|
|
发表于 2016-5-27 10:45:05
|
显示全部楼层
本帖最后由 yamuling 于 2016-5-27 12:54 编辑
Q：Am I right to assume your theoretical end-game DAC uses half band (full band?) filters with +1M taps? Or will your WTA filter still be a better approximation/reproduction of the analog wave due to the taps not being infinite? If I understand you well, this is where we walk into the ceiling of what is possible with 16bit coefficients - so more (or infinite) taps stop making a difference due to this limitation. Do these 16bits refer to the recording (so for 24bit high-rez recordings these numbers would be higher) or to the filter / processor architecture?
I'm looking forward to your +1M taps DAC somewhere in the future :-) As computational power keeps rising, we might actually be able to pull it off. Or am I dreaming your dream too loudly now?
A：The 16 bits refer to the accuracy of the coefficients against the ideal sinc function - so if we want to get an impulse response that has 16 bit accurate coefficients (that is the truncation of the coefficients occurs when the coefficient is below 16 bit in level), then you need about 1M taps. But what level is audible? Hugo is about 12 bits accurate against the ideal, Dave is about 14 bits - and there is a substantial difference in SQ between the two. But at what point does increasing tap length make no difference? Nobody knows the answer to that, but I will find out! There is also the issue of oversampling level, but that's another story...
Q：Dear Mr Watts, what does 12 bits accurate against the ideal mean?
A：The ideal impulse response, the sinc function gradually converges to zero, so the future and past samples no longer have a significant effect. With the 12 bits accuracy, then one can say that the WTA impulse response is roughly the same as the ideal sinc function to a 12 bit accuracy, as the coefficients are now smaller than 12 bits. There will come a point when the coefficients are so small they no longer have an audible effect, and I certainly have not hit that point yet.
找到了，不过完全没说要升级呀，这还是在发布半年之前的讨论...只是在讨论一些极限理想值 
更新隔壁yeziwong的解释：
这个精度说的是数字滤波器和理想滤波器（理想的滤波器的时域波形是一个sinc函数，因为频域是一个方波，是一个非因果系统，不可能完全实现，是能无限逼近）之间的接近程度，tap越多，数字滤波器越接近理想值，当然硬件开销会越大，并且tap数越多，后面的系数越接近于0，所以精度的提高和硬件开销最终是一个指数关系
这个和DAC的精度是两码事
|
|
[复制链接]
