| Electronics > Projects, Designs, and Technical Stuff |
| dac noob is confused |
| << < (3/6) > >> |
| Wimberleytech:
--- Quote from: radiolistener on March 03, 2019, 12:19:18 am --- --- Quote from: Wimberleytech on March 02, 2019, 10:58:03 pm ---...and what, pray tell, is a 32bit DAC? It does not exist...pretty sure. --- End quote --- it exists, for example AK4499. According to the datasheet it has 32 bit and has: - SNR 140 dB for mono - SNR 137 dB for stereo - SNR 134 dB for quadro yes, I know that SNR=140 dB is corresponds to 23-bit ideal DAC, but we're talking about real DAC, it's a bit worse than ideal :) --- End quote --- Hell, you can call it 64bit if you want but nothing can resolve 64bits or 32bits for that matter (dunno, maybe at 0K you can, but that is damn cold). These are "marketing bits." |
| radiolistener:
--- Quote from: Wimberleytech on March 03, 2019, 12:35:46 am ---Hell, you can call it 64bit if you want but nothing can resolve 64bits or 32bits for that matter (dunno, maybe at 0K you can, but that is damn cold). These are "marketing bits." --- End quote --- no, I don't think this is marketing. Any ADC or DAC has ENOB lower than it's bit width. This is because ENOB and SNR depends on jitter, linearity, noise and many other things. Which for real ADC/DAC will never be the same as for theoretical ideal ADC/DAC. Real DAC/ADC always will be a little worse than theoretical ideal DAC/ADC. For real DAC/ADC there always will be some jitter, non-linearity, noise and other stuff, which will leads to SNR degradation. But these bits are fair. Just check, you will not be able to get SNR=140 dB on any of existing 24 bit DAC. But this 32 bit DAC provides SNR=140 dB. |
| Wimberleytech:
--- Quote from: radiolistener on March 03, 2019, 12:49:35 am --- --- Quote from: Wimberleytech on March 03, 2019, 12:35:46 am ---Hell, you can call it 64bit if you want but nothing can resolve 64bits or 32bits for that matter (dunno, maybe at 0K you can, but that is damn cold). These are "marketing bits." --- End quote --- no, I don't think this is marketing. Any ADC or DAC has ENOB lower than it's bit width. This is because ENOB and SNR depends on jitter, linearity, noise and many other things. Which for real ADC/DAC will never be the same as for theoretical ideal ADC/DAC. Real DAC/ADC always will be a little worse than theoretical ideal DAC/ADC. For real DAC/ADC there always will be some jitter, non-linearity, noise and other stuff, which will leads to SNR degradation. But these bits are fair. --- End quote --- Come on...32bits is meaningless. At least 8 bits are noise (and that is being kind to the DAC). Lets say you have a 5V reference. The LSB is about a nanovolt. I can design you a DAC that has 32bit input but you will never get 32bit performance. The fact that there are 32 digital bit at the input means nothing. |
| radiolistener:
--- Quote from: Wimberleytech on March 03, 2019, 01:01:05 am ---Lets say you have a 5V reference. The LSB is about a nanovolt. --- End quote --- AK4499 uses 9V reference and it's noise is about 1 uV. It means SNR = 1/9000000 = 139 dB. Which is close to the datasheet value SNR=140 dB. So, it looks fair. What is wrong here? --- Quote from: Wimberleytech on March 03, 2019, 01:01:05 am ---I can design you a DAC that has 32bit input but you will never get 32bit performance. --- End quote --- but this DAC is designed as 32 bit and has 32 bit performance. Yes, it is not good enough like theoretical model of ideal 32-bit DAC (which doesn't exists in reality), but it works good and gives performance much better than 24 bit DAC. |
| SiliconWizard:
Note that a sigma-delta DAC can be implemented with pretty much any bit width - it's a very scalable and 100% digital (except the analog front-end obviously) topology. You can make a 64-bit DAC for that matter, it's just pretty much a matter of silicon area and max clock frequency. You will always get a lower ENOB than the DAC bit width, but increasing the bit width is a pretty simple way of increasing the ENOB without having to resort to overly complex digital filtering, which is not only tricky but can also take a lot more area (more transistors) than barely increasing the bit width and using much simpler digital filtering. Thus it may be a way of actually getting a given performance level with reduced power consumption for instance. Marketing a 32-bit sigma-delta DAC as 32-bit is indeed kind of misleading as it's mainly a technical choice and certainly doesn't have 8 more ENOBs than good 24-bit sigma-delta DACs, but it's not a lie per se. |
| Navigation |
| Message Index |
| Next page |
| Previous page |