Hi Tom, I'm not really (infact not at all)a DSP expert but as I understand it the answer to your question actually has more to do with the filter at the output of your DAC and sampling rate than anything else. Additional bits in your D-A do not as I understand it buy you a better reconstructed output but greater dynamic range. An over-sampling D-A converter interpolates it's output between sucessive samples based upon a particular filter function (it doesn't just output the same voltage 8/16 sample times). The idea is to offload some (most) of the filtering from the analog output. Sampling theory says that ANY complex waveform with spectral components up to 1/2Fs (half the sampling frequency) can be re-constructed using by a number of independant sinusoids. The important thing to note is that these sinusoids must always be lower than the nyquist frequency (obviously otherwise there would be spectral components higher than the nyquist frquency which we've already dis-allowed). Essentially so long as your filter can re-create a sinusoid at the nyquist frequency and below you can build any waveform you like as long as it doesn't contain spectral components above that frequency. If you just want to produce a single sinusoid at a single output level then 1 bit resolution is plenty. Think of it like this, if you had more bits and you were sampling an (exactly) 20Khz sinusiod at (exactly) 40K samples/sec what would your data look like ? It would consist of 2 different values dependent upon where in the waveform the sample was taken but they would always be the same 2 values (just like a 1 and a 0). The very least you should expect your filter to do is get rid of any "steps" in the output. If you picture such a waveform it should be fairly noticable that the steps themselves must introduce some unwanted high frequencies into your output (think of each step as half a square wave, not really accurate but you get the idea, that rising or falling edge is pretty nasty spectrum wise). Conceptually the filter on an D-A is not just there to filter out the high frequency components introduced by the D-A steps but also to re-build the sampled data. Your real problem as far as I can see is that you need to re-produce some waveforms (such as a square wave or sawtooth wave with a fundamental of 20Khz) that have spectral components in excess of 20Khz. A square wave or triangle for example is theoretically comprised of an infinite set of odd harmonic sines' STARTING at the fundamental frequency. In order to re-construct a decent looking square wave you might choose to sample up to the 7th harmonic meaning you'll need around 300Khz bandwidth or 600K samples/sec. If however you change the impulse response of your filter to corespond to your desired waveshape you could probably build your tone generator a lot more easily. An alternative might be to look at a DDS synthesizer... As I say I am not a DSP expert by any stretch of the imagination, if there's something I've missed hopefully some body can fill in the gaps... Regards, ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Mike Cornelius Internet: mike@bytethis.com.au Byte This Interactive Phone: +61 2 9310-2157 PO Box 1342 Strawberry Hills FAX: +61 2 9319-3948 NSW 2012 Australia URL: http://www.bytethis.com.au ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -----Original Message----- From: pic microcontroller discussion list [mailto:PICLIST@MITVMA.MIT.EDU]On Behalf Of Brandon, Tom Sent: Wednesday, April 05, 2000 11:50 AM To: PICLIST@MITVMA.MIT.EDU Subject: Audio DAC Accuracy [Tech] I am looking at a project where I would like to synthesise various audio waveforms. Looking at sine, square etc 20Hz-24kHz.etc. My question is in regards to the accuracy of DAC needed. Now, typically Audio applications use 16+ bit at 44.1 and up. I have no problems with the sample rate, it has to be at least double the highest frequency of interest and close to double will produce very low quality sine waves (e.g. CD audio, 20kHZ sine @ 44.1kHz = ~2.2 samples\cycle) hence most audio converters use 8 to 16x oversampling so you're looking at about 350kHz (44.1 @ 8x). But what effect will bit depth have on the reproduction? Obviously it will introduce larger step sizes and thus less smooth curves. If you're reproducing complex signals I could see it effecting the reproduction\capture of appropriate sines. Slight nonlinearities may not be able to be filtered out as easily due to the complex source. But, Ig all you're producing is single oscillators then what sort of accuracy would be needed? Also, for the same project I'd like to use 1 DAC to drive a few SHAs to do mutlichannel. I've seen one SHA capable of >12bit (an Analog devices 16bit SHA) based on DNL and this was only single channel. Does anyone know of a multi channel SHA with a DNL suitable for 14+ bits ideally with settling time to suit 1-2MHz. Also, I'm still looking for a DAC capable of 14+ bits @ >1MHz at a reasonable price. Thanks, Tom.