The DDDAC uses the TI PCM1794 with defeated digital filters, claiming to be a non-oversampling DAC. More info.
That is an excellent question. Following, is my highly oversimplified and non-authoritative explanation of what is essentially happening. While both sigma-delta-modulators (SDMs) and oversampling digital reconstruction filters (ODFs) operate at multiples of the input sample rate they have distinctly different functions.
The function of an SDM is to reduce the quantization noise (the quantization error) of the converter. This is achieved by spectrally relocating the quantization noise/error out-of-band, which, for audio, is to the ultrasonic band. Now, imagine a 1-bit quantizer based SDM DAC reproducing some arbitrary D.C. level. The DAC's 1-bit quantizer can correctly produce only two specific levels. All levels in-between would exhibit extreme amounts of quantization noise/error if it were not for the SDM unit relocating that noise/error to the ultrasonic band. If instead of a D.C. level the DAC were to output an A.C. signal, the SDM would dynamically adjust the ultrasonic quantization noise so that each next sample is accurately converted in analog level.
The function of the ODF is to remove (well, mostly remove) the repeating bands of ultrasonic images created by the sharply switching output of the DAC quantizer unit when reproducing an A.C. signal. These ultrasonic images are not to be confused with the ultrasonic noise created by an SDM unit. These images are repeating copies of the desired signal except shifted up in frequency and falling in amplitude. The ODF's job is to remove these repeating ultrasonic images. It does this by sharply low-pass filtering the desired signal band which results in additional samples being sent to the quantizer and converted. All that these additional samples do is to reconstruct the fundamentally sine based original audio signal.
Should an SDM based DAC convert a signal without benefit of an ODF (NOS) the output will have the same non-ringing impulse response, the same repeating ultrasonic image bands, and the same uncorrected high treble response droop characteristics of any other NOS based zero-order-hold DAC. However, there will also be SDM created ultrasonic quantization noise present. That's my conception, but as I said, it's non-authoritative.
I did know about dddac's TDA1543 NOS, from years back. It has that charac. "NOS sound". But I'd prefer carefully-implemented OS, even with older DFs such as SAA7220.
Don't know why dddac would pursue NOS for delta-sigma DACs other than for novelty or proof-of-concept. I've never heard such a setup, so I'll refrain from further comment. I can say that dddac seems to be a lone supporter of this topology.
Oversampling, with delta-sigma DACs, is partly used for the same reason OS is used for multi-bit DACs: gentler analog filtering.
In addition (I suspect), to increase the sampling rate so that the higher-speed DS modulator stage more easily "accelerates" to speed. E.g., for PCM Red Book (at 44.1khz) data to be "accelerated" to >1MHz speed of DS modulation is probably asking too much (= noise). But if one up/ovesamples before the DS mod stage, this eases the transition to the higher-speed DS modulation.
Don't know why dddac would pursue NOS for delta-sigma DACs other than for novelty or proof-of-concept. I've never heard such a setup, so I'll refrain from further comment. I can say that dddac seems to be a lone supporter of this topology.
Oversampling, with delta-sigma DACs, is partly used for the same reason OS is used for multi-bit DACs: gentler analog filtering.
In addition (I suspect), to increase the sampling rate so that the higher-speed DS modulator stage more easily "accelerates" to speed. E.g., for PCM Red Book (at 44.1khz) data to be "accelerated" to >1MHz speed of DS modulation is probably asking too much (= noise). But if one up/ovesamples before the DS mod stage, this eases the transition to the higher-speed DS modulation.
I believe that his objective was to obtain NOS sound character from a SDM based D/A chip. Why from a SDM based chip? That's simply because they are the only type of audio application specific D/A in current production, as far as I know. The key insight behind the ddac, IMO, is the recognition that oversampling digital signal reconstruction filtering is a function different from sigma delta modulation.
Why no support from major chip manufs?
*I know that ESS does conduct fairly extensive subjective evals.
Right. However, while the major chip manuf's allow end users of hardware manufs to choose between certain OS filters ("Sharp", "Slow", "MP", etc.) they do not allow "OFF". Why? Because the lab measurements (at "OFF") start to suck? Or do their own* subjective tests reveal flaws/compromises with NOS?
*I know that ESS does conduct fairly extensive subjective evals.
I feel fairly certain that the engineering departments of major DAC chip producers all accept the proven scientific principles of sampling theory - which is that the objectively correct reconstruction filter has a sharp frequency cut-off, and is preferably linear phase if a digital filter, to boot.
Soft filter roll-off modes were initially provided on-chip because they offered a shortened latency of the filter processing engine and intended for use in mastering. The specialty audio industry came to feature the soft filter modes as often subjectively preferable to the sharp. The DAC chip marketing folks picked up on that customer use and, like any good supplier, they began addressing that general market desire with the additional option of minimum phase filter modes.
I have no knolwedge of what the various DAC producers do or don't do as far as subjective chip evaluation. Having once been a middle manager in the IC industry I strongly suspect that there is no formal DAC chip subjective evaluation program within most vendors.
