Digital that sounds like analog

DonH50 · Dec 1, 2012

Glitch, very hand-waving explanation: Think of DACs as switching small units of voltage, or charge, that is passed to the output. Because switches in the DAC do not turn on and off instantaneously and in perfect synch, there is a small period of time when both are on, or (depending on the design) both are off. Parasitic elements like stray capacitance around the switches and in the switching devices themselves cause some energy to "bleed off" to the output during these times, and the result is a fast "glitch" until the switches are fully off/on and stable. Depending upon the design and signal, these glitches can be very large (1/2-fullscale) and although they are very fast (because the switches change state quickly) it is putting wideband noise into the output at the clock rate of the DAC. Furthermore, because the amount of charge depends upon how many switches are changing state, which is in turn a function of the signal. That means the glitch energy may depend upon the signal, which due to various factors in the DAC adds signal distortion as well as noise to the output.

HTH - Don

LL21 · Dec 1, 2012

DonH50 said:
Glitch, very hand-waving explanation: Think of DACs as switching small units of voltage, or charge, that is passed to the output. Because switches in the DAC do not turn on and off instantaneously and in perfect synch, there is a small period of time when both are on, or (depending on the design) both are off. Parasitic elements like stray capacitance around the switches and in the switching devices themselves cause some energy to "bleed off" to the output during these times, and the result is a fast "glitch" until the switches are fully off/on and stable. Depending upon the design and signal, these glitches can be very large (1/2-fullscale) and although they are very fast (because the switches change state quickly) it is putting wideband noise into the output at the clock rate of the DAC. Furthermore, because the amount of charge depends upon how many switches are changing state, which is in turn a function of the signal. That means the glitch energy may depend upon the signal, which due to various factors in the DAC adds signal distortion as well as noise to the output.

HTH - Don

Thank you. That is helpful! In the end, i would NOT say i am here to learn...that would be presumptuous of me. I am more here to 'appreciate' and get a 'sense' of what is inside a DAC, which may help (on the edge) in how i look at my digital options. Ultimately, i will still listen first and foremost because one thing i do appreciate for sure is that ultimately all of these various elements are complex and must be designed and implemented by someone who has to make choices out of all these various elements to come up with his/her DAC...and it is all of these choices that a few email trails are never going to help me fully understand...only appreciate.

thus, i listen to the finished product with more appreciation of what may have gone into it (thanks to people like you and Opus111 and others)...and let my ears tell me if i like it.

opus111 · Dec 1, 2012

Especially for you Don, here's an expanded view of the passband, together with the group delay. I'm not sure what happened in that the design is supposed to be equiripple, but the lowest ripple is half the amplitude of the others.

Note this is a simulated response with almost ideal inductors - very high Q. The real inductors have lower Q and it turns out passband ripple is quite sensitive to Q. For Lloyd, Q means quality and its a measure of how ideal an inductor is. A low loss inductor has a high Q.

LL21 · Dec 1, 2012

opus111 said:
Especially for you Don, here's an expanded view of the passband, together with the group delay. I'm not sure what happened in that the design is supposed to be equiripple, but the lowest ripple is half the amplitude of the others.

View attachment 6866

Note this is a simulated response with almost ideal inductors - very high Q. The real inductors have lower Q and it turns out passband ripple is quite sensitive to Q. For Lloyd, Q means quality and its a measure of how ideal an inductor is. A low loss inductor has a high Q.

You're even predicting my questions! Great!

LL21 · Dec 1, 2012

so here are 2 questions before i sign off for the day:

1. what are your technical focuses in achieving digital that sounds like analog? (rank them in order of what you think is most important)
- lack of phase aberrations
- the lack of Glitch
- low jitter
- low distortion
- low intermodulation
- other???

2. What are your key implementation focuses to achieve this? (again ranked in order)
- NOS v SD
- analog filter design
- kind out output tube
- shielding/isolation/rfi/etc.
- quality of power transformer
- other???

flez007 · Dec 1, 2012

Lloyd, it is 2AM your time right now, do you ever sleep? (or want to volunteer to help Jack as moderator when all of us are sleeping?)

opus111 · Dec 1, 2012

I see Don has posted some very helpful stuff on glitching

I'll add to that by talking a little about DAC architectures - this particularly relates to audio DACs in that there are three broad families of multibit (for which I sometimes use the rather sloppy characterization 'NOS') DACs. The three families are resistor string, resistor ladder and segmented current source. The first of these gives a voltage output, the second is normally current output and the third always current output.

