Cost and competence + some other reasons
There are several reasons including:
1. Design for measured performance (which is not necessarily the same as subjective performance)
2. Design for cost (good filters typically cost much more to implement, think die area as one such example)
3. Latency (FIR filters give you latency as a function of their lenght which end users typically don't understand). IIR filters typically require higher precision and better understanding of the problem at hand.
4. Competency of designers (typically math heavy not necessarily audio experts). Indeed, the combination of math and audio seems to be a rarity indeed so the matematicians win.
5. Competency of buyers (often pretty minimal with some notable exceptions, but are they heard in their organizations?)
5b. You can't have your cake and eat it! Tradeoffs must be made and for some reason such tradeoffs are usually made "wrong" on the component level + few have the wish to make the filters themselves for several reasons including competency and cost.
6. Multibit: The process used for the analog output is not very well suited for the digital front end which makes it even harder to meet item 2 above. Trimming etc makes this much much worse again.
7. Low bit depth: Everything is now much easier for the digital designers as they can work on a digital process but that forces other corrections to have to take place.
8. Design for integration (i.e integrate stuff together that doesn't necessarily make sense wrt sound quality)
9. Make tradeoffs to suit specialist applications such as mobile devices, PC's etc. Usually means low power, small size, low cost.
10. Try to "solve" other problems by implementing features not desired for pure sound quality reasons.
Indeed, for those into photography, multibit sensors are analogous to CCD imagers in that the light is captured AND processed in a more "direct" fashion. For CMOS sensors, there is much more processing going on to compensate for operating in the "wrong domain" to begin with (as is the case of single bit or even low number of bit DAC's which are most common these days). Having said that, both methods can achieve excellent results for both audio and photography.
There are several reasons including:
1. Design for measured performance (which is not necessarily the same as subjective performance)
2. Design for cost (good filters typically cost much more to implement, think die area as one such example)
3. Latency (FIR filters give you latency as a function of their lenght which end users typically don't understand). IIR filters typically require higher precision and better understanding of the problem at hand.
4. Competency of designers (typically math heavy not necessarily audio experts). Indeed, the combination of math and audio seems to be a rarity indeed so the matematicians win.
5. Competency of buyers (often pretty minimal with some notable exceptions, but are they heard in their organizations?)
5b. You can't have your cake and eat it! Tradeoffs must be made and for some reason such tradeoffs are usually made "wrong" on the component level + few have the wish to make the filters themselves for several reasons including competency and cost.
6. Multibit: The process used for the analog output is not very well suited for the digital front end which makes it even harder to meet item 2 above. Trimming etc makes this much much worse again.
7. Low bit depth: Everything is now much easier for the digital designers as they can work on a digital process but that forces other corrections to have to take place.
8. Design for integration (i.e integrate stuff together that doesn't necessarily make sense wrt sound quality)
9. Make tradeoffs to suit specialist applications such as mobile devices, PC's etc. Usually means low power, small size, low cost.
10. Try to "solve" other problems by implementing features not desired for pure sound quality reasons.
Indeed, for those into photography, multibit sensors are analogous to CCD imagers in that the light is captured AND processed in a more "direct" fashion. For CMOS sensors, there is much more processing going on to compensate for operating in the "wrong domain" to begin with (as is the case of single bit or even low number of bit DAC's which are most common these days). Having said that, both methods can achieve excellent results for both audio and photography.
That's funny. I built a NOS DAC, and when I upgraded it to 4x OS it sounded MUCH better! 
I think, in theory, and often in practice, oversampling or digital filtering can improve things.
For the DIYer, NOS is easy to implement. There's no digital filter to interface adding tons of jitter to the signal, and the clock frequencies are very low, and thus easy to use and inherently low in jitter. Also, you can implement unfiltered NOS, to great effect - this is something that obviously can't be permitted in mass production.
So, if you convert to NOS, on a mid priced player, it will most probably be perceived as an improvement.
Also, what DAC did you use, and what was it replacing?
However, if you are careful with your clocks, and ensure that the master clock is low in jitter then I feel oversampling is superior - for obvious reasons.
I think, in theory, and often in practice, oversampling or digital filtering can improve things.
For the DIYer, NOS is easy to implement. There's no digital filter to interface adding tons of jitter to the signal, and the clock frequencies are very low, and thus easy to use and inherently low in jitter. Also, you can implement unfiltered NOS, to great effect - this is something that obviously can't be permitted in mass production.
So, if you convert to NOS, on a mid priced player, it will most probably be perceived as an improvement.
Also, what DAC did you use, and what was it replacing?
However, if you are careful with your clocks, and ensure that the master clock is low in jitter then I feel oversampling is superior - for obvious reasons.
Beacuse its cheap the first. In audio world is like this you have quantisation noise and it much easy to put digital filter beacuse if you must put some large coils and the you have phase distorsion especially in high frequncies. but I listen in NON Os. Digital filter improve 6dB signal to noise ratio. Transformer in out (to cut all frequencies up to 20000 Hz solve this problem) I dont have but I tr,y results is very very fine.
Well yeah, there is a conflict between correctness (number of taps in the filter, number of bits in the accumulator, coefficients quantization, quality of dither, etc) and the cost of silicon used to implement all this stuff.
Take the filter in the CD723 for instance, it's the worst of the worst, 16 bit output, no dither, roundoff errors all over the place, quantization noise... and you can easily hear it, all detail is gone, ambience disappears, etc. Why ? Cost.
Some really stupid filters like some SONY parts will not realize the math can yield values over 16 bit full scale (surprise !!!) and instead of clipping as it should do, or gracefully scale down and dither properly, will cause a digital overflow which is basically a large spike in the opposite direction the signal was originally going.