Last edited:
is that possible ?
i don't have EE background so correct me if i made any mistake. my understanding of sigma delta DACs is they need 2x oversampling to get 1 bit of resolution. modern delta sigma DACs are probably 4 - 8 bit so that's why they call it "multibit delta sigma". because reducing bit depth made quantitation noise, they use noise-shaping technique to push the noise out of audio band and send it to ultrasonic area. then, by using some filters (some are very steep which IMHO made new problem) they remove the noise. and that's how they achieve higher bit depth.
http://www.diyaudio.com/forums/digital-source/15439-how-does-delta-sigma-dac-work.html
i don't have EE background so correct me if i made any mistake. my understanding of sigma delta DACs is they need 2x oversampling to get 1 bit of resolution. modern delta sigma DACs are probably 4 - 8 bit so that's why they call it "multibit delta sigma". because reducing bit depth made quantitation noise, they use noise-shaping technique to push the noise out of audio band and send it to ultrasonic area. then, by using some filters (some are very steep which IMHO made new problem) they remove the noise. and that's how they achieve higher bit depth.
http://www.diyaudio.com/forums/digital-source/15439-how-does-delta-sigma-dac-work.html
I suspect that confusion (possibly, my own ) stems from the fact that there are two different oversampling ratios involved in SDM + ODF DACs. The ODF unit has the lower rate of the two, often x8. The SDM has the higher rate, x64 for example, which then would be 8 times the oversampling digital filter's output rate. So, while both the digital filter and the sigma delta modulator operate at some multiple of the native input rate the job of the ODF is to REMOVE ultrasonic image content, while the job of the SDM is to ADD (relocate) ultrasonic quantization noise. I don't know whether that expalanation helps.
) stems from the fact that there are two different oversampling ratios involved in SDM + ODF DACs. The ODF unit has the lower rate of the two, often x8. The SDM has the higher rate, x64 for example, which then would be 8 times the oversampling digital filter's output rate. So, while both the digital filter and the sigma delta modulator operate at some multiple of the native input rate the job of the ODF is to REMOVE ultrasonic image content, while the job of the SDM is to ADD (relocate) ultrasonic quantization noise. I don't know whether that expalanation helps.
DF better for RedBook?
With respect to an ODF used prior to a DS dac (PCM1794, etc.), I can understand why an ODF would be needed for RedBook PCM (16/44.1).
But at higher rez (24/192), I'm not sure how much benefit an ODF will have.
Then, again, I do not have the years-accumulated wisdom or theoretical/R&D experience of the major players (TI, AD, Cirrus, ESS).
Frankly, I think the multi-bit hype, that has crescendo'd the past several years (especially on diy and vintage-audio forums), is strange and mostly unsupported by the audiophile community at large.
With respect to an ODF used prior to a DS dac (PCM1794, etc.), I can understand why an ODF would be needed for RedBook PCM (16/44.1).
But at higher rez (24/192), I'm not sure how much benefit an ODF will have.
Then, again, I do not have the years-accumulated wisdom or theoretical/R&D experience of the major players (TI, AD, Cirrus, ESS).
Frankly, I think the multi-bit hype, that has crescendo'd the past several years (especially on diy and vintage-audio forums), is strange and mostly unsupported by the audiophile community at large.
I've neglected to address an important point. Oversampling also opens up ultrasonic spectrum space for the SDM unit to relocate the quantization noise. The greater the oversampling ratio the more spectrum is opened and the less is the intrusion of the relocated noise in to the upper portion of the desired signal band. Without sufficient open (unused) spectrum the relocated quantization noise could significantly raise the signal band noise floor. The below link is to a nice graph on the dddac site that shows exactly that effect occurring. It shows a 40dB increase in the noise floor for a 44.1k sample rate by 20kHz, verus a 192k sample rate. I assume that the depicted sample rates are all without benefit of oversampling.
http://dddac.com/pictures/pics1794/test_quantization_noise_floor_44_96_192fs_large.png
http://dddac.com/pictures/pics1794/test_quantization_noise_floor_44_96_192fs_large.png
Last edited:
i don't see oversampling is something bad. on the contrary it's a good thing. IMHO the evil is in the filters.
in this following experiment you'll see how different filters causing different phase problem and time smearing (ringing) which we may perceived as "digital sound". although it's done on music files using software and much more steep filters than used in delta sigma DACs, it shows us who's the bad guy
Ringing False: Digital Audio's Ubiquitous Filter Page 2 | Stereophile.com
in this following experiment you'll see how different filters causing different phase problem and time smearing (ringing) which we may perceived as "digital sound". although it's done on music files using software and much more steep filters than used in delta sigma DACs, it shows us who's the bad guy
Ringing False: Digital Audio's Ubiquitous Filter Page 2 | Stereophile.com
Oversampling is an integral aspect of the digital reconstruction filter's functioning.
You're touching on a very controversial area here. Sampling requires band limiting the desired signal. Because CD supports only just enough spectrum to carry the audio band it requires sharply band limiting the desired signal. It is the sharp band limiting which produces the infamously ringing filter impulse response. While sampling theory is fully satisfied with sharp band limiting there is not universal agreement as to the affect on human subjective perception using dynamic signals such as music.
you are correct. no wonder i feel like i'm not the only one. well, instead of opening that infamous can i better be a silent reader. lots of financial interests here
you are correct. no wonder i feel like i'm not the only one. well, instead of opening that infamous can i better be a silent reader. lots of financial interests here
No, no, please feel free to open any cans you wish to. Use this forum to express your opinions, to ask questions and to share your findings. No one knows everything, although, I suspect that one or two of us here would like everyone else to believe that they do. My view is that we are all here to both learn and to teach, depending on the topic and the participants on a given thread.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.