Digital that sounds like analog

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
A very good question - I've run into a bit of difficulty over nomenclature with my designs.:)

The 2X filter you refer to isn't so much a filter, rather a hybrid DAC-filter and as such has no digital filter. So I sometimes call it NOS for the absence of any digital filter. But I suppose its better called 'MOS' for 'Minimal Over Sampling'. The DACs themselves run at 88k2 but only receive the exact same data as comes off the CD. In this thread I'm a bit undecided whether to pursue the MOS design further - it depends if a much simpler NOS sounds as good as MOS when combined with the filter I've been talking about. if so then perhaps I'll sideline the MOS design for the time being. But then MOS plus the same filter might well turn out to be even better...
 

dallasjustice

Member Sponsor
Apr 12, 2011
2,067
8
0
Dallas, Texas
What about discrete resistors instead of DAC chips? Don't many of these chips apply their own digital filters which could work against the NOS/low noise design philosophy? Also, I didn't see minimizing pre-ringing on the list of design goals and I thought I'd throw that out there. I'm a big believer in the evils of pre-ringing. :D
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
Me too but let's clear up the term pre-ringing i,e not the Gibb's Effect which occurs near 20KHz but rather the pre-echo of typical linear phase filters used in most DAcs
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
What about discrete resistors instead of DAC chips?

If you mean like the TotalDac then I can't see what problem that solves that isn't addressed by integrated DACs at a much lower price because they can use laser trimming to get the precise resistor values. The problems of R2R DACs for audio in my mind are too steep a hill to climb.

Don't many of these chips apply their own digital filters which could work against the NOS/low noise design philosophy?

Most do nowadays but I'm specifically avoiding the newer chips.

Also, I didn't see minimizing pre-ringing on the list of design goals and I thought I'd throw that out there. I'm a big believer in the evils of pre-ringing. :D

Oh I am a frim believer in minimum phase filtering (no pre-ringing) too, so that comes as standard whether I go MOS or NOS.
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 22, 2010
3,947
306
1,670
Monument, CO
Your comment about self-dithering prompts me to ask - how to ensure that the dither is indeed of the correct amplitude and prob. density function in the loop?

Good question. Thre are various rules in texts on DS converters by e.g. Candy, Norsworthy, Temes, etc. You need enough to break up tones and mask the quantization noise without significantly raising the overall noise floor, hurting dynamic range, or affecting lopo stability. You can estimate it, simulate, measure it, but remember the original reason for dither was to eliminate tones in the digital filters with LF and DC inputs, not to mask the queantization noise and make the noise floor more "analog" as is done now. That is different than insuring stability. Then there is the whole subject of using "colored" dither to provide the benefits without reducing in-band SNR...
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 22, 2010
3,947
306
1,670
Monument, CO
What about discrete resistors instead of DAC chips? Don't many of these chips apply their own digital filters which could work against the NOS/low noise design philosophy? Also, I didn't see minimizing pre-ringing on the list of design goals and I thought I'd throw that out there. I'm a big believer in the evils of pre-ringing. :D

Matching discrete resistors, and their interfaces (wiring), to 16+ bits seems like a nightmare to me.

Pre-ringing can't happen with the analog (causal) filters opus111 is discussing here. That is an artifact of digital filters...

MOS means metal-oxide-silicon transistors to me... Of course, NOS means new-old stock or nitrous oxide, too. Too many TLAs, too few letters. :)
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
Good question. Thre are various rules in texts on DS converters by e.g. Candy, Norsworthy, Temes, etc. You need enough to break up tones and mask the quantization noise without significantly raising the overall noise floor, hurting dynamic range, or affecting lopo stability.

Yes I've read a lot about use of dither to eliminate tones - in these cases the dither though is often low level but OOB tones. Like what Philips did on their 'Bitstream' parts back in the mists of time. But I've never read anything about ensuring correct dither PDF and amplitude to eliminate noise modulation. I'm guessing one reason for this is that the designers were always using FFTs to examine the performance of their loops. So they were accustomed to looking in the frequency domain.

You can estimate it, simulate, measure it, but remember the original reason for dither was to eliminate tones in the digital filters with LF and DC inputs, not to mask the queantization noise and make the noise floor more "analog" as is done now. That is different than insuring stability. Then there is the whole subject of using "colored" dither to provide the benefits without reducing in-band SNR...

Perhaps this is just a terminological difference but I'm unaware of using dither to mask quantization noise, rather than decorrelate it from the wanted signal.
 