Up until last year I was aware of no audio DACs (by which I mean boxes, not chips) which were based on resistor string technology. Then the Metrum Octave made quite a splash where the designer had scratched off the ID of the DAC chips and used the mystery element in his marketing. Of course that was like a red rag to a bull to me so I embarked on a quest to discover what the mystery DAC chip could be, based on the marketing claims and pictures I found of the internal layout. It turns out the Octave uses a TI 16bit resistor string DAC (actually multiple DACs in parallel) - voltage output. its the only instance I know of of this technology in an audio DAC and its received glowing reviews so there may well be potential for optimizing resistor strings in audio applications.

TI, having bought Burr Brown once had plenty of resistor ladder DACs in its portfolio, but is overall discouraging new designs with these chips. The most notable is the PCM1704 which has 'NRND' status with the suggestion that any new designs use one of their newer S-D type chips. But discerning audio customers aren't the only ones who prefer the R2R technology over the newer stuff, and TI hasn't announced a last time buy on the 1704 even though it seems right on the verge of doing so.

The only company that's made segmented current source DACs for audio is Philips (or NXP as its now known). These include the almost legendary TDA1541A and also the lower cost, lower power and lower performance TDA1543 and TDA1545.

The reason I've introduced these architectures is that they have quite different performance in terms of glitch. The fact that discerning listeners prefer some kinds over others indicates to me that probably glitching makes an audible difference. This has taken quite a long time to dawn on me because audio DACs don't specify glitch energy in the datasheets.

LL21 · Dec 1, 2012

flez007 said:
Lloyd, it is 2AM your time right now, do you ever sleep? (or want to volunteer to help Jack as moderator when all of us are sleeping?)

i tend to work 7 days, but on the weekends, tend to work with music on and also leave WBF on in the background...when something strikes my fancy, i just post in.

So i get to work, listen to music and every once in a while post a bit...the best of all worlds!

opus111 · Dec 1, 2012

lloydelee21 said:
1. what are your technical focuses in achieving digital that sounds like analog? (rank them in order of what you think is most important)
- lack of phase aberrations
- the lack of Glitch
- low jitter
- low distortion
- low intermodulation
- other???

Interesting question. Let me pontificate a bit before re-ranking your categories. The no.1 challenge in digital design I've found to be noise control. Digital systems are high speed and consequently high noise. We need to prevent that noise getting into our audio as its the primary cause of digititus, the harshness and greyness by which some listeners recognize a digital system is playing. So to rank your list its easy to say what's relatively unimportant - the things that don't contribute to the digititus. At the bottom then come jitter and phase aberrations. If by distortion you mean THD that would have to come next. Which leaves intermodulation followed by glich right at the top. But its hard to separate these last two. Perhaps if we characterize glitch as just another form of noise modulation then it becomes easier - noise modulation is the king of the hill and triumphs over intermodulation.

2. What are your key implementation focuses to achieve this? (again ranked in order)
- NOS v SD
- analog filter design
- kind out output tube
- shielding/isolation/rfi/etc.
- quality of power transformer
- other???

This one is a little easier. Because noise modulation is the key flaw of SD, then it has to be NOS right at the top. The remaining things stem from fixing up noise modulation issues that arise from NOS DACs, in particular glitching. So filter design is key to keep the glitches out of the downstream analog stages where they work their evil ways in generating digititus. I'm a total ignoramus on tubes, so I'll delete that one, there won't be any tubes in my designs - at least not in my lifetime. Shielding/isolation/rfi is a systems level category, everything is done with rfi in mind because that's the primary noise in a digital system. Any kind of power transformer can be fixed up with suitable countermeasures so its the lowest on the list. So you pretty much nailed my priority order first time, once tubes are deleted

DonH50 · Dec 1, 2012

Thanks! It looks like the last ripple is off a bit. Maybe you have a component value a little off, but I also seem to have a vague memory about something like this happening for odd and/or asymmetric filters (different input/output impedances). I have my old grad filter text handy but it has been so long it probably wouldn't help much (and it was a lousy text anyway).

That does remind me to ask why you chose a 100 ohm output load instead of hi-Z?

Yup, high Q is needed for the filter to be close to the ideal. And, the equations don't get any prettier when you add loss elements...

I talked a little bit about various DAC architectures and showed some plots in my old DAC tutorial threads. Resistor strings can be segmented, unless you mean pure unary (2^N - 1 resistors for N bits). Ladder (R-2R) and current source (most of mine of been current sinks due to the device technology) can be segmented or not; most are to improve performance and relieve matching concerns. The more you segment a DAC, adding switches that cover smaller parts of the whole, the less the error requirements and the easier it is to match.

