Oversampling: "Internal" vs. "External"

[hollowman]

DAC chips like PCM1796 seem to offer internal OS filters (e.g. PCM1796 offers 8×).

Outboard oversampling filter chips also exist, such as DF1706. Sales literature from TI notes "The DF1706 is the ideal companion for Texas Instruments's PCM1704 24-bit audio Digital-to-Analog (D/A) converter." BUT, the PCM1704 supposedly features internal OS, anyway. So, why would one opt for a dedicated OS filter? Is this scenario similar to the receiver vs. amp+pre-amp or receiver vs. integrated-amp "debate"?

One more question:

The CURRENT prices of newer DAC chips, like the aforementioned PCM1796, are a LOT cheaper (5.90 USD) than older ones like PCM1704 (30.74 USD). Features/specs-wise, the newer chips SEEM to be "superior". And even TI, on their PCM1704 product literature page, states "Product is Not Recommended for New Design (NRND) ... Replaced By: PCM1796".

How good are the newest chips compared to the older and respected models, such as PCM1704?

[jean-paul]

Originally posted by hollowman
BUT, the PCM1704 supposedly features internal OS, anyway.

How good are the newest chips compared to the older and respected models, such as PCM1704?

PCM1704 does not feature internal oversampling AFAIK. The reason for using external OS chips was more practical as integration was harder in the past. Also some manufacturers were better in making DAC chips while others were better in making OS filters. For instance Burr Brown DAC chips were used mostly with NPC OS chips. Nowadays integration of functions and costcutting are more important. Sometimes the OS filter and the technical operation of the DAC part are too related to eachother ( as with delta sigma ) to separate them. Manufacturers of DAC chips also tend to integrate OS filters and many other functions in their chips to be able to offer a more versatile product. A low component count "total" solution makes it easier and thus less expensive for OEMs to design and build new devices.

On paper the newer TI multibit Delta-Sigma DAC chips are better. They're also smaller and cheaper to make. It seems these factors are key features for success in audioland. How good it sounds is of less importance Some do sound very good, some are only mediocre in sound quality. The former Burr Brown company had a strong belief in audiophile quality and this was proven with their R2R DAC chips which have a good reputation because of this. Because the older multibit chips itself and their technique are obsolete/unobtanium they have gained a cult status by now.

[hollowman]

Originally posted by jean-paul


PCM1704 does not feature internal oversampling AFAIK.

TI's datasheet claims it does (8x @96KHz):

http://focus.ti.com/lit/ds/symlink/pcm1704.pdf

More below...

Originally posted by jean-paul


The reason for using external OS chips was more practical as integration was harder in the past. Also some manufacturers were better in making DAC chips while others were better in making OS filters. For instance Burr Brown DAC chips were used mostly with NPC OS chips. Nowadays integration of functions and costcutting are more important. Sometimes the OS filter and the technical operation of the DAC part are too related to each other (as with delta sigma ) to separate them. Manufacturers of DAC chips also tend to integrate OS filters and many other functions in their chips to be able to offer a more versatile product. A low component count "total" solution makes it easier and thus less expensive for OEMs to design and build new devices.

On paper the newer TI multibit Delta-Sigma DAC chips are better. They're also smaller and cheaper to make. It seems these factors are key features for success in audioland. How good it sounds is of less importance Some do sound very good, some are only mediocre in sound quality. The former Burr Brown company had a strong belief in audiophile quality and this was proven with their R2R DAC chips which have a good reputation because of this. Because the older multibit chips itself and their technique are obsolete/unobtanium they have gained a cult status by now.

Do you (anyone reading this) have any personal preferences -- all things being equal -- with respect to non-OS vs. OS vs. US (upsampling) vs. Bitstream/MASH for standard (Red Book) CD playback?

I seem to find a lot of NOS DIY/kits projects nowadays. But my albeit limited experience with a NOS from diyparadise (Monica2) has been not all that great.

I was surprised to see this favorable mention of the TDA1547 Philips Bitstream DAC because MASH/Bitstream chips have been dissed by so many designers.