NorthStar

Member
Feb 8, 2011
24,305
1,323
435
Vancouver Island, B.C. Canada
---- http://en.wikipedia.org/wiki/Dither

* Check the Rotel RCD-991 CD/HDCD player (PMD-100 digital filter/HDCD decoder IC):
- PCM-63 dual Burr-Brown 20-bit D/A converters (differential balanced).
- Low-jitter digital circuitry.
- User-selectable dither aids in "voicing" to best match a particular system.
- Substantial 5-segment power supply.
- +++
 

NorthStar

Member
Feb 8, 2011
24,305
1,323
435
Vancouver Island, B.C. Canada
And the relevance of this Rotel CD player is? Bob - when you have suggestions please indicate why you are suggesting.

---- "Dither" Richard, simply dither. ...That Rotel CD player offers that feature (user-selectable: 10/11 different modes).

And if I impeded again in your thread I apologize to you, and won't be posting here no more. :b
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
I disagree there - discussing something means having a conversation, a dialogue - just a cite a discussion doth not make. Particularly when the cite appears to pay zero attention to context...:p

And if I impeded again in your thread I apologize to you, and won't be posting here no more.

Ah that's a conditional apology :cool: You didn't notice your imposition of an irrelevance then? ;)

Since you've brought our attention to one of the better (IMO, I've not stopped to listen to it) CD players out of Rotel, I'll just mention that I don't propose to use the PCM63 because its architecture is based on R2R which in my estimation is too prone to glitching - hence higher noise modulation. Its otherwise a good chip though and potentially better sound-wise than one of its multibit successors.
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 22, 2010
3,947
306
1,670
Monument, CO
Yes I've read a lot about use of dither to eliminate tones - in these cases the dither though is often low level but OOB tones. Like what Philips did on their 'Bitstream' parts back in the mists of time. But I've never read anything about ensuring correct dither PDF and amplitude to eliminate noise modulation. I'm guessing one reason for this is that the designers were always using FFTs to examine the performance of their loops. So they were accustomed to looking in the frequency domain.

Perhaps this is just a terminological difference but I'm unaware of using dither to mask quantization noise, rather than decorrelate it from the wanted signal.

1. I am not aware of dither's (noise decorrelation's) application to reducing noise modulation either. Not something I have studied. Anxiously awaiting your findings...

2. Yes, I have gotten into trouble with this terminology before. When I first applied dither to a radar system converter back in the 80's it was described as "masking" quantization spurs, and I have used the term to describe noise decorrelation ever since, sorry.


p.s. I have had problems using ceramic caps in analog filter applications as they are pretty sensitive to voltage and can cause nonlinearities in the signal. See e.g. http://www.edn.com/design/analog/44...itor-becomes-a-0-33--F-capacitor?cid=EDNToday for a recent overview.
 

Kindhornman

New Member
Dec 3, 2012
21
0
0
Los Angeles, California, USA
Richard,
My question here is why you are using a passive elliptical filter rather than an active filter? Wouldn't this as in any passive filter cause losses that could be avoided with an active section?

The second part of this question is why the elliptical filter and its obviously more complex solution to a high order filter. I would have thought that you would have used a Saleen-Key type of filter with perhaps an LR filter topology. Couldn't you do that with a few sections of Saleen-Key and just gone with a higher order filter like four sections of 2nd order LR filters and had 8th order? Wouldn't this eliminate the passband ripple on the lower side and allowed you to start the filter at 20Khz instead of the 17Khz that you a starting at? I know that right now this isn't to be a commercial product but if it did become one the fact that the listed upper freq. cutoff would be below the standard of the industry listing of 20hz to 20Khz cause a marketing problem?
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
Hi Steven - welcome to WBF :)

My question here is why you are using a passive elliptical filter rather than an active filter? Wouldn't this as in any passive filter cause losses that could be avoided with an active section?

This is really an excellent question and one which harks back to what Steve asked much earlier on - what makes my DAC sound 'analog'? Firstly a little background - I've been working on DACs specifically for perhaps 3 years now. I started out with an S-D design but in optimizing that I noticed that the faster the opamp I used for I/V conversion, the better sound. By faster I mean more gain-bandwidth, higher slew rate. This prompted me to look at the nature of what was coming out of the DAC chip - I found very fast edges with rise times of the order of 1nS (its a guesstimate as my scope isn't that fast). Reflecting on this it wasn't surprising as the DAC itself is a small geometry CMOS part - CMOS digital logic has similar characteristics if you buy the 'advanced' stuff. This discovery got me thinking - how to 'tame' such high speeds?