A pure binary DAC would need each element (switch cell) to match to an lsb; for a 16-bit DAC that is 1 part in 65,536 (2^16, about 0.0015%). One using 8 segments for the MSB needs only 16-8 = 8-bit matching, 1 part in 256 (2^8, 0.39%). A full-unary design using one switch for each step (lsb) only needs to match cells to 50%, but 65,536 is a LOT of cells.

You consider noise modulation to be the primary problem with delta-sigma DACs? Not questioning, asking.

Thanks - Don

opus111 · Dec 1, 2012

The original filter I didn't design from scratch, it was one I modified based on what I think was a design for an FM tuner filter I found on one of the electronics industry online newspapers. So it originally cut off around 15kHz - i scaled the frequency so the first stop band zero moved to 22kHz (I think it was previously 19kHz to reject the FM stereo sub-carrier frequency) and this is how its ended up.

DonH50 · Dec 1, 2012

OK, thanks.

I added some to my previous post while you were typing.

Do you have a filter design program? There are some on the net. I think I have some but they depend upon Mathcad or Matlab (not sure which, probably have them for both). That would make life easier if you are hand calculating, or using tables.

NorthStar · Dec 1, 2012

-- Oh wow, I just discovered this thread!

Now I just need to take the time and read it. :b

_______________

EDIT: This is a great thread Richard, and Don, and Lloyd, ...

opus111 · Dec 2, 2012

Don - about the 100ohms. Firstly I'd have to look at the original filter I picked up but I guess it must have had some kind of a load as this one has equal source and termination resistors. The post amplifier I'm using isn't a standard opamp as I prefer something with a bit more GBW - I found AD603 was available quite cheaply here so I'm using that. Its designed for RF or IF amplification and has an in-built 100ohm load, but also very valuably it has a DC controlled volume element.

About S-D DACs, yes I consider therre's an elephant in the room in that respect which nobody but ESS wants to talk about. ESS has done us a great service in prizing open the can of worms, but still there are a significant number of listeners who prefer multibit (particlarly TDA1541A) to the ESS Sabre on redbook. There's so much investment in product by TI, CS and ADI in the digital audio market I doubt we'll ever get an admission of the problems out of them even though ESS has published at least one graph showing (what I take to be) noise modulation issues with an ADI part. Noise modulation with a few parts is quite evident from some of the FFT plots they do include - the give-away is the higher noise floor at higher signal levels.

As regards filter design programs, I do have MATLAB but haven't used it so far for analog filter design, I have used it for FIRs. I've been relying on the tables in the Williams book for non-elliptics and would like to find an earlier edition which does include the elliptics. They got put into a PC program in the latest edition presumably to make way for more detail on digital filtering but I for one felt the book to be better just focussing on analog filters.

JackD201 · Dec 2, 2012

DonH50 said:
You consider noise modulation to be the primary problem with delta-sigma DACs? Not questioning, asking.

My question as well since there's a plethora of different algorithms out there, many custom ones, precisely to mitigate this. Koch, Meittner, Paychev and De Luca make very analog sounding machines with S-D and PWM, some of the most analog I've ever heared so I am curious.

opus111 · Dec 2, 2012

Analog tape has noise modulation - its the main chance that digital has of beating analog so it has to be licked

That some people have custom algs to mitigate noise mod is news to me (other than ESS) - got any links?

LL21 · Dec 2, 2012

opus111 said:
Analog tape has noise modulation - its the main chance that digital has of beating analog so it has to be licked

That some people have custom algs to mitigate noise mod is news to me (other than ESS) - got any links?

AS you know there is a parallel digital discussion ongoing where you have been participating...i wanted to bring something from Tomelex to Atmasphere here to get your thoughts since this is a technical thread on digital approaching/beating analog discussion as well. Here is the quote below:

tomelex said:
...now heres what you said that interested me, i am technical in ANALOG (not digital expert at all) audio and electronics and have built amplifiers and stereo processors and so you can go deep as you want, but from the quote below:

IMO vinyl exceeds digital in most of these regards; the problem areas are the THD and SN ratio. Digital has low THD but very high **inharmonic** distortion, that is to say distortion based on the scan frequency rather than fundamental tones withing the recording itself. Anyone with a recording of a sweep tone can find evidence of this inharmonic distortion; as the sweep tone sweeps, 'birdies' will be heard at the same time. These are intermodulations caused by the distortion of the conversion process and they exist in very measurable quantities in any digital gear.

