I came across a link today to an article about the merits of a non-OS DAC over an OS DAC. Of course, I'm referring to the filter part of the DAC. Anyway, this article doesn't make much sense to me: http://www.sakurasystems.com/articles/Kusunoki.html What is your take on this? Is the author correct when he says that 16-bit accuracy is not possible if you oversample? I don't believe that this article was very well written. It didn't convince me that non-OS is the way to go. Let me know what you guys think. -Larry
Non-ovesampling DACs vs. Oversampling DACs for audio - your feedback requested
Started by ●February 6, 2004
Reply by ●February 6, 20042004-02-06
In article fMTUb.19737$w_5.11049@newssvr24.news.prodigy.com, Larry McFarren at invalid@address.com wrote on 02/06/2004 16:35:> I came across a link today to an article about the merits of a non-OS DAC over > an OS DAC. Of course, I'm referring to the filter part of the DAC. Anyway, > this article doesn't make much sense to me: > > http://www.sakurasystems.com/articles/Kusunoki.html > > > What is your take on this? Is the author correct when he says that 16-bit > accuracy is not possible if you oversample? I don't believe that this article > was very well written. It didn't convince me that non-OS is the way to go. Let > me know what you guys think. >i'm going over this, but my first take is that the above statement is incorrect in the most meaningful sense. the only reason i might consider conventional DACs is so you can time-multiplex them between channels. r b-j
Reply by ●February 6, 20042004-02-06
robert bristow-johnson wrote:> In article fMTUb.19737$w_5.11049@newssvr24.news.prodigy.com, Larry McFarren > at invalid@address.com wrote on 02/06/2004 16:35: > > >>I came across a link today to an article about the merits of a non-OS DAC over >>an OS DAC. Of course, I'm referring to the filter part of the DAC. Anyway, >>this article doesn't make much sense to me: >> >>http://www.sakurasystems.com/articles/Kusunoki.html >> >> >>What is your take on this? Is the author correct when he says that 16-bit >>accuracy is not possible if you oversample? I don't believe that this article >>was very well written. It didn't convince me that non-OS is the way to go. Let >>me know what you guys think. >> > > > i'm going over this, but my first take is that the above statement is > incorrect in the most meaningful sense. the only reason i might consider > conventional DACs is so you can time-multiplex them between channels. > > r b-j"The only reason for audio, I imagine. For servos, other criteria rule. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●February 7, 20042004-02-07
Jerry Avins <jya@ieee.org> wrote in message news:<40246155$0$20044$61fed72c@news.rcn.com>...> robert bristow-johnson wrote: > > > In article fMTUb.19737$w_5.11049@newssvr24.news.prodigy.com, Larry McFarren > > at invalid@address.com wrote on 02/06/2004 16:35: > > > > > >>I came across a link today to an article about the merits of a non-OS DAC over > >>an OS DAC. Of course, I'm referring to the filter part of the DAC. Anyway, > >>this article doesn't make much sense to me: > >> > >>http://www.sakurasystems.com/articles/Kusunoki.html > >> > >> > >>What is your take on this? Is the author correct when he says that 16-bit > >>accuracy is not possible if you oversample? I don't believe that this article > >>was very well written. It didn't convince me that non-OS is the way to go. Let > >>me know what you guys think. > >> > > > > > > i'm going over this, but my first take is that the above statement is > > incorrect in the most meaningful sense. the only reason i might consider > > conventional DACs is so you can time-multiplex them between channels. > > > > r b-j > > "The only reason for audio, I imagine. For servos, other criteria rule. > > JerryA brief read indicates numerous dubious claims. 1) The argument that oversampled DACs have greater sensitivity to jitter is completely wrong. He is ignoring the fact that jitter sensitivity is proportional to the average step-size from one sample to the next (for example, if there is no change at all in the DAC output, then obviously it doesn't matter when this happens). The analysis goes like this; a) assume random jitter. The instantaneous error at each sample point is proportional to the CHANGE in output level times the instantaneous time error. These errors must be summed in an RMS fashion to get the total error magnitude. b) The CHANGE in output level falls in direct proportion to the oversampling ratio (assuming your worst case signal is always 20KHz full-scale). c) The instantaneous jitter error events occurs more often for higher oversampling rates, but ... d) The jitter spectrum is spread over a larger bandwidth. Point "b" gives audio-band noise FALLING LINEARLY with oversampling rate.. Point "c" gives a full-spectrum error magnitude that rises as the SQUARE ROOT of oversampling ratio... Point "d" gives an in-band noise component that FALLS by the SQUARE ROOT of oversmapling ratio. So "c" and "d" cancel each other out, and the result is that jitter errors FALL LINEARLY WITH OVERAMPLING RATIO. Now if you want to be pessimistic, you could assume that the jitter errors were not random, but had long-term correlations. This potentially eliminates the oversampling reduction factor ("d"), resulting in in-band jitter errors that fall only by the square-root of the oversampling ratio. The author of the article in question seems to assume that every DAC output event starts from 0, goes to the full value, and returns to zero again. If this were the case, then point "b" would be eliminated. But DACs are not designed this way (I have some credentials in this area). Most DACs have switched-capacitor output filters to remove sigma-delta noise, and the step-to-step change is mostly dominated by the interpolated signal itself. Some higher-end DACs use multi-bit noise-shapers that output directly, with no filtering. Such DACs have a bit higher step-to-step change statistics for small signals, but for large high-freqiuency signals (always the worst-case for jitter) the step size is still dominated by the interpolated signal itself. By the reasoning presented in the paper, there would be no oversampling DACs on the market with more than 16-bit performance for "common" amounts of jitter. The fact is that you can buy such converters from numerous manufacturers at ridiculously low prices, and measure their performance with clock jitters as high as 500 ps, and still easily exceed 16-bit performance. Bob Adams