One thing I found with the sound quality, once I'd installed a very fast opamp (LM6172) in the I/V stage was that the size of the feedback capacitor made a difference to the sound - smaller was definitely better. If you look at (say) TI's datasheet applications for their PCM179X DACs you'll see they use something around 2n2. I found something less than 10pF sounded best with the LM6172. But then the LM6172 can cope with such a small cap because its slew rate is in the thousands of V/uS. When I tried with such a small cap on the NE5532 it gave really odd sound effects, like birdies on an FM tuner. It could be fine with music playing but in gaps between tracks the birdies would start up.

To explore what might be the cause of this, I fired up LTSpice and tried simulating. I couldn't provoke any untoward behaviour, no matter how hard I tried (and I tried for days). So whatever this effect was, it certainly wasn't reflected in the normal opamp models. I gave up with modelling and decided the only way I could make progress was to carry on with listening.

Another 'data point' I came across was Lynn Olson's report of measuring the output of his PCM63 into passive I/V with a spectrum analyser. That's a piece of kit I don't yet have so I was very interested in what he found. Which was that the spectrum even from this relatively slow chip extended to 50MHz. This finding correlated nicely with what came out of my (much faster) S-D DAC chip.

The conclusion I drew from all this was that standard opamp circuits really could not cut the mustard at the output of DAC chips. When I think about what opamp I'd be prepared to use for audio, opamps with GBW < 10MHz I tend to think of as a bit suspect. So that's a ratio of audio bandwidth to chip bandwidth of 1:500. Applying this rule of thumb to Lynn's DAC output spectrum, I'd need an opamp with GBW of 25GHz. Not really going to happen. I needed to tame the edges before sending the DAC's signal into any amplifying stage.

(I'll carry on with this saga in a subsequent post and of course get on to your other questions).
 

LL21

Well-Known Member
Dec 26, 2010
14,411
2,509
1,448
Hi Steven - welcome to WBF :)



This is really an excellent question and one which harks back to what Steve asked much earlier on - what makes my DAC sound 'analog'? Firstly a little background - I've been working on DACs specifically for perhaps 3 years now. I started out with an S-D design but in optimizing that I noticed that the faster the opamp I used for I/V conversion, the better sound. By faster I mean more gain-bandwidth, higher slew rate. This prompted me to look at the nature of what was coming out of the DAC chip - I found very fast edges with rise times of the order of 1nS (its a guesstimate as my scope isn't that fast). Reflecting on this it wasn't surprising as the DAC itself is a small geometry CMOS part - CMOS digital logic has similar characteristics if you buy the 'advanced' stuff. This discovery got me thinking - how to 'tame' such high speeds?

One thing I found with the sound quality, once I'd installed a very fast opamp (LM6172) in the I/V stage was that the size of the feedback capacitor made a difference to the sound - smaller was definitely better. If you look at (say) TI's datasheet applications for their PCM179X DACs you'll see they use something around 2n2. I found something less than 10pF sounded best with the LM6172. But then the LM6172 can cope with such a small cap because its slew rate is in the thousands of V/uS. When I tried with such a small cap on the NE5532 it gave really odd sound effects, like birdies on an FM tuner. It could be fine with music playing but in gaps between tracks the birdies would start up.

To explore what might be the cause of this, I fired up LTSpice and tried simulating. I couldn't provoke any untoward behaviour, no matter how hard I tried (and I tried for days). So whatever this effect was, it certainly wasn't reflected in the normal opamp models. I gave up with modelling and decided the only way I could make progress was to carry on with listening.

Another 'data point' I came across was Lynn Olson's report of measuring the output of his PCM63 into passive I/V with a spectrum analyser. That's a piece of kit I don't yet have so I was very interested in what he found. Which was that the spectrum even from this relatively slow chip extended to 50MHz. This finding correlated nicely with what came out of my (much faster) S-D DAC chip.

The conclusion I drew from all this was that standard opamp circuits really could not cut the mustard at the output of DAC chips. When I think about what opamp I'd be prepared to use for audio, opamps with GBW < 10MHz I tend to think of as a bit suspect. So that's a ratio of audio bandwidth to chip bandwidth of 1:500. Applying this rule of thumb to Lynn's DAC output spectrum, I'd need an opamp with GBW of 25GHz. Not really going to happen. I needed to tame the edges before sending the DAC's signal into any amplifying stage.

(I'll carry on with this saga in a subsequent post and of course get on to your other questions).