Phase response is another matter; in many traditional digital recordings there are terrible phase problems brought on by the use of the brickwall filter! Later digital recording that take advantage of higher scan frequencies (88KHz being a good example) don't need quite so brutal a filter and thus sound better when knocked down to Redbook.

What we can take from this is that these things are not cast in concrete!*some* digital is better in phase than *some* analog and also the other way 'round

Thoughts, Opus111...seems like there is consensus above on intermodulation distortion...question is then on your thoughts as relates to the subsequent comments about Phase? I know Zanden makes a big deal of this point in their description of their analog filter and their own focuses on design?

opus111 · Dec 2, 2012

Yes I read that over on the other thread and couldn't figure out what Atmasphere meant and as I hadn't been totally satisfied with another earlier interaction with him, I decided to let it be rather than engage it. I really don't know what 'scan frequency' means here - perhaps sample frequency but if so then perhaps he's talking about NOS DACs with their imaging frequencies. Which is what this filter is designed to reject!

I do have some experience with digital recording and I have never come across the effect that Atmasphere is describing, its totally alien to me. I have heard 'birdies' but only from the earliest S-D type DACs (Philips Bitstream SAA7320) and then only at low level. As far as I'm aware there is no inherent distortion in the conversion process and if there were I'd see it in my digital recordings. As for phase problems - potentially there are but nowadays nobody is using 'brick wall' filters (like I'm presenting here) rather they use digital brick wall filters which give zero phase problems, at the expense of other potential problems (pre-ringing).

Lastly the comments about knocking down to redbook aren't consistent with my understanding of digital either. Since redbook is 44k1 then there's always a need for a fairly steep filter - rather like this one I'm presenting - to prevent anything above 22k getting in. The steep filter can be digital (in the case where an oversampling ADC is used) or analog.

LL21 · Dec 2, 2012

opus111 said:
Yes I read that over on the other thread and couldn't figure out what Atmasphere meant and as I hadn't been totally satisfied with another earlier interaction with him, I decided to let it be rather than engage it. I really don't know what 'scan frequency' means here - perhaps sample frequency but if so then perhaps he's talking about NOS DACs with their imaging frequencies. Which is what this filter is designed to reject!

I do have some experience with digital recording and I have never come across the effect that Atmasphere is describing, its totally alien to me. I have heard 'birdies' but only from the earliest S-D type DACs (Philips Bitstream SAA7320) and then only at low level. As far as I'm aware there is no inherent distortion in the conversion process and if there were I'd see it in my digital recordings. As for phase problems - potentially there are but nowadays nobody is using 'brick wall' filters (like I'm presenting here) rather they use digital brick wall filters which give zero phase problems, at the expense of other potential problems (pre-ringing).

Lastly the comments about knocking down to redbook aren't consistent with my understanding of digital either. Since redbook is 44k1 then there's always a need for a fairly steep filter - rather like this one I'm presenting - to prevent anything above 22k getting in. The steep filter can be digital (in the case where an oversampling ADC is used) or analog.

Thanks...so lets take your earlier comment about priorities in NOS/Multi-bit DACs:

- where do you think DACs are missing something that has prevented them from more definitively beating analog for [may of us] who still find something about tape and even LP to be superior?

Obviously that is the question everyone is asking themselves in digital playback everyday...curious as to where you think something is missing?

opus111 · Dec 2, 2012

lloydelee21 said:
- where do you think DACs are missing something that has prevented them from more definitively beating analog for [may of us] who still find something about tape and even LP to be superior?

Well I think the DACs (by which I mean the chips) to do the job probably already exist, they just don't get engineered correctly and marketed optimally so don't reach customers' hands. If I didn't hold out the hope that digital could beat analog then I'd not be doing what I'm doing here. I guess I like a really big challenge but also none of the arguments of the analog lovers about what's wrong with digital make sense to me, so I can't see a fundamental problem with digital per se, as a technology. Its just the implementations which suck.

As regards what's missing - its what digital adds which is the problem, not what it takes away.

Digital that sounds like analog

Member Sponsor & WBF Technical Expert

Well-Known Member

Banned

Well-Known Member

Well-Known Member

Member Sponsor

Banned

Well-Known Member

Banned

Member Sponsor & WBF Technical Expert

Banned

Member Sponsor & WBF Technical Expert

Member

Banned

WBF Founding Member

Banned

Well-Known Member

Banned

Well-Known Member

Banned

Similar threads