theoretical representative behaves the same. It is just the most logical and the most understandable (uhm, for those who understand ... nothing meant by this !).

 

All the best,

Peter

 

 

[Paul R]

For an analog driven speaker? Sure, it is a very logical setup.

 

For a digital data stream? I am not so convinced of that. Digital data streams I understand.

-Paul

 

 

[jkeny]

Paul,

It's the analogue you listen to. Work back from that - the speakere needs Pos & neg voltage to vibrate - this comes from the amplifier( Pos & neg voltages)which is driven by the ..... You get the point - it goes all the way back to the digital representation of the signal - it is recorded from analog (pos & neg) to digital (pos & neg representation of the waveform).

 

You are correct that the negative has to be artificially represented in digital by two's complement binary coding specifically to address negative values.

 

Doe sthis help?

 

[manisandher]

This is Nelson Pass's take on this:

 

http://www.passlabs.com/pdf/articles/seclassa.pdf

 

Mani.

 

[Paul R]

Indeed, this paragraph much more accurately states what I was groping around in the dark trying to put into words:

 

 

For reproducing music as naturally as possible, push-pull symmetric

operation is not the best approach. Air is not symmetric and does not

have a push-pull characteristic. Sound in air is a perturbation around a

positive pressure point. There is only positive pressure, more positive

pressure, and less positive pressure.

 

 

(Emphasis is mine.)

 

Incidentally, I remember reading the article he refers to from 1977.

 

-Paul

 

 

[The Computer Audiophile]

Hi Guys - Don't mean to barge in but I must say your knowledge sharing is much appreciated and way above my skill level.

 

Thanks!

 

[PeterSt]

Hi again Paul,

 

I can imagine that you read your own thoughts in there, but I really think it isn't so.

Also, I think the article is hardly related to anything of the subject here.

 

The "push pull" from amplifiers is about two active components creating a single wave. One side is pushing, the other side is pulling at the same time. Look at the "two men sawing a tree" example from the article, and compare this with one man doing that from one end (single end :-) only. So, this is about the better control from one side (nothing works against the mechanism), though taking (much) more power to do it. Fine.

 

How the story about air sneaked in that article, I don't know, but I guess some mystique is always good. It will be true though, but how it relates to SE being better because air behaves the same ? ... but again, fine.

 

To keep within the context of the story (for those who like that), the loudspeaker driver will be a single ended source. Nice ... (but put a driver at the opposite wall of the room, and when at the proper position for phase (impossible for all frequencies) you'll have a push pull setup. Good.

 

The general pressure of air will be 1 ATO. Or whatever. This is always positive. Good.

 

When my speaker driver pushes forward the pressure increases by a fraction. The larger the driver / the larger the excursion, the more that will be so. When - half of the frequency later - my driver pulls back, pressure first will become the standard (1 ATO) again (zero crossing), and then will get negative - relative to the base. Whether the negative and positive are not linear (not as much but opposite) I don't know. Nice story. Should be correct.

 

Fact is though, when 0 voltage is fed to a loudspeaker driver, it will be at rest. Nothing moves, no pressure differences will be created;

When you feed the driver with a negative voltage (try a 6V battery with plus and minus switched) the driver will pull back. As long as you leave the battery on, it will stay at this pulled back state and the amount of pulling back was denoted by the voltage (-6 in this case);

Take out the battery, and the driver will be in the rest position again.

Put the plus to plus and minus to minus, and the driver will push forward one time. Again it will stay there, as long as the battery is full.

 

This is how it was setup long ago, and it will not change. The whole chain anticipates on that, and maybe it is so that magnetic tape works with plus only (I just don't know) and has to be corrected for that reason.

 

The (also) fact that all operates under an already positive air pressure doesn't tell much, because everything will be operating relative to that anyway.

 

Lastly - and not to make it more complicated, but because it just is so which *is* of some importance for understanding - notice that when a loudspeaker driver comes forward (plus) within the moving forward (at the plus level) it draws back again but NOT crossing zero volt, and forward again, say, 1000 times. This is because the main excursion direction is directed by the lowest frequency it can take (and which is in order at that time), while all higher frequencies it can also take, ride on that. Thus, at moving forward for the bass frequency it 1000 times goes back and forth all while being positive. The same happens at pulling back, now all occurring at negative V.

 

So you see, it just is there, and won't work otherwise.

 

One additional thing I found, possibly of some interest :

Many "Lounge" / "Ambient" labels (often sets of albums, like Buddha series) make use of the phenomena I described, in an illegallish fashion. They keep up the positive voltage for a longer time than nature dictates, so the driver will keep on being pushed forward all the time, within that range actually playing normal music. Notice that this technically can be done, and implies nothing more nor less than what you suggested.

What about that ... (and somehow this implies a trance).

 

Regards,

Peter

 

 

 

[Paul R]

Peter - there are several points I would like to comment back to you on.