You also noted delta-sigma superiority with newer TI DAC chips. Do you (anyone) have a general pref. between delta-sigma vs. parallel? The folks at LessLoss seem to highly favor parallel over sigma-delta; specifically, they use the PCM1704 DAC with DF1706 OS filter. But I'm not sure how much of this is marketing and how much is "reality".

In many ways, I keep coming back to square one as far as the "NOS vs. OS vs. US vs. Bitstream/MASH for Red Book CD playback" debate is concerned: Scratching my head! To resolve this I'll ask the following questions:

- Is there a *general* consensus on this issue (well, as far as the myriad DIY audio forums are concerned)?

- Or do DIYers feel that the DAC itself isn't THAT important. Not that it's a non-issue but, rather, one should, instead, pay equal/greater attention to the analog or I/V stage?

[martin clark]

To look at the problem sideways, the one major difference between the three approaches you outline is what happens out-of-band.

NOS results in full-scale aliasing in the audioband, and images at harmonic frequencies above the audio band, along with a 3dB droop in response across the audioband (no sinx/x compensation). It results in an easy-on-the-ear sound like a vinyl disc that needs washing. Personally - I've heard it, and really don't like it. The big gain is simplicity to DIY, and somewhat reduced jitter sensitivity.

MASH/bitstream offers great linearity when measured, but to achieve this it produces massive amounts of wideband noise that is 'shaped' or shovelled above the audioband where it is assumed to the least damage. How much noise? Well the total energy contained is the same as the wanted signal... but at HF filtering is theoretically easy with just RC elements. These dacs have by far the greatest sensitivity to jitter, which complicates the choice.

Multibit dacs produce the minimum amount of out-of-band noise, and have moderate jitter sensitivity (which depends on oversampling rate) but are difficult to trim for monotonicity (linearity) and expensive to fabricate - hence the difference in prices, and why the true 16-bit dacs have all but dissappeared. From the manufacturers POV, if your customers are buying parts based on %THD and cost, and you can get great figures for both with a one-bit switch and some clever noiseshaping on a cheap CMOS process, then why struggle to make linear R2R dacs in expensive bipolar processes?

So it's a real trade-off. I do think that the attention paid the whole analogue stage design (incuding PSUs and I/V) can make far more difference overall. The real art IMO is control of that out of band noise - and how the system that follows copes with it. But extracting the last 5% that each approach has to offer is far harder, which means direct comparisons are hard to make. I've heard great and nasty examples of each approach...

[analog_sa]

Originally posted by martin clark
It results in an easy-on-the-ear sound like a vinyl disc that needs washing. Personally - I've heard it, and really don't like it. The big gain is simplicity to DIY

What a wonderful and accurate description. I have certainly heard better highs from cassette decks.

[ACTIVA]

Hi Hollowman,

You may refer to this tread on the DAC chips discussion :
New to DACs; After Some Research, is Texas Instruments PCM1794 the best DAC?

The Bi-CMOS / Bipolar R-2R ladder type DACs like the PCM1704, PCM1702, PCM63, AD1862 and TDA1541A are / were less cost sensitive designs and optimum solely on the performance area. Hence their higher cost and also need an external FIR digital filter.

The Sigma Delta /PDM /MASH types are CMOS in general and need noise shaping / over-sampling circuitry in order to work properly, therefore the digital filter is integrated in the chip itself. The Sigma Delta types have improved greatly through the years and come with impressive specs, such as the PCM1792 / PCM1794 and AD1955.

Some people still prefer the R-2R PCM1704 than the PCM1792/1794 types on subjective sound quality. But do remember your SPDIF receiver, PCB layout, I/V converters, LPF, power supplies are as important as the DAC chip itself.

Regards
Max

[hollowman]

Originally posted by ACTIVA
You may refer to this tread on the DAC chips discussion :
New to DACs; After Some Research, is Texas Instruments PCM1794 the best DAC?

The Bi-CMOS / Bipolar R-2R ladder type DACs like the PCM1704, PCM1702, PCM63, AD1862 and TDA1541A are / were less cost sensitive designs and optimum solely on the performance area. Hence their higher cost and also need an external FIR digital filter.

