[sdiy] Re: Inexpensive 16 bit DAC?..other useful thoughts..

[Bob Weigel]

   There is also a way to build an accurate DAC with even cheaper 
parts...provided you don't need the conversions to take place too 
quickly :-).  I haven't seen this done but I'm sure it probably has 
been..but take say a pic chip with 16 ins and an out.  Use the ins for 
data to be converted and the out use to drive a duty cycle output...in 
other words write the pic program to read the 16 bit number and feed it 
to the timer every 64,000 how ever many it is counts.    The output then 
charges a small capacitor through a resistor going to the input of a 
buffer amp, having an appropriate RC time constant so as to give a 
smooth value at that point.
     Very useful and cheap way to do it for low frequency applications.  
I think I'll incorporate that idea in some places in my dream synth 
design in fact :-).  But anyway, obviously the faster the processor the 
higher the frequency response but I'm not seeing it go much beyond 1Khz 
or so before it leaves the realm of 'cheap'.    Unless one can retro an 
old 350mhz pc processor to work with minimal facilities outside.   I 
know people get buys on some of those.
     Anyway I know there are lots of smoothing technicques.  Still when 
you drop to a very quiet passage, and intricate details in piano and 
strings just...are not there...all the kings dithering and all the kings 
modelling the natural content shall not restore thereto :-)  Raymond had 
a good idea there.  Bang/bit baby :-).  12 bits sounds great when it's 
full head room.  Even 10 sounds great usually so...what better thing to 
do that use the 16 bits with 10 or 12 of them as the data and the 
remainder as an exponent?  It's really wasteful to go straight 16 but it 
is less complicated of course.  -Bob

Antti Huovilainen wrote:

> On Sun, 24 Apr 2005, Bob Weigel wrote:
>
>> In boards where effect calculations are done, we know it's good to 
>> keep 24bits going in the digital real to keep enough headroom not to 
>> loose quality at points.  Once things are being converted back to the 
>> analog realm 16bits starts to sound quite adequate.   12 bits with 4 
>> bit exponent however...sounds better.  I use a Kurzweil K1200 and one 
>> of the things I love
>
>
> Properly designed systems use dithering to fix this. Quantization 
> noise sounds quite horrible because it is correlated to the wanted 
> signal (indeed, quantization distortion is a much better term). By 
> adding some (+- 0.5 LSB) random noise before quantization, the error 
> is not correlated to the signal anymore and the distortion is eliminated.
>
> With dithering, the bitdepth of the DA converter only determines the 
> noise floor of the signal. The situation can be improved even more by 
> shaping the noise power to be mostly in areas where ear is less 
> sensitive. This way some additional 10 dB perceptual SNR improvement 
> can be had.
>
> You can try the results if you have a program that can use variable 
> dithering and quantization levels. Using foobar2000 (mp3 player), I 
> found that 8 bits is quite acceptable with modern overcompressed pop 
> music and for some cases even 6 bits would be enough!
>
> Antti
>
> "No boom today. Boom tomorrow. There's always a boom tomorrow"
>   -- Lt. Cmdr. Ivanova
>
>