 

First, the way a speaker produces sound from electrical energy is simply a very clever trick, not a law of nature.

 

Second, sound is not equal to electricity, going from one to the other requires a conversion, and I wondered why certain choices were made in that conversion. Those choices were choices, not the only way to do it.

 

And third, I am not arrogantly reading something into an article that isn't there.

 

Audio engineering and audio sound reproduction is a very specialized subset of a much broader field. As such, it will have accepted practices and assumptions. I am interested in what those practices are, but I am not comfortable accepting an assumption "because that is the way it is." This stuff is engineering, not a voodoo science.

 

In terms of audio data, the current formats were chosen, at least in part, because of the processor, memory, and storage considerations of at least 20 years ago. It is quite possible that they are still the best fit, but perhaps they are not.

 

I am interested in opinions on that, but just saying "that's the way it is!" isn't what I was looking for.

 

Don't take that as a personal slam - I know you have a software product of some kind out there related to audio and I have not looked at it nor am I making any criticism of it.

 

 

 

[PeterSt]

Hi Paul,

 

I 100% understand how you think and what you are looking for, but still it is as it is ! I mean, suppose I would change this all and e.g. work with positive "data" only ... next I could let work my DAC the same (and I really would have performed the "crossover" solution which was talked about in this very thread ... weren't it that the PCM1704 internally just does that ! (you may look at the data sheet of it)) ... but

 

whatever DAC could work with my software, or whatever playback software could work with my DAC ?

 

NONE !

 

So this is why it is as it is, and most probably won't change. Nothing will be compatible.

 

Furthermore there's no real use(case) to it, besides me too would be very happy when the 2s complement wasn't in the middle of my life (as a programmer) all the time.

So, if my opinion wasn't clear already :

Software *can* use positive only; A DAC anticipating on that *can* work with positive data only. Even a loudspeaker would be able to produce sound with positive data only (see my Ambient example), although it would loose half of its intrinsic power, and may carry a large® distortion in the mean time.

 

In either case I am not sure anymore what the conversation is about;

You wanted to know the how, maybe the why, and I guess now you know the how and why maybe you are going to change the world. Ehh, why ?

Only kidding, but I *am* lost on what this is about now.

 

Kind regards,

Peter

 

[jkeny]

Paul,

I think what you are questioning is the idea of how acoustic waves are generated. Considering the basic concept of the loudspeaker hasn't changed since it was invented then this might be a difficult exercise. I can't think of another way of generating acoustic waves, can you?

 

[regal]

I'm definitely a Multibit guy, I prefer the good ole PCM63K, AD1862, and PCM1704K's to the newer DAC's. However I don't like NOS filterless DAC's. Give me a good PM100 or SM5842 OS DAC anyday (with a nonNFB analog stage). Every NOS DAC I have tried can't reproduce frequencies accurately above 14khz, I don't understand why but with my system a filterless NOS DAC will play a 18khz tone as mainly 14khz, for example. I think this is why when you see frequency sweeps of NOS DAC's they drop off above 14khz, when playing music this lends to a huge boost in treble at 14khz then a sharp drop-off.

 

[giro]

Hello,

Just signed up to CA. I am very intriqued in mulitbit. I started researching in my spare time when I wanted to know more about the digital to analog conversion process. Starting with history makes sense to me, is it not wise to stick/work with designs made by the people who invented digital in the first place?

 

The reasons I favour mulitbit:

 

I see bitstream as TDA1540 on roids. My opinion.

From what I've read, it would appear noise shaping was formulated to overcome the problem phillips faced with the 14bitter after sony having declared 16bit as the standard (perhaps a marketing /move tactic on sony's part - contending).

 

If it wasn't for this, we might not have bitstream today at all, right?

The fact bitstream oversamples x256 aswell.

I agree with a statement on another forum:

"I think with NOS DACs what makes them sound analog is they use the lowest possible sample rate and that means the lowest possible glitching. ... - running oversampled increases the rate of glitch production by the oversampling ratio."

 

Another point regarding the western economic model, which remains mostly unchanged and applies to everything produced for the mass consumer market / consumer culture,

Inflation makes it so that quality is increasingly getting harder to afford, on average.

Bitstream was a cheaper alternative to the multibit, perfect for the mass market.

That says it all for me really.

 

Also, our ears are non linear transducers -

The main reason I think vinyl sounds the way it does, over modern digital is because of the integral linear distortion factor inherant with vinyl players.

Ears respond to modulation, they enjoy fluctuation and variation - all modern dacs today are designed to be as transparent as possible (and rightly so) and most are 'inherently linear and monotonic over thier entire range', so it's down to who ever composes the music to insure a varied sound.

 

I personally think (analog domain) non linearities associated with vinyl and mulitbit is what makes them sound the less digital and 'warmer'.