A cheap FPGA like XC3S250E ($13, qty 1) can spit out 1.2 GMAC/s which is much more that what you'd need to correctly implement an oversampling filter. Or a Blackfin DSP (so cheap !). But those are still too expensive for our dear kit manufacturers... Just stick a $2 codec and be done with it...
I found hope in the ESS Sabre DAC because at last the management seems to have let the IC designer actually do some of the stuff right even if it costs half a mm² of extra silicon and a few extra package pins ! I hope this chip sells by the millions so we can get it cheap.
And about opamps, there's nothing wrong with opamps, although of course badly implemented opamps will suck like everything else !
Take the filter in the CD723 for instance, it's the worst of the worst, 16 bit output, no dither, roundoff errors all over the place, quantization noise... and you can easily hear it, all detail is gone, ambience disappears, etc. Why ? Cost.
Some really stupid filters like some SONY parts will not realize the math can yield values over 16 bit full scale (surprise !!!) and instead of clipping as it should do, or gracefully scale down and dither properly, will cause a digital overflow which is basically a large spike in the opposite direction the signal was originally going.
A cheap FPGA like XC3S250E ($13, qty 1) can spit out 1.2 GMAC/s which is much more that what you'd need to correctly implement an oversampling filter. Or a Blackfin DSP (so cheap !). But those are still too expensive for our dear kit manufacturers... Just stick a $2 codec and be done with it...
I found hope in the ESS Sabre DAC because at last the management seems to have let the IC designer actually do some of the stuff right even if it costs half a mm² of extra silicon and a few extra package pins ! I hope this chip sells by the millions so we can get it cheap.
And about opamps, there's nothing wrong with opamps, although of course badly implemented opamps will suck like everything else !
Oh well, those intersample overs.... my Yamaha CDX-596 had tons of it... and it dithers to 20bits with a free running pseudo random register, which makes the effects from these overs really confusing (not plain wrap-arrounds, gives a complete "hash" on the scope display). It took me a while to figure that out...
BTW, is there a list of how commonly used DACs react on intersample overs, since those are very likely to be found on many older recordings from the zenit of the loudness war...
- Klaus
BTW, is there a list of how commonly used DACs react on intersample overs, since those are very likely to be found on many older recordings from the zenit of the loudness war...
- Klaus
Okay, so a wide spread.
But, what were the clocks like on these?
I believe NOS works better with a poor clock. Its less sensitive to jitter due to the lower frequencies involved.
Also, you've taken the step into (I assume passive) IV without opamps, which may also be contributing to the dramatic difference you observe.
Don't get me wrong. I'm very pleased that your NOS DAC works so well, and you're happy with it. I was very happy with mine, but I've had a great improvement by introducing 4xOS. Mind you, further improvements occurred when I worked on the power supply (oversampling isn't everything).
I don't believe mid-range mass production can afford attention to detail that DIYers can.
But, what were the clocks like on these?
I believe NOS works better with a poor clock. Its less sensitive to jitter due to the lower frequencies involved.
Also, you've taken the step into (I assume passive) IV without opamps, which may also be contributing to the dramatic difference you observe.
Don't get me wrong. I'm very pleased that your NOS DAC works so well, and you're happy with it. I was very happy with mine, but I've had a great improvement by introducing 4xOS. Mind you, further improvements occurred when I worked on the power supply (oversampling isn't everything).
I don't believe mid-range mass production can afford attention to detail that DIYers can.
You see, that's interesting, because I perceive the exact opposite.
I do feel that NOS is 'more relaxed' in sound (perhaps less metallic as Bernhard put it).
When I first upgraded to oversampling, I was surprised how much bass was missing, but this was more than made up for with extra detail in the treble.
I did eventually get used to the lack of bass, there is plenty there - its just less prominent - probably more accurate.
The biggest problem I have with NOS is the 3dB loss at 22kHz, or something like that, I can't recall the figures exactly, but higher frequency audio components are attenuated somewhat. I feel you lose the detail and imaging information when you move to NOS.
My DAC is easily convertible. I might try NOS again to remind myself of the differences.
I do feel that NOS is 'more relaxed' in sound (perhaps less metallic as Bernhard put it).
When I first upgraded to oversampling, I was surprised how much bass was missing, but this was more than made up for with extra detail in the treble.
I did eventually get used to the lack of bass, there is plenty there - its just less prominent - probably more accurate.
The biggest problem I have with NOS is the 3dB loss at 22kHz, or something like that, I can't recall the figures exactly, but higher frequency audio components are attenuated somewhat. I feel you lose the detail and imaging information when you move to NOS.
My DAC is easily convertible. I might try NOS again to remind myself of the differences.
I like NOS DAC sound because it sounds like vinyl, i have Motown LP record it has strong bass and punch, I couldn´t get such sound from digital but 4 x TDA 1543 sounds very similar, i think it has to do with 4x converters it made full sound. I don´t think audiophiles will be satisfied with the sound, it is not much detailed, but sounds exactly how music should sound. DAC has plenty of bass no shortage. I didn´t measure frequency but highs sounds like turntable and airy.
John007 said:
Don't get me wrong here, but I don't thing that you should compare low- to mid-end CD-players with the stuff you have now, and specially not compare those with a decent, modern non-oversampling DAC that are put in high-er end CD-players, stand-alone DAC, or the better DIY kits. Same goes for opamps. Make a fair comparison, or make none at all.
I never heard a decent OS DAC sound metallic, nor flat or dull.
If well designed, both kinds can sound very good, if not, they both suck.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.