Barely following the technical stuff...but enjoying the read as your story of how your thinking on DACs has evolved...in that regard, way cool! Keep going!
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
Continuation of an answer to Steven's question

The solution that suggested itself based on the experience related in the last post was to use a passive filter. At first I tried series SMT (1206 size) inductors between a simple TDA1545 DAC and the amplifier stage which used an AD603. This constitutes a 1st order filter - with about six such inductors in series I was fairly sure I heard an improvement in the sweetness of strings through this arrangement, an encouraging result. I then experimented with cheaper parts called 'ferrite beads' instead of inductors. The inductors were wirewound and about 10X the price of ferrite beads, so I tried with a dozen or so ferrite beads in series and this produced the rather surprising result of reducing the 'detail' in the sound. But with somewhat longer listening, this new 'reduced detail' sound didn't actually lack information, nothing was missing just 'details' no longer stood out. So I began to consider that 'detail' is an artifact of excess HF noise being fed into solid-stage circuits and the ferrites were somehow absorbing this noise. I'm so far unclear what frequency band the noise might occupy but my guess is above 5MHz as the ferrite beads have little effect below this. The detail-free sound was all the more relaxing and engaging - typically 'analog' so the ferrite bead string stayed put - a highly cost-effective upgrade.

This taught me that active circuits have an extreme sensitivity to HF noise where sound quality is the desired result. Hence in a high noise environment, passive components are the only parts that can be used until the HF is tamed - solid state will always compromise the SQ. The AD603 is considerably better suited to amplifying HF signals than the average high-speed opamp, as it has a GBW of the order of 1GHz. Perhaps the susceptibility to RF is a reason why many DAC builders choose valve output stages - I've never designed with valves so I don't know how susceptible they are to this HF noise problem, my suspicion is they're much less vulnerable. Valves though are out of the question for me, hence passive filtering it has to be.

Its quite possible to create an active version of this filter design by using opamp gyrators in place of the inductors to implement what are called 'FDNRs' and I may well try this experiment. The opamp solution won't be cheaper but I do predict it won't sound as good. It will be physically much more compact and lighter so if the compromise in sound isn't too great it could suit portable applications. I'd still use the string of ferrite beads though prior to any active filter stage to tame the worst of the HF hash.

I'll move on to the later question in a subsequent post - just mention here that passive filter loss is real (6dB in this case) but a small price to pay for getting the true 'analog' sound quality.
 

opus111

Banned
Feb 10, 2012
1,286
3
0
Hangzhou, China
The second part of this question is why the elliptical filter and its obviously more complex solution to a high order filter. I would have thought that you would have used a Saleen-Key type of filter with perhaps an LR filter topology.

Sallen-Key filters are active only - they use a unity gain buffer. In theory the buffer could be a single transistor but I think the filter would be significantly less than ideal then. Given that I've ruled out active filters on SQ grounds, there are indeed other options for passive filters.

Up until I'd built the elliptic filter I'd been using a 4th order Butterworth with two series Ls and two shunt Cs. That's not too dissimilar to the LP section of an LR4 crossover. I was also using 3 ferrite beads to tame the excess 'detail' figuring that HF was getting through above the series L's self-resonant frequencies, where they begin to look capacitive. On swapping over to the elliptic it was instantly no contest at all, the soundstage depth improvement was unmissable and so the 4th order filter is going to be confined to the spares box. One surprise is that the elliptic doesn't need the ferrites adding in order to sound perfectly sweet. The 4th order Butterworth is much shallower in roll-off, I chose an f-3dB around 30kHz so as not to eat too much into the audio passband.

At first the passband ripple of the elliptic concerned me but I rather think its probably inaudible, when compared to the response irregularities of my speakers. I shall be continuing to listen over the coming weeks though and it anything shows up I'll update the thread accordingly.

Your question about the marketing problem of having an HF roll-off at 17kHz rather than 20kHz is an interesting one that I'll chew over. I rather think this could be seen as a plus if I market it as an apodizing filter which attenuates aliasing which might have been included due to half-band filters being employed somewhere in the digital chain. Thinking about it that might be a reason for the soundstage improvement over the 4th order Butterworth.
 

About us

  • What’s Best Forum is THE forum for high end audio, product reviews, advice and sharing experiences on the best of everything else. This is THE place where audiophiles and audio companies discuss vintage, contemporary and new audio products, music servers, music streamers, computer audio, digital-to-analog converters, turntables, phono stages, cartridges, reel-to-reel tape machines, speakers, headphones and tube and solid-state amplification. Founded in 2010 What’s Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals, we enjoy learning about new things and meeting new people, and participating in spirited debates.

Quick Navigation

User Menu

Steve Williams
Site Founder | Site Owner | Administrator
Ron Resnick
Site Co-Owner | Administrator
Julian (The Fixer)
Website Build | Marketing Managersing