I might add, if I wanted to define what warm sounds like to someone a good example was the main menu soundtrack on gran tourismo 4.

I don't know how they do it but that's what I picture warm as.

 

just my 2c ...

[13mh13]

My students need answers and mine aren't clear and concise enough!

Can someone, very succinctly, describe the difference between single-bit and multi-bit delta-sigma modulation?

 

Thanks!

[audiventory]

My students need answers and mine aren't clear and concise enough!

Can someone, very succinctly, describe the difference between single-bit and multi-bit delta-sigma modulation?

 

Thanks!

 

Hi 13mh13,

 

1. DSD = 1 bit PCM + Noise

 

Noise shaped (most part its energy shifted out of audible frequency range).

 

2. DSD ADC/DAC technically simpler than PCM due absent of voltage matrix.

Less efforts for better sound in audio applications.

 

3. PCM ADC/DAC better suitable for wide band applications. As example, N x MHz bands in telecommunications.

 

Simple about DSD How work sigma delta modulation in audio

 

Best regards,

Yuri Korzunov

[13mh13]

Hi 13mh13,

 

1. DSD = 1 bit PCM + Noise

 

Noise shaped (most part its energy shifted out of audible frequency range).

 

2. DSD ADC/DAC technically simpler than PCM due absent of voltage matrix.

Less efforts for better sound in audio applications.

 

3. PCM ADC/DAC better suitable for wide band applications. As example, N x MHz bands in telecommunications.

 

Simple about DSD How work sigma delta modulation in audio

 

Best regards,

Yuri Korzunov

Korzunov .... respectfully, I have no idea what your three points (or that CONFUSING linked page ) have to do with the query.

It's not about "DSD"!! Single-bit or multi-bit modulation is found in ADCs/DACs that incorporate one or the other in their core architecture.

 

Let's re-state the query a bit more specifically:

If the on/off state of each PWM pulse only needs one (1) bit, as per its orig. design, why did some designers feel more than one bit (i.e., more than one amplitude or voltage level) needed to incorporated (i.e., multi-bit delta-sigma)? After all, the old (= costlier and less linear) multi-bit architecture -- i.e., tradit. old-school multi-bit, non-D-S PCM, with all its $$ precision resistor trimmings -- were what TI, Wolfson, AD, Crystal/Cirrus, AKM, et. el, were trying to move away from ... right?

 

In any case, I have found some insights here and here. Alas, these are hardly succinct, as per my original request, but they are somewhat informative.

 

Anyone else want to take a crack?

[Jud]

Korzunov .... respectfully, I have no idea what your three points (or that CONFUSING linked page ) have to do with the query.

It's not about "DSD"!! Single-bit or multi-bit modulation is found in ADCs/DACs that incorporate one or the other in their core architecture.

 

Let's re-state the query a bit more specifically:

If the on/off state of each PWM pulse only needs one (1) bit, as per its orig. design, why did some designers feel more than one bit (i.e., more than one amplitude or voltage level) needed to incorporated (i.e., multi-bit delta-sigma)? After all, the old (= costlier and less linear) multi-bit architecture -- i.e., tradit. old-school multi-bit, non-D-S PCM, with all its $$ precision resistor trimmings -- were what TI, Wolfson, AD, Crystal/Cirrus, AKM, et. el, were trying to move away from ... right?

 

In any case, I have found some insights here and here. Alas, these are hardly succinct, as per my original request, but they are somewhat informative.

 

Anyone else want to take a crack?

 

So on the planet where you live, this is supposed to encourage other people to want to help you?

[Melvin]

So on the planet where you live, this is supposed to encourage other people to want to help you?

 

Jud .. a succinct answer I must say ;-)

[tranz]

Google is your friend.

[Tony Lauck]

 

Let's re-state the query a bit more specifically:

If the on/off state of each PWM pulse only needs one (1) bit, as per its orig. design, why did some designers feel more than one bit (i.e., more than one amplitude or voltage level) needed to incorporated (i.e., multi-bit delta-sigma)? After all, the old (= costlier and less linear) multi-bit architecture -- i.e., tradit. old-school multi-bit, non-D-S PCM, with all its $$ precision resistor trimmings -- were what TI, Wolfson, AD, Crystal/Cirrus, AKM, et. el, were trying to move away from ... right?

 

Anyone else want to take a crack?

 

There are problems making high quality 1 bit modulators due to the difficulties dithering them:

 

http://sjeng.org/ftp/SACD.pdf

http://www.robertwannamaker.com/writings/rw_phd.pdf

 

A single bit implementation is much cheaper to implement because it does not require precision balancing of circuitry associated with the individual bits. However, techniques were found to dynamically balance mismatched components, making multibit implementations cost effective. This appeared in the high end in the dCS "Ring DAC" but is now available in many multi-bit sigma-delta DAC chips such as the ESS SABRE products.