Yes the volume is fully digital, as I get better transparency that way. Analogue volume controls have noise and linearity problems, and more importantly small signal resolution problems. Every metal to metal interface has oxides, and this has small signal non-linearities, and this degrades the perception of depth. I have known about this since the early 80's, but could never measure a problem, just hear it. The good thing about Dave was with the discovery that I needed noise shapers that worked at 350 dB to get the best depth, then I have been able to run simulations and measure the performance of this problem - and I confirmed that extremely small errors did have an influence on depth perception - so now I have a rule of thumb number of 350 dB performance - that's why we can't measure these tiny problems but can hear them. It was interesting that working with digital solved an understanding problem that was purely analogue.
Anyway, simply doing volume controls digitally won't solve the problem, as conventional noise shapers are actually worse at resolving small signals than analogue. Not only that, how the truncation issue is dealt with is also important, as this also degrades depth performance. So now I have Dave's noise shapers running at 350 dB, I needed the WTA filter outputs and the volume OP to also work at 350 dB performance too. Today, when I design I now formally test for this by ensuring that it will perfectly reproduce a -301 dB signal which ensures its much better than 50 bits accuracy. Here is a 24 bit WTA output after noise shaping:
The flat noise floor section at -385 dB is actually a simulation/FFT issue. The -301 dB signal should be -301.02999 dB - it came out at -301.030 dB, so the fundamental is perfectly preserved amplitude. Even though its a 24 bit output, it is accurate to more like 64 bits.
Now I get this performance by using noise shapers (and I had to go to silly lengths to get this level of performance - before Dave I worked to 200dB or 32 bit accuracy)
My next headache is with Davina and getting perfectly preserved small signals but with 44.1 and 16 bit - in this case, I can't use noise shaping like this, so its going to be an interesting challenge maintaining depth perception.
看Rob写的,貌似现在数字音量精度和失真比模拟音量小不少......
|