The Sigma Delta /PDM /MASH types are CMOS in general and need noise shaping / over-sampling circuitry in order to work properly, therefore the digital filter is integrated in the chip itself. The Sigma Delta types have improved greatly through the years and come with impressive specs, such as the PCM1792 / PCM1794 and AD1955.

Some people still prefer the R-2R PCM1704 than the PCM1792/1794 types on subjective sound quality. But do remember your SPDIF receiver, PCB layout, I/V converters, LPF, power supplies are as important as the DAC chip itself.

Thx for everyone's input.

I've ordered a couple of the chips noted in this thread. Upon testing them in various DAC "back-ends", I'll report back.

I generally feel that newer components perform better than older ones, regardless of technology. (I stress GENERALLY!). I think my experience is correct because:

- mathematical/topological designs evolve (for the better)
- manufacturing technology/techniques improve
- manufacturers -- even the major ones -- do pay some attention to reviews and online forum discussions

OTOH, newer models MAY be compromised because of:

- "cheaper" (inferior) materials (e.g. (semi)conductor grades) and components usage
- miniaturization ... may lead to signal-quality loss
- manuf's change of direction/focus. Some have hypothesized that TI's incorporation of BB may have led to sound-quality compromises. The popularity of "low"-fidelity formats like MP3 (and lack of sales of high-rez formats such as SACDs, DVD-A's, and Red Book CDs) heavily influence corporate decision-making.
-etc.

Technological evolution is often three steps forward, two steps backward -- and these steps are rarely concurrent.

[martin clark]

I'd broadly agree with all that. The change to using 'multibit' modulators in bitstream is making for some exceptional specs, with a true 20-21bits of resolution being recovereable in some instances. That's a pretty astonishing achievement!

On miniaturisation I'd suggest that - while this does make it harder for us amateurs - the benefits of much smaller current loops far outweight the inconvenience. For example it offers much more effectiveness in PSU decoupling. Compare some of the new SOIC/SSOP pinouts with the old SAA7220 digital filter, where everything inside shares the one supply - and the supply and ground pins are as far away from each other as possible, on a 24-pin DIL...

[hollowman]

Originally posted by martin clark
I'd broadly agree with all that. The change to using 'multibit' modulators in bitstream is making for some exceptional specs, with a true 20-21bits of resolution being recovereable in some instances. That's a pretty astonishing achievement!

On miniaturisation I'd suggest that - while this does make it harder for us amateurs - the benefits of much smaller current loops far outweight the inconvenience. For example it offers much more effectiveness in PSU decoupling. Compare some of the new SOIC/SSOP pinouts with the old SAA7220 digital filter, where everything inside shares the one supply - and the supply and ground pins are as far away from each other as possible, on a 24-pin DIL...

Unfortunately, I lack the technical background to comprehend (and time-effectively further investigate/Google) some of your points. To that end, can you/anyone further elaborate on:

"the benefits of much smaller current loops far outweigh the inconvenience".

"The change to using 'multibit' modulators in bitstream is making for some exceptional specs, with a true 20-21bits of resolution being recovereable in some instances. That's a pretty astonishing achievement!" Which specific DAC chip(s) are being referred to here?

Thx again.

[martin clark]

Well decoupling is both a science and a black art, but fortunately there's a really excellent article by Guido Tent which offers some insights for the rest of us.

http://www.tentlabs.com/Products/Com...decoupling.pdf

As for resolution - I'm not really up to speed on current 'best specs', sorry. But the AKM parts used in the Slim Devices transporter just tested in Stereophile turned in better than 20-bit resolution on test (see latest issue), and I bet there are other supplier's parts in the same ballpark.

Just how good this is? Well if you set the full-level output at 2v, the CD standard, then true 21-bit resolution (ie a noise floor 21bits below fullscale) means the noise voltage is about 0.9uV. That's about the same noise in the audio bandwidth as you'd measure across a 4.7Kohm resistor just sitting